Découvrez Les IA Experts
 |
Nando de Freitas | Chercheur chez Deepind |
 |
Nige Willson | Conférencier |
 |
Ria Pratyusha Kalluri | Chercheur, MIT |
 |
Ifeoma Ozoma | Directrice, Earthseed |
 |
Will Knight | Journaliste, Wired |
Profil AI Expert
Non Disponible
Les derniers messages de l'Expert:
2024-03-01 00:00:00 CAFIAC FIX
2024-03-11 00:00:00 CAFIAC FIX
2023-05-22 18:50:25 @DanielHadas2 @AltMiddlePeds @LegoLas44941816 @Medical_Nemesis Oh yes. It was certainly incongruous to watch the people who denounced Theranos come out in favor of asymptotic testing for the said virus...
2023-05-22 18:38:27 @DanielHadas2 @LegoLas44941816 @Medical_Nemesis The downward trend in screening is why the USPSTF recommendation was so jarring, even to medical professionals.
2023-05-22 18:37:55 @DanielHadas2 @LegoLas44941816 @Medical_Nemesis We reached a level of peak screening in the early aughts, and it's progressively been less since. Still more than it should be, IMO, but definitely less.
2023-05-22 18:25:35 @DanielHadas2 @LegoLas44941816 @Medical_Nemesis I'm not sure this is true. Prostate screening is a counterexample. Widespread prostate screening has stopped. Mammography has been more stubborn, but outside the US it is certainly trending in the "less" direction.
2023-05-22 15:46:28 @Medical_Nemesis @DanielHadas2 @LegoLas44941816 The science cannot fail, it can only be failed.
2023-05-22 15:44:20 @RexDouglass I agree with you here. As in so many other cases, "the only winning move is not to play."
2023-05-22 15:43:23 @DanielHadas2 @LegoLas44941816 @Medical_Nemesis Also an odd tweet because it is not even clear if cancer screening saves statistical lives.
2023-05-22 15:40:17 @RexDouglass If you want to really be a hard ass, reject the paper with that reason rather than declining the review.
2023-05-19 19:00:00 CAFIAC FIX
2023-05-21 19:00:00 CAFIAC FIX
2023-04-24 13:28:52 @drjohnm @Sensible__Med @f2harrell @kaulcsmc @djc795 @brophyj If people are arguing about the p-values, the intervention isn't effective.
2023-04-23 23:51:55 @mraginsky I mostly followed this: https://t.co/k3FTxTiJ5x 1.5oz Elijah Craig 1/4 tsp matcha powder 4g demerara sugar 10 drops Regan's bitters Whisk the bourbon, matcha, and sugar until combined. Then add to old-fashioned glass with an ice cube and bitters.
2023-04-23 23:42:52 @mraginsky Tonight I went with matcha old fashioned. https://t.co/8bMYCaTzOA
2023-04-23 23:27:55 @mraginsky A+
2023-04-23 00:01:05 @mraginsky Manifest it, Max.
2023-04-22 23:52:45 @mraginsky Saving this for later as I am lacking the right citrus. Made a Monte Cassino instead. https://t.co/qSiG6VXQCo
2023-04-22 23:33:28 @mraginsky The epicurious recipe is paywalled. Can you share your proportions here?
2023-04-21 00:00:01 CAFIAC FIX
2023-04-19 17:26:01 @RexDouglass Robustness checks are the absolute worst and decidedly detract from validity. All these do is "flood the zone," giving authors far too much rhetorical slack to argue away valid critique.
2023-04-19 16:09:09 @therealrthorat https://t.co/DSmOfWfeML
2023-04-19 15:55:28 @therealrthorat Agree the exact number isn't important. But since he has such a precise estimate, there must have been some systematic method behind it. I just want to know what that was.
2023-04-19 13:59:49 In this 2010 commentary, David Sackett declares “there were only 347 reports of randomized trials” in medicine through 1955. 347 is so oddly specific! How did he come up with that number? https://t.co/1euS0HzItv
2023-04-18 13:22:53 @DanielHadas2 EBM is not the only knowledge system that has been hijacked in this manner (3/3).
2023-04-18 13:22:45 @DanielHadas2 But when assembled into a bureaucratic system, we end with a framework that is both dehumanizing and gameable by corporate and institutional interests. (2/x)
2023-04-18 13:21:28 @DanielHadas2 I think what we see is what always happens with "frameworks for rigor." The initial idea isn't bad (we should be careful and conservative with medical treatments and do experiments to ensure they work). (1/x)
2023-04-18 13:20:30 @DanielHadas2 I enjoyed that podcast when I listened to it, but it describes a caricature seen on Twitter. The Canadian School they describe always said that personal clinical experience was an essential part of EBM.
2023-04-13 13:37:16 @phl43 Oh, for sure. But that's just more fuel for the Science Isn't Real fire.
2023-04-13 13:26:48 @phl43 TBF, did you read that "narcan as moral hazard" paper? They ran a regression discontinuity on an outcome with exponential growth. It's a great Science Isn't Real (TM) case study.
2023-04-12 16:09:04 @RexDouglass Do you have a favorite "MC is incoherent" paper? And you are right, that once you say it, it's sort of obviously incoherent. But so many still people believe in FDR control!
2023-04-12 16:02:52 It’s a fixable flaw, but with social media platforms, the initial design decisions matter a lot!
2023-04-12 16:02:32 Substack Notes is weird: to follow someone, you either have to subscribe to their newsletter or rely on the algorithm. That’s a fatal flaw.
2023-04-12 13:58:53 @RexDouglass Is there a standard paper on the adversarial setting? I would cite your tweet moving forward, but that also goes against cultural conventions...
2023-04-11 15:22:43 @joftius I swear, I had this record on when I read your tweet between sets... https://t.co/76Z029ztKD
2023-04-11 15:18:54 @joftius Baleen diet, Josh. Get the TikTok account. It will be legendary.
2023-04-09 21:03:46 @drjohnm He's much scarier online than off. (Most of us are). Would you mind if I forwarded it to him?
2023-04-09 20:23:41 @drjohnm Great post! Did you send it to Andrew?
2023-04-09 14:03:15 @RexDouglass Totally. And think of how this must look with code on github.
2023-04-09 13:55:51 @RexDouglass If anything, this LLM craze has made me appreciate the internet. We take for granted the scale and spread of what's a web search away.
2023-04-09 13:47:22 @KordingLab Open letters are the worst form of science.
2023-04-09 02:20:16 @SebastienBubeck @DimitrisPapail You have a funny response to being caught flat-out lying.
2023-04-09 02:10:00 @SebastienBubeck @DimitrisPapail I'm not trying to be helpful. I'm asking you to stop being a charlatan.
2023-04-09 01:51:34 @hyonschu Inserting "assisted suicide" for "treatment" in that ramble...
2023-04-09 01:50:00 @hyonschu You have the best prompts. https://t.co/I3tuq56O74
2023-04-09 01:45:45 @SebastienBubeck @DimitrisPapail Just FYI, you've been going around telling people that it's impossible that there are tikz unicorns on the internet. You may want to pause to think about how you give your talk. https://t.co/7buuDxedY8
2023-04-08 21:40:45 https://t.co/xYA2kQWvcm
2023-04-08 21:40:02 For the jaded internet hipsters, here's a long thread of programmatically animated ducks. https://t.co/4anGku28OA
2023-04-08 02:26:49 @DimitrisPapail @begusgasper @mraginsky @damekdavis https://t.co/dyPvA5Q5Eh
2023-04-08 02:13:20 @DimitrisPapail @damekdavis @begusgasper No, Beamer has been shite since people were using overhead projectors.
2023-04-07 20:11:33 @yoavgo Because some people can't turn down the limelight.
2023-04-07 19:13:00 @ml_barnett @citre_piotto Moreover, you didn't address Colkit's argument about colinear confounders.
2023-04-07 19:05:53 @ml_barnett @citre_piotto This is an editorial statement, not a scientific result. Doing a fatally flawed and hacky meta-analysis does not prove that individual studies are flawed.
2023-04-07 18:31:46 @ml_barnett I read the paper. What do you find wrong with Colkitt's argument?
2023-04-07 17:37:45 @Apoorva__Lal We have great causal inf. folks here! In addition to Peng, Avi Feller, Erin Hartman, Sam Pimentel. They humor my heterodoxy...
2023-04-07 17:12:26 @Apoorva__Lal My take is the enterprise is doomed and can't be fixed by methodology. It is rooted in a statist bureaucratic ideology that arose much more recently than we acknowledge. But hey, I live in a CS department, so I can throw stones at stats and econ and PS from the engineering quad.
2023-04-07 17:09:02 @Apoorva__Lal superpopulation is always implausible and is a story we tell ourselves for sensemaking. but there are other stories that are equally valid and potentially more useful.
2023-04-07 17:04:44 @Apoorva__Lal Even though you have attributed my nihilism to doing too much machine learning, it actually comes from reading too much David Freedman and too many bad observational studies.
2023-04-07 17:03:42 @Apoorva__Lal Even in design-based land, people are often too flippant about their models and adjust for some preposterous quantities.
2023-04-07 17:02:32 @Apoorva__Lal Yeah, I'm fine with the design-based view, though computing things correctly is often a pain. The superpopulation view is funny because we know it's always insanely wrong (nothing is a fucking GLM), but we pretend like it's true and declare significance for publication glory.
2023-04-07 16:58:52 @Apoorva__Lal @cameron_pfiffer https://t.co/r7km9EKaBt
2023-04-07 16:58:33 @Apoorva__Lal @cameron_pfiffer Even in ML, there's a split between those who believe in data-generating distributions and those who are nihilists and believe in nothing.
2023-04-07 16:57:30 @Apoorva__Lal "Formally, clustered standard errors adjust for the correlations induced by sampling the outcome variable from a data-generating process with unobserved cluster-level components." Who in the social sciences is actually sampling from a well-prescribed data-generating process?
2023-04-07 16:56:13 @cameron_pfiffer @Apoorva__Lal Disclaimer: I don't believe any standard errors.
2023-04-07 16:52:22 @Apoorva__Lal Why do social scientists believe in robust/clustered SEs? I've never understood the argument in favor of them.
2023-04-07 03:16:16 @Apoorva__Lal The dirty secret is it sucks at Nintendo.
2023-04-07 01:44:13 Alright... time to get back to posting. https://t.co/4eKtAAQmDY
2023-04-06 16:45:34 @jessicadai_ I heard @jessicadai_ was working on it.
2023-04-06 00:46:16 @DanielHadas2 @jelistrop If I tell everyone in treatment to wear masks, and don’t say anything to control. And then I conduct phone surveys asking people if they felt sick six weeks later, what do we conclude from the results?
2023-04-06 00:44:59 @DanielHadas2 @jelistrop Yes. You’re examples are good here. If the treatment isn’t well defined and the outcome is nebulously measured, it doesn’t matter if you randomized.
2023-04-03 20:31:06 @halhod those are retweets. the "retweeted by" label apparently got fucked when he pushed the doge logo.
2023-04-03 16:27:15 @Apoorva__Lal Eh, Misha is a wonderful and smart person but has taste that I never quite get. It's all good. This paper by Sasha, Sasha, and Misha is a contemporary Russian tour-de-force. https://t.co/rnwlQT5PYT
2023-04-03 16:10:24 @Apoorva__Lal I’ll have to read this, but why is it in PNAS? And also, I’m pretty sure I nearest neighbors is also consistent.
2023-04-03 12:58:29 @apsmunro @Ikwane @zeynep @WHO Wait, I thought the incidence was 1 in 5. https://t.co/9Tf1hbliTY
2023-04-03 12:57:27 @walidgellad @apsmunro @zeynep @WHO Zeynep's favorite argument is the fallacious appeal to statistical power. https://t.co/R127pSssPo
2023-04-02 20:57:52 @KordingLab This may be one of your best tweets.
2023-04-02 20:52:37 @LeoAparisidL @RexDouglass @phl43 Yes, good one. I think that counts.
2023-04-02 19:38:48 @RexDouglass @phl43 Are there *empirical* studies in econ. or sociology that were "surprising" and have been rigorously replicated? (3/3)
2023-04-02 19:37:51 @RexDouglass @phl43 Similarly, I was under the impression that the evidence for lemon markets was pretty thin. (2/3)
2023-04-02 19:37:40 @RexDouglass @phl43 There are great. Can we dig in more? Arrow's theorem is surprising, but it's a math statement about systems, not an empirical finding. (1/3)
2023-04-02 15:00:56 @phl43 Can you think of *any* surprising results in the human-facing sciences that turned out to be replicable?
2023-04-02 14:14:56 This 40-year-old Harper’s article by Steven Levy on the disruption caused by… spreadsheet programs is enthralling, especially as a metaphor for our new AGI office assistants. (H/t @kevinbaker) https://t.co/XQwWrrnLkV
2023-04-02 13:09:26 @npparikh Half seems generous.
2023-04-02 01:41:31 @LuisOala @npparikh Hah, an enormous collaboration of people who hate each other but begrudgingly work together because it's too expensive to build their own equipment for the experiments they desperately want to run.
2023-04-02 00:54:10 @npparikh The Manhattan Project is also a horrible metaphor. FFS, they built a fucking weapon of mass destruction.
2023-04-01 18:06:24 @kevinbaker As you said, ML people like myself needed to talk to others about the human impact of our tech. 2/2
2023-04-01 18:05:49 @kevinbaker I remember this slightly differently. I think in 2016 myself and my colleagues thought AI worked. But given the accumulated evidence today, this conclusion was suspect. 1/2
2023-04-01 15:50:25 @benedictevans @kevinbaker Possibly, but I worry tech criticism gets stuck assuming everyone who uses the internet is a gullible mark.
2023-04-01 14:07:22 RT @RogueRad: The chap who prophesised the extinction of radiologists can’t rule out that AI will wipe out humanity
2023-04-01 14:07:19 @RogueRad EXACTLY.
2023-04-01 13:56:43 @yoavgo I also like how they decided to use the ranker published by Weibo. https://t.co/owOv3Q4YdW
2023-04-01 13:55:24 @jvmncs @yoavgo Hah, no I get that. And I bet it was fun to write too. But why did Twitter need to pay someone to write that?
2023-04-01 13:53:15 @yoavgo I like the hyperparameters in the ranking algorithm. https://t.co/3cfF0tpGnA
2023-04-01 13:50:33 @yoavgo I agree that people will find a bunch of absurd hard-coded rules that likely push the bounds of legality. But did you look at the ML architecture? Or the custom ML serving engine built in Rust for an algorithmic feed everyone hates?
2023-04-01 02:16:30 @kevinbaker LLMs are just computer programs designed to assist with language-related tasks. They don't have tongues.
2023-04-01 02:10:58 @kevinbaker Mostly agree, though some of the takes are dumber than anticipated. https://t.co/LO5yl61GbK
2023-03-31 23:20:31 @SergeyFeldman No sir. I do not.
2023-03-31 23:16:57 ML Twitter, what do you think about THE ALGORITHM? It's basically what I expected: a haphazard pile of bullshit held together with twine.
2023-03-28 20:36:32 For folks interested in ML4Health jobs, my friend @venkmurthy is looking for a data scientist to work on exciting applications of machine learning to computational imaging in healthcare. https://t.co/O9deyYwS0B
2023-03-28 15:38:23 @hyonschu Oh goody, look at this long list of action items. https://t.co/hQIPdAPMh8
2023-03-28 15:29:24 @hyonschu Not bad, but it's not particularly self aware, is it?
2023-03-28 15:24:12 @hyonschu https://t.co/jmQ9qtWcDX
2023-03-28 13:59:41 @DanielHadas2 Well, before this, the killer app of ML was algorithmic feed generation on social media. I'm still on the fence as to which is worse.
2023-03-28 13:48:18 @Noahpinion @JSEllenberg You need to log off sometimes dude.
2023-03-28 13:43:54 @JSEllenberg You spent your youth shit posting on economists-turned-pundits on a widely open internet forum run by a narcissistic idiot billionaire?
2023-03-28 13:41:48 @DanielHadas2 Sigh. https://t.co/CboWLjqYlu
2023-03-28 13:31:56 @RexDouglass Facts.
2023-03-28 13:14:38 @jaycaspiankang I don't expect Stanford people to know better, but I think Matt Y was one of the first to share it.
2023-03-28 13:12:03 @jaycaspiankang Also, hall of fame of extremely stupid charts! Even the data bros who are sharing that garbage should know that x-axis is criminal.
2023-03-27 21:28:37 @minilek @ruthstarkman @timnitGebru Wait, is the question why Moore's Law is different than whatever OpenAI people want you to believe about chatbots?
2023-03-27 20:28:06 @fhuszar Finally, we have solved world peace.
2023-03-27 17:20:36 @jwherrman I've been on Macs for so long I had no idea this was a thing. I wonder what the AB test looked like that greenlighted that "feature."
2023-03-27 17:16:05 @jwherrman "For now" is the sad but true part.
2023-03-27 15:41:57 @RuxandraTeslo These "experts" study how simple models work in canned settings. And then they go and write a paper by playing with a software artifact and making sweeping claims without having any idea about what's in the artifact nor having any conception of the associated cognitive science.
2023-03-27 15:41:10 @RuxandraTeslo Right, but this is the illusion of expertise. You have the same expertise as they do with regards to this particular subject of study (the potential of what chatGPT can do in applications).
2023-03-27 15:34:03 @RuxandraTeslo Because it reads like a high school science project? I do like this figure, however. https://t.co/TBZbMHyqs6
2023-03-27 15:29:28 @RuxandraTeslo This, as a recent example, is an embarrassing paper by people who used to be considered "serious" researchers. https://t.co/Qj43RJ3MwV
2023-03-27 15:29:12 @RuxandraTeslo LLM research has devolved into unsubstantiated claims with zero rigor and all evidence being cherry picked case studies.
2023-03-27 15:22:43 Funny that this is now true both inside and outside ML research. https://t.co/ni85iIcVjx
2023-03-27 15:08:50 @kevinbaker "Exponential growth is always hard to fathom." Or something...
2023-03-27 15:06:33 @kevinbaker This discourse will only get stupider.
2023-03-27 13:37:08 @DanielHadas2 Quetelet (and later Yule) believed that you could understand "social physics" and describe the cause-and-effect of nature's natural experiments by regression. This is the intellectual leap that gets us into most of the social science messes we still deal with today. 3/3
2023-03-27 13:35:44 @DanielHadas2 Using population statistics to answer *why* questions was just becoming popular in the late 1800s. Quetelet started the trend, and a famous regression by Yule in 1899 "found" that welfare caused poverty. 2/x
2023-03-27 13:33:27 @DanielHadas2 Fascinating thread. I have to read this book. I'm speculating here, but what stands out to me in your summary is that Durkheim never used statistics to explain *why* a pattern occurred, but instead to illuminate *that* a pattern occurred. 1/x
2023-03-27 13:16:22 @dnunan79 @GidMK If an RCT says exercise doesn't prevent chronic disease, does this mean we shouldn't recommend exercise?
2023-03-26 14:21:34 @RexDouglass I'll just leave this here... https://t.co/s77F5voH7u
2023-03-26 14:16:47 @RexDouglass Have you seen the "paper" from Microsoft research claiming "first contact" with real artificial intelligence?
2023-03-26 13:58:15 @RexDouglass What about the people with credentials doing exactly the same thing?
2023-03-25 17:49:00 @Apoorva__Lal https://t.co/3i5GCOhQux
2023-03-25 14:48:37 It is ironic that the company promising to automatically generate all of your back-end code can’t build a secure web service. https://t.co/UmXYeCjzbb
2023-03-25 02:16:57 @DimitrisPapail https://t.co/VOcPEeepk4
2023-03-24 13:54:05 @RuxandraTeslo Did you have to edit the code it returned or did it do what you wanted on the first try?
2023-03-24 13:24:53 @RuxandraTeslo Could you share an example? I'm in a similar boat with 90% of my stackoverflow time spent on matlplotlib.
2023-03-23 13:34:56 @mraginsky That second one is for me, right?
2023-03-23 13:28:52 @mraginsky Did you make that image with midjourney?
2023-03-23 13:24:02 @SeanTrende But how did those predictions about driverless car pan out?
2023-03-22 23:08:03 "AI doomerism is AI boosterism under a different name." https://t.co/qQtJpZUcX7
2023-03-22 16:48:16 @nanjiang_cs @jasondeanlee @fhuszar Based on the rest of the thread, I think Ferenc is just talking about the difference between in and out of distribution behavior. Easy to have two models that are the same in-distribution but different out-of-distribution.
2023-03-22 14:06:57 Super excited to dive into this brand new book on the history of the fetishization of data by @chrishwiggins and @nescioquid. https://t.co/lEVI31ukOL
2023-03-22 13:32:37 @Medical_Nemesis Clearly they didn't science hard enough, because none of the p-values are less than 0.05.
2023-03-22 00:28:43 @jaycaspiankang That sucks! Our lights have been flickering but we're still on here in the south side. For now.
2023-03-21 23:56:51 @jaycaspiankang I walked home with my wife in this nonsense, and she asked: "Is this an atmospheric river?" No "Is it a pineapple express?" No "Is it a bomb cyclone?" No "Whatever it is, it sucks."
2023-03-21 19:26:29 @DimitrisPapail Dude, come on. https://t.co/IdPxknYCJj
2023-03-21 16:41:26 @thanhnguyentang @CsabaSzepesvari Thanks! Also, I love your Twitter bio and share the sentiment. I show this Venn diagram on the first day of any ML class. https://t.co/Gz6fYr55wG
2023-03-21 16:38:56 @CsabaSzepesvari Thank you! Your feedback on early drafts was invaluable!
2023-03-20 15:48:28 @filippie509 And I know search results aren't perfect, but can't you still use Google the way you describe?
2023-03-20 15:40:22 @DanielHadas2 The argument was always more manipulative: you were saving _other_ people's lives.
2023-03-20 13:31:50 @RexDouglass I can't get inside the heads of those who think that they can statistics their way out of a bad experiment. I don't like being cynical, but I worry that most know better and do it anyway because it makes publishing easier.
2023-03-20 13:06:01 @fhuszar it's at least a year too late for that.
2023-03-19 22:37:17 @PandaAshwinee @yoavgo What's the original momentum paper?
2023-03-19 15:08:22 @yoavgo talking to people is very underrated.
2023-03-19 15:07:58 @yoavgo in no particular order: talking to people, surveys, blogs, news articles, biographies, text books, lecture notes, alumni newsletters, archives...
2023-03-19 14:41:30 @yoavgo I have been doing a lot historical research over the past 1/2 decade and have found citation networks to be only a small piece of the puzzle of intellectual history.
2023-03-19 14:38:06 @yoavgo People who want to know what happened before Adam are also cheated.
2023-03-19 14:33:35 @yoavgo People don't cite papers from 5 years ago, because they are obsessed with credit assignment of papers from arxiv last week. I always joke that you no longer have to cite papers if they are more than two years old.
2023-03-19 14:31:39 @yoavgo But that doesn't need to be a citation. It can simply point to the github. The idea that we need a "citation" indexable by google scholar is a bizarre contemporary quirk in our field.
2023-03-19 14:29:02 @yoavgo Twitter, obvs. I'm only half joking.
2023-03-19 14:28:12 @yoavgo But it's also a demonstration of how citations are cultural. No one ever cites Cauchy when they run gradient descent. WTF are they citing this Adam paper which just adds momentum to RMS Prop?
2023-03-19 14:26:56 @yoavgo I also tend to believe that the important papers are very rare, and it's ok if most papers have "middling, unmeasurable impact." 3/3
2023-03-19 14:26:33 @yoavgo For highest impact, you usually can tell without looking at a citation graph. 2/3
2023-03-19 14:26:20 @yoavgo For the purpose of intellectual history, citations are obviously great. You can trace many ideas backwards through the literature. 1/3
2023-03-19 14:20:20 @yoavgo To what end? For the sake of understanding the lineage of ideas or for deciding hiring and promotion?
2023-03-19 14:16:23 @yoavgo The most cited paper in all of optimization research. https://t.co/548lawEsGs
2023-03-19 14:15:53 @yoavgo I think citation patterns tell us something about the social networks of research, but I don't think it's worth reading much into their statistical patterns.
2023-03-19 14:09:54 @yoavgo Disagree. Ideally we'd stop using h-index all together. That we count citations as a contextless measure of "impact" is the problem.
2023-03-19 14:04:41 @yoavgo But if we only cite good papers and not flag planters, how can we hire people based on their h-indices?
2023-03-18 23:38:58 @jkuroda @benedictevans Availability, but at what cost?
2023-03-18 20:51:17 @carlosgr_nlp @yoavgo I more fear that as we make bullshit generation more efficient, we will be asked to produced more of it. No time will be saved, but bullshit will abound.
2023-03-18 20:08:57 @carlosgr_nlp @yoavgo Have you heard of Jevons Paradox?
2023-03-18 20:07:26 @jrk So I don't suffer the same fate: which cursed building was this?
2023-03-18 17:46:58 @Vaishaal I realize that my thread and question are conflating the issues of automation and reliability. Should have phrased differently...
2023-03-18 17:46:36 @Vaishaal Airplanes are super reliable, but they also require a ton of human intervention to maintain the reliability.
2023-03-18 17:44:33 @wa3id Very under-appreciated point. It's a negative correlation. You only notice things when they break. The things that work are invisible.
2023-03-18 16:41:52 @RoxanaDaneshjou @zacharylipton I believe you, but how can you really know for sure?
2023-03-18 16:39:41 @yoavgo yeah, I can also see it. I want to do sociology research to learn about their use-cases/values.
2023-03-18 16:33:31 @zacharylipton This is how @RoxanaDaneshjou refers to her kid.
2023-03-18 16:32:26 @volokuleshov The safety of air travel is terminally under-appreciated. There have only been 2 fatal accidents on all U.S. air carriers in the past decade.
2023-03-18 16:29:48 @yoavgo Research aside, I'm curious to see who will pay 20 bucks a month to assist with their bullshit writing/reporting.
2023-03-18 16:25:13 @KordingLab @benedictevans Uh, really? I was told otherwise. https://t.co/qAqY3blHoE
2023-03-18 16:20:02 @Vaishaal This reference suggests 1 failure per 1000 years. Is that 8 9s? https://t.co/cZRTuZupAC
2023-03-18 16:14:07 @KordingLab @benedictevans ...or hurt for that matter. But I want examples of *artificial* systems. I.e., made by people.
2023-03-18 16:11:34 @vinodg Do you mean the elevator? I agree, didn't used to be autonomous, but now is the only form of Level 5 autonomy in all of mobility.
2023-03-18 16:01:52 @LuisOala Reliability of cloud storage is incredible, right?
2023-03-18 15:19:16 @seanAugenstein 4. refrigerators
2023-03-18 15:15:28 @seanAugenstein My list so far: 1. elevators 2. home heating 3. well water pumps
2023-03-18 15:00:09 @daugman Water pump is a great example.
2023-03-18 14:57:58 @daugman I'm curious about that. How often do those engines need to be inspected by a mechanic?
2023-03-18 14:53:59 @seanAugenstein When I say 5 9s of automation, I want something that can be up for 365 days without a human inspecting or fixing the system. (Slightly different than what 5 9s means in high availability computing.)
2023-03-18 14:52:50 @seanAugenstein Undeniably an amazing innovation. They still need frequent human intervention though, right? That is, people inspect these engines multiple times per year?
2023-03-17 20:30:43 @ajlamesa @alexanderrusso Just FYI from someone trapped here: it was weird before covid, and it is weird after covid. And all of this nonsense felt nuts at the time.
2023-03-16 17:48:23 @ml_angelopoulos ?
2023-03-16 16:15:18 @KordingLab Who's the bloater? https://t.co/AeWEzOCJ4f
2023-03-16 16:10:39 @danilobzdok @KordingLab Maybe the right metaphor @KordingLab is that LLMs are the cordyceps of machine learning research.
2023-03-16 16:09:59 @danilobzdok @KordingLab Or this thread? https://t.co/hEqnqCuNfq
2023-03-16 16:08:41 @danilobzdok @KordingLab I mean, wtf is happening here? https://t.co/5rqhfhn8Kf
2023-03-16 15:42:13 @KordingLab It's because brains and DL have been proven to be mutually exclusive.
2023-03-16 14:30:19 @RoxanaDaneshjou I worry some people non-ironically think this would be good...
2023-03-16 14:27:54 @DimitrisPapail ?
2023-03-16 13:56:30 @Apoorva__Lal I'd like a less whimsical button.
2023-03-16 13:47:43 Coming soon to a faculty meeting near you! Replace "email" with "recommendation letter" or "ad hoc report." https://t.co/kxabrJ5W5m
2023-03-15 19:06:49 @npparikh @geomblog https://t.co/tWrMwd07XA
2023-03-15 18:55:39 @npparikh @geomblog Suresh should also unfollow him!
2023-03-15 18:49:06 @npparikh Unfollow Nate Silver. You have nothing to lose... at all.
2023-03-15 15:14:40 @soumithchintala @OpenAI Placing the burden on everyone else to reproduce the unreproducible is the wrong model.
2023-03-15 15:13:57 @RexDouglass Putting the burden on everyone else to reproduce the unreproducible is the wrong model.
2023-03-15 15:13:49 @RexDouglass We have seen time and time again in science that major breakthroughs are just simply wrong (cold fusion? more recently: amyloid hypothesis? room temperature superconductors?).
2023-03-15 15:12:21 @RexDouglass Can I quote tweet Soumith with "the only response to snake oil is to demand transparency?"
2023-03-15 13:32:47 @yoavgo I agree except it's not the hardest problem.
2023-03-15 13:30:03 @Bernstein Why do you think Time allowed all of the sources for this article to remain anonymous? It reads like standard Twitter gossip.
2023-03-15 13:10:04 @lrntzrsc @gabrielpeyre
2023-03-14 18:59:23 @DimitrisPapail Are the Star Wars movies research?
2023-03-14 18:47:37 @RexDouglass But can it play chess? And does that prove it has an internal representation?
2023-03-14 18:34:43 @benmschmidt @andriy_mulyar It probably signals they are going to submit it to the NeurIPS conference. Not clear *why* they want a conference pub, but the whole AGI movement confuses me.
2023-03-14 18:31:42 @benmschmidt @andriy_mulyar Totally wild! And only partially because the NeurIPS template is the worst.
2023-03-14 16:19:43 @james_e_b_ It doesn't make sense in academia either, but here we are!
2023-03-14 14:37:27 @mraginsky What about the people who use threads AND show more? They must have something good to say, right? https://t.co/mwTLMKDt0t
2023-03-14 13:39:40 @TAH_Sci @professor_dave @apsmunro Thanks. But it's a stretch to say this is a study in favor of masking considering that the recommendation is for powered respirators. (Also, I think your tagging of Alasdair didn't work)
2023-03-14 13:20:36 @TAH_Sci @professor_dave @apsmunro Can you point to studies? I have really seen nothing here.
2023-03-14 13:03:04 @JSEllenberg Do you count characters, or do you just watch the wheel next to the reply button and make sure it's never more than half full?
2023-03-14 13:02:09 @JSEllenberg Or, you know, people could write a blog when they needed 281 characters. Remember blogs? We could bring those back.
2023-03-14 12:59:41 @JSEllenberg Or people could post a teaser and then link to their extended Facebook post.
2023-03-14 12:56:01 It wouldn't be Twitter if half of the people on this website weren't very wrong. https://t.co/dS5zGHBkF9
2023-03-13 18:43:02 @Raamana_ 99% of the time you get one extra word that was completely unimportant. The remaining 1% of the time you get endless word salad that would be embarrassing on Facebook.
2023-03-13 18:18:56 If you see "show more" at the end of a tweet, what do you do?
2023-03-13 15:50:38 @adamcifu What we really need now is #medtwitter's take on bank runs.
2023-03-13 15:39:21 @adamcifu Though perhaps this should make you consider why you are following Topol in the first place...
2023-03-12 22:48:59 @MonicaGandhi9 @zeynep @jfcsoup Monica, there is no need to apologize to that bully. She is not worth your time.
2023-03-12 22:45:03 @jaycaspiankang Obviously true. But which group of people has grown more popular after embracing Twitter?
2023-03-12 19:59:44 @Apoorva__Lal I mostly agree that fancy bases don't buy you much with small tabular data, but the kernel trick still helps. :)
2023-03-12 19:59:30 @Apoorva__Lal Hah. Thanks! And also... https://t.co/h49mLlYLj8
2023-03-12 19:08:01 @Apoorva__Lal When I was in graduate school, people were ashamed to say they did AI. But shamelessness went out of fashion, so here we are. It's still funny that the only thing in AI that works is nonparametric estimation of E[y|x].
2023-03-12 18:57:00 @Apoorva__Lal I like this one. https://t.co/hyQBHV4c6m
2023-03-12 18:54:41 @lauretig @Apoorva__Lal @a_strezh Wait, how did I miss this? Someone must have a screenshot.
2023-03-12 16:12:26 @DanielHadas2 I agree (and lived) with everything here. But why was this particular group of people in particular liberal American cities so willing to dive in on this in the first place?
2023-03-12 14:27:56 @PaulMainwood FWIW, there was this one: https://t.co/DyrkS1isJV
2023-03-12 02:59:55 @RexDouglass And what's even more fascinating is the anti-anti-paper by David Freedman, Jamie Robins, and Diana Petitti. I'd have expected Freedman to take the other side, and I think their rebuttal doth protest too much. https://t.co/Qf5wtM8vDg
2023-03-12 02:57:54 @RexDouglass The condensed version in the Lancet is a faster read: https://t.co/kxx4TDU8BX
2023-03-12 02:51:09 @RexDouglass I love this description, by the way.
2023-03-12 02:47:02 @RexDouglass There was a moment when Cochrane was really challenging conventional wisdom. Have you read the meta analysis on mammography? It wasn't solely responsible, but the review pushed the needle against over-screening.
2023-03-12 02:35:50 @RexDouglass Sure sure, I wasn't even talking about masks anymore. Just about the bizarre rituals that the Cochrane Collab uses that pose as "rigor" even though they are clearly cultural and adapted to a variety of biases and influences.
2023-03-12 02:24:45 @RexDouglass Yeah, I agree. I also think there's something pernicious about this "absence of evidence" bullshit that Cochrane and EBM people love. It provides mealy mouth bullshit that is a bit too Pharma friendly.
2023-03-12 02:13:57 @RexDouglass We'll OpEd our way to the truth.
2023-03-12 01:59:46 @RexDouglass They go at the top! https://t.co/ABDM5l44YQ
2023-03-12 01:57:31 @RexDouglass
2023-03-11 21:30:59 @walidgellad I would love to hear your design for this. I don't see how you could do a compelling design. What effect size do you imagine? How many hospitals would be needed to account for cluster effects?
2023-03-11 21:20:01 @drvictoriafox I forgot to add school closures. And to be clear: All of these ideas, including masks, failed.
2023-03-11 21:17:20 @drvictoriafox There are also no wars about test and trace, social distancing, one way signs in the supermarket aisles, circuit breaker lockdowns... only one stupid intervention has survived.
2023-03-11 17:09:16 @sdbaral @zeynep @LizHighleyman I know you know that constructive conversations on this topic aren't possible. But I appreciate your consistent, calm outreach.
2023-03-11 15:53:57 @walidgellad @PunditPandemic Honestly, I would very much like to read any sort of study that shows this.
2023-03-11 15:41:35 @Neoavatara @walidgellad @PunditPandemic @davidzweig @cochranecollab @nytimes @NYTScience Which studies have you seen that shown benefits for N95s? This was one showed that, at least for hospital workers, there was no difference over surgical. https://t.co/DyrkS1isJV
2023-03-11 14:58:59 @jaycaspiankang He'll be the last man standing with laser eyes in his profile pic.
2023-03-11 14:52:55 @jaycaspiankang Some players are keeping it real. https://t.co/1kisWWcK5d
2023-03-11 14:50:37 @walidgellad @davidzweig @Neoavatara @cochranecollab @nytimes @NYTScience Tufekci is a sophist and a bully. B. Stephens is a knee jerk paleoconservative hack. We should all agree that the NYT Op Ed page shouldn't be leading the discourse on anything.
2023-03-11 14:49:11 @walidgellad @davidzweig @Neoavatara @cochranecollab @nytimes @NYTScience I was referring to a different review, but that's not the point. The point is this whole argument is making everyone stupid and evidence based medicine is not some panacea of epistemology. https://t.co/RfMLmCMXNL
2023-03-11 14:40:02 @walidgellad Substack exists for a reason.
2023-03-10 23:06:27 @phl43 Part of the reason it collapsed was because the startup/VC world was already in a shaky space. But hard to guess how a big collapse of tech startups would impact the broader economy.
2023-03-10 22:06:59 @apsmunro @TAH_Sci Completely agree. I’m just saying that there is a lot of probability space between those two regimes.
2023-03-10 21:10:18 https://t.co/KUod40LfCZ
2023-03-10 20:08:39 @TAH_Sci @apsmunro The stars have to align for BvM to apply though. And it's usually hard to quantify how much evidence is enough to hit that asymptotic limit.
2023-03-10 20:07:54 @TAH_Sci @apsmunro Yes. Even the most religious EBM zealot certainly doesn't apply Bayes theorem.
2023-03-10 19:59:55 @TAH_Sci @apsmunro Even in the best case scenario that we are Bayesian robots, evidence can only move priors so much. I've pushed @apsmunro a bit on this before, but the difficult part of assessing evidence is accepting people can have wildly different priors.
2023-03-10 19:52:30 @TAH_Sci @apsmunro The answer is always priors, but I wish we'd be more up front about this. Because this means it's not about "uncertainty" and "absence of evidence"
2023-03-10 19:41:11 @apsmunro I agree. But I can say the exact same thing about masks. Why are they different?
2023-03-10 19:21:40 @apsmunro What's your take on ivermectin?
2023-03-10 19:10:55 @subsix848 Oh yeah. You can already see it in the quote tweets.
2023-03-10 19:10:20 @bergerbell @subsix848
2023-03-10 17:13:30 Re: SVB and FDIC: I'm very curious about who picks up the phone... is this a therapy hotline or what? https://t.co/03N0sVdo5R https://t.co/uKRSqoKkbz
2023-03-10 16:08:29 @VPrasadMDMPH @CBoy287 All of the randomized trials were underpowered except for the ones I agree with. Also, all of the randomized trials run about masks during covid showed masks worked. https://t.co/ZwNd89YkkR
2023-03-10 15:49:22 @CBoy287 @VPrasadMDMPH Tufekci's writing on long COVID is more detached from reality than her writing on masking.
2023-03-10 15:41:18 @rak3re The Science (TM) is clear, Raffi. The Science is clear.
2023-03-10 15:12:53 @Neoavatara @DemFromCT Work is an imprecise word that lets anyone conclude they are correct. That's my TedY talk
2023-03-10 14:46:59 @DanielHadas2 @bergerbell Don't worry, Prof. Hadas, we'll test and trace our way out of this misinformation.
2023-03-10 14:41:08 @jaycaspiankang The terminally online feels beg to differ. https://t.co/dUb1LAbtMG
2023-03-10 14:30:47 @jaycaspiankang Phones clearly bad. But what about Roblox and Minecraft on a gaming PC? The data shows that's good.
2023-03-10 14:28:39 @rak3re You should reread that PNAS paper. It's incredible. I pulled some quotes here: https://t.co/V061WHBXP9
2023-03-10 14:18:16 @rak3re Her Long Covid advocacy is equally belligerent and inaccurate.
2023-03-10 14:17:32 @rak3re She's a coauthor here: https://t.co/5VMLkiPlRi
2023-03-10 14:16:44 @ajlamesa https://t.co/2bdgSINRTs
2023-03-10 14:14:07 @ajlamesa https://t.co/4sjsasiObn
2023-03-10 14:13:33 @ajlamesa Let's do some pull quotes for the advice of Tufekci endorses https://t.co/rGGY11tMxF
2023-03-10 14:12:08 @ajlamesa Remember, the only good evidence review is the one Tufekci endorses and coauthors: https://t.co/k8ULw4XDWa
2023-03-09 16:48:44 @bergerbell I agree with you, but the fake news crew has always been pro censorship. Their argument can be compelling at first glance: new technology has enabled certain harmful viral information to achieve new reach not possible before. But it's ahistorical and rests on flimsy evidence.
2023-03-08 15:17:05 @AdanZBecerra1 @ADAlthousePhD @djc795 @yudapearl @drjohnm @ChristosArgyrop @soboleffspaces @PWGTennant @MariaGlymour Sure, but the RCTs in Development Economics do not cast RCTs in a good light: they spend millions of dollars to harass people in LICs to answer obvious questions. It's pretty gross stuff.
2023-03-08 15:12:59 @ADAlthousePhD @djc795 @AdanZBecerra1 @yudapearl @drjohnm @ChristosArgyrop @soboleffspaces @PWGTennant @MariaGlymour And two years after awarding a fake nobel prize for RCTs, they awarded a fake nobel prize for shoddy observational studies!
2023-03-08 15:12:09 @ADAlthousePhD @djc795 @AdanZBecerra1 @yudapearl @drjohnm @ChristosArgyrop @soboleffspaces @PWGTennant @MariaGlymour We all know that the econ nobel prize is a fake nobel prize.
2023-03-08 00:57:05 @scritic I strongly agree with that!
2023-03-08 00:56:21 @RexDouglass Do I need a rigorous statistical formalism to set policies that govern a society? More nuanced... but again, they can only be part of a much bigger picture. 2/2
2023-03-08 00:56:12 @RexDouglass The alternatives are nuanced! At the highest level, do I need a rigorous statistical formalism to tell me how to live a good life? No. But sometimes I feel like "The Science" thinks I do. 1/2
2023-03-08 00:52:18 @RWJE_BA I don't think common sense is a bad option for many things in our every day lives. But of course, I agree that common sense and conventional wisdom are very far from infallible. We must consistently interrogate and reinterrogate our beliefs. RCTs play a key role here.
2023-03-08 00:49:28 @maartengm @DimitrisPapail I keep seeing people arguing both sides of this, and that's what I find most perplexing.
2023-03-08 00:48:15 @rdnowak 100% agree. Though I still see plenty of work on "causal identification from data alone..."
2023-03-07 19:34:48 @scritic How would such engagement with pragmatics look in STS? I ask knowing this is a conversation we will find difficult to have on Twitter. Regardless, I'd love to hear more about what you mean here.
2023-03-07 19:05:03 @phl43 Your are completely right and yet Econ gave TWO Nobel Prizes to this "revolution" in the past 4 years.
2023-03-07 18:34:28 @tadesouaiaia Hah, which one are you?
2023-03-07 17:55:44 @Apoorva__Lal Oh, and have you read this one? Might be up your alley. https://t.co/hbZCE4xMEt
2023-03-07 17:52:58 @DanielHadas2 I chose this list partially by the quality of the writing. I think you'll enjoy Leamer and Gigerenzer. And I'm personally quite fond of Freedman's hard boiled style.
2023-03-07 17:36:41 @Apoorva__Lal I highly recommend "Empire of Chance" which was written by Gigerenzer, Daston, Porter, and others. It's the Avengers of statistical history. https://t.co/t71Su254Ea
2023-03-07 17:31:08 @Apoorva__Lal That part is hilarious, especially if you read the section with a German accent in your head. Have you read much Gigerenzer?
2023-03-07 16:59:36 We have 50 years of cogent critiques of statistical regressions in causality. But instead of causing introspection and pause, these critiques merely beget more torturous techniques and entrenchment. Why? 10/10
2023-03-07 16:59:05 Certainly, all of these critiques individually have flaws. Some of their proposed fixes didn’t pan out at all. None are perfect. But taken as a whole, there’s no way around the core message that statistics assumes causation, never proves it. 9/10
2023-03-07 16:58:45 Gigerenzer, 2004. "Mindless Statistics." https://t.co/vqHHV7FHUM 8/10
2023-03-07 16:58:25 Altman, 1994. "The scandal of poor medical research." https://t.co/bgiPde86BL 7/10
2023-03-07 16:57:33 Freedman, 1991. "Statistical Models and Shoe Leather." https://t.co/QS5NX9SDaP 6/10
2023-03-07 16:56:04 Lieberson, 1985. Making It Count The Improvement of Social Research and Theory. https://t.co/sNv9Zy68mO 5/10
2023-03-07 16:55:08 Leamer, 1983. "Let’s take the con out of econometrics." https://t.co/sY2dr9XJVT 4/10
2023-03-07 16:54:47 Meehl, 1978. "Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology." https://t.co/VDmjpzKFrJ 3/10
2023-03-07 16:54:19 We have known for at least 50 years that this is wrong, and yet academics continue to push this illusion. Here are some of my favorite critiques… I’ll give 6 because I don’t believe in 93 part threads. 2/10
2023-03-07 16:53:06 There was a thread on here last week where two Turing award winners are jovially promoting the idea that mathematical statistics is better at determining causation than common sense. 1/10
2023-03-07 16:28:29 @joshua_saxe @JSEllenberg The phrase "emergent nonlinear internal representation" is exactly what I'm talking about.
2023-03-07 16:23:38 @zeynep @Theophilus_TP @NateSilver538 @AlecMacGillis @davidzweig He's a professor of machine learning at Oxford. Is it your sinecure at the New York Times that makes you important and him a rando?
2023-03-07 16:04:48 @joshua_saxe @JSEllenberg No, I don't think all of psychology and sociology necessarily are hand waving. In fact, I've been pretty deep in those fields lately, and I've learned a ton! But it is certainly true that there is much less extremely good work than extremely mediocre work.
2023-03-07 16:01:24 I shelved this thread yesterday when this site was broken, but now @phl43 and @RexDouglass are nudging me into posting it today... https://t.co/Kj6DCgYQ8p
2023-03-07 15:55:31 @phl43 I am a broken record on this, but the conclusion then has to be don't mandate. https://t.co/K9ojg0lEAR
2023-03-07 15:53:57 @phl43 I agree with almost all of this, but the more sophisticated a method, the more assumptions the method necessarily makes.
2023-03-07 15:46:50 @phl43 Worth a read: https://t.co/vqHHV7FHUM
2023-03-07 15:44:09 @phl43 Absolutely. And this is especially when it comes to quantitative methods. Gigerenzer is spot on in calling quant methods "statistical rituals." Go through the motions and get your Nature paper.
2023-03-07 15:34:15 @Theophilus_TP @zeynep @NateSilver538 @AlecMacGillis @davidzweig You don't even need a variant. You just need some hospital strain. The new memo is that anything can be cured by lockdown. https://t.co/vZsTfksvCJ
2023-03-07 15:23:12 @zeynep @Theophilus_TP @NateSilver538 @AlecMacGillis @davidzweig By your logic, we should lock down the UK now, then. Have to lock down to save the hospitals, right?
2023-03-05 16:39:03 How many times can this same article be written? It's getting silly at this point. https://t.co/qu1EmKGITI
2023-03-05 15:33:27 @DanielHadas2
2023-03-05 15:24:02 @DanielHadas2 I'd like to believe that isn't true. But I definitely think that this is a common dismissal of the critiques.
2023-03-05 15:04:58 @DanielHadas2 The thing I can't wrap my head around: we have had incredibly cogent critiques of mathematical regressions for over 50 years, but instead of causing introspection and pause, these critiques merely beget more torturous techniques and entrenchment.
2023-03-05 14:36:52 @DanielHadas2 "We preregistered our analysis plan." The idea that math is better at determining cause than common sense has ruined the human facing sciences.
2023-03-05 10:00:00 CAFIAC FIX
2023-03-02 22:00:00 CAFIAC FIX
2023-02-28 03:02:33 @ben_golub @DKThomp Course Hero vs. Cochrane. Who do you have?
2023-02-28 03:02:12 @rak3re @BallouxFrancois @DKThomp https://t.co/ksJmt1fdR8
2023-02-28 03:00:05 @DKThomp That you trust the dude who works for a cheating-as-a-service company over the Cochrane Collaboration says a lot.
2023-02-27 23:02:20 @jkuroda Don't worry, Jon. I'll put yours on my webpage.
2023-02-27 22:04:19 @AnilMakam @dr_dmorgan Depends on what you mean by most. We still have plenty of masks at the Berkeley Bowl and Whole Foods in Berkeley. I am sure it will go away eventually, but this intervention has a long tail. And agreed about academics: I have a weekly meeting at UCB that still mandates masks.
2023-02-27 20:56:46 @deepcohen I approve. I read all emails as if they are ransom notes anyway.
2023-02-27 20:37:29 @DimitrisPapail exactly.
2023-02-27 20:35:53 @DimitrisPapail @PandaAshwinee I just had someone claiming the opposite! We now have equipoise and must run an RCT.
2023-02-27 20:35:03 @GidMK @KelseyTuoc @CYirush @Jabaluck @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab This is intervention bias. If the most methods are not effective, then you may have to accept that doing nothing may be better than deploying said methods.
2023-02-27 20:09:04 @JSEllenberg @dr_kkjetelina @QuaveEthnobot The fungus difficulty on plague inc. is so much harder than virus or bacteria!
2023-02-27 19:51:40 @KelseyTuoc @CYirush @Jabaluck @GidMK @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab I mean, if you just want to reply with a bunch of ad hominem, that's fine. But it suggests to me that you are not approaching this question "free from priors."
2023-02-27 19:50:57 @GidMK @KelseyTuoc @CYirush @Jabaluck @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab There is also the third alternative: don't close schools, don't mandate masks. And it doesn't matter, but that 10% reduction was also never shown for school closures.
2023-02-27 19:45:51 @dcmayo Hah, I agree that this is strong evidence. Yikes.
2023-02-27 19:25:56 @GidMK @KelseyTuoc @CYirush @Jabaluck @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab If the study is hard to run, and the effect size is small, and the power calculations tell you that you need 1M people, why is the conclusion not obviously: "This effect of this intervention is marginal at best, and we should focus our limited efforts elsewhere?"
2023-02-27 19:23:58 @PandaAshwinee I just had to send an email to 4 younger professors as they had all recently moved institutions. All 4 listed their emails this way!
2023-02-27 19:19:26 Has anyone ever done a study to see whether listing your email handle (AT) school (DOT) edu actually reduces spam?
2023-02-27 19:11:47 @KelseyTuoc @CYirush @Jabaluck @GidMK @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab You could also read the associated blog post: https://t.co/MA0G2wwUMC
2023-02-27 19:11:20 @KelseyTuoc @CYirush @Jabaluck @GidMK @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab What part of my question is confusing? If you would like more context, I'd encourage you to read this article: https://t.co/CVVvkYnIqD
2023-02-27 19:04:49 @KelseyTuoc @CYirush @Jabaluck @GidMK @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab But if the most significant difference between the treatment and control arms is the size of the arms, don't you think drawing any causal conclusion is going to be suspect? https://t.co/idZchSiTpf
2023-02-27 19:00:21 @KelseyTuoc @CYirush @Jabaluck @GidMK @PaulGlasziou @johnkriby1 @CaulfieldTim @voxdotcom @DFisman @cochranecollab But why were the groups of different size when the experiment design set out to have them be equal size?
2023-02-27 18:30:37 @apsmunro @casertron3000 If you think the pre test probability of any easy pandemic intervention is 0, then you are in the same place as before the pandemic. If you think the swiss cheese model is accurate, then you claim we need better evidence. Either way, we get nowhere.
2023-02-27 18:29:34 @apsmunro @casertron3000 The problem with the Bayesian approach is that people with different priors come to completely different conclusions. Which is where we are...
2023-02-27 15:18:21 @apsmunro I was with you up to that last point. I haven't seen high quality evidence that clinical evidence ends many debates.
2023-02-27 14:56:39 @whippletom @Darren__Beattie @snj_1970 The other issue with big RCTs: they become too big to fail. If you spend millions of effective altruism dollars on a mask RCT in Bangladesh, you must return with something to show it was worth it.
2023-02-27 14:56:06 @whippletom @Darren__Beattie @snj_1970 Economists love big RCTs because they love to think about marginal effects. But most marginal effects aren't real or come with marginal harms.
2023-02-27 14:54:52 @whippletom @Darren__Beattie @snj_1970 If one does a power calculation and it says the RCT will need 1M people, actually doing the 1M person RCT is the wrong conclusion. Instead, maybe we should step back and think about whether this intervention is going to be worth the investment.
2023-02-27 14:51:47 @CYirush @GidMK @KelseyTuoc @PaulGlasziou @johnkriby1 @Jabaluck @CaulfieldTim @voxdotcom @DFisman @cochranecollab Yes. But when you ask these questions, the reply back is "No no no, let's fight about stats in a 76 page appendix and never tell you raw numbers. Because that's how good experimental science works."
2023-02-27 14:50:25 @whippletom Me too! Even as a scientist, I did not appreciate how science was an irrational mess until engaging with inter-pandemic scholarship. This led to a lot of reflection.
2023-02-27 14:42:32 @R_Hughes1 @whippletom @snj_1970 @WesPegden @Jabaluck Don't worry, he'll be here in a second to write a ten tweet thread that doesn't answer your question.
2023-02-27 14:41:14 @whippletom Completely agree with you about meta-analyses more generally. They can be useful! But no matter what sort of "rigor" they claim, meta-analyses are biased, editorial methods applied to science. They deserve much more skepticism than the RCTs they are built upon.
2023-02-27 14:31:13 @whippletom @R_Hughes1 @snj_1970 I find this flow chart from our paper to be illustrative. It shows that the biggest differences occur at enrollment and symptom reporting. Both of which are highly susceptible to bias. https://t.co/8FamARJSca
2023-02-27 14:29:32 @whippletom @R_Hughes1 @snj_1970 That's 100% correct. The study staff knew the treatment assignments and may have been more zealous in enrolling the treatment arm. If interested, see this paper for more details. https://t.co/CVVvkYnIqD
2023-02-27 14:26:16 @R_Hughes1 @whippletom @snj_1970 @WesPegden It's super odd! And it cannot be attributed to being about "econ vs epi." It's a willful omission because the counts were so close. Look at this table in the appendix. There is an obvious next line that is omitted. https://t.co/NJChCNBtkE
2023-02-27 01:00:00 CAFIAC FIX
2023-02-19 21:47:50 @yoavgo Thank you for saying this out loud. Because it is true.
2023-02-19 21:47:28 @yoavgo As a PhD Student, you can write 6 papers per year!
2023-02-18 22:25:08 @GaryMarcus @kaalam_ai @ARGleave The original paper was really bad and borderline fraud. It was just attacking a really stupid misconfiguration of KataGo playing rules that no one played. No one who knows Go would have declared their claimed wins as wins.
2023-02-18 22:23:26 @kaalam_ai @GaryMarcus @ARGleave Let me add, that I thought the initial paper was borderline academic fraud, and I find this approach of making loud blatant misclaims reprehensible. Exploiting the "there's something wrong on the internet" nerd bug is not a particularly ethical approach to research.
2023-02-18 22:15:55 @kaalam_ai @GaryMarcus @ARGleave They have made a major revision, adding a skilled amateur Go player, Kellin Pelrine, to the paper. This version came out in January: https://t.co/yPJ9PaRmwT Pelrine is the player who defeated KataGo in the FT writeup. Do you have the same objections to the new version?
2023-02-18 15:37:19 @BasseyE No, that tracks. When I was in Madison, we never did our faculty dinners there, but now thinking maybe we should have...
2023-02-18 14:50:57 @calebhannan @jaycaspiankang @joshgondelman Oh, yes. The prestigious "What Car? 2023" Award. We're all very familiar.
2023-02-18 14:43:56 @RexDouglass What does chatGPT give? (I'm joking, I'm joking.)
2023-02-17 16:38:56 @damekdavis Delete the app? Or wait two weeks.
2023-02-17 16:34:48 @damekdavis oh, it's the same chamber.
2023-02-17 15:43:45 I see we've reached the "Chatbots are rebooted versions of Weizenbaum's ELIZA after all" phase of the discourse. A simple decision tree would have predicted we'd get here.
2023-02-16 23:04:21 @yoavgo There’s a kooky philosophy thing too. But mostly the business thing.
2023-02-16 23:00:43 @npparikh Been this way since 1956. And the choice to oversell and hype and obfuscate has always been deliberate and cynical.
2023-02-16 14:16:34 @RexDouglass Did you see this experiment they ran at NeurIPS? It's definitely very random! The data is consistent with "50% are clear rejects. Of the rest, accept on a random coin flip." First run in 2014: https://t.co/jp67PPeUWJ Replicated in 2021: https://t.co/WjI0pIaKJp
2023-02-14 23:39:58 @sdbaral @Jabaluck @BallouxFrancois @adsquires @akm5376 Totally. But what Francois was showing in his thread that we are able to make sense of a messy world in some cases. It's a bummer that most interventions turn out to be ineffective, but somethings do work and we can find great evidence for those!
2023-02-14 23:35:42 @Jabaluck @sdbaral @BallouxFrancois @adsquires @akm5376 Or most of the effect comes from staff bias? Hard to say!
2023-02-14 22:51:06 @MikeIsaac Even the demo was riddled with crazy errors! But for some reason no one checked. https://t.co/rwPoeGAGaB
2023-02-14 17:04:13 @zacharylipton you were looking for such a list a while back, no?
2023-02-14 17:03:43 A great thread on the evidence base of public health interventions. Salient points for those who think about evidence: (a) most examples are pooled analyses of observational studies (b) the interventions are associated with reducing risk by more than a factor of 2. https://t.co/xv5NxjGWjr
2023-02-14 13:58:17 Terrifyingly hilarious overview of an insane number of mistakes in last week’s Bing/ChatGPT demo. Why did Google lose 10% of their value for a technicality, but Microsoft threw up 50 minutes of bullshit and no one noticed? https://t.co/4J1oQ2cwc4
2023-02-13 17:14:13 @GYamey @benryanwriter @LizHighleyman This is an awful thread that essentially uses the Bangladesh RCT to conclude masks work. It is the opposite of a systematic review, but gets the clicks on the twitters.
2023-02-12 18:06:54 @mraginsky @MelMitchell1 @yoavgo yet.
2023-02-12 18:05:30 @mraginsky @MelMitchell1 @yoavgo Hah. Now waiting for you to do this to a Thomas Friedman column.
2023-02-12 18:04:01 @MelMitchell1 @mraginsky @yoavgo There are these new unbounded "show more" tweets... (may god have mercy on our souls).
2023-02-12 18:03:00 @mraginsky @MelMitchell1 @yoavgo
2023-02-12 18:01:00 @mraginsky @MelMitchell1 @yoavgo 3000 words will always miss a lot of complexity and nuance of X, no?
2023-02-12 17:59:25 @MelMitchell1 @yoavgo @mraginsky Isn't science is always more nuanced than what can be portrayed in a New Yorker article?
2023-02-12 17:56:42 @MelMitchell1 @yoavgo The other two are certainly critiques, and likely what @yoavgo was referring to.
2023-02-12 17:53:33 @joelgrus @yoavgo Eh, I find zero joy or delight interacting with it. That doesn't mean Ted and I are wrong, and those who enjoy it are right.
2023-02-12 17:48:05 @MelMitchell1 @yoavgo I saw @mraginsky's thread, and I didn't see this as a "critique" but more an addendum for those who wanted to get into technical weeds. Max, correct me if I'm wrong there. Let me read the others.
2023-02-12 17:30:41 @yoavgo I hate to ask, but where are the criticisms? I'm only seeing lots of people in my timeline critiquing the criticisms. (I'm interpreting this to mean I have a well curated Twitter timeline. Though I'd prefer if there was more talk about aliens.)
2023-02-12 17:28:37 @stanislavfort https://t.co/qLH5C9WPqh
2023-02-12 17:18:02 And that's it... I might add Chiang's chatbot essay for next year's students, but my guess is the chatbot bubble will have burst by then.
2023-02-12 17:17:03 4) David Leslie's scathing review of Stuart Russell's book. https://t.co/VfTbK9jhfx
2023-02-12 17:15:24 3) @mysdick's history of our conceptions of artificial vs natural intelligence. https://t.co/tJ0JzgLxCC
2023-02-12 17:13:50 2) Maciej @Pinboard Cegłowski's Superintelligence: The Idea That Eats Smart People. https://t.co/I532c9ms6a
2023-02-12 17:12:34 1) Ted Chiang's critique of the threat of superintelligence. https://t.co/hqgQYBpWA8
2023-02-12 17:11:22 I have a recommended reading list for Artificial Intelligence, and it hasn't changed since 2019. I give this list to my grad students, but all of the articles are broadly accessible if you're interested. Very short .
2023-02-12 16:41:50 @mraginsky @zacharylipton I didn't consider that a take, but more a teaching opportunity. It was a good thread!
2023-02-12 16:41:08 Ted Chiang's recent exploration of chatbots is excellent, but this 2017 essay on the absurdity of superintelligence had the Silicon Valley AI evangelists dead to rights. https://t.co/hqgQYBpWA8
2023-02-12 16:38:03 @Pinboard Is Jack Clark an AI? Or are his milquetoast AI fanboy takes naturally generated?
2023-02-10 18:57:03 @BenMazer I cannot click like on this tweet, Dr. Mazer.
2023-02-10 18:53:52 @BenMazer People envision two main applications for AI in medicine: 1) Improving imaging 2) "automating the processing of unstructured data into diagnostics, prognostics, clinical predictions, etc." (1) is already here, but it's not as sexy as (2) I guess? https://t.co/lXc2LKbOR1
2023-02-10 18:52:39 @BenMazer The other thing that's weird about this conversation is most imaging machines have some kind of AI in them already. But you don't bill for the AI, because it's part of the package deal for the software service.
2023-02-10 18:49:57 @BenMazer What's wild is there are ancient CPT codes for AI already. Here's an AI imaging device from the 1990s. It uses neural nets. Its CPT code is 0174T. This has existed for a while, but people were less hyped about AI until recently... https://t.co/reNgvb27sO
2023-02-10 17:22:12 @aminkarbasi @DimitrisPapail Is this Amin or a chatbot predicting single words to confound me?
2023-02-10 16:55:41 @aminkarbasi @DimitrisPapail Exactly what?
2023-02-10 15:55:24 @jaycaspiankang Every time I think I have some novel hot internet take, I realize Max or @jwherrman wrote about it 5-10 years ago. My punditry is no better than the chatbots.
2023-02-10 15:43:09 That 22B number comes from the Wikipedia itself. It looks like they are using bzip2, so maybe an intrepid hacker could squish it even smaller? https://t.co/pkTR5bRJtK
2023-02-10 15:39:26 @jaycaspiankang 5 years ago Max Read called out a similar thing happening with engagement and SEO. Now throw text into the infinite iteration of fake generation. https://t.co/EcD5HgPKnq
2023-02-10 15:36:01 @Pinboard I still see your tweets and savor each one like a fine confection.
2023-02-10 15:34:16 I'm sure you're all tired of reading about chatbots, but maybe we can consider Ted Chiang's essay on compression and blur the last word on the subject. GPT3 has over 175B parameters. The wikipedia can be losslessly compressed into 22B bytes... https://t.co/knecbagBB9
2023-02-09 18:28:28 @cameronmarlow Weren't boomers parents in the 80s? So this music was the annoying synth pop the kids were listening to? In any event Iiiiiii'lllll beeeeeee goooooooooooonnnnnee
2023-02-09 18:21:07 @cameronmarlow DX-7
2023-02-09 15:58:08 @jwherrman summarizes what baffles me about the new LLM web wars: Google and Bing could easily make their search engines suck less without any fancy new technology. https://t.co/AkFRXaNBel https://t.co/aRvyiZNOjs
2023-02-09 14:56:49 Google friends, I know you have the compute. I want to hear what happens when you upscale to 48KHz!
2023-02-09 14:56:20 The idea is so simple: generate a very low resolution (3KHz) sketch of audio, and then use a cascade of neural nets to do super resolution and upscale it to 24KHz. https://t.co/4a4wPXa2fU
2023-02-09 14:55:45 Super cool work out of Google yielding the best sounding “generative neural music” I’ve heard yet. I had convinced myself that diffusion models would not be able to generate coherent audio. I was very wrong! https://t.co/PYrZGjYXrD
2023-02-09 14:30:52 @sdbaral This website has existed since 2009 and can solve any calculus problem. And yet calculus education has not changed at all. So yes, the meaning of chatGPT remains very unclear. https://t.co/hcO0jEYM59
2023-02-09 14:22:13 @apsmunro @adamcifu The fruit is always only low hanging in hindsight.
2023-02-09 00:34:48 @phl43 You going to download the Edge Browser?
2023-02-08 21:05:43 @MedCrisis ChatGPT makes far graver errors, right? So I don’t think that’s it.
2023-02-08 20:06:40 @MikeIsaac why do you think it is that chatGPT makes infinitely many errors but doesn't catch the same heat?
2023-02-07 17:47:02 @MikeIsaac But wait until GPT4! (that's actually coming, for better or for worse)
2023-02-07 17:45:13 @MikeIsaac @jesteinf Not just complaints. Lawsuits. OpenAI does not fear lawsuits (they learned this from Elon). Or government regulators for that matter! Google, Facebook, Apple, MSFT do.
2023-02-07 17:37:34 @MikeIsaac GPT3 or chatGPT? GPT3 was released in 2020, before LAMDA and well before chatGPT. we can't play down that what gets people excited about chatGPT is the parlor trick more than the "AI technology."
2023-02-07 17:26:47 @MikeIsaac Did you already memoryhole the guy who got fired for going public claiming Google's LAMDA chatbot was sentient? LAMDA was released in 2021.
2023-02-07 16:50:31 @JacobAShell What is your evidence for this claim?
2023-02-07 16:29:26 @DanielHadas2 @JacobAShell The Knowledge Machine is great. A gen pop book, but I think the core argument is spot on. That said, my gripe is that while Strevens delights in highlighting the irrationality of scientists, he doesn't follow through in describing the negative consequences of the irrationality.
2023-02-07 16:13:50 @DanielHadas2 @JacobAShell Using this narrow view of sense making in big bureaucratic organizations is foolish and dangerous.
2023-02-07 16:13:04 @DanielHadas2 @JacobAShell I read Strevens as explaining Kuhn. If all science is drudgery, why does anyone do it? Strevens' argument explains why more monkeys are better than fewer. But, it also explains the danger you highlight: scientists think you can resolve all conflict with "more experimentation."
2023-02-07 15:35:36 @anish_koka @drjohnm I had written out a multi-tweet reply, but this nails what I wanted to say.
2023-02-07 14:54:10 @Medical_Nemesis This gives the argument too much credit. Why do we think that an N95 permanently attached to one's face prevents transmission of an infection?
2023-02-07 14:26:17 @DanielHadas2 But Stevens' argument considers only the benefits and not harms of such scaling of science. As you lay out in your thread and elsewhere, there are serious problems with creating institutions like THE SCIENCE where the monkeys all must fall in line. 6/6
2023-02-07 14:25:26 @DanielHadas2 If that’s the right model for science: stupid humans flailing about in a world of infinite complexity, with lucky hits followed by feeding frenzies, then you want more scientists, not less. 5/x
2023-02-07 14:25:03 @DanielHadas2 In the space of this random and chaotic search, sometimes a scientist monkey gets lucky and finds something that works. Then a bunch of other scientist monkeys jump on that direction, and a paradigm emerges. 4/x
2023-02-07 14:24:44 @DanielHadas2 The world is infinitely complex, and insight alone is unlikely to fuel discovery. We'd be better off with a million scientist monkeys pipetting at a million lab benches to cast a wider net for possible ways forward. 3/x
2023-02-07 14:24:14 @DanielHadas2 Why do scientists do the drudge work? Strevens pointed out that this is because scientists love conflict and engage conflict by doing more experiments. 2/x
2023-02-07 14:23:52 @DanielHadas2 Michael Strevens has a fascinating counterargument to this which I think has some credence. Kuhn highlights how, though we love to hold up the Einsteins and Newtons, most of science is sheer drudgery. 1/x
2023-02-07 14:12:05 @hsom113 @snj_1970 @R_Hughes1 We do not!
2023-02-07 14:11:47 @dnunan79 @drjohnm 4. Oh, also they assumed fixed effects between the treatment-control pairs! Nonsense. So what do their CIs mean? There were so many design decisions that you could get the p-value of the primary outcome to range from 0.03 to 0.9. Which is the right size of the CI?
2023-02-07 14:10:32 @dnunan79 @drjohnm In the Bangladesh study: 1. the CIs came from stata's "robust cluster vce" command. 2. They assumed an implausible linear model to control for covariates. 3. They conditioned on randomness that balanced covariates, but didn't correct for this.
2023-02-07 14:08:44 @dnunan79 @drjohnm Also, the confidence intervals in mask studies (and cluster RCTs more generally) are not real! In a perfect RCT with no interference and clear binary outcomes, there's a reasonable meaning to the CIs. In these mask studies, the CIs are more qualitative than quantitative.
2023-02-07 14:05:10 @snj_1970 @hsom113 @R_Hughes1 I agree with this, but I'd also add that this discussion of behavior distracts from the possibility that masks really aren't effective enough to curb transmission. Even if you are trained and fit test, you don't wear an N95 alone in a BSL3 lab.
2023-02-07 03:49:28 @KelleyKga @walidgellad @VPrasadMDMPH Do you remember how you weren't allowed to wear the N95s with the valves because the initial story was all about source control? I remember because Californians had excess valved N95s from the recent terrible wildfire seasons.
2023-02-07 03:47:29 @KelleyKga @walidgellad @VPrasadMDMPH The rise of the respirator was interesting, because for the first year it was all about cloth and surgical masks. Redfield claimed in front of congress that a surgical mask might provide more protection than a vaccine. https://t.co/tr1znRriqV
2023-02-07 03:20:29 @walidgellad @VPrasadMDMPH At this point I'm only still reading this mask stuff out of amazement for the tribal fighting. Why don't we see the same sorts of kooky studies or religious devotion to test-and-trace, social distancing, etc? Why the fetish and polarization around this particular intervention?
2023-02-07 00:32:05 @bergerbell @james_e_b_ In any event, it's real healthy here in our universities.
2023-02-07 00:31:40 @bergerbell @james_e_b_ This poor reporter over the weekend was harrassed by an academic and her followers after asking about the Cochrane Review. The academic was the one responsible for getting the beaches closed in California: https://t.co/Wg4gBHAr8f
2023-02-07 00:24:07 @VPrasadMDMPH This one is written by academics. Usual suspects, obviously. And they cite the MMWR phone interview! It's great. https://t.co/00tpqxsShp
2023-02-06 20:23:54 @MikeIsaac @GuilleCummings It would be wild if true, and great demonstration of incompetence. "Following" is the barebones functionality of this stupid website, and engineering that from scratch wouldn't be that daunting: https://t.co/aBSw1DGAQv
2023-02-06 20:14:00 @GuilleCummings @MikeIsaac Are you sure? Because that should be impossible. If you are actually missing tweets on Following, that's a major major bug.
2023-02-06 20:06:07 @MikeIsaac Maybe the lesson is algorithmic timelines are always opaque, gameable bullshit and we'd be better off without them?
2023-02-06 17:54:01 @yoavgo Stop insulting my cat.
2023-02-06 17:44:32 @KurtLass1 So many things were botched at Dulles!
2023-02-06 17:25:20 @fisherra97 A valid and well argued rebuttal.
2023-02-06 16:31:25 @DimitrisPapail TBF I'm ready to do one on Occulus.
2023-02-06 15:23:03 I also like to imagine how long this video would be in our TikTok era. With some jump cuts and faster pacing, 10 minutes would be reduced to 30s.
2023-02-06 15:22:23 Another favorite example of technology that failed to catch on, watch this video envisioning the people-movers at Dulles Airport delivering you in luxury to rockets you'd take to the moon. https://t.co/0zvr78OMUz
2023-02-06 13:58:48 @npparikh To be serious for a second, useful is one of my least favorite words. In context, "useful" is usually indistinguishable from "used." Yet "useful" is supposed to have a positive connotation while "used" is neutral.
2023-02-06 03:03:54 @rsalakhu There are 152 dog breeds in ImageNet, but my cat had decided that whichever ones it sees, he should run away. Up a tree if possible. Smarts.
2023-02-05 23:51:47 @damekdavis @ArthurB I am available for expert testimony at an eminently reasonable rate.
2023-02-05 23:36:07 @jkuroda I still hate the Apple Pencil, and some part of that dislike comes from residual bad experiences with styluses for 30 years.
2023-02-05 23:34:43 Which AIs can force you to remember that you are not the most important creature in the house?
2023-02-05 23:34:25 Let's list the ways! Can a Large Language Model walk across your laptop keyboard when you are writing an important work email? Can a chatbot knock things off high shelves? Can an AI image generator get hair all over your favorite chair?
2023-02-05 23:33:36 I almost always agree with Neal, but must dissent with this take. Cats are much more useful than any existing AI. https://t.co/mAJctU8e5p
2023-02-05 21:59:10 @ajlamesa The most condescending mask thread I've seen in years. All of the plebes out here are too unsophisticated for his no-true-Scotsman non sequitur. Mind boggling.
2023-02-05 21:56:41 @bergerbell @AstorAaron @AlausBezdalius You don't even need studies, my friend! You just have to assert that they work! All that is required is common sense of course! Or just think about completely ludicrous thought experiments! https://t.co/U7K0tZpxi5
2023-02-05 21:03:26 @npparikh Couldn't disagree more. Do you have a cat?
2023-02-05 17:06:32 @ml_barnett Yes. My concerns are that in health care 1) most data is junk 2) most outcomes are poorly defined 3) prediction is a very small part of care and treatment.
2023-02-05 16:57:54 @ml_barnett I'm bullish on image processing but bearish on the rest. That said, I'm looking forward to watching how these things develop and to continued conversations about the potential of the technology.
2023-02-05 16:14:57 @TheBarbarienne @TAH_Sci https://t.co/sCaVvFBGGV
2023-02-05 16:14:24 @AggieInCA https://t.co/GEvVBQs1DD
2023-02-05 16:07:28 @ml_barnett I have a follow-up question: in your mind, what are recent examples of fancy AI/ML methods adding any value to health care?
2023-02-05 16:05:29 There were prototypes for all of this technology in 1993. So why did these not pan out?
2023-02-05 16:04:31 My fave: *Self-driving cars.* This problem is really hard for too many reasons. Just build trains, people. https://t.co/Z7JZfa52cc
2023-02-05 16:04:05 *Virtual reality.* Like wearables, people don’t want to sit around with an annoying headset. But Meta’s stock is through the roof this week, so maybe we’re finally turning the corner. https://t.co/1OfPtmDMvS
2023-02-05 16:03:47 *Everyone will wear computers.* It turned out that people wanted tricorders, not wearables. https://t.co/XUvdR1hThU
2023-02-05 16:03:30 The AT&
2023-02-05 15:04:41 @TAH_Sci I'm obsessed with technological fortune telling. Most techno-utopian forecasts are wrong! I did my graduate work at the MIT Media Lab and saw plenty of wild predictions there that never came to pass. Why was AT&
2023-02-05 00:33:35 @NICU_doc_salone @KelleyKga @JenniferNuzzo But what if there's no bang?
2023-02-04 21:35:18 @phl43 I think the private sector gets off too easily for both inertia and trendchasing. Certainly computer science, you see the worst ideas of academia borrowed and reinforced by the industrial research labs.
2023-02-04 21:23:15 @phl43 God, so much of academia. Recommendation Letters Ad Hoc Committees for Promotion Teaching Calculus to everyone...
2023-02-04 21:21:23 @NICU_doc_salone @KelleyKga @JenniferNuzzo Nuzzo is mysterious as her opinion seemed clear pre-covid, and no good explanation has been presented for her shift. 2/2
2023-02-04 21:21:16 @NICU_doc_salone @KelleyKga @JenniferNuzzo Yes, I agree. And I can even think of several PH voices on Twitter that spoke out against masking. But in California, the majority of PH embraced widespread masking, and none of the leaders were particularly active on Twitter. 1/2
2023-02-04 18:55:25 @Apoorva__Lal @cameron_pfiffer Yeah, this makes sense. There are too many methods! And we keep coming up with more of them.
2023-02-04 17:39:20 @Apoorva__Lal @cameron_pfiffer Eigenfaces is great as (a) it's compelling and visual but (b) reveals that if you don't align everything perfectly, PCA returns garbage.
2023-02-04 17:30:53 @Apoorva__Lal What is the typical introductory example in social science?
2023-02-04 16:49:28 @moreisdifferent What was the scientific reason to think fluvoxamine prevented covid?
2023-02-04 16:20:58 @anish_koka Don't cut the economists slack with that casual "perhaps!"
2023-02-04 16:20:13 @EvanPhD @politicalmath This argument is wrong. There is plenty of evidence that Influenza is airborne. https://t.co/yJH5Jgtci5
2023-02-04 15:58:55 @F_Vaggi are there studies on acupuncture in the Lancet?
2023-02-04 15:53:33 @BenMazer The chemistry-based argument against Theranos was also immediately evident for the exact same reason. (I know you knew better at the time, so this is just for the sake of argument here.)
2023-02-04 15:51:29 @F_Vaggi I'm saying that the broader community considers cupping studies are only worthy of publication in specific edge journals. Whereas other similarly silly interventions will be published in more journals with more prestige.
2023-02-04 15:49:56 @F_Vaggi The paper ended up in "Journal of Traditional Chinese Medical Sciences." https://t.co/xDyVtmJW2w
2023-02-04 15:48:37 @F_Vaggi This is a poster, not a review.
2023-02-04 15:46:10 @BenMazer So it's about cultural conventions, not evidence.
2023-02-04 15:45:08 @F_Vaggi That's right. But there is no Cochrane review of cupping. Or at least I couldn't find one. Please correct me if I'm wrong.
2023-02-04 15:40:50 Most American medical professionals outright dismiss the value of cupping or homeopathy. But why don’t they instead say “we lack high quality evidence on the benefits?”
2023-02-04 15:40:31 When there is a lack of evidence of effectiveness, why does establishment faith in some interventions persist while others are casually dismissed?
2023-02-03 19:30:24 @KelleyKga Do you remember what the marketing pitch was here?
2023-02-03 19:00:56 @MikeCoopForster @TheLawyerCraig Oof, you couldn't pay me enough to watch That 90's Show. Not gonna do it. Not at this juncture.
2023-02-03 18:50:58 @j_foerst The medical records seems to be a regulatory issue more than a technology one, no? Legitimately curious: how does that work with the NHS in the UK? If you go into a random hospital seeking treatment, are they able to pull your medical history?
2023-02-03 18:13:40 Yes?
2023-02-03 18:13:21 "Have you ever borrowed a book, thousands of miles away?" "Have you ever paid a toll without slowing down?" "Have you ever watched the movie you wanted to, when you wanted to?" "Have you ever studied with a classmate thousands of miles away?"
2023-02-03 18:11:43 Wait, hold up, these were directed by David Fincher and narrated by Tom Selleck?
2023-02-03 18:07:26 Who among us has not sent a fax from the beach?
2023-02-03 18:04:59 These 1993 AT&
2023-02-03 17:54:36 @damekdavis You know how the saying goes, it's always advisable to throw good money after FTX money.
2023-02-03 17:51:49 @contrarian4data Yeah, I get it. It's disgusting. But tweeting at them doesn't change their behavior, it only makes them worse. And retweeting them gives them more attention.
2023-02-03 17:49:39 @contrarian4data Wouldn't you be better off if you blocked J3r3my K@mil and G@vin Y@m3y? I don't know what is the upside of fighting with those dummies at this stage.
2023-02-03 16:29:35 @excel_wang I'd also quibble with their randomization inference (the only green dot), but fair enough.
2023-02-03 16:20:47 @jaycaspiankang But Google owns YouTube. There’s very little separation between the two. (i have no desire to read farhad’s argument for why that doesn’t matter. )
2023-02-03 16:19:36 @fmanjoo @nusionconnect @nytopinion No need to read a story that starts with such a dumb premise.
2023-02-03 01:42:55 @MmeBlackBalloon @erichhartmann Thank you! Much appreciated. (Though I think many people would disagree with the hyperbole part...)
2023-02-02 21:20:51 @jaycaspiankang Yesterday Bomani Jones flagged this one of Randy Moss yelling about coin toss strategies. It is must see television. https://t.co/XaBVAXsCfb
2023-02-01 22:42:54 @DanielHadas2 @Medical_Nemesis I'm a fan of flashy socks in general, so I'd be all in favor of evidence based assessments in support of awesome anklewear.
2023-02-01 17:16:50 @Apoorva__Lal is that in a published paper or a preprint?
2023-02-01 16:45:18 @DanielHadas2 @Medical_Nemesis But let's step back from this: is it *THAT* much more absurd than any other study of masking?
2023-02-01 16:44:30 @DanielHadas2 @Medical_Nemesis And yet they spent a lot of money and took a lot of time to get us a report that the null hypothesis was not rejected.
2023-02-01 16:44:17 @DanielHadas2 @Medical_Nemesis This is insane for multiple reasons: why is this an outcome we care about? And why is this better than studying the effect of studying whether magic crystals reduce the number of people who tell us they had headaches?
2023-02-01 16:42:48 @DanielHadas2 As @Medical_Nemesis has pointed out, the hypothesis is completely insane: if we give people two cloth masks each, then the fraction of people who tell us they had covid symptoms by a phone survey will go down.
2023-02-01 16:42:06 @DanielHadas2 This paper does a great service in highlighting the fundamental limits of evidence based medicine.
2023-02-01 04:19:39 @JSEllenberg My impression is most people who have heard of Epic wish they had never heard of Epic.
2023-01-31 15:37:20 @dnunan79 @dense_evi Exactly this! Well stated.
2023-01-31 15:36:31 @dnunan79 @dense_evi What I'm saying is we have an intervention that: a) sounds implausibly weak once properly unpacked b) only admits bias susceptible designs c) has no case studies demonstrating efficacy Why do we then conclude "there remains massive uncertainty of efficacy?"
2023-01-31 15:34:35 @dnunan79 @dense_evi Right, we agree here. But there's something tricky I'm trying to unpack that is probably impossible to explain properly over twitter. So I apologize for my lack of clarity in the brevity of tweet format.
2023-01-31 15:32:35 @dnunan79 @dense_evi My concern with this framing is it focuses us too much on methodology and not on the hypotheses being absurd. The main instrument in masking studies is giving people masks. The idea that this will CAUSE a drop in respiratory infections is preposterous once you unpack it!
2023-01-31 15:15:45 @dnunan79 @dense_evi But the "=" in your statement is an assumption, not a fact. The EBM position "we can say nothing because the error bars are large" is an assumption. Why is the assumption "people have been arguing about this for decades and we see no strong evidence of efficacy" invalid?
2023-01-31 14:51:07 @dnunan79 @dense_evi Why after all of this does this mean that "we have to settle with uncertainty" rather than "This intervention, as currently conceived, does not work?" 3/3
2023-01-31 14:50:14 @dnunan79 @dense_evi Despite of the interest, no one is able to design a "high quality" RCT. Every study design conceived is susceptible to bias. Despite the bias, the best estimated odds ratios in randomized trials are less than 1.1. 2/
2023-01-31 14:49:54 @dnunan79 @dense_evi I don't think this is the correct takeaway. Here we have an intervention and a committed group aiming to show it works and a committed group aiming to show it does not. Equipoise on steroids. 1/
2023-01-31 14:45:38 @politicalmath @EdwardLHamilton hair dressers!
2023-01-31 14:14:33 @Medical_Nemesis This is the closest I've seen you come to cursing, and I'm here for it.
2023-01-30 22:31:08 @damekdavis https://t.co/sQvrbpeKuN
2023-01-30 01:00:00 CAFIAC FIX
2023-01-23 17:08:06 Having a specific aesthetic is not a flaw of electronic art, it’s a feature. But it’s only a feature when it enables artists to push against the extremes of the constraints of the tool.
2023-01-23 17:07:38 The Fairlight CMI, designed as a general purpose sampler, had an idiosyncratic eerie sound that everyone fell in love with again last summer. https://t.co/klYa180vQb
2023-01-23 17:07:00 Yamaha’s DX-7, based on a Stanford patent for broadly expressive FM synthesis, is best known for its 10 presets that defined the 1980s. https://t.co/R7ZzobYPl6
2023-01-23 17:06:21 In theory, subtractive synthesis can produce any timbre. But the best subtractive synthesizers sound like nothing else. Case in point, Moogs. https://t.co/jQP507dWrZ
2023-01-23 17:05:53 I can only view this through the lens of electronic music. There, “general purpose” synthesizers only become popular when they create a niche, narrow sound.
2023-01-23 17:04:25 I’m happy someone else noticed the blatant prog rock bias in AI art. These tools should let you lean hard into the aesthetic! https://t.co/x7nkMRrVuj
2023-01-23 02:47:42 Football is wonderful and also unbelievably stupid. https://t.co/D6XWXbIbkz
2023-01-23 02:36:14 @conorsen This chart makes no sense.
2023-01-23 00:38:36 Robbie Gould vs. The Yips feels unfair.
2023-01-22 22:49:15 @FO_ASchatz Curious: Do your metrics have them with a better shot this year than last? They seemed really dominant last year.
2023-01-22 22:23:24 @drewmagary Notebook?
2023-01-22 15:56:02 @Petrarca_Fr @walidgellad @anish_koka To be fair, the Nature Comms Article is from 2020.
2023-01-22 15:22:54 @walidgellad @anish_koka https://t.co/McYkhdowoP
2023-01-22 15:22:38 @walidgellad @anish_koka I mean, there's even a name for this, right? https://t.co/CNPUTVRIXc
2023-01-20 20:56:05 @jwherrman @npparikh @katecrawford @NYMag Of course, plenty of signs point to them having done a bunch of other things behind the scenes to actually make this work: https://t.co/ugAcOFDud7
2023-01-20 20:54:49 @jwherrman @npparikh @katecrawford @NYMag The fluency is still "surprising" to me, a jaded old man who has been in proximity of AI wankers for too long.
2023-01-20 20:54:17 @jwherrman @npparikh @katecrawford @NYMag They are shadowy, so we never know exactly what they are doing. But the claim here is they built an algorithm that takes text and simply "predicts the next word" and this suffices to generate an impressively engaging dialogue agent.
2023-01-20 20:51:46 @jwherrman @npparikh @katecrawford @NYMag In the past, the novelty of these parlor tricks wore off very quickly or were eclipsed by the work of other teams. But for the last couple years, I think they've been onto something, and there's something that feels very different about chatGPT.
2023-01-20 20:50:20 @jwherrman @npparikh @katecrawford @NYMag The only thing I'd add here is that we have to treat OpenAI differently from all other AI producers. This crew has been all about the showmanship and parlor tricks from the outset.
2023-01-20 19:29:23 @npparikh @katecrawford @jwherrman @NYMag I don't know, Neal. I've seen many people who work on AI absolutely astonished by chatGPT. I find it perplexing, but such reactions are not rare in my experience.
2023-01-16 23:28:44 @james_e_b_ https://t.co/AZC7KJvtF0
2023-01-16 20:40:09 Planning a major overhaul of my research program so I can write sentences like "Lightning has fascinated and terrified humankind since time immemorial." The control lightning with lasers! https://t.co/wP9T0lnzgO
2023-01-16 19:44:34 @def_chris_suter Do you recall what the interaction looked like? What I attribute to the new regime is the slovenly ugliness.
2023-01-16 17:36:00 @_kunal_talwar_ on mobile or in browser or both?
2023-01-16 17:01:19 @drewmckevitt Either this person said what the reporter wanted said, or (a) his first reaction to boring, boilerplate, pablum is "this is so good that there's no way my students wrote this" and (b) he respects his students so little that he tells the story to the New York Times.
2023-01-16 16:54:45 @drewmckevitt Why?
2023-01-16 16:52:36 @drewmckevitt Because if it's true, this article is a scathing indictment of higher education. The professor clearly isn't educating his students and thinks so lowly of them.
2023-01-16 16:37:51 @drewmckevitt Do you think the first story---where the best essay in the class was by the chatbot---is hyperbole?
2023-01-16 16:35:04 Adventures in app design: the option to switch between For You and Following has been available for several years but was obscurely placed in your timeline. The new, non-subtle, clunky, ugly buttons make the option unavoidable.
2023-01-15 19:54:19 @yoavgo What does that stand for? I only can come up with vulgar possibilities.
2023-01-15 16:34:03 @nitashatiku The best part is he still hasn't deleted the tweet.
2023-01-15 00:07:07 @can you don't block ads?
2023-01-14 15:38:41 @DanielHadas2 The Novavax vaccine is "not novel" technology. And yet it is as effective against infection as the mRNA vaccines and has a similar adverse effect profile. Imagine if it had been produced a bit faster and mandated with the same furor: what would the story be for why it was bad?
2023-01-14 15:16:28 @Medical_Nemesis The blanket assertion "this thing works" is an obnoxious construction too often thrown around on science and medical social media. Effectiveness can only be defined in context.
2023-01-12 15:37:14 In honor of the 60th bday of @madsjw, continuing the human pyramid legacy of Wisconsin optimization research. The Dude abides. https://t.co/SGnb6Jc8Xg
2023-01-11 16:21:22 @mraginsky Sadly true. But still!
2023-01-11 13:55:24 I'm terminally online and thought I'd seen everything, but I'm amazed by the exponential growth and frenzy of the anti-gas stove meme.
2023-01-11 13:52:23 @awgaffney In which way are induction stoves "safer?" Is it in the same way that wearing two masks is safer than wearing one to prevent an ILI?
2023-01-11 13:41:46 @ben_golub This argument is disappointing in 2023. If a human acts negligently with a car, there is a system (yes, imperfect) set up to punish the human. What is the recourse against Tesla for releasing a fraudulent product that harms people and removing this product from the road?
2023-01-10 22:18:14 @jaycaspiankang “I signed up to work at an Singularity Worshipping Death Cult, not fucking Microsoft.”
2023-01-10 18:37:16 @james_e_b_ @R_Hughes1 @BQuilty Literal poetic justice.
2023-01-10 18:18:11 @yoavgo I love these image-gen tools but don't think of them things as particularly "AI"-esque. They a super neat combo of (a) great image search (b) a "happy accident" blend tool. The art still comes from artists' interaction and iteration with the tools. What's your take?
2023-01-10 18:03:20 @adhamelarabawy It is illegal to stop suddenly on the highway in California. https://t.co/6gQaq8ek0c
2023-01-10 17:40:23 @yoavgo I agree. It's not a good example. I don't like this style of argument per se, but a more Moravec-style argument would be: written language is only 6000 years old, whereas animals have been rapidly navigating through space for millions of years.
2023-01-10 17:38:03 @AVMiceliBarone @visarASU In SF: https://t.co/PFxo2QPUNa
2023-01-10 17:36:26 @AVMiceliBarone @visarASU Unmanned in Austin: https://t.co/IhuGuubK70
2023-01-10 17:30:40 @AVMiceliBarone @visarASU No, waymo runs without safety drivers.
2023-01-10 17:25:45 @yoavgo That we need billions of dollars in compute and the entire internet to write essays like a fourth grader?
2023-01-10 17:24:49 @visarASU @AVMiceliBarone They do night drives in SF too. And apparently they are testing in Austin, but I've seen some scary videos there. Again, a lot of regulatory loopholes allow these rather unsafe vehicles on our streets.
2023-01-10 17:19:48 @AVMiceliBarone @visarASU Small area, and absurdly low volume. No one actually rides Waymo.
2023-01-10 17:17:34 @yoavgo Oh, I see what you mean. Simply put: Moravec's Paradox is tasks that are really simple for people are often really hard for computers and vice versa.
2023-01-10 17:13:02 @yoavgo Yeah, I mean, people drive with an elementary stereo camera setup, while blasting music, checking text messages, talking on the phone... I don't think it's the sensing.
2023-01-10 17:07:39 I know this is going to read like an outrage tweet, but this is actually a question about our federal regulatory system: Why is this deadly and dangerous technology allowed on the road? Why is there no regulatory recourse? https://t.co/k139PCd0rt
2023-01-10 17:02:25 @yoavgo I don't know what the problem is! I base almost all of my self-drivng skepticism on the fact that I can't concretely tell you why it's hard, and no one else can either.
2023-01-10 16:56:36 @yoavgo Yes. 100%. It's surprising how discontinuous the jump was.
2023-01-10 16:54:42 @yoavgo Full-self driving remains my favorite instantiation of Moravec's Paradox.
2023-01-10 16:54:08 @yoavgo I mostly agree. There's the clean and clear path to level 5 autonomy: elevators. But no one can tell you what is needed to get actual fully autonomous driving or even articulate clearly why the task is hard. You'll get a bunch of mealy mouthed answers about edge cases.
2023-01-10 16:48:13 @KordingLab @VenkRamaswamy Oh come now, very false. But I appreciate the congrats.
2023-01-10 16:40:58 @jaycaspiankang It's also awesome how there's no recourse against this crappy technology. If it kills people, then the blame gets passed to the driver, not to Tesla.
2023-01-09 22:07:45 I am fascinated that most replies tell me LLM tasks are easier than self-driving, not that the respective demos are more impressive. https://t.co/4tNkkg83XM
2023-01-09 16:41:07 @RidleyDM It was hundreds of research teams and decades. Weirdly, the necessary technology for Google photos existed in 2015 when he released this comic. So it wasn't a particularly timely example. Goes to show not even the most clever of us can generally know what is currently possible.
2023-01-09 16:37:31 @RidleyDM You are missing the point. Everyone is happy to accept death. The problem is that thinking "all I need to do is be safer than humans" completely misdirects the engineering effort. I'll write a thread about this because it's important.
2023-01-09 16:36:23 @RidleyDM How much research, compute, and data went into building the feature he describes in that cartoon?
2023-01-09 16:31:05 @RidleyDM Now add step 3: would be more energy efficient if you strung them together with a tether, and had the lead car power the platoon... Voila. 2/2
2023-01-09 16:30:43 @RidleyDM I'm aware of the use case. Now pitch me the design where there are never accidents. We did this at Berkeley in the 90s. the plan was: 1. remove all humans, or give the self-driving cars their own lanes 2. have the cars platoon for energy efficiency and safety. 1/2
2023-01-09 16:25:51 @RidleyDM The reason they're not is because of this mistargeting: if engineers thought about what would be required to make self-driving cars maximally accident-free, they would design trains.
2023-01-09 16:22:04 @RidleyDM Ironically, they said the same thing about self-driving cars: rather than saying "no accidents are acceptable" they approached the problem as "being better than humans." Fatal flaw in the approach.
2023-01-09 14:51:26 @visarASU I don't know, watch this video. There was a ton of Google hype in 2012 of their tech being both safe and ready to ship. https://t.co/mx5Sycy8G7
2023-01-09 14:47:07 This is a cool demo! https://t.co/Fx9EnauBoS
2023-01-09 14:42:06 In which ways are the large language model demos of 2022 more impressive than the self-driving car demos of 2012?
2023-01-09 14:23:21 @wesyang The 2013 demos of self-driving cars by google were far more impressive than the gpt demos of 2023.
2023-01-09 14:17:39 https://t.co/1ZbDInGLBF https://t.co/oIBtr3YKgK
2023-01-09 03:07:22 @JSEllenberg This game was so awesome: https://t.co/GYKoS246iS
2023-01-09 01:48:45 @Apoorva__Lal Did you watch this concert? It's completely insane. https://t.co/K6Su2ziHnJ
2023-01-09 01:10:04 @JSEllenberg I loved watching him at Wisconsin, but why is he your favorite? The 21st century has had otherworldly football players, and I'm not sure JJ is even my favorite defensive player.
2023-01-08 22:27:10 @mraginsky No way. That's hysterical. I probably should have said The Medici. Or, for the heads, Rajun Cajun.
2023-01-08 22:24:01 @mraginsky Harold's Chicken Shack.
2023-01-08 20:00:05 @viveksworld Sigh. But do the jets?
2023-01-08 19:21:54 He’s still in the league, ChatGPT! https://t.co/2D6QXykXKu
2023-01-08 17:23:34 @ben_golub @itaisher @vitaly_nwu No, I definitely don't believe this. But I think there's an authority granted to theories with numbers, even if they are less predictive than those without numbers. And if those number-based, data-based theories don't make better predictions, the authority is unearned.
2023-01-08 15:56:59 @ben_golub @itaisher @vitaly_nwu I find this a bit confusing on two fronts: (a) What is the evidence that these predictions are directionally/occasionally right? (b) What is the value of predictions are directionally/occasionally right if we can't qualify the extent to which they are wrong?
2023-01-08 15:52:42 @ben_golub @itaisher @vitaly_nwu Deciding rational choice is the best way to craft policy is a choice in of itself. And this choice must agreed to by all of those engaged in the decision making. Only if a community agrees on rational choice decision making can game theory can be applied as a solution. 3/3
2023-01-08 15:50:49 @ben_golub @itaisher @vitaly_nwu For such complicated participatory decision problems, there are myriad ways to decide an outcome. Rational choice strategy is *one possible approach*. But there are alternatives. 2/3
2023-01-08 15:49:52 @ben_golub @itaisher @vitaly_nwu Let me see if I can clarify the last point: Often times, when there are many stakeholders involved in some tense decision process, you need to (a) define boundaries of the problem to be addressed (b) justify the end policy to all stakeholders (c) resolve conflicts. 1/3
2023-01-08 15:33:47 …composing follow-up tweet on observational causal inference… https://t.co/YebsSnZOHy
2023-01-08 14:07:47 @AlexJohnLondon 100% agree. But what do you make of the use of these mathematical representations to describe humans and their societies?
2023-01-08 14:05:28 @sivavaid Yes. The military applications proceeded the application to econ by only a few years (all done in the 1940s). But by 1950 there was a sharp and fascinating split between the econ and nat sec camps, with both sides interested in very different questions.
2023-01-08 14:01:39 @yomritoyj That's bold! Why do you think theory didn't help in these cases?
2023-01-08 14:00:55 @Raamana_ Can't write follow-up thread... have to formulate my analogous tweet about observational causal inference first...
2023-01-08 13:47:45 @ben_golub @itaisher @vitaly_nwu Now we turn back to human economies: Many times when a system of rules need to be devised, stakeholders consense that a rational solution is best and then apply game theory to design the rules. But deciding that a rational solution is best *requires consensus.* 4/4
2023-01-08 13:46:31 @ben_golub @itaisher @vitaly_nwu On the other hand, for understanding how to *design* systems of multi-agent rational optimizers, game theory is very powerful. 3/4
2023-01-08 13:46:05 @ben_golub @itaisher @vitaly_nwu It’s nearly universally accepted that game theory does not describe/predict how humans make decisions. And the patches (behavioral, evolutionary, etc.) don’t fix this. 2/4
2023-01-08 13:45:22 @ben_golub @itaisher @vitaly_nwu First off, great thread. A very fair response to my shit stirring. But I want to push back a bit, and I think Itai is doing the same: I didn't reject the broad definition of game theory. I rejected the specific application of understanding humans and human economies. 1/4
2023-01-08 00:22:42 @mraginsky @DanteKalise I'm getting misinterpreted: even in the 50s, everyone except some eccentrics knew that classical game theory wasn't going to work like that. but all of the proposed "fixes," including behavioral econ, etc. haven't fixed the inherent problems.
2023-01-08 00:13:10 @mraginsky @_onionesque On point!
2023-01-08 00:12:46 @mraginsky @samhaselby Yes, Machine Dreams is great! @eigenstate also recommended "More Heat than Light." I haven't read that one. Have you?
2023-01-08 00:03:58 @_onionesque What's AGI?
2023-01-07 21:27:46 @DanteKalise Every day. But as Astrom always reminds me: PID is anything but simple!
2023-01-07 20:00:19 @ivan_spartakfan @TACJ Yes, these and the auction example are salient. Because none are "natural" human decisions. Instead, stakeholders consensed that a rational solution was best and then applied game theory to design the rules. But deciding that the rational solution was best required consensus.
2023-01-07 19:33:53 @DanteKalise This is a fascinating report for so many reasons.
2023-01-07 15:58:50 @samhaselby No, will read! I worry that some of the work highlighted here, where econ decided it was better to do crummy biostatistics, is as problematic as game theory. (Though for different reasons). But let me read this in full before I really start shit talking. Thanks!
2023-01-07 15:51:24 @MKocher8 @samhaselby What's an example of folk rationalism?
2023-01-07 15:50:51 @samhaselby It's particularly interesting to me to read von Neumann saying such things, given the hindsight of how fundamental physics stalled out post WWII.
2023-01-07 15:50:03 @samhaselby Yes, that's misconception 1. One of their arguments for *why* no one had been able to render econ into math is a weird analogy with 19th-20th century physics. https://t.co/Y1y5fVVT5T
2023-01-07 15:44:42 @TACJ That's a good, even if somewhat niche, example. I think you can make a compelling case that theory has helped auction design.
2023-01-07 15:37:40 @RidleyDM Are you aware of much legal scholarship on this? I think the language of game theory gets bandied about, but does anyone ever try to compute a Nash equilibrium when negotiating a settlement?
2023-01-07 15:18:48 @samhaselby Passages like this highlight many misconceptions. https://t.co/ueIy881ApN
2023-01-07 15:17:21 @RidleyDM I want to emphasize the "as a means to understanding humans and human economies" part. There, I'd argue total failure even though this was the initial intent of GT.
2023-01-07 15:16:26 @samhaselby I've been rereading von Neumann and Morgenstern (the book that formalized and popularized GT), and they really set out to mathematicize econ. in the same why that Newton mathematicized physics. It's a bold program! But one can already see in the intro why the program was doomed.
2023-01-07 15:07:44 @ifihadastick The problem is that almost all of social scientific study has shown that GT fails to capture how people make or frame economic decisions, even when presented with economic costs.
2023-01-07 15:05:37 @samhaselby Do you recall what his objections were? I could also reach out to him, offline. But the more I dig into it, the worse it seems.
2023-01-07 14:55:02 We don’t talk enough about how game theory, as a means to understanding humans and human economies, was a total failure.
2023-01-07 14:07:10 @davidpapineau @phl43 Did you see the paper where masks in schools cure structural racism? https://t.co/wQey0cGMry
2023-01-05 22:37:25 @KordingLab Oh, my department is horrible about this. We are probably the worst institution in the field for paper spamming. So many terrible examples you could list! But even if you ask those who shall not be named, they'll tell you it's bad.
2023-01-05 22:25:05 @KordingLab Is this a good development?
2023-01-05 17:49:14 Underrated short film about AI that no one mentioned in my replies: "Be Right Back" (S2E1 of Black Mirror). It's perhaps a bit too real given our current capabilities, but I rewatched it this week and it remains exceptional. https://t.co/8f1htLTH9u
2023-01-05 14:32:54 @jamescham @loriberenberg The Breakfast Club is not about AI, but the performances in High Art are so robotic that it is.
2023-01-04 23:15:57 @loriberenberg @jamescham Clearly it's: 3. Short Circuit 2. High Art 1. War Games
2023-01-04 23:15:39 @loriberenberg @jamescham How does short circuit make it on there but War Games not! Criminal! We need a list of best movies about AI starring Ally Sheedy.
2023-01-04 18:12:36 @AnilMakam Yup. And all the way up to 500K for "married filing jointly."
2023-01-03 20:09:23 @FarhadMohsin1 @hyperactve For the purposes of this thread, I am only considering movies about AI. Loved Arrival and Moon, though.
2023-01-03 19:34:43 @yoavgo evergreen.
2023-01-03 19:28:38 @ducha_aiki https://t.co/X1AqcUHSkc
2023-01-03 19:15:59 @ducha_aiki T2>
2023-01-03 19:14:43 @ducha_aiki This will forever be iconic, but I don't need more than the gif. https://t.co/ZkdG3oCc2I
2023-01-03 19:13:21 @ducha_aiki I've watched the three movies I mentioned several times each. I've watched The Matrix once. It was visually stunning, especially for its time. It was peak 90s pastiche. But the story about the AI was and is ridiculous, and I had no desire to ever experience it again.
2023-01-03 18:45:13 @chriswolfvision NB It's a very American 80s movie, starring Steve Guttenberg AND Ally Sheedy.
2023-01-03 18:35:06 @chriswolfvision Despite adoring District 9, I couldn't get into Chappie. I thought Short Circuit was better (also 80s, but also not as good as the three movies I listed).
2023-01-03 17:02:28 @yoavgo @DimitrisPapail There will never be a useful definition of AI-complete as intelligence is a perpetually moving target. https://t.co/tJ0JzgLxCC
2023-01-03 16:43:19 @be335651783 More generally, I personally prefer artificial life stories to superintelligent AI stories. They are always better commentary on the human condition.
2023-01-03 16:42:01 @be335651783 And all 3 of the movies I listed are as much about artificial life as they are AI: Blade Runner is clear. In Terminator, we see what AI thinks humans are like. In War Games, the AI is modeled after Dr. Falken's dead son.
2023-01-03 16:41:11 @be335651783 I know it's sacrilege, but I never could get into 2001. Probably my least favorite Kubrick movie.
2023-01-03 15:13:19 @thepacketrat Hah, that's a deep cut!
2023-01-03 14:58:17 To be clear, the best movies about AI are 1. Blade Runner 2. The Terminator 3. War Games
2023-01-03 14:56:24 Why were the best movies about artificial intelligence all made in the early 80s?
2023-01-03 14:13:10 @GaleMorrisonEd @TwitterSupport It would be nice if Twitter made this easier and more transparent, but instead they insist on adding useless features like showing the number of views a tweet has.
2023-01-03 14:08:49 @GaleMorrisonEd @TwitterSupport Go to the top of your timeline, and click on this icon. Then select Latest Tweets. https://t.co/MDvG6WlYyo
2023-01-03 14:01:00 @docmilanfar A perfect list for those who aim to remain indecisive in 2023.
2023-01-03 00:28:48 @MUBNiazi Well, I do not plan on reading any of that guy's books in 2023.
2023-01-02 23:29:10 @bergerbell . Thanks, I'm now dead.
2023-01-02 23:26:07 @bergerbell I mean, seriously, Geneve, fuck this guy: https://t.co/iJfydyv4H6
2023-01-02 23:25:16 @bergerbell What's the point of following dweebs like Walter Bragman and Phil Stencil at this stage of the game? The only people who care about what these mediocrities write are the people they insult. Classic trolling.
2023-01-02 18:31:05 @andreavhowe @mraginsky I think you have to read them sequentially. That said,The Stand is 1200 pages, and I think the Dark Tower Series is only about 4000 pages total. So it's like reading The Stand 3 times or something?
2023-01-02 17:37:49 @neuralreckoning After which book should I stop? Book 3? (I read 1-3 30 years ago...)
2023-01-02 17:33:14 @mraginsky Does The Wheel of Time also feature Randall Flagg?
2023-01-02 17:30:01 @JSEllenberg I just finished The Gunslinger for the first time in 30 years and it was fucking awesome. I see no choice but to go all in. Before that, I reread Salem's Lot, which didn't hold up as well but was still reasonably terrifying.
2023-01-02 17:23:42 Are we still sharing high brow reading lists? Because I'm going to finally finish reading The Dark Tower in 2023.
2023-01-02 00:01:47 @JSEllenberg And the 2023 Vikings are the worst 12-4 team of all time.
2023-01-01 20:34:06 @Pinboard Dude. We're all on mastodon now. The singularity happened on Twitter and all replies are now generated by AI.
2023-01-01 16:57:59 @pseudnonymus @JacobAShell Right. We already have web search that doesn't speak like a bad TV psychic.
2023-01-01 16:56:48 @JacobAShell My work is AI adjacent and I'm often lumped into the "AI researcher pile." But I find AI insufferably boring as well. Most AI research is telling bad stories, inspired by arm chair psychology and bad sci fi, about industrial automation demos.
2023-01-01 16:08:17 @KordingLab What a ridiculous set of options. HAPPY NEW YEAR!
2023-01-01 15:49:43 @KordingLab https://t.co/sK89kDajJo
2023-01-01 15:49:11 @KordingLab The human raters are the fucking authors.
2023-01-01 15:44:38 @KordingLab WTF? The abstract of the linked paper says the opposite. https://t.co/CsdDn2s7JT https://t.co/eecIw39cWL
2022-12-31 22:15:47 @DanielHadas2 @james_e_b_ The overproduction of covid science was so depressing. The sheer volume was overwhelming, and yet, to very close approximation, none of the results hold up longer than 3 months.
2022-12-31 21:55:44 @DanielHadas2 @james_e_b_ Indeed, we instead have seen studies on excess deaths, arguing strongly that the benefits of NPIs have been marginal at best. 2/2
2022-12-31 21:55:10 @DanielHadas2 @james_e_b_ Yes, this is an excellent point. There have been no new articles over the past year about the benefits of closures. You'll still see mask studies (sadly), but you are correct to be surprised by the lack of recent studies on NPI effectiveness. 1/2
2022-12-31 21:36:57 @DanielHadas2 @james_e_b_ Anyway, I'll stop there. Most of these articles are ghastly, but there is an entire industry of writing shitty papers about the joys of NPIs.
2022-12-31 21:36:29 @DanielHadas2 @james_e_b_ https://t.co/5bWLvLH3G7
2022-12-31 21:35:36 @DanielHadas2 @james_e_b_ https://t.co/XFHyRnN0PM
2022-12-31 21:31:44 @DanielHadas2 @james_e_b_ https://t.co/9HUe9PA3la
2022-12-31 21:30:36 @DanielHadas2 @james_e_b_ https://t.co/IxI9KU4xDI
2022-12-31 21:30:13 @DanielHadas2 @james_e_b_ Papers in Science/Nature alone: https://t.co/im9Z100U5o
2022-12-31 21:29:08 @DanielHadas2 @james_e_b_ I'm not sure what you have in mind, but there have been hundreds of studies on this. They are all of highly dubious quality, of course, but there is no shortage of them.
2022-12-31 21:11:27 @jaiagreen @besttrousers @GrantEHaines @LizHighleyman @senorhettler Yes, I agree, practical differences are hard to come by. Do you have the link to this study, by chance? The estimates on R are always complicated as they tend to assume simplified dynamics and immune naive populations. So I'd be curious to see how they reached 19%.
2022-12-31 17:41:47 @besttrousers @jaiagreen @GrantEHaines @LizHighleyman @senorhettler This is absolutely not true. Trivially, Vvccines are clearly effective in randomized trials. But out of dozens of trials on masking, none show effectiveness at the community level. (And don't bring up Bangaldesh: that one does not demonstrate effectiveness.)
2022-12-31 16:14:23 @Theophilus_TP @ajlamesa Right! Why is everything so much more elegant in Latin?
2022-12-31 16:01:56 @ajlamesa His assertions don't even match reality and are just based on his usual bad gut instinct: hospitals were "overwhlemed" in New York *after* lockdown. And hospitals were "overwhelmed" in Los Angeles... in January 2021, almost a year after CA's even harder and earlier lockdown.
2022-12-31 15:58:46 @Neoavatara It's a hand-wavy argument that feels right but no evidence is provided. Hospitals *were* overwhelmed in New York in March 2020. And then they *were* overwhelmed in Los Angeles... in January of 2021. So how did the uniform NPIs of March 2020 help?
2022-12-31 15:52:27 @ajlamesa This thread is an example the worst kind of causal reasoning, and a kind that has been popular since the beginning: "If our desired outcome occurred, then it must be because of our actions."
2022-12-30 04:46:17 @graduatedescent @BarneyFlames To be fair, Eric has many plausible null hypotheses. But regardless, a 50% overlap in the acceptance lists of two committees is not a favorable signal.
2022-12-30 04:25:19 @CalvinMccarter https://t.co/7YEAWlTBtB
2022-12-30 04:24:44 @graduatedescent @BarneyFlames My null (and Eric Price's proposed null) is "~50% of the papers are obviously bad, and the rest are accepted at random with probability ~50%." This is consistent with both iterations of the experiment.
2022-12-30 04:11:56 @Raamana_ I really wish I had clarified in advance: I'm against pre-publication peer review. Non-anonymous, post-publication peer review obviously cannot be removed from research.
2022-12-30 03:57:04 @Raamana_ I need to write something longer about this... I try very hard not to do Twitter threads longer than 5 tweets. But more to come in 2023. :)
2022-12-30 03:33:46 @graduatedescent @BarneyFlames How are you evaluating this based on the numbers on that blog?
2022-12-30 03:05:06 @graduatedescent @BarneyFlames What’s your null hypothesis? I’d ideally like to run a non parametric test on the real data.
2022-12-30 02:13:12 @mbeisen @mrdrozdov @robtibshirani Who picks who gets to be on your rapid response committee?
2022-12-30 02:12:34 @mrdrozdov @mbeisen @robtibshirani They propose replacing the authority of peer review with the authority of some NAS like committee. This goes in the opposite direction of what I value.
2022-12-30 02:02:27 @mrdrozdov @mbeisen @robtibshirani It proposes “a rapid-review service for preprints of broad public interest” by “a diverse contingent of scientists ready to comment on new preprints.” How is this not gatekeeping?
2022-12-30 01:56:32 @ajlamesa Sure, but attributing those things to “public health” is revisionist history.
2022-12-30 01:52:33 @DanielHadas2 For whatever it is worth, I had the same experience.
2022-12-30 01:51:54 @DanielHadas2 You will get called a conspiracy theorist for such an assertion, but it’s such a strikingly important point.
2022-12-29 23:50:01 @Apoorva__Lal @jugander We have a similar bloat problem in CS but do the opposite: a 10 page published version, and then a 100 page "non-reviewed" version. LOL.
2022-12-29 23:49:13 @Apoorva__Lal @jugander 2022: https://t.co/WjI0pIaKJp
2022-12-29 23:48:28 @Apoorva__Lal @jugander There were actually randomization experiments at Neurips that suggest reviewing for "prestige" is indistinguishable from random. 2014: https://t.co/jp67PPeUWJ 1/2
2022-12-29 23:36:33 @phl43 Read the replies to my tweet here and be depressed that nothing will ever change. https://t.co/cc5YQhRiNf
2022-12-29 23:30:27 @pfau I think we can do better! (than peer review, not than democracy, lol.)
2022-12-29 23:29:07 @jugander The thread stimulated good conversation, but it's not what I was after: I wanted refs to academic literature on PR that concluded the system was beneficial. No one sent me any links! Considering how so much research shows PR is not helpful, what is the take away?
2022-12-29 13:26:55 @pfau I 1000% agree! For all its warts, I think Twitter has actually been positive this regard. But there would be a lot less anger and misunderstanding and a lot more productive engagement if the academic convos on twitter were done IRL.
2022-12-29 00:20:44 @mraginsky @DimitrisPapail @peter_richtarik @icmlconf @lastpositivist I'm old enough to have been there, and I don't think the reviews materially changed either way. (Bad before, bad after)
2022-12-29 00:16:03 @erik_paulson All of the papers I've seen suggest that the CS conference program selection process is equivalent to a random number generator.
2022-12-29 00:15:10 @SeanTrende Unfortunately, the academic literature on peer review suggests the idealized version doesn't exist. And in math, the move to pre-publication review on arxiv has been uniformly embraced.
2022-12-29 00:10:25 @kaulcsmc @DrToddLee @EricTopol @NEJM Thanks!
2022-12-29 00:10:00 @mraginsky @DimitrisPapail @peter_richtarik @icmlconf @lastpositivist Had to bring it to the top of the timeline. I worry that our grumpy predilections may have avoided engaging with the positive case. https://t.co/cc5YQhRiNf
2022-12-29 00:07:53 I've read dozens of papers on and reviews of peer review, and not one suggests that peer review adds value. Has anyone read or written a positive case for peer review?
2022-12-29 00:06:20 @mraginsky @DimitrisPapail @peter_richtarik @icmlconf @lastpositivist I've read dozens of papers on and reviews of peer review, and not one suggests that peer review adds value.
2022-12-29 00:04:00 @kaulcsmc @DrToddLee @EricTopol @NEJM Which observational studies do you have in mind here?
2022-12-28 23:55:10 @DimitrisPapail @peter_richtarik @icmlconf None of the ones I have in mind are ML journals, if that's what you are asking. The ML journals occupy a rough space, and JMLR could really innovate and lead by adopting a radical alternative to peer review.
2022-12-28 23:53:57 @MountainOfMoon @DimitrisPapail @peter_richtarik @icmlconf Twitter is the best. https://t.co/lAf8oBDHqV
2022-12-28 23:53:10 @MountainOfMoon @DimitrisPapail @peter_richtarik @icmlconf LOL, apparently you just have to type reinforcement learning and provide a link and you get retweeted by a reinforcement learning bot. https://t.co/7LjohYW0DE
2022-12-28 23:51:50 @katrosenfield @emmaogreen And they all show up in the replies to her tweet, further validating her reporting.
2022-12-28 23:43:46 @DimitrisPapail @peter_richtarik @icmlconf There are, and many have far more reach and influence than ICML. Shocking!
2022-12-28 23:36:50 @DimitrisPapail @peter_richtarik @icmlconf To be clear: I'm also in favor of abolishing peer review as for journals as well.
2022-12-28 03:21:16 @RexDouglass @jon_mellon @zacharylipton Yikes. Away from my library proxy at the moment, but will read this one later.
2022-12-28 03:06:50 @RexDouglass @jon_mellon @zacharylipton In general, I’m a Freedman zealot and believe you will never establish causation with regression. But I want to at least find a few examples where the OR is big enough to turn heads. The best example is the British Doctors Study and there is no DAG there.
2022-12-28 03:02:34 @Apoorva__Lal @RexDouglass @jon_mellon @zacharylipton @analisereal Hah, I’m with you. But again, to be specific: what’s an example in Econ or other social sciences where the OR is 10 or more? I’m specifically looking for published estimates of huge magnitude.
2022-12-28 02:22:44 @jon_mellon @RexDouglass @zacharylipton Yes, Interrupted time series are the worst!
2022-12-28 02:20:23 @RexDouglass @jon_mellon @zacharylipton But people don’t tend to publish these in scientific venues, do they? I would like specific examples if possible. Just to be clear: I’m mostly agreeing with you! But I am also curious about the academic landscape of such results.
2022-12-28 01:03:12 @RexDouglass @jon_mellon @zacharylipton Hmm, but even the best-case odds-ratio asserted by the authors there was 1.1. Do you have an example of a study with an OR of 10 or more that turned out to be false?
2022-12-28 00:36:21 @RexDouglass @jon_mellon @zacharylipton I don't disagree here, but do you think determining efficacy is _easier_ for treatments with small effect sizes?
2022-12-27 23:53:01 @paulesq711 @feelsdesperate In retrospect, the lockdown strategy may have delayed covid, but it prevented covid nowhere. And in retrospect, the harms to society---and especially to children---were not worth it. 5/5
2022-12-27 23:51:40 @paulesq711 @feelsdesperate California locked down earlier and harder than most of the US in Spring 2020 before there were many cases. We had the longest school closures in the US. We then had the worst wave in the US in Jan 2021. 4/x
2022-12-27 23:51:13 @paulesq711 @feelsdesperate Czechia was a "European success" story for locking down early and masking. They ended up with one of the highest death tolls in Europe. 3/x
2022-12-27 23:50:22 @paulesq711 @feelsdesperate 3. we knew in Feb 2020 that there would be multiple waves. there was a lot of denial about these facts that all ended up being true. it wasn't more data that was needed, it was more acceptance that this was not something we could control. 2/x
2022-12-27 23:50:04 @paulesq711 @feelsdesperate I want to respectfully challenge your position here, because I think we need more reflection on this: 1. we knew in Feb 2020 that the the old were 10000x more at risk than the young. 2. we knew in Feb 2020 that if the disease left China it would be come endemic. 1/x
2022-12-27 23:45:39 @DanielHadas2 @drvictoriafox @casertron3000 @JamesSurowiecki No, you misunderstand him. Surowiecki is never allowed to be wrong, but politicians are always wrong. This exchange between Zac and James back in 2020 is hilarious and emblematic. https://t.co/fNBIs1zib4
2022-12-27 23:41:22 @feelsdesperate @paulesq711 No. In retrospect, given how this played out, school closures, stay at home orders, etc. should never have happened.
2022-12-27 23:26:43 @jon_mellon @zacharylipton Rabies vaccine is not 100% effective (i.e., some people tragically still contract rabies even after vaccinated), but this is another good example of a very large effect size.
2022-12-27 14:36:10 @ajlamesa 2nd waves only happen if you behave badly. This is why Germany never had a bad covid wave. Oh wait. https://t.co/LJ0iEVyhOE
2022-12-27 14:28:13 @ajlamesa According to wisdom-of-crowds guy, 2nd waves only happen if you lift mask mandates. https://t.co/OYzU4NZgTN
2022-12-27 08:27:36 @ShalitUri @zacharylipton 100%. And our acceptance of this fact doesn’t rely on precise estimates of risk ratios.
2022-12-27 08:06:55 @ShalitUri @zacharylipton Vaccine efficacy is a great example for Zach’s question: can t we “believe” the RRs that come from observational studies considering the volatility of disease dynamics?
2022-12-27 07:59:00 @zacharylipton The effect was so large that they were considered “discovered” with case study experiments in the 1910s. Here’s a talk I gave if you want to learn more vitamin history. :) https://t.co/TQUq5XK3fA
2022-12-27 07:56:59 @zacharylipton I’d be shocked if a compelling example exists. But if you find one let me know! My mantra is “Large effects don’t need statistics. Small effects are illusory.”
2022-12-27 06:40:54 @zacharylipton Antibiotics and vitamins were pre “modern” causality. Do they count in your mind? Give me a starting year and I’ll see if I can think of anything since then.
2022-12-27 06:38:22 @zacharylipton Vitamins provide a near infinite reduction in the risk of acquiring a deficiency disease.
2022-12-27 06:37:20 @zacharylipton We never ran an RCT on antibiotics before their widespread adoption.
2022-12-27 06:33:47 @zacharylipton “Smoking causes cancer” consistently yields an odds ratio over 10. “Seat belts prevent car accident fatality” fairly robustly has an odds ratio of around 2.
2022-12-27 06:28:36 @zacharylipton Would you be ok with odds ratios? Or are you asking specifically about absolute risk?
2022-12-27 00:20:33 @tdietterich I don't want to argue about technical innovation here. I am more interested in why we impulsively assign credit to "AI" when always behind the curtain are underpaid workers in LMICs.
2022-12-27 00:17:29 @tdietterich It says that the bots were too crude, so they hired contract labor. I would not call a decision tree read by a person "automation." This is just often what is given to contract workers to avoid having to pay for training,
2022-12-27 00:02:53 Re LRT: it's still wild to me that so many think that moderation is done with automated "algorithms," when in reality it's farmed out to inexpensive, poorly trained labor in the Philippines.
2022-12-26 23:54:06 Machine learning at scale in action, everybody! https://t.co/DDFItqPjkJ
2022-12-26 01:38:29 @DanielHadas2 What if they were already broken and Covid merely revealed their irreparable state?
2022-12-26 01:10:31 @EWidera Nice. A few other super tasty ones with designs I like... Sibilla, Cynar, Punt E Mes, Angelino. https://t.co/BARjIGm0mR
2022-12-26 00:35:11 @EWidera Do you like amari? They have pretty bottles. A crowd pleaser is Averna. If you want something more smoky and less sweet, Sfumato. https://t.co/BW18GBw5xV
2022-12-26 00:18:09 @JamesSurowiecki @MicroGuy18 Why does it matter if they were done with covid? Many public health departments are now arguing that masks prevent the spread of flu and RSV. We have decades of research showing this is not true, but that doesn't seem to matter. https://t.co/THHjL1BNzS
2022-12-26 00:14:00 @JeffBullard16 @JamesSurowiecki @truthartbeauty This is the study. Looking forward to hearing James' insights as to why this RCT is also underpowered. https://t.co/DyrkS1isJV
2022-12-26 00:11:39 @JeffBullard16 @JamesSurowiecki @truthartbeauty James doesn't care. I'm sure he'll tell us it was woefully underpowered.
2022-12-24 23:59:27 @excel_wang Yes, the editor heuristic is essentially "Of all of the people who might be able to understand this paper, which ones can exert some pressure over to actually going to finish this review?"
2022-12-24 23:50:48 @excel_wang You can't interpret the result this way. What if only shoddy reviewers were nudged by the paltry 100 dollars? Moreover, you say "potentially." There is no evidence that first choices are better than seconds for assessing paper quality.
2022-12-24 23:38:09 @excel_wang Increasing the rate of acceptance and decreasing the median review time is not "fixing" peer review. In fact, if you read the discussion of outcome 3, the paper argues that the paid reviews are not substantively different.
2022-12-24 02:50:04 @tallulahforster @drvictoriafox Oh sure. Certainly Yamey's analysis is baseless. But I think folks on both sides of the spectrum make the mistake of arguing their point based on "the data."
2022-12-24 02:42:03 @drvictoriafox @tallulahforster People will comb over the data for years, but, based on our experience with previous pandemics, I doubt we'll learn much in the end. As you said, pandemic policy leans more on values than "what works."
2022-12-24 02:30:22 @drvictoriafox @tallulahforster We all made a mistake buying into this "worse/better" narrative, as if we could assess winners and losers by staring at shoddily aggregated death time series. The only thing following curves enabled was political mudslinging.
2022-12-24 00:04:49 @bergerbell They were much worse than this. People who were outspoken surveillance critics *strongly advocated* for using these surveillance tools to "fight covid." A horrifying about face.
2022-12-23 23:50:58 @DanielHadas2 The power of the scientific method rapidly diminishes as it moves from the inanimate world to human behavior. Science, as it turns out, has a hard time with the singular and the unverifiable.
2022-12-23 23:43:18 @DanielHadas2 But the caveat, of course, is that most disputes in science are settled using this approach!
2022-12-23 23:42:22 @DanielHadas2 Yes, I agree with this as well. The worst sort of inquiry comes from "My team believes X, so I must run an experiment to prove X is true."
2022-12-23 23:35:02 @DanielHadas2 This is a nice way of putting it. We must clarify the *kind* of politics that scientists engage with in the "name of science." Petty fights over competing theories: unavoidable. Asserting policies of particular ruling parties are demanded by The Science: avoidable.
2022-12-18 15:44:13 @ajlamesa Love an article that starts with "There are so many reasons to wear a mask right now." And yes, our humble Bay Area still boasts a ton of masking.
2022-12-17 21:36:14 @kevin_bowen @bendreyfuss Twitter adamantly claims that their shadowbanning does not affect your followers seeing your tweets. They could be lying, but no one has provided evidence of this.
2022-12-17 21:26:20 @bendreyfuss The other part of the theory is that shadowbanning affects the algorithmic timeline. But it is unclear who wants to see these tweets from people you don't follow that are liked by people you follow.
2022-12-17 20:28:15 @SeanTrende Your experience doesn’t count unless you write a 37 tweet thread.
2022-12-17 19:45:50 @RWerpachowski @tdietterich @rajiinio We have far more than five 9s for elevators. My question is "what enables the reliability of elevators?"
2022-12-17 19:10:38 @tdietterich @rajiinio It depends on what you mean, right? Do elevators deliver five 9s because we have mechanistic models?
2022-12-17 19:06:20 @rajiinio Chat bots are one of the lowest stake applications you can imagine. And they consistently generate buzz for our AGI compatriots in SF.
2022-12-17 18:51:15 @drvictoriafox @VPrasadMDMPH Have to add that this person is not just an editor, but a senior editor of health and medicine! Wild.
2022-12-17 18:42:57 @rajiinio Yes. And also for most of the RL applications targeted, we need five 9s of reliability. But machine learning consistently only delivers one 8.
2022-12-17 16:32:24 @DanielHadas2 I thought this reporting was insightful and solid, though still incomplete: https://t.co/kyxU0GoCWd
2022-12-17 16:31:40 @DanielHadas2 Maybe Elon will let me into TwitterHQ to do some analytics. LOL. More seriously though: if it's in silicon valley's interest to prevent people from seeing their tech doesn't work, how can one do fact finding?
2022-12-17 15:57:38 @DanielHadas2 But the shadowbanning, while potentially harmful, seems to have been ineffectual. And that highlights the broad incompetence of tech nerds more generally. This manifest incompetence should influence how we think and talk about social media. 5/5
2022-12-17 15:57:00 @DanielHadas2 I certainly agree that the opinions of the old Twitter management was indistinguishable from the editorial board of the NYTimes. This remains problematic. The outright banning of people and topics is inexcusable. 4/5
2022-12-17 15:56:30 @DanielHadas2 What matters to me is how the NYTimes reporters find quotes for their stories. These tend to be exclusively pulled from one side of a niche group of warring fanatics. 3/5
2022-12-17 15:55:55 @DanielHadas2 Twitter corp. foolishly spent a ton of time engineering for the normie, but what keeps this site relevant is the interaction of the media with a broader class of, for lack of a better word, influencers. 2/5
2022-12-17 15:54:52 @DanielHadas2 Perhaps. But normies on Twitter mostly follow celebrities and sports... https://t.co/PlSSII129s 1/5
2022-12-17 15:38:11 @DanielHadas2 Having followed covid twitter for far too long, this is just preposterous. Twitter is all about fighting with "the enemy." That's what makes this site work: put a bunch of people who hate each other in close proximity and have them virtually scream until their fingers bleed.
2022-12-17 15:35:15 @DanielHadas2 @moveincircles I'm a big @moveincircles fan. Which essay of her's do you have in mind?
2022-12-17 15:34:10 @DanielHadas2 I believe that there is something powerful about the engagement design of social media, but the coding done on top of the initial design is all clearly incompetent once investigated.
2022-12-17 15:33:12 @DanielHadas2 @moveincircles These companies need you to believe that their algorithms do something. If not, they go bankrupt. This is already happening to Facebook.
2022-12-16 23:52:43 @DrToddLee @VPrasadMDMPH I'm trying to figure out why they excluded 100 patients because they tested positive for influenza. Seems to defy the logic of a test-negative control design.
2022-12-15 21:04:44 @JorgenHarris @phl43 Why do peer review have to be part of this process if a working paper is openly available for feedback? All of the evidence I have seen, even in econ, suggests that peer review adds no additional value.
2022-12-15 19:41:32 @contrarian4data @kerpen @ifihadastick @rt7683 @jsm2334 I think I'm missing what you want me to see... I still see a 5x risk reduction.
2022-12-15 19:39:24 @contrarian4data @ifihadastick @rt7683 @jsm2334 @kerpen What's wrong with this table?
2022-12-15 17:48:30 @ifihadastick @contrarian4data @rt7683 @jsm2334 @kerpen yeah, I agree with you about the counting in the peds trials. that shouldn't be allowed.
2022-12-15 17:24:38 @ifihadastick @contrarian4data @rt7683 @jsm2334 @kerpen How do you see this in Table S30 that @contrarian4data posted? They start counting at randomization.
2022-12-15 16:46:14 @contrarian4data @ifihadastick @rt7683 @jsm2334 @kerpen I'm not sure what to read off here? In the Moderna data, that's a 78% efficacy against any infection. The peds trials were much lower efficacy. When rephrased in terms of risk reduction. 78% efficacy is 5x risk reduction. 30% is only 1.4x.
2022-12-15 15:03:11 @DimitrisPapail @damekdavis Forget it, Dimitris. It's Twitter.
2022-12-15 14:54:31 @damekdavis The academic promotion process is rotten and broken and needs to be completely redone. But bureaucracy can never shrink, so we are stuck with a metastatic death spiral of bean counters, LORs, and ad hoc committees.
2022-12-15 14:34:01 @j_mcgraph CS has to clean its own house first. I have been surprised to learn that code and data release is not required for some of the major "AI" conferences.
2022-12-15 14:30:43 @TheNickFoy True, that's one of the main tenets in The Church of The Science.
2022-12-15 14:26:23 @TheNickFoy The tweet above the quoted one is arguably worse: no one has a moral obligation to test for covid.
2022-12-15 02:41:47 @necozay @BasseyE Zhang may be the exception that proves the rule here. What would have happened had he posted it to arxiv? Reading accounts of how the Annals processed this paper, it seems a bit lucky that they ended up truly reading it. https://t.co/clumnxAnos
2022-12-15 01:24:38 @damekdavis Yes, there is a nihilistic hole we quickly fall down, but I was asking about relative importance. It's clear that data transparency is far more important than peer review for everything except pleasing bean counters on promotion committees.
2022-12-15 01:22:02 @Raamana_ Yes, I'm ok with trying to suss out the bad reasons even if there are no good reasons. Inertia and general academic conservatism is certainly a partial explanation.
2022-12-15 00:54:16 @BasseyE Also, you should read this argument against peer review which is fantastic: https://t.co/ClI8wgNMrx
2022-12-15 00:53:15 @BasseyE Counterpoint is that peer review is barely 60 years old.
2022-12-15 00:39:25 The hivemind selected the correct answer. Can someone now explain why Option B is always mandatory yet Option A too often remains optional? https://t.co/WwwXPVmwV1
2022-12-14 17:43:29 Context for Option 2, a must-read essay on the utility of peer review. https://t.co/ClI8wgNMrx
2022-12-14 17:42:42 Some context for Option 1. @phl43 proposes a new rule of thumb. https://t.co/ElxnH8LV1b
2022-12-14 17:41:40 Which is more important for "advancing science?"
2022-12-14 17:37:06 @JLimHospMD @Medical_Nemesis I limit the number of anonymous accounts I follow, but I make a major exception for @Medical_Nemesis. Definitely one of the most thought provoking accounts on this website.
2022-12-14 17:32:45 @phl43 Spoiler: this is what the essay I linked proposes as well. It's also my personal preference for what we should do. And I would still very much like to hear your arguments, as they only strengthen the case. Looking forward to reading!
2022-12-14 17:24:59 @phl43 Usually when I raise this argument, someone replies "well what is your constructive suggestion to fix it?" I finally read a compelling answer and would love to hear your take: https://t.co/ClI8wgNMrx
2022-12-14 17:23:43 @phl43 This is a good rule of thumb! Now what does it say about medicine?
2022-12-13 15:43:16 @rajiinio @KordingLab Maybe this is a more appropriate distinction: Google/Deepmind is obsessed with Science and Nature papers. OpenAI occasionally publishes conference papers, but usually just goes straight to a massively hyped, flashy demo. My claim is the latter has been more influential.
2022-12-13 15:33:59 @rajiinio @KordingLab GPT-x is not an improvement in any scientific sense. It's an improvement of mindshare and marketing. OpenAI built an incredibly successful API and hype machine on top of very unsurprising technology.
2022-12-13 15:11:22 @rajiinio @KordingLab I don't agree. BERT was really bad at text generation, and it seems like the "masking" idea wasn't important. And, most importantly, Google is terrible at making flashy demos like Open AI. GPT etc. are ELIZA-Bots on an infinite supply of Azure credits and SV hype.
2022-12-13 04:09:48 @mraginsky @KordingLab Yours is 100% better, but man, what is that saying about chatgpt's concoction?
2022-12-13 03:54:46 @mraginsky @KordingLab Eh, I couldn't resist. But man, this sounds disgusting. https://t.co/zPLr6OKN7U
2022-12-13 01:47:34 @mraginsky @KordingLab Hahah, dear god.
2022-12-13 01:14:31 @mraginsky @KordingLab Bartender Max, invent me a cocktail called "Big Data."
2022-12-12 23:47:00 @KordingLab Others clearly catch up (stable diffusion produces more interesting art than DALLE2), but I think it's important to keep in mind that there is a single company driving this general trend. Why do you think that is? And why are people so captivated?
2022-12-12 18:55:25 @KordingLab But these are all copying OpenAI, no? OpenAI is always the first mover. And OPT-175B and galactica are embarrassingly poor models.
2022-12-12 18:38:18 @KordingLab When you say large models, it seems to me that you are envisioning "OpenAI projects." Am I wrong? Can you point to something made by someone else?
2022-12-12 17:49:10 @KordingLab big data! speaking of old fashioned!
2022-12-12 17:34:51 @KordingLab my editor told me to never use the word modern in a name. you'll be old fashioned in 2 months.
2022-12-12 17:20:43 @KordingLab LAME sounds right to me. I don't know why we obsess about the size so much. It fetishizes waste and wealth.
2022-12-12 16:30:26 @KordingLab Why are you using Stanford branding for defining the future of education?
2022-12-12 14:24:20 @phl43 Eh, what's the point of granting people that NPIs have "some" effect? Every intervention does "something." But almost all don't do much towards desired ends. (And once you accept this, the conclusion that science isn't real follows in about two lines.)
2022-12-12 14:17:56 @KordingLab What's a foundation model?
2022-12-11 22:17:10 @rwjdingwall @guardiannews This quote is by a 77 year old "clinical professor emeritus" at UC Berkeley who, as far as I can tell, has written 5 academic papers ever, and none on face masks. Aka, according to the Guardian, an expert.
2022-12-11 15:07:30 @TheNickFoy @KelleyKga Nick, if he hadn't blocked you maybe you could have gotten him to also explain why he thought Theranos was a great company and why he figured AI would have replaced radiologists by 2020.
2022-12-11 01:42:52 @can @jack It used to switch back weekly, but I don't think Twitter has tried to switch me back to Home for several months.
2022-12-11 01:31:21 @can @jack I'd be curious (a) which percentage of AUs stick with Latest Tweets (b) which percentage of AUs know you can switch to Latest Tweets. I should call my neighbor Schellenberger and see if he can get me some analytics.
2022-12-10 23:27:16 @ZacBissonnette Saying that this person "pushed" school closures is an understatement. Check out her quotes in this amazing article: https://t.co/6CEz8m2gkW
2022-12-10 17:33:09 @ADAlthousePhD "explorative post hoc analyses" is a wonderful collection of words.
2022-12-10 04:05:30 @alirahimi0 I had to google it. And then I was similarly disappointed.
2022-12-09 22:09:48 I had a fun time talking to @jaycaspiankang about why ChatGPT is boring, and how it's a reflection of us all being pretty boring too. https://t.co/BVDhe1aDLY
2022-12-09 18:11:18 @AnilMakam @walidgellad Very very generous!
2022-12-09 17:13:42 @alchemytoday @R_Hughes1 @GYamey @DrNeilStone Oh, come on, man. This is not an argument. Twitter has no business. They have lost money for years. The best thing for Twitter's "business" in the last ten years was Trump. Banning Trump was bad for its business. So was that censorship?
2022-12-08 14:19:13 @dnunan79 My heuristic: It has "e-values" in the sentence, so be very suspicious!
2022-12-08 14:18:21 @phl43 Bureaucracy never shrinks, and bullshit always begets more bullshit. But what do you think about the fact that Twitter didn't collapse when all the software engineers left?
2022-12-08 14:09:11 @JacobAShell @DanielHadas2 100% this. We're seeing the same thing at the UC. If you can get away with charging full price for sitting in your bedroom with a laptop, why couldn't you get away with simply giving everyone an A who's on your roster?
2022-12-08 13:00:00 CAFIAC FIX
2022-12-07 08:00:00 CAFIAC FIX
2022-11-15 15:25:31 @VPrasadMDMPH @davidemccune @R_Hughes1 @DanielHadas2 @AlastairMcA30 @apsmunro @ADC_BMJ There are funny things that seem to happen in scientific innovation. There are exceptions, but most often we see(a) you find a huge effect (say, gleevec), and everyone uses the treatment.(b) you find a null effect, and people argue that you did the experiment poorly.
2022-11-15 15:23:04 @VPrasadMDMPH @davidemccune @R_Hughes1 @DanielHadas2 @AlastairMcA30 @apsmunro @ADC_BMJ Right, and that's what's so wild. I think most oncologists know this, but there are vested interests and inertia that keep screening practices going regardless.
2022-11-15 15:21:25 @VPrasadMDMPH @davidemccune @R_Hughes1 @DanielHadas2 @AlastairMcA30 @apsmunro @ADC_BMJ How large would a study need to measure all cause mortality reduction from colonoscopy? Millions at least, no?
2022-11-15 15:20:29 @VPrasadMDMPH @davidemccune @R_Hughes1 @DanielHadas2 @AlastairMcA30 @apsmunro @ADC_BMJ Perhaps I could rephrase: RCTs *can* change practice, especially when the results are dramatic. However, they are not sufficient to change practice, especially when the effect sizes are tiny and the events are rare.
2022-11-15 15:15:33 @VPrasadMDMPH @davidemccune @R_Hughes1 @DanielHadas2 @AlastairMcA30 @apsmunro @ADC_BMJ I don't know! We've had so many RCTs on cancer screening, and the debate rages on no matter what they end up finding, no?
2022-11-15 14:58:20 @davidemccune @DanielHadas2 @AlastairMcA30 @apsmunro @R_Hughes1 @ADC_BMJ Daniel is arguing something else, and I think it's worth engaging with: has there ever been equipoise for masking children?Given that we knew from early 2020 that children were at low risk, that wearing masks goes against many people's value, that there may be other harms...
2022-11-15 14:52:45 @DanielHadas2 @davidemccune @AlastairMcA30 @apsmunro @R_Hughes1 @ADC_BMJ Taking it to genital mutilation is too extreme, but I agree with the general argument. Perhaps a better analogy: We don't do RCTs on corporal punishment.
2022-11-15 02:49:25 @ElGee35 @fperrywilson @EpiEllie @NEJM I read the appendix. The proposed DiD model is preposterous and inappropriate: pandemic growth is nonlinear and depends on multiple unreported factors.In fact, many including the *author of the used DiD method* have critiqued using DiD models when evaluating pandemic policies.
2022-11-15 01:59:59 @JDHaltigan @AllenShrugged @VPrasadMDMPH @walidgellad @AnilMakam Tweet 12 https://t.co/EpUypbm3Px
2022-11-15 01:55:05 @JDHaltigan @AllenShrugged @VPrasadMDMPH @walidgellad @AnilMakam you don't even find the word "exponential" mentioned in the supplement.
2022-11-15 01:51:10 @JDHaltigan @VPrasadMDMPH @walidgellad @AnilMakam It also models the effect of masking on case rates as linear, when it's clear it can't be. Anyway, without the stupid DiD, this paper is just like the 50M plots @ianmSC has been posting for 3 years. Except Ian has 50M plots, and they have one.
2022-11-15 01:45:59 @JDHaltigan @VPrasadMDMPH @walidgellad @AnilMakam I think the argument about models just muddies the water. Of course any other analysis will not find an effect. Their DiD model is completely implausible, but it makes it look "well done" and "rigorous."
2022-11-15 01:40:25 @walidgellad @AnilMakam Note that in the article, the DID model assumes that the difference between masked and unmasked should be linear (page 5 in the supplement).
2022-11-15 01:36:52 @walidgellad @AnilMakam In that thread he claims it's "because exponential growth is weird."
2022-11-14 14:48:55 @JacobAShell It's called Cryptozoology, sir. Please stop centering the Sasquatch in your discourse.
2022-11-14 01:41:22 @BranchWestyn @ShiraDoronMD @PunditPandemic I look forward to reading.
2022-11-14 01:39:57 @JSEllenberg Effective Altruism, AGI risks, alignment, Jordan. You can't escape that nonsense out here.
2022-11-14 01:02:44 @BranchWestyn @ShiraDoronMD @PunditPandemic I'm arguing that even if we ran an RCT, the results are preordained.
2022-11-14 00:52:38 @BranchWestyn @ShiraDoronMD @PunditPandemic But the important point is this: if the people doing the study were skeptical of masks, this study would find a null effect.If the people were positively dispositioned towards masks, the study would find a tiny significant effect.
2022-11-14 00:51:59 @BranchWestyn @ShiraDoronMD @PunditPandemic Yes, it would be cluster randomized by classroom or school. There would have been a "mask promotion" campaign in the school, but not a mandate. And students would be individually consented.This wouldn't test mandates, but would measure spread in the presence of encouragement.
2022-11-14 00:28:34 @ShiraDoronMD Even if we disagree that masks (and especially mandates) work, I still think this thread is very wise.
2022-11-14 00:24:44 @ShiraDoronMD @PunditPandemic @BranchWestyn We could have done what they attempted to do in the Bangladesh mask study. Ask for consent on a per-student basis and only look for infections in those who agree to consent.
2022-11-14 00:02:02 @anish_koka I say we publish this person's incoherent econometric analyses in the NEJM. Who's with me?
2022-11-13 20:45:36 @james_e_b_ @ajlamesa @paulkrugman Someone (maybe me) really should write up a history of the "blueprint for reopening" in California. It was like the war on terror with equity metrics. A bunch of technocratic bullshit that solved nothing.
2022-11-13 20:44:38 @james_e_b_ @ajlamesa @paulkrugman Newsom made draconian rules that forced many private schools to close. We had this color coded system with arbitrary metrics that were impossible to achieve. Eventually pressure caused him to drop the mandates against private schools.
2022-11-13 18:07:53 @djc795 @VPrasadMDMPH A colleague sent me this one yesterday. https://t.co/Xve6gtMXCj
2022-11-13 17:57:38 @djc795 @VPrasadMDMPH Hi David, could you share the other observational study you had in mind?
2022-11-13 15:15:00 @davidclarkphel1 @samhaselby It's all on the wikipedia. SBF went to Crystal Uplands. Ellison went to Newton North.But you are right, it's a bit wild that I know both of these high schools and their reputations.
2022-11-13 15:12:43 @anish_koka Good thing I have all of this gold buried in my backyard.
2022-11-13 15:05:42 @samhaselby There's something dysfunctional at both of those places that is different from the ivies. Both are much more driven by techno-utopianism. And this whole FTX scandal is "when gifted and talented programs go horribly wrong."
2022-11-13 15:02:51 @samhaselby I agree with your broader point, but it's worth noting SBF went to MIT. And Caroline Ellison went to Stanford, though her parents were MIT Econ Profs. Both MIT and Stanford are not Ivy League but are certainly dominated by the progressive neoliberalism of which you speak.
2022-11-11 21:18:29 @ADAlthousePhD Joking aside, that is a beautiful lift.
2022-11-11 21:12:32 @ADAlthousePhD Pretty good signoff, but sumo is cheating.
2022-11-11 13:24:46 @sdbaral virtue signaling, wishful thinking, and irredeemably confounded observational studies.
2022-11-10 18:26:46 @apsmunro This stuff stops when prestigious journals stop publishing it. The past week has shown we're still a ways off from that, unfortunately.
2022-11-10 13:56:23 @schmangee @KelleyKga @ajlamesa @alexanderrusso @TracyBethHoeg Eleanor J. Murray, Sc.D., M.P.H, aka epiellie.
2022-11-10 13:45:43 @katrosenfield Less is definitely more. Removing some of the existing policies would be good as well. Their misinformation policies are draconian but enforced based on vibes.
2022-11-10 13:01:40 @politicalmath Very sad to see Twitter zealots being platformed by NEJM
2022-11-10 12:40:18 @alexanderrusso Wow, this article is absurdly biased. Raifman, who is an unhinged masking zealot, is just quoted as a partial expert with no mention of her past advocacy, whereas Tracy Hoeg is quoted as a "fierce critic of school masking."
2022-11-10 12:29:55 @zoonotic1 @AaronRichterman @BallouxFrancois @ENirenberg I assure you that I, @AaronRichterman, @excel_wang , and @TAH_Sci know what sensitivity analyses are and also know they are never dispositive.
2022-11-10 03:05:20 @AaronRichterman @BallouxFrancois @ENirenberg The issue is that close contacts didn't need to test if they were masked. I don't see how this resolves that.
2022-11-10 02:59:13 @AaronRichterman @BallouxFrancois @ENirenberg What exactly does that clarify in your mind?
2022-11-10 01:11:45 @DanielHadas2 @TShampling @Medical_Nemesis Such an important point. And I think you can see a lot of this thought from some of the EBM crowd, even as they use the language of data. Many are advocating medical conservatism and humility, but using the language of “science.”
2022-11-09 23:15:47 @vineettiruvadi @tdverstynen @KordingLab Have fun on mastodon. Muting you now.
2022-11-09 23:14:31 @vineettiruvadi @tdverstynen @KordingLab Did you read this paper, or are you just being annoying?
2022-11-09 23:12:06 @vineettiruvadi @tdverstynen @KordingLab Why did you tag me on this?
2022-11-09 23:11:43 @matthewshaw1111 @AAMortazavi This paper came out as a preprint and the criticism was pretty damning. I'm astounded this got published in NEJM.Vinay's blog in particular pulled no punches.https://t.co/rKgXGqUSWS
2022-11-09 15:02:17 Look folks, R doesn't lie. This scatter plot reveals I'm right. https://t.co/fsqOLykjcY
2022-11-09 14:01:18 @thackerpd They were wrong in the other direction this time. Nothing has been fixed. The response rates are so low that it’s just guessing.
2022-11-09 13:55:50 @thackerpd Or it's just random guessing by pollsters who can't get anyone to answer their polls. We can pick whichever answer makes us feel better.
2022-11-09 13:52:31 Polls are usually wrong and they can be wrong in any direction. That's what we call analysis, folks. https://t.co/UI7mpqr4r8
2022-11-09 01:15:59 @DimitrisPapail What if it's Bayesian, bro?
2022-11-08 21:58:33 @andreavhowe I've lived here for 10 years and still haven't gotten used to the absurdity of the California ballot.
2022-11-08 21:55:29 @andreavhowe Mistaking them for more ballot propositions, I voted no on all of them.
2022-11-08 19:00:37 @AnilMakam Sadly, everyone who votes in CA knew what you meant without the key prepositional phrase.
2022-11-08 17:59:07 @JLimHospMD Apparently, the NY Times and CNN both need to hear this.
2022-11-08 16:53:16 @ajlamesa Voting in California shows how this is true. Our system is completely insane. It's built to make machine democrats undefeatable.
2022-11-08 16:09:10 @ducha_aiki @JFPuget Now I think I don't understand again. The continuous-label problems on Kaggle are, in bulk, much more noisy and "hard" than the discrete-label ones. My colleagues and I did a meta-analysis a few years back.
2022-11-08 16:02:37 @JFPuget @ducha_aiki Hmm, but this is not the same as saying that classification problems are easier than regression problems. You are saying that using regression-mode in XG-Boost works better than classification-mode on certain regression problems on Kaggle. Yes?
2022-11-08 15:39:07 @walidgellad For what it's worth, I liked your proposal to bring back snail mail.
2022-11-08 15:36:20 @walidgellad What is the solution in your mind? Especially that in this EHR data-mining case, it's a single PI causing the problem.
2022-11-08 15:15:50 @ducha_aiki Hmm. I don't see why it would be any different for those two different model classes. Perhaps @JFPuget has an example in mind.
2022-11-08 15:07:53 @KelleyKga @walidgellad At some point we need to call out the PI by name: Ziyad Al-Aly. This is the person behind all of these VA EHR data mining papers. They are consistently highly questionable, opaque, and unverifiable. Yet he consistently promotes them hard in the media.
2022-11-08 15:02:28 Spoiler alert. https://t.co/7ypI6VmzmG
2022-11-08 15:01:53 @djhsu @itsrainingdata @ducha_aiki
2022-11-08 15:01:25 For those interested, @itsrainingdata reminds us that the answer is in Chapter 6.7 of "A probabilistic theory of pattern recognition" by Devroye, Györfi, and Lugosi. https://t.co/tCKi3tKvBt
2022-11-08 14:59:08 @djhsu @itsrainingdata @ducha_aiki What's the other one?
2022-11-08 14:56:56 @itsrainingdata @djhsu @ducha_aiki LOL. And the answer is revealed. https://t.co/7C2UxN6UTI
2022-11-08 14:50:24 @djhsu @ducha_aiki Yes, I think this is the answer. And we could make this statement more formal if we wanted.
2022-11-08 14:42:42 @DanielHadas2 @Medical_Nemesis I think they understand, I've spoken with many of them. It's just too easy for scientists to get sucked into fighting data with data.(I certainly have been guilty myself)
2022-11-08 14:29:02 @ducha_aiki @roydanroy Got it. That makes sense!
2022-11-08 14:28:05 Fascinating conversation here. And it leaves us with a rephrasing of the question @ducha_aiki was asking: "Why do ML models trained to predict discrete labels work so much better than those trained to predict real-valued labels?" Thoughts? https://t.co/6S2SUyqpMt
2022-11-07 18:42:15 @ducha_aiki Yes, exactly. It could also be "I'd like to know the mass, but I am only observing a bunch of forces and accelerations."
2022-11-07 18:38:11 @giant_hornet LOL, that's a great answer! I love it.
2022-11-07 18:37:35 @ducha_aiki Not a stupid question at all. I'm asking more about what you care about in the end. Using B for A suggests that you really only cared about A in the first place. Does that make sense?
2022-11-07 18:29:07 How about a poll?
2022-11-07 18:28:10 Option C: something else! (Please reply to this thread).
2022-11-07 18:27:58 Option B: Regression is when I propose a model of how Y depends on X, and I use data to find the parameters of this model.
2022-11-07 18:27:45 Option A: Regression is when I solve a supervised learning to predict Y from X when Y is real-valued.
2022-11-07 18:27:34 I know everyone has left for Mastodon, but this tweet has inspired an interesting conversation on ML Twitter. It left me asking: What does “regression” mean to you in the context of machine learning?https://t.co/jKK78POci4
2022-11-07 17:57:47 @KordingLab @elonmusk that last part is accurate. as we all know, progress in academia happens one funeral at a time.
2022-11-07 17:49:53 @KordingLab @elonmusk Let my people go!
2022-11-07 17:48:03 @KordingLab @elonmusk wait, was monkeypox the boils? holy shit.
2022-11-07 17:44:45 @KordingLab @elonmusk I dunno, I'm most excited for the raining fire...
2022-11-07 17:39:50 @KordingLab You don't want to hang around for the 10 plagues first?
2022-11-07 14:51:32 @DimitrisPapail @qberthet @jasondeanlee @shai_s_shwartz @djhsu not silly at all, IMO...
2022-11-07 14:37:29 @jaycaspiankang It's human supremacist to presume you can know what the AI assumes from its lived experiences.
2022-11-07 14:29:49 @jaycaspiankang Maybe if you put it in quotes and were mocking one of the weird hashtag-resistance grifters on here?
2022-11-07 14:27:06 @qberthet @jasondeanlee @shai_s_shwartz @djhsu Don't want to hash it out on Twitter? I just had my first cup of coffee, so I'm ready to wax philosophical!
2022-11-07 14:21:24 @jasondeanlee @shai_s_shwartz @djhsu This conversation has drifted far away from the original question: why does classification work so much better than regression? I have a lot of thoughts on this, but I'd like @djhsu and @jasondeanlee's takes.
2022-11-05 20:18:31 @samhaselby 100%. I couldn't agree more.
2022-11-05 20:14:23 @samhaselby That's interesting! Additionally, our non-socialized health care loves to treat everyone as if they are "sick until proven healthy" and this influences the committees that draft guidelines towards intrusive over-caution.
2022-11-05 20:08:11 @samhaselby But the question of who we take advice from is a hard one. I advocate for more voices rather than fewer as specific disciplines have too many blindspots to lead the way for any issue of broad societal importance. 4/4
2022-11-05 20:07:05 @samhaselby Health economics research is problematic in the same way that most economics problematic: it works backwards from desired policy outcomes and uses complex math to shoehorn data to a desired conclusion.3/x
2022-11-05 20:06:31 @samhaselby I also thought her *advocacy* about opening schools was important and more right than wrong.But that doesn't mean I want to take health care advice from her or from any health economist.2/x
2022-11-05 20:04:13 @samhaselby Understood, and I empathize. Oster's pre-pandemic focus was on pregnancy, and I tend to agree with her assessment that the US guidance for pregnant mothers is exceptionally conservative.1/x
2022-11-05 19:56:00 @samhaselby The interaction between health *policy* and health *care* is messy. Example: Much of the advice of personal physicians comes from national guidelines of the CDC and other orgs. If the CDC advice differs from Euro advice, and your doctor parrots CDC advice, what do you do?
2022-11-05 19:43:03 @samhaselby In what way is the field of Health Economics devoid of relevant medical expertise?
2022-11-05 19:31:08 @excel_wang exactly this.
2022-11-05 19:29:39 @samhaselby Could you clarify? Because people shouldn't take advice from economists about anything? Or because public or private health recommendations must come from those with particular credentials?
2022-11-05 19:12:27 @edsbs And he had such an obvious path he could have taken instead: Belichick beat the Bills in a wind storm calling only 3 passes... uh, less than a year ago.
2022-11-05 16:03:57 @madsjw I say chuck em!Lauren says "It has a lot of salt, how can it go bad?"
2022-11-05 15:53:27 @madsjw wait, boss, you moved that sauce from your other house?
2022-11-05 15:34:26 @DanielHadas2 Can you say more about what you mean by this? When has science not been about prestige and authority?
2022-11-05 15:25:39 @drjohnm What if we did a trial of trials?
2022-11-05 14:55:15 @ajlamesa More: https://t.co/umXo0G6GcBI remember this as when I started getting scared about the authoritarian consequences to come.
2022-11-05 14:53:14 @ajlamesa Fauci was saying junk like this in March 2020.https://t.co/1B9eKzpZoo
2022-11-05 14:20:28 @jasondeanlee @shai_s_shwartz Right and I'd add:- No ERM method actually outputs probabilities.- Probabilities are not the only way to capture uncertainty
2022-11-05 14:10:12 @ajlamesa Not just schools. California locked down before many but had a deadly and devastating fall '20- winter '21. In Europe Czechia was a model for proactive lockdown in summer '20, ended up with a terrible wave that winter. Proactive lockdowns were demonstrably terrible policy.
2022-11-05 14:03:21 @jaycaspiankang "bro, you said you were going to give me your house."
2022-11-05 13:59:26 @jaycaspiankang No gavel banging detected.
2022-11-05 02:43:41 @AaronGoodman33 @VPrasadMDMPH On Mastodon you compose "toots." I am not making that up.
2022-11-04 16:54:08 @elizamondegreen Hmm. Latest Tweets doesn't use an algorithm. You should never see tweets from people you don't follow unless one of your followers is retweeting them.
2022-11-04 15:20:45 @scottlincicome We need to run an AB test to know for sure.
2022-11-04 13:48:44 @venkmurthy Also, Twitter thrives because reporters are here, not because it has some magic algorithm.
2022-11-03 18:53:43 @ajlamesa Now people like to say the early anti-mask rhetoric by PH folks was a "nobel lie" to save PPE for HCWs. But the record in March 2020 suggests that no one thought mask mandates were a useful intervention to slow a respiratory infection.
2022-11-03 18:51:44 @ajlamesa It's wild that we forget that most PH people thought masking was completely useless. Check out this early nature paper trying to estimate interventions. No masking policies are even considered!https://t.co/im9Z0ZIKRg
2022-11-03 18:50:27 @ajlamesa It's a worthwhile thought experiment for sure. I remember here that *everything* was closed around March 15 by Newsom's executive order. The mandatory masking followed that by about a month.
2022-11-03 18:15:44 @ajlamesa Korea had state mandated mass surveillance contact tracing. They also had long school closures that were longer than much of the United States.
2022-11-03 18:13:50 @ajlamesa But this was because of Japanese law, not because of mask culture. China, Taiwan, and Korea also have a "mask culture."
2022-11-03 13:39:18 @anish_koka The problem is, since Topol is raising it, the concern must be bullshit.
2022-11-03 13:20:52 @KordingLab Gedanken experiment: Who wins in a fight? T-Rex or Mastodon? I think that will answer your question.
2022-11-01 16:47:17 @KrauthBen But the memes. She had so many memes.
2022-11-01 15:25:30 @samhaselby The honest answer:She wrote a lot about opening schools and minimizing covid restrictions on kids. This made the left hate her.She then wrote a naive piece in the Atlantic calling for "amnesty" between both sides of the covid war. This made everyone hate her.
2022-11-01 15:22:28 @samhaselby Putting this in Haselby-land: She's an economist at Brown, and also the daughter of two economics professors at Yale.
2022-10-31 15:03:37 @JacobAShell @DorianAbbot ... Or you could just amplify the rantings of an anonymous account whose shtick is cowardly ranting against the cowardice of academia.
2022-10-31 15:02:49 @JacobAShell @DorianAbbot You could also email the two professors who run the group and ask for comment: https://t.co/hgJiDtB7Pc
2022-10-31 15:01:48 @JacobAShell @DorianAbbot Sure, let's first ask why the event is still listed as happening on the group's website: https://t.co/O1KzGNoJfk
2022-10-31 14:57:57 @AlecMacGillis @KatherineEban @jeffykao The report relies on mistranslated, misdated Chinese documents and numerous other factual errors that have been pointed out all over Twitter. But it was a compelling read, so everyone had a good time, I guess.
2022-10-31 13:54:49 @DanielHadas2 @Stanford Curiously, the event remains on the webpage as uncanceled. Did they forget to update their website?https://t.co/O1KzGNoJfk
2022-10-31 03:16:41 @SarahTheHaider @Absolutionis He's reading your tweets! https://t.co/Lc89u1YwSa
2022-10-30 20:08:37 @politicalmath I would love to hear from Tesla employees as to whether they receive similar mandatory training emails.
2022-10-30 15:51:34 @JSEllenberg @AuerbachKeller That is a definition we can agree upon!
2022-10-30 13:53:41 @JSEllenberg @AuerbachKeller What is misinformation?
2022-10-29 16:25:31 @DanielHadas2 @JacobAShell Do you two have any comment on how the latest report "bombshell" requires mistranslating Chinese? (I feel like this is Daniel's lane)
2022-10-29 01:41:21 @ajlamesa And no evidence of significant transmission on public transport!
2022-10-28 13:49:42 @TheNickFoy I mean, legs are coming soon, Nick. Legs are coming. https://t.co/lcluFhTEmB
2022-10-27 17:25:41 @fhuszar @maosbot Don't disagree there. And also that AI Alignment is a nonsense rationalist fever dream based in bad science fiction.
2022-10-27 17:24:30 @fhuszar @maosbot On it!
2022-10-27 17:11:52 @fhuszar @maosbot 300 people working on "The AI Alignment Problem" is 300 too many.
2022-10-27 14:10:32 @quantum_aram Not today at least...
2022-10-27 13:46:19 @AChandra_TO @phl43 LOL. I need to get Derek Thompson to apologize.But right: I think the "peer review" canard is only used when convenient.
2022-10-27 13:41:43 @AChandra_TO @phl43 I don't know, MSM runs with preprints when it suits them.Just last week, the Economist picked up that bizarre Lab Leak preprint written by three conspiracy theorists.
2022-10-27 13:38:51 Even Elon can't escape. “The best minds of my generation are thinking about how to make people click ads.” https://t.co/0tDidHYwLn
2022-10-27 13:35:24 @phl43 Interesting perspective from computer science: CS has convinced deans and managers that *conference* papers can and should be counted as journal publications. Pro: As a result, there are no prestigious CS journals.Con: Our conference review system is a shitshow.
2022-10-27 13:31:51 @phl43 Agreed, I have been reading variations of this anti-journal argument for over a decade and no one can form a cogent rebuttal.Unfortunately, one reason journals continue to exist is publications are required for promotion in academic and industrial research.
2022-10-27 13:25:11 @phl43 I don't agree that social media ends up being better than journals (both have warts), but I also can't believe how many journals get away with charging *authors* thousands of dollars per paper. That alone is an argument for their end.
2022-10-27 00:39:52 @VPrasadMDMPH @johndefeo One of the authors thinks Monkeypox was lab leaked.
2022-10-26 14:02:04 Hell has frozen over: I wrote a blog post saying a bunch of nice things about neural nets (and art?). https://t.co/wGTqgcZG2N
2022-10-25 19:36:23 @FO_ASchatz This can happen even with no statistics:(Rush score, pass score)Team 1: (4,1)Team 2: (3,3)Team 3: (1,4)Team 2 is not best in either score, but it has the best added scores.
2022-10-25 19:05:42 @jaycaspiankang There was an earthquake?
2022-10-24 16:30:30 @LaurentLessard @CSSIEEE Uh, that's totally bizarre. What a waste of speaker time.Unthawing from our frozen virtual pandemic time is going to take a while to shake out.
2022-10-24 16:26:28 @LaurentLessard @CSSIEEE Wait, are there going to be videos instead of talks at the in-person conference?
2022-10-24 14:14:36 @DanielHadas2 We have also created a bizarre literature on Long Covid. Long Covid arose on Facebook and became popularized by a poorly reported story by Ed Yong. It’s a perfect example of science not only behaving like the media, but taking its queues from the media. 3/3
2022-10-24 14:14:20 @DanielHadas2 Since then, we have learned about vaccines. We can quibble with policies, but these papers are good. There have been good papers on treatment as well, in particular from the RECOVERY trial in the UK. 2/3
2022-10-24 14:13:39 @DanielHadas2 Yeah, the following the science meme certainly added too much importance to the science.It's remarkable that we basically knew everything about this virus by May 1, 2020. And we knew a lot by March 1, 2020. 1/3
2022-10-23 14:18:58 Solid advice for "data scientists," no matter which topic they are studying. https://t.co/iGR2neHSKO
2022-10-23 04:03:53 @jaycaspiankang Tarik Glenn is pretty cool.
2022-10-23 04:03:21 @jaycaspiankang Curious who you have as the 2nd cool person after Marshawn Lynch.
2022-10-22 21:13:38 @JacobAShell @DanielHadas2 @Spear_Owl But it could alternatively be social-media-driven mass formation psychosis not cooperating with medicine's preexisting analytical categories.
2022-10-22 21:04:29 @JacobAShell @Spear_Owl @DanielHadas2 Though they were influenza pandemics, rather than covid pandemics, the global mortality associated with the 1957 and 1968 pandemics are both of similar magnitude to covid. The main difference being Asia was hit harder and North America and Europe significantly less so.
2022-10-22 20:51:56 @JacobAShell @DanielHadas2 @Spear_Owl Post-viral sequelae are indeed a thing, and teasing out their etiology has always been a challenge. Here's one survey from 1987: https://t.co/7igETSURw5
2022-10-22 20:35:30 @JacobAShell @DanielHadas2 @Spear_Owl No, I know. That's why I wrote it. I didn't mean it as an attack but as a lament: we're globally stuck doing obviously stupid things when common sense clearly dictates we should do the opposite.
2022-10-22 20:33:13 @DanielHadas2 @Spear_Owl @JacobAShell I see. Thanks for the clarification.I think it's interesting that I see the opposite direction, and I should think more about why.
2022-10-22 20:28:59 @JacobAShell @DanielHadas2 @Spear_Owl I do not disagree at all. But there are many other things I wish were a global consensus: that we shouldn't infringe on the habitats of wildlife, that humankind does not yet possess the ability to stop all respiratory viruses,...
2022-10-22 20:27:24 @JacobAShell @DanielHadas2 @Spear_Owl In what way are the symptoms odd?
2022-10-22 20:26:37 @DanielHadas2 @Spear_Owl @JacobAShell How does one understand "the science" without "the people and the politics?"
2022-10-22 20:26:05 @DanielHadas2 @Spear_Owl @JacobAShell There is a clear dichotomization in politics between those who are willing to entertain the lab leak and those who are not.
2022-10-22 20:25:32 @DanielHadas2 @Spear_Owl @JacobAShell Here in this thread, @JacobAShell is saying that the virology community looks guilty, hence, without understanding their arguments, they clearly must be wrong on "the science."
2022-10-22 20:25:17 @DanielHadas2 @Spear_Owl @JacobAShell We don't have to belabor into this point, but don't you think ad hominem is exactly what the lab leak debate is?
2022-10-22 20:14:16 @DanielHadas2 @Spear_Owl @JacobAShell The only way that scientific paradigm shifts happen is with new experiments that slowly chip away at an infinite landscape of possibility.Every once in a while we find something amazing, but usually, we just muddle through minutiae while we wait for the new revolution.
2022-10-22 20:13:27 @DanielHadas2 @Spear_Owl @JacobAShell My read of the post-Kuhn landscape is that there is no genuine scientific aporia but only politics.
2022-10-22 20:04:20 @DanielHadas2 @Spear_Owl @JacobAShell But there is no way to prove it either way! There will never be new data. We're all stuck scrutinizing mirages and arguing from our political corners.
2022-10-22 20:02:04 @JacobAShell @Spear_Owl @DanielHadas2 Or maybe, just like Latour told us, scientists are a bunch of crazy weirdos, and they shouldn't be out on Twitter showing the world their whole asses?
2022-10-22 20:00:00 @Spear_Owl @DanielHadas2 @JacobAShell But that thread has input from dozens of other scientists. It's not just Andersen and Rassmussen.And you can assert "there's a cartel here suppressing science!."And we will continue to fight because no one can do a real experiment to prove anything about the fucking lab leak.
2022-10-22 19:58:08 @Spear_Owl @DanielHadas2 @JacobAShell Yeah, I don't disagree. He and Angela Rassmussen both behave like children on here, and I don't understand why they can't see how harmful this behavior is.
2022-10-22 19:53:54 @JacobAShell @Spear_Owl @DanielHadas2 I think everyone on this stupid website protests too much.If anything, I'm more Latourian than I was at the start of the pandemic. But the outsider critics fall into the same traps as the insiders. No one is immune.
2022-10-21 20:22:02 @rdnowak @DimitrisPapail @amirvaxman_dgp Barron showed that functions f where the Fourier transform \tilde{f} satisfied the following inequality were well approximated by a small sum of sigmoidal functions. He also showed the number of required functions to achieve an eps-approx in L2 scaled like 4 B^2/eps^2. https://t.co/QUV74jG7Xh
2022-10-21 20:15:51 @rdnowak @DimitrisPapail @amirvaxman_dgp I agree with Rob. Barron's results were describing the sorts of functions that were well approximated by neural nets. (Cybenko's argument is certainly tidier.)
2022-10-21 16:06:14 @BenMazer @drvictoriafox @GYamey He didn't say either of these things. I understand that Ioannidis is a convenient punching bag and you don't want to engage with what he was saying in good faith.
2022-10-21 15:57:11 @BenMazer @drvictoriafox @GYamey Everyone was running around in fear, as you have admitted you were yourself. Ioannidis was asking "do we have the data to justify the fear?" Someone had to ask. Look at all the stupid shit we did because people were unreasonably afraid!
2022-10-21 15:13:58 @KevinBardosh Was the discussion unplanned and arose at another event? In order for this to be a productive conversation, it seems like thoughtful mediation would be necessary.
2022-10-21 15:06:21 @KevinBardosh I wonder how many (if any) of these discussions have occurred.
2022-10-21 14:13:45 @DanielHadas2 Your threads are too good. I look forward to my daily meditation on them.I hope you collect them in one place at some point.
2022-10-20 16:02:29 @teow_matthew @PrincetonUPress Thank you so much!
2022-10-20 15:18:24 @awgaffney Respectfully, perhaps our approach to influenza vaccination also lacks sufficient evidence. The flu vaccine has far fewer AEs than the covid vaccine, so the risks are lower. But the benefits of broad flu vaccination remain uncertain.
2022-10-19 16:46:43 RT @PrincetonUPress: Happy Pub Week to Patterns, Predictions, and Actions by Moritz Hardt &
2022-10-18 15:15:09 @unsorsodicorda @PrincetonUPress I don't think they have immediate plans, but I don't think it hurts to let @PrincetonUPress know there's demand!
2022-10-18 15:14:43 @rasbt @PrincetonUPress Thank you!
2022-10-18 14:18:30 @improveai @PrincetonUPress Thank you!
2022-10-18 14:10:55 @FO_ASchatz The football gods have given me no choice. I must respond with this GIF. https://t.co/vRshf94uSA
2022-10-18 14:07:20 @FO_ASchatz How frequently do we get Jets-Giants?
2022-10-18 14:05:14 Patterns, Predictions, and Actions is out today in hardcover and ebook! Published by @PrincetonUPress.https://t.co/iasUe6mbY8 https://t.co/M9rgBxPWc1
2022-10-18 13:21:18 @DanielHadas2 @MmeBlackBalloon @therealcaitjan "Best answered" being the critical part. There are many drugs that obviously work wonders and likely don't need RCTs, but we force them all into an RCT-approval framework because miracle drugs are a vanishing fraction of the treatments pharma tries to get to market.
2022-10-18 13:11:06 @MmeBlackBalloon @DanielHadas2 @therealcaitjan Yes, and since RCT evidence focuses on the average, our modern human-facing sciences have lost sight of individuals. But there are undoubtedly many forms of evidence, and there is no reason these other forms should have "lower value" in some rigid epistemological hierarchy.
2022-10-18 13:06:02 @DanielHadas2 @MmeBlackBalloon @therealcaitjan Yes. And though they are not perfect, RCTs have a huge benefit: they provide a critical regulatory framework to remind ourselves that most interventions don’t work at scale.
2022-10-18 01:57:29 @therealcaitjan @DanielHadas2 @MmeBlackBalloon @ThSchlich This article by Schlich is a great reference, thanks! I look forward to reading this.
2022-10-18 00:52:26 @DanielHadas2 @MmeBlackBalloon @therealcaitjan I hope to write up a more comprehensive survey on the RCT, its scope, and its criticism. It's on my near-term todo list. 7/7.
2022-10-18 00:51:24 @DanielHadas2 @MmeBlackBalloon @therealcaitjan In the social sciences, the critiques are more scathing. In particular, Nancy Cartwright has written extensively about how RCTs are miscast as a gold standard, and Angus Deaton's "Randomization in the Tropics" skewers the application of RCTs in development econ. 6/x
2022-10-18 00:49:22 @DanielHadas2 @MmeBlackBalloon @therealcaitjan John Ioannidis, an EBM diehard, has voiced concerns about how the RCT framework ended up being hijacked by Pharma interests.https://t.co/yWSgP1A0b45/x
2022-10-18 00:48:27 @DanielHadas2 @MmeBlackBalloon @therealcaitjan For an example of a medical critique, Starzl, a pioneer in transplant surgery, has a great paper called “Randomized Trialomania” that details a significant delay in approving a quality therapy for transplant patients due to the rigidity of the RCT confirmation paradigm. 4/x
2022-10-18 00:48:11 @DanielHadas2 @MmeBlackBalloon @therealcaitjan And even in places where they are the “gold standard,” they leave scientists wanting. In the medical sciences, there’s an acceptance that RCTs for approving drugs are deeply flawed, but “the best thing we’ve got.” 3/x
2022-10-18 00:47:32 @DanielHadas2 @MmeBlackBalloon @therealcaitjan RCTs have very limited scope: they are clearly the only game in town for drug approval. They determine how ads get placed in your internet experience. And they have modest influence policy in development economics. But it's not as if *all* science is driven by RCTs. 2/x
2022-10-18 00:45:45 @DanielHadas2 @MmeBlackBalloon @therealcaitjan This is indeed a favorite topic of mine. There’s a huge body of criticism out there, and I’m not sure where to start. I also don't want to spam your replies with my annotated bibliography. 1/x
2022-10-16 14:36:17 @anish_koka OK, not sure I want to work through this calculation on Twitter as this always goes badly, but I think they are saying:17 deaths in screened (11,843 people)unadjusted risk = 0.14%.157 deaths in control (56,365)unadjusted risk = 0.28%.RRR in PP is 0.52.
2022-10-16 14:17:02 @anish_koka Oh, interesting. Do you think the number 17 in table S4 is the actual count? I suspect the number comes from applying their "inverse probability weighting" analysis. https://t.co/LFVYDSOTm8
2022-10-16 14:01:40 @anish_koka What do you mean by raw per-protocol? How did you get the 0.21%?
2022-10-07 16:12:12 @GaleMorrisonEd @schmangee Don't disagree with what you write about EVs, but my point was expanding on @schmangee 's complaint about the lack of regulatory teeth in self-driving. These companies have killed multiple people because of negligence (and arrogance) yet have suffered no sanctions.
2022-10-07 13:16:34 @schmangee The CEO who greenlighted and brags about a product that consistently kills people by driving them into the sides of trucks in Florida is the richest man alive.
2022-10-06 16:09:24 @EugeneVinitsky Stratego is underrated.
2022-10-06 16:08:59 @yisongyue come on, man.
2022-10-06 16:02:01 ETA: I realize this is likely a question more appropriate for reddit...
2022-10-06 16:01:33 It's been a few years since I've seen a big splashy story about AI playing video games (I think the last was AlphaStar). Anyone have any pointers to more recent developments?
2022-10-03 14:36:02 @bergerbell @ajlamesa But out to where?
2022-10-03 14:24:57 @AshleyRindsberg @g_shullenberger This is @g_shullenberger's beat, but how is this not just conspiracy theory 101?There are clear places where the US government throws money after horrible things with clear paper trails.The obsession with Peter Daszak by a niche part of conservative media muddles the cause.
2022-10-03 14:21:42 @wesyang @g_shullenberger @AshleyRindsberg The NIH gave 33 million dollars to these people at Duke to invent arbitrary technocratic metrics to keep schools closed or masked: https://t.co/Hw3ItFjxjN
2022-10-03 14:17:32 @g_shullenberger @AshleyRindsberg Gotta love the big conspiracy expose that starts with funneling "hundreds of thousands of dollars."
2022-10-03 13:32:57 @apsmunro We also need to reflect on why "stay home when sick" was not considered an ethical control arm.
2022-10-03 13:30:29 @DanielHadas2 @baeolophus1 @Dr_Tad Yes, and school closures were a comparable policy disaster to the Iraq War. But in both cases, bad ideas perpetrated by supposed opposite ends of our political spectrum.
2022-10-02 16:40:42 @g_shullenberger Don't forget "history is on our side."
2022-10-01 14:36:44 @tdverstynen @KordingLab On Twitter, everyone has an idea of what "public health consensus" was, except no one ever cites evidence for this consensus. Where is your evidence for schools being high transmission sites? And that the transmission there was "much higher than restaurants?"
2022-10-01 02:05:08 @mraginsky @ylecun @kamalikac @DimitrisPapail it's good, but I strongly disagree with the assertion that "If you don’t check arXiv for new papers every day, then you are really missing out."
2022-09-30 21:59:42 @Raamana_ @KordingLab Gregg Gonsalves
2022-09-30 21:56:58 @BenMazer Hmm, these studies were influential, but the Oregon and Bangladesh ones were both deeply flawed. That they changed no one's mind and were used as flimsy evidence for policy makes policy making look more capricious, motivated, and arbitrary. Not a great argument for more RCTs!
2022-09-30 18:00:31 @mraginsky @DimitrisPapail @kamalikac You apparently have not dipped your toes deep enough into STOC/FOCS peer review...
2022-09-30 17:49:07 @BenMazer In your opinion, what are the great success stories of RCTs informing public policy?
2022-09-30 17:47:35 @mraginsky @DimitrisPapail @kamalikac I don't think that STOC and FOCS are particularly great conferences. They are certainly not ones I've ever been excited to attend. @DimitrisPapail also didn't mention them in his list.
2022-09-30 17:17:24 @DimitrisPapail @mraginsky @kamalikac Interesting that you and @mraginsky give different answers than what I would say:Only in CS are conference pubs counted as journal pubs. In all of the conferences you like, a pub/presentation does not count as "a dean-approved credit" on your CV.
2022-09-30 16:55:06 @DimitrisPapail @mraginsky @kamalikac Sure, but why? These are all effects, not causes.
2022-09-30 16:49:29 @DimitrisPapail @mraginsky @kamalikac ITA is great. As are the SIAM conferences. But something salient differentiates them from the ML conferences, no?
2022-09-30 16:06:11 @KordingLab I'm sorry, this call for civility lacks introspection.The tweeter spent much of his time advocating for harmful policies like school closures in the name of THE SCIENCE.The person he's tweeting about spent most of his time on Twitter harassing women and spitting vitriol.
2022-09-30 15:34:04 @DimitrisPapail Comrade, you have nothing to lose but your chains.
2022-09-30 13:53:09 @ShaleenDeep @kamalikac @DimitrisPapail The contradictions in your tweet are revealing.
2022-09-30 13:20:12 @kamalikac @DimitrisPapail It's very industry heavy, has an unclear scope, has a perplexing review process. And I have had multiple people tell me that they have no idea how to evaluate ML candidates who apply with 17 ICLR papers.
2022-09-30 13:19:19 @kamalikac @DimitrisPapail ICLR is an interesting case for us. It came out of nowhere, and then fueled by Deep Learning hype and industry money, ML people decided we should consider it equivalent to all of the other conferences. But why?
2022-09-30 13:04:51 @DimitrisPapail Interestingly, I checked the webpage, and there is no rule saying you must submit to ICLR.
2022-09-29 16:31:52 We were promised self-driving cars and cures for cancer. Instead, we got automatic clip art generation. When do we stop hyperventilating about the imagined danger?https://t.co/5JCHy0FGPE
2022-09-29 14:16:23 @edsbs https://t.co/xsTsTrKk4A
2022-09-28 23:10:47 @tetraduzione @AlexGDimakis @thegautamkamath Curious what you think.There are few things I've added to papers that have annoyed more people.
2022-09-28 13:57:01 Surprisingly milquetoast!I miss the days when Pitchfork give a 9.4 to an esoteric noise band and the review would be lamentations on the author's last breakup. https://t.co/o3EJXPxekC
2022-09-27 20:49:46 @jaycaspiankang Also adding:Pla DaekBird and BuffaloTeni East Kitchen
2022-09-27 20:46:03 @jaycaspiankang Oh man, when are we going to Steve's?Have you tried the hole-in-the-wall ramen shack next door?
2022-09-27 14:06:53 @jaycaspiankang Considering that I deleted several draft replies to this for 10 minutes, I'm fairly sure this isn't freedom.
2022-09-26 16:04:58 RT @bwhitman: Ever dream of 100 synthesizers in a room, each with its own speaker, communicating together and individually addressable?Ple…
2022-09-26 15:58:15 @kamalikac @ajlamesa @YIMBY_Princeton Speakers are allowed to remove their masks in a lot of places! It's my favorite exception. Allowed to remove your mask to drink water or speak.Anyway, Kamilika, sorry UCSD is being stupid.
2022-09-26 14:57:20 @kamalikac @ajlamesa @YIMBY_Princeton Are speakers allowed to remove their masks? (This is my favorite exception to masking policies, as that's the most likely time to expel virus.)
2022-09-26 14:37:10 @ajlamesa @YIMBY_Princeton Even this fall, as @YIMBY_Princeton points out it's only in classrooms! Which is *bizarre*. Why classrooms?I'll never forget when activists at Berkeley claimed there was "no evidence that transmission was not happening in classrooms." Pure paranoia.
2022-09-26 14:36:04 @YIMBY_Princeton @ajlamesa Depends on the country. Spain and Italy were pretty strict about masking, even in University settings. UK considerably less so. Scandinavia: none at all.
2022-09-26 14:26:32 @YIMBY_Princeton @ajlamesa There is a huge contingent of covid maximalists on Twitter, and they are remarkably consistent about amplifying absurdity. Offline, their influence is smaller. But my son tells me half of his 4th grade class still wears N95s, so they are real.
2022-09-26 14:15:24 @YIMBY_Princeton @ajlamesa I don't understand these rules at all, but I think it's because the most neurotic and vocal about covid safety are old professors and librarians who seem to fear students.
2022-09-25 23:40:03 @matthewshaw1111 @contrarian4data Carbs are delicious.
2022-09-25 23:39:29 @contrarian4data For the first part of the bio: "It is difficult to get a man to understand something, when his salary depends upon his not understanding it."For the 2nd part of the bio: apparently this person is drawn to tribal fads and brags about it.
2022-09-25 23:34:02 @contrarian4data https://t.co/kNueZdyuEv
2022-09-25 14:06:32 @jaycaspiankang I agree, but why does it feel like everyone is now phoning it in?
2022-09-25 14:01:19 @jaycaspiankang Maybe we just ran out of things to be outraged about? Is that focusing on the positive?
2022-09-24 23:02:59 @dnunan79 The bailey part of the argument is where people try to use this to defend observational methods with matched controls. But *those* studies are almost always data mining inferior to RCTs.
2022-09-24 23:00:46 @dnunan79 So you cannot really refute the argument, but you have to explain why in many situations RCTs are the best thing we have. There is a narrow, but incredibly important scope for RCTs where other evidence won't cut it.
2022-09-24 22:59:57 @dnunan79 It's a motte and bailey, so tricky to refute.1. It's certainly true that RCTs aren't the only type of evidence.2. It's also true that RCTs are not only way to safety test mission critical systems.3. It's also true that many medical breakthroughs happen without RCTs.
2022-09-24 14:21:46 @walidgellad @apsmunro Understood. Though your point about EHR mining being useless without chart review is worthy of consistent repetition.
2022-09-24 14:17:46 @apsmunro @walidgellad Also, aren't these cohorts quite different? If you are scouring EHRs and you see the ICD-10 code for covid, is that just a positive test? That is, could the covid cases be more likely to be incidental as opposed to those coded as "other respiratory infection?"
2022-09-23 16:27:24 @moultano That's way too deliberate and clever to be AI generated. Don't get me wrong, I love the sorts of things that are generated with DALL-E-2 and Midjourney, but none of that work would be considered clever.Correct me if I'm wrong with a good example or two.
2022-09-22 18:59:26 @maestro_rayo @DanielHadas2 @casertron3000 Honestly, I remember Lipsitch being far more vocal at the beginning of the pandemic. I wonder why he decided to stop engaging as much.
2022-09-22 18:42:32 @DanielHadas2 @casertron3000 @maestro_rayo I've been looking for the article where Hanage downplayed the possibility of a 2nd wave. I hadn't remembered it being this early.
2022-09-21 13:34:29 @DanielHadas2 Perhaps a more appropriate metaphor is not a god but a golem.https://t.co/C0DsYOWE5M
2022-09-20 15:44:30 @KordingLab beyond false.
2022-09-19 13:30:54 @ajlamesa I'd bet most of these places no longer offer plastic straws.
2022-09-18 22:28:55 @FO_ASchatz Are the Broncos terrible?
2022-09-18 19:28:27 @WesPegden @Jabaluck @VPrasadMDMPH To underscore this point: In the Pfizer vaccine trial, there were 8 symptomatic positives in treatment and 162 in control. Even if there was a 2x underreporting bias in treatment (there wasn't), the efficacy would still have been over 90%.
2022-09-17 18:45:20 @jaycaspiankang But what about your newly adopted hometown Cal Bears? https://t.co/XeUV58sH6B
2022-09-09 01:58:31 @CYirush @feelsdesperate I dunno, CAP spends much of his time here talking about how academia is a toxic mess of egomaniacs and virtue signalers. How should we pick and choose which professors are legit?
2022-09-07 20:32:37 @feelsdesperate I think china is more responsible for the murk than our institutions, but our team hasn’t helped either.
2022-09-07 20:21:20 @feelsdesperate Again, because I know it’s your thing, I like the lab leak controversy as it’s a purely hyperreal scientific debate. There is no way to uncover new evidence, so we just get likes and retweets.
2022-09-07 20:19:56 @feelsdesperate Just like people’s awareness of the location of the WIV, the bay locality claims seem to be BS: https://t.co/ozVjjOkclr
2022-09-07 20:16:27 @feelsdesperate I’ve seen plenty of arguments for “no” on both questions. The lab leak stuff is funny because the scientists on both sides are such weirdos on Twitter.
2022-09-07 20:05:02 @feelsdesperate Curious, and I know you love to poast, but why do you think that person is credible?
2022-09-06 13:59:52 @KevinBardosh Right, the paper reads much more like a position paper on observational methods. If NPIs really worked, you'd expect the signal to stare you in the face at this point. That we are left quibbling over hacky stats methods that tease out illusions of signals is informative.
2022-09-06 13:24:53 @james_e_b_ You had a pithy tweet about this a while back, essentially saying that anyone can look at the data and see that NPIs failed to control covid so there is no need for fancy counterfactual modeling.
2022-09-06 13:14:38 @KevinBardosh My main concern is that meta-analysis is inappropriate when the papers reviewed are almost all observational studies. (to oversimplify: garbage-in, garbage-out)
2022-09-06 13:11:42 @james_e_b_ Have to be careful with those numbers: this is a meta-analysis which takes other (bad) papers and then, using cherry picking, comes up with some aggregated estimate of impact. But if all of the papers put in the model are garbage, then the estimates you get out are...
2022-09-03 19:37:03 @The_OtherET 4.4 yards per pass attempt is what we all want from Big 10 football.
2022-09-03 19:35:25 Though I also fondly recall being in the stands for this one where when Montee Ball and James White ran for 300 yards. https://t.co/rd3bWnoNrv
2022-09-03 19:30:20 This box score is giving me so much midwest nostalgia... https://t.co/40r8HQDesd
2022-09-03 14:16:24 @JacobAShell @the_midwits @feelsdesperate It's complicated, as part is also pandering to college kids in our new role as adult day cares, but this is interpretation makes a lot of sense. Especially seen from the sciences, which went all in on DEI as anywhere else on campus.
2022-09-02 18:10:54 @LizHighleyman @Neurofourier I dunno, he's guessing as much as everyone else is.
2022-09-02 14:24:07 @zeynep @walidgellad @k_stephensonMD For better or for worse, they have a history of formulating advice to "minimize confusion" without evidence that their advice actually minimizes confusion. They express similar thinking behind the recommendation of flu shots for all in 2010: https://t.co/3n3PI7y4dI
2022-09-01 19:22:16 @contrarian4data What makes you think Will Stancil's IQ is over 50?
2022-09-01 19:00:44 @JacobAShell @keefer1369 @Clarksterh @willyl0wman @feelsdesperate @daily_barbarian Weird, no? They found SARS2 progenitors in Laotian bats, but there's no evidence these ever passed to humans. It's humbling how little we understand about viral pathogenesis.https://t.co/Wsz4PqmPH1
2022-09-01 18:26:11 @keefer1369 @JacobAShell @Clarksterh @willyl0wman @feelsdesperate @daily_barbarian But we don't even have the bat predecessor for SARS1...
2022-09-01 16:15:02 @feelsdesperate @JacobAShell @keefer1369 @willyl0wman @daily_barbarian That's right. It's only part of the puzzle.
2022-09-01 16:04:35 @keefer1369 @JacobAShell @willyl0wman @feelsdesperate @daily_barbarian I think you're both right here: As @keefer1369 says, the correlation between density and the initial spread is low enough to reject the null hypothesis that population density explains the spread. But as @JacobAShell, the way the evidence is presented is confusing.
2022-09-01 15:32:35 @JacobAShell @keefer1369 @willyl0wman @feelsdesperate @daily_barbarian Very interesting. Do you have a means to overlay these two maps? My GIS data wrangling skills are terrible: I tried to overlay the images you posted and can't see the right features to align them.
2022-09-01 15:14:10 @JacobAShell @keefer1369 @willyl0wman @feelsdesperate @daily_barbarian I have seen others argue that the market is neither the most crowded market nor in the most dense/busy part of the city. I don't know how to verify these claims about human activity in Wuhan.
2022-09-01 14:51:33 @willyl0wman @feelsdesperate @JacobAShell @daily_barbarian You: assert something verifiably incorrectMe: "sorry, that's wrong"You: aha, this just proves my point.
2022-09-01 14:49:25 @JacobAShell @willyl0wman @feelsdesperate @daily_barbarian I don't disagree with you, but the idea that it's a 10 minute walk from the lab to the market is simply wrong.
2022-09-01 14:42:35 @willyl0wman @feelsdesperate @JacobAShell @daily_barbarian This is wrong. It's 10 miles away and on the other side of the Yangtze river. https://t.co/q78VBCU5Jk
2022-09-01 14:27:28 @feelsdesperate @JacobAShell @daily_barbarian Hah, I 100% endorse this reaction!
2022-09-01 14:26:23 @feelsdesperate @JacobAShell @daily_barbarian Yes, of course it could. But what's your evidence for an earlier leak? There is none.The evidence for the zoonotic spillover isn't great, but it's probably as good as what we have for "Covid is airborne."The evidence for the lab leak is entirely speculative.
2022-09-01 14:13:37 @JacobAShell @feelsdesperate Yup.
2022-09-01 14:13:05 @JacobAShell @feelsdesperate @daily_barbarian Knowing how science establishment works, there is no way to disprove that scientists have been selectively publishing papers that suggest a zoonotic origin. The only angle of the lab leak that persuades me is the structural angle. 2/2
2022-09-01 14:11:06 @JacobAShell @feelsdesperate @daily_barbarian The scientific evidence for zoonosis is stronger, but one could dismiss this as "virologists circled the wagons and decided to only publish papers in favor of zoonosis." 1/2
2022-09-01 14:07:25 @feelsdesperate @JacobAShell This email exchange actually casts doubt on the lab leak theory. They thought it was suspicious, looked more, and then decided their suspicions were wrong. That's what's supposed to happen!
2022-09-01 14:04:45 @feelsdesperate @JacobAShell The anti-establishment folks are too credulous to the lab-leak folks. Their evidence is weak.But I am fascinated by how conspiracy theories become conspiracies rather than conventional wisdom, which is why @daily_barbarian is the perfect person to cover this story.
2022-08-28 14:58:01 @schmangee He still has surprising influence. One of my colleagues just sent Wachter's tweets to our faculty mailing list, using them as evidence for why we shouldn't return to in-person meetings.
2022-08-28 14:45:26 @schmangee It remains distressing how much influence this clearly neurotic hypochondriac has over Bay Area covid policy. The SF Chronicle has him on speed dial.Appeals to authority have been both damaging and deeply revealing throughout this crisis.
2022-08-28 14:11:19 @feelsdesperate @DanielHadas2 Listen, if you two had just masked up, millions would have been saved.
2022-08-27 20:33:16 @daily_barbarian I made this graph using @benmschmidt's interactive data analysis tool. https://t.co/UnK3DaGvJ9
2022-08-27 20:32:45 @daily_barbarian It's not remotely close! https://t.co/O3AFedLLvN
2022-08-27 20:02:46 @james_e_b_ yes, it needs to be a good, dry curacao or the cocktail will be too cloying. I use Pierre Ferrand.
2022-08-27 19:57:03 @james_e_b_ Something really bonkers: equal parts smoky scotch, green chartreuse, and campari.Negroni variations are my favorite.
2022-08-27 19:56:22 @james_e_b_ A few ideas:1) sub Punt E Mes for the vermouth. More bitter and warming.2) sub dry curacao for the vermouth. More citrus and refreshing.3) here's a full variation: do equal parts rye, cynar, and cardamaro.
2022-08-27 16:18:39 @james_e_b_ Do you want a different liquor to replace the vermouth? Or are you looking for a cocktail that's only an amaro + gin?
2022-08-27 14:36:43 @bergerbell @PunditPandemic In the hyperreal, we're all anonymous accounts.
2022-08-27 14:24:29 A great thread breaking down the unfathomable prices charged by "prestige" "open access" journals. I will never understand why the scientific community abides such egregious rent seeking. https://t.co/vzfD655VHI
2022-08-27 14:20:33 @venkmurthy The solution has been there all along: just get rid of the f@#%$*% fees!The main journal in machine learning, Journal of Machine Learning Research, has no publication fees and is 100% open access.
2022-08-27 14:18:18 @ifihadastick I mean, you just have to go to the next tweet in his timeline: https://t.co/yGTg2Nzqzr
2022-08-27 14:10:00 @DanielHadas2 @james_e_b_ Relevant passage: https://t.co/ODRVNVjG87
2022-08-27 14:09:19 @DanielHadas2 @james_e_b_ This is a great article (with references) that is right up Daniel's alley: https://t.co/kHaQVdan1B
2022-08-27 14:06:01 @DanielHadas2 @james_e_b_ It's also not at all clear that the vaccines in 1957 were effective or if we were just seeing geographic disparity in flu burden.(Aside: James and I seem to be the only people interested in the history of the '57 pandemic.)
2022-08-25 17:28:20 @KameronDHarris I'd be all in favor of all of academia moving to an arxiv overlay journal model. This is the way.
2022-08-25 17:02:20 Important points about the new executive order on open access. It is absurd that when I author a paper in a biomedical journal, I am charged thousands of dollars for it to be open access. When I author a paper in a machine learning journal, it is *free*. https://t.co/1L4qbgnenH
2022-08-25 16:20:13 Cool thread by @tomgoldsteincs relating diffusion models used in DALLE2 to image denoising.It left me wondering: Has anyone made a similar diffusion model for audio yet?https://t.co/Q2KNbDVqts
2022-08-25 13:49:19 @politicalmath Unsurprisingly, it was by @AlecMacGillis https://t.co/HBfrdcRHmZ
2022-08-25 13:04:41 @DanielHadas2 As you point out, even if we could do it, the ethical problem arises and we're stuck.While it is counterintuitive, I maintain that pandemic mitigation does not have a technocratic solution. Rather, it is a stress test of societal values, and few came out looking good.
2022-08-25 12:54:22 @DanielHadas2 As much as I respect the evidence-based medicine crowd, their reliance on this one crude epistemological tool is an Achilles heel.
2022-08-25 12:53:19 @DanielHadas2 Unfortunately, you can do this with RCTs either. RCTs require repeatability: the treatment must be similar in the experiment and in the field. Running an RCT on an novel pathogen provides low quality evidence.
2022-08-25 12:48:14 @schmangee 22000 likes!
2022-08-25 12:46:37 @lizzywol While I am also annoyed at this salary, there are many other useless administrators here who make a comparable wage.And our campus is in a structural deficit for renovating its football stadium (they took on $500M in debt).The fiscal problems of public higher ed run deep.
2022-08-24 16:06:46 For the departments, it's not like just because we teach more students that somehow we get unlimited resources from campus. (I was sold on the Field of Dreams concept of department growth for a decade here. It didn't happen.)
2022-08-24 16:05:48 For the students: If you, for example, come to Cal and all your CS classes have over 500 students in them, is that more valuable than one of those accelerated $10K coding bootcamps?
2022-08-24 16:04:46 It is not clear to me that this trend, which is based on calculated economic decisions, is in the mutual interest of students OR CS departments.
2022-08-24 16:03:40 Using @benmschmidt's interactive exploration, I made this plot. CS is growing faster than all of the other STEM fields. Biology is growing pretty rapidly too. But it's not like all of STEM is exploding. https://t.co/7KdGcxSNhx
2022-08-24 16:02:08 This image is part of a longer, fascinating thread by @benmschmidt. But diving into the sciences also reveals some striking trends. https://t.co/bLz4eRdFcu
2022-08-24 15:48:12 @benmschmidt Oh wow, this looks amazing. Thanks!
2022-08-24 15:45:46 @benmschmidt In my department, we jointly teach EE and CS, but our EE classes have not seen the same exponential growth in enrollment as the CS ones.
2022-08-24 15:45:02 @benmschmidt One follow up: could you break out the science majors you are tracking (including CS) and make a similar plot to this one?I'd be really curious to see how the interest in the different STEM disciplines are fluctuating over time. https://t.co/AJ971uNyFp
2022-08-24 15:43:33 @benmschmidt Oh yes. I'm of the opinion that if you have to include "science" in the name of your program, you definitely doth protest too much. I was curious if the CS curve was overwhelming the other STEM curve, or if they were in fact growing in parallel. Seems like the latter!
2022-08-24 13:38:26 @benmschmidt What does this graph look like if you remove CS from the "science" curve? From this particular plot, it looks like the growth in STEM is driven almost entirely by growth in CS.
2022-08-23 13:54:23 @politicalmath @reason @mkhammer Same experience here in Berkeley. Endless bait and switch.
2022-08-19 22:59:31 @CBoy287 100% agree.Even in the 65+ I would have hoped to see an odds ratio larger than 2. The OR is only 1.5, and that is again not accounting for confounding.
2022-08-19 22:50:51 @CBoy287 Thanks!
2022-08-19 22:45:59 @CBoy287 Would you mind posting a link to where they post this data? Thanks!
2022-08-19 16:04:37 @zeynep Just an FYI: Twitter has admitted that they are prioritizing labeling anything that might "could lead to increased exposure or transmission." @KelleyKga managed to actually get a reply from their tech support.https://t.co/IJrWwWcBsY
2022-08-18 15:35:11 @RufusSG @BallouxFrancois @MmeBlackBalloon @raefneerg @andyaschmidt @boynamedshannon @NJ_Mama_Bear True, but with effect sizes of such absurdly large magnitudes, you don't need very large sample sizes.
2022-08-18 13:57:56 @MAbsoud my bad... the thread linked to two lancet papers, and I had clicked on the wrong one.
2022-08-18 13:48:49 @MAbsoud From where did you extract this figure? I couldn't find it in the paper nor in the supplementary appendix.
2022-08-17 15:23:42 @hoofnagle Yup. My son has been going to an aftercare program at the Willard Clubhouse for 5 years now. There's a blatant anti-child sentiment to the opposition of expansion of these facilities.
2022-08-17 13:26:42 @hoofnagle Be explicit, they oppose space that more *children* can use. The movement against Willard Park is *explicitly* anti-children. We've been to some of the public meetings on this space, and the attendees are predominantly retirees who want fewer kids in the park. Super gross.
2022-08-16 18:34:03 @DanielHadas2 @montaigne01 @Brkr_Morant @ladoodles @KelleyKga @BallouxFrancois @politicalmath He says in that article "they've returned to normal in the UK," but can't quite make the final jump to "we can do it here too." He'll get there though.
2022-08-16 18:20:05 @DanielHadas2 @montaigne01 @Brkr_Morant @ladoodles @KelleyKga @BallouxFrancois @politicalmath J G Allen wasn't quite as extreme as Wen, but he's also shifted pretty significantly.
2022-08-16 14:35:41 @jaycaspiankang @bubblesdepot @freedarko They're already all over Liz Phair.
2022-08-16 14:35:05 @folk_future @jaycaspiankang They already found Modest Mouse.https://t.co/9TyA4G6TBY
2022-08-16 13:51:26 @VPrasadMDMPH It's being shared by the same people who shared the insane mask study including data from sars1. Pushing bad science is part of the agenda.
2022-08-16 13:22:24 @daily_barbarian The lab leak is my favorite example of how there is no real debate in covid land, just dichotomized team sports.The covidians are too dismissive, but the covidiots are far too credulous. It's all just evidence-free yelling.
2022-08-16 02:23:50 @DanielHadas2 @KelleyKga @BallouxFrancois @politicalmath But if no one talked about covid, would it even exist?
2022-08-16 00:40:50 @KelleyKga @DanielHadas2 @BallouxFrancois @politicalmath I'd respect them if they just decided that you could no longer discuss covid at all on here. At least that would be a principled stand, and I have no doubt the world would be better off.
2022-08-16 00:36:59 @KelleyKga @DanielHadas2 @BallouxFrancois @politicalmath I still can't get over that "anything to reduce spread" email they sent you. How do they evaluate that? What on earth could that even mean?
2022-08-16 00:15:29 @anish_koka the responses are all from the model!
2022-08-16 00:10:28 RT @stochastician: "Zero to nazi in two lines" -- I give you Foundational Models.
2022-08-16 00:10:22 @TAH_Sci I'm guessing covid twitter?
2022-08-16 00:06:09 I mean, I know they always say "read the fine print," but in this case they really aren't joking. https://t.co/KFmOIMig8r
2022-08-16 00:05:18 A cordial first interaction with Meta's GPT-3 clone. https://t.co/XlTHFdBEfy
2022-08-15 21:12:15 @jaycaspiankang @StephMBurdick the gavel has been banged.
2022-08-15 21:06:43 @jaycaspiankang @StephMBurdick But I like yelling.
2022-08-15 16:18:09 RT @bwhitman: I wrote a little about the best company I ever knew. Our dear Canopy lives on in surprising ways. ps: more like this soon! …
2022-08-15 13:22:10 @daily_barbarian Love dudes with 200K Twitter followers talking about how other's "only real communities are online."https://t.co/PHyiZj2hpq
2022-08-14 23:14:51 @DanielHadas2 @feelsdesperate @bergerbell After the Iraq war, there was no change in our punditry.After Jon Stewart "eviscerated" Tucker, Tucker just moved to Fox News and is now more influential than Stewart.This systemic dysfunction goes way beyond lockdown, and I'd love to read you expanding on this theme.
2022-08-14 23:06:20 @DanielHadas2 @feelsdesperate @bergerbell This is true. I had dinner with friends who had no idea about the age gradient of covid deaths. And I don't blame them for not knowing! Their businesses were crushed by lockdowns, and they're just now getting stable income again.
2022-08-14 22:38:52 @DanielHadas2 @feelsdesperate I think that the endless dichotomization that @bergerbell talks about is also what makes this so complicated. Sunetra Gupta, John Ioannidis, and Alex Berenson agree, but normies only see the cuckoo guy who thinks marijuana causes schizophrenia and hence dismiss the argument.
2022-08-14 14:28:59 @feelsdesperate Given her academic webpage, I think Ellie is quite self aware. She is trying to be a "science explainer" through gif littered articulations of the most facile arguments conceivable.
2022-08-14 14:25:30 @DanielHadas2 Great thread. @phl43 makes a similar point about testing:https://t.co/GRZnjDIGndThe whole NPI artifice is built on ethically dubious assumptions.
2022-08-14 14:18:47 @feelsdesperate Yup. Ellie is my favorite example of the genre as she can say "We didn't have schools in the middle ages, so what's the big deal of closing them now?" and suffer zero repercussions.
2022-08-14 14:14:05 @anish_koka @DrJBhattacharya @TracyBethHoeg @Accad_Koka Who let this guy back on Twitter?
2022-08-14 14:09:49 @phl43 This is all correct, but how many infections have actually been even delayed by testing? Testing interventions have less evidence supporting them than face masks.
2022-08-14 14:06:03 @feelsdesperate It's wild that your obviously correct take is posted from an ironic anonymous shitposting account but the aggressively stupid QT is from a real person's profession account. I love the discourse.
2022-08-11 22:06:18 @KelleyKga I'm amazed you actually got them to admit this in writing. Well played!
2022-08-11 21:51:22 @BenMazer Do we have any evidence for the value of N95s as source control?
2022-08-09 15:31:57 @samhaselby I just don't see how these are more impressive than the recovery act, Dodd Frank, and the ACA. Sure, he had 59 senators, but what's the difference between 59 and 50 when there is 100% polarization?
2022-08-09 15:19:03 @samhaselby Honest question here: What has he accomplished legislatively that impresses you?
2022-08-08 15:17:37 @politicalmath Did you see this? https://t.co/LTVLLK7lC0Twitter covid censors are out en force this week.
2022-08-07 14:24:54 @NICU_doc_salone @sdbaral @MmeBlackBalloon @colourmeamused_ @CarloDallapicc1 @jwilcox79 @LizHighleyman @dr_imogen Just wanted to add another small note of support here. I definitely appreciate your perspectives.I recall that you shared the BMJ article on the evidence for aerosol transmission, and I was shocked by how poor the evidence base was.
2022-08-07 00:38:49 @AliceFromQueens My mistake to engage.I don't understand the staying power of this conspiracy theory, but fighting about whether 911 was an inside job is equally as fruitful.
2022-08-07 00:37:44 @AliceFromQueens You can nitpick words, but the interview is him ranting accusing people of taking actions without evidence. This is passage demonstrably inaccurate and reveals Sachs to be engaged in a pissing contest. https://t.co/lK4HR1zbFi
2022-08-07 00:28:08 @AliceFromQueens you ended your tweet with "prestigious medical journal, The Lancet," hence arguing to authority rather than the substance of the interview, which was a bunch of hearsay and gut feelings.
2022-08-07 00:26:53 @AliceFromQueens Why does that matter? It's all a bunch of big ego dweebs picking petty fights and seeking the approval of those who dole out the research dollars, be they philanthropists or agency bureaucrats.
2022-08-07 00:19:27 @AliceFromQueens Let's also be clear that the Lancet has been known to publish rather dodgy work with some shady lack of COI declaration. https://t.co/HoQAvo2EER
2022-08-06 17:36:17 @glastris @jaycaspiankang but shouldn't address *why* all of these jobs are going to students with bachelor's degrees and not those with high school only? if it's only blind credentialism that leads to better outcomes, we should address this irrational faith in credentialism.
2022-08-06 16:17:25 @thackerpd @DrJBhattacharya @VPrasadMDMPH Respectfully, Paul, what they say is true.
2022-08-06 15:55:12 @DrJBhattacharya @VPrasadMDMPH I'd agree, Jay. We shouldn't be blaming the paper which is a fine case study. We should be blaming the Twitterati who love to retweet papers without reading them.
2022-08-06 15:21:13 @stephmurrayyyy @schmangee Iannattone practices in Montreal, though. And there is a large contingent of loud, Canadian medical-adjacent folks on Twitter who are surprisingly obsessed with permanent pandemic theater. Not sure what explains that.
2022-08-06 13:25:22 @VPrasadMDMPH @zchagla I do think it would still be valuable to show that classrooms are not driving transmission of disease on campus. But for that, they would have needed a control group.
2022-08-06 13:21:02 @Saikmedi @AbraarKaran A control group would have been invaluable in this study as it is pretty clear that transmission does not happen inside of classrooms no matter what you do.
2022-08-05 14:04:11 @DanielHadas2 Oh yeah. And it's concentrated. Covid monomania still reigns supreme in Berkeley.This week, we had a major protest over a stupid issue that dates back 50 years. All of the kids who broke construction equipment and fences in an outdoor park wore some sort of covid mask.
2022-08-05 14:02:31 @DanielHadas2 https://t.co/mX6rPoGxdEHuh, turns out this group is vanishingly small, yet got disproportionate coverage.
2022-08-05 14:01:51 @DanielHadas2 https://t.co/rGjgqlYn7zOnly one in five parents of children under 5 plan to vaccinate a child right away.Whoops, ended up being well under 1 in 25.
2022-08-05 14:01:13 @DanielHadas2 That the New York Times has not figured out at they have vastly overestimated the size of their base shocks me.Their articles seem to think covid monomania is a mainstream position, but are always off by nearly an order of magnitude in their estimates.
2022-08-05 13:56:50 @DanielHadas2 That editorial was maddening, and I agree that it's because it is written for the NY Times readers who are still paralyzed by covid terror. It's amazing that he commends the UK, who is doing none of this bullshit, and still recommends nonsense like masking post infection.
2022-07-26 20:25:11 @samhaselby I get the 2nd part, but what's libertarian about nudge theory?
2022-07-25 15:30:27 @daily_barbarian this is undeniably the only correct take.
2022-07-25 13:56:39 @politicalmath Fair enough. As I said, I look forward to reading your substack on this.
2022-07-25 13:48:13 @politicalmath Sure, but the point stands. Are you sure this wasn't a similar "crisis" to the "ventilator shortage?"I understand the second guessing of sclerotic regulatory systems, but they are too easy to blame as they have vast, grave responsibilities.
2022-07-25 13:38:05 @politicalmath I'm sure you'll write more about it, but I'd be curious to know which steps you think would have actually mattered. What would 160K masks/day have achieved?In my assessment, the only correctable mistake that led to excess deaths was the delayed vaccine roll out.
2022-07-25 13:27:59 @KordingLab https://t.co/Md6chZ0RiJ
2022-07-24 13:36:49 @anish_koka I do enjoy that both sides are certain they are right.
2022-07-23 17:05:28 @BallouxFrancois @DrJBhattacharya @MartinKulldorff But much closer to 1957 or 1968 than 1918, no? This seems to be not widely acknowledged.
2022-07-23 01:28:08 @anish_koka the science is quite clear!
2022-07-22 14:59:56 @mgtmccartney @dnunan79 @IBonnx @TomPMarshall I've been recently doing a lot of research on how we "test" software. I worry that if the EBM folks knew what we were up to in computer science, they'd all have heart attacks.
2022-07-22 14:49:53 @afternoondelete Even more recently, they weren't tracking the number of times you listened to downloaded mp3s in itunes.
2022-07-22 14:45:58 @afternoondelete Also this just reflects that we can now easily track what people listen to, no?
2022-07-22 14:44:01 @dnunan79 @mgtmccartney @IBonnx @TomPMarshall Hmm, I don't agree with that last part. Bridges operate in open, uncertain systems with lots of variability, and that's why they sometimes collapse.The reasons RCTs are not necessary in engineering are hard to summarize in a tweet.
2022-07-21 14:03:19 @adamcifu The main text has passages that should give us pause as well. https://t.co/emszahesvU
2022-07-21 13:28:55 @zeynep Have you ever played with Twitter's reporting tool? It's ridiculous.https://t.co/rpKl5LamAw
2022-07-21 12:57:02 @phl43 Yes. Something like:-the null is true.-run 1000 experiments but file-drawer most those that don't reject the null.-published results are ~20 p<
2022-07-21 02:51:50 @Saikmedi @anish_koka He quotes Michael Mina as guessing 300K/yearHe quotes Céline Gounder as estimating steady state of 100-250K/year.These are all guesses from the usual suspects. The only actual numbers were in that block quote, and I don't know if they are correct.3/3
2022-07-21 02:50:14 @Saikmedi @anish_koka 2nd, there is this quote with no links to sources. https://t.co/Sadx8QqlZb
2022-07-21 02:49:34 @Saikmedi @anish_koka There are some unlinked claims in the article:Trevor Bedford says "we can expect that every year, around 50 percent of Americans will be infected and more than 100,000 will die." 1/3
2022-07-20 17:42:08 @JobbingLeftieH I am amusing myself trying to figure out how the flow is supposed to work in this flow chart. https://t.co/oQCZj53l7I
2022-07-20 17:40:49 @JobbingLeftieH Ugh, I read it, and this is exactly what the paper says! It's astounding. It's "abandon EBM principles if the mood suits you."Wild stuff.
2022-07-20 15:49:13 @james_e_b_ And not every school has only 18 students.https://t.co/89rqfGJeQD
2022-07-20 14:26:04 @phl43 I'd go a step further: I think you can take perfectly good RCTs and meta-analyze their findings into fakeness.
2022-07-20 14:18:24 @phl43 Someone really needs to write the "all meta-analyses are fake" paper.
2022-07-20 14:00:09 @TAH_Sci @kenkenkenken99 @apsmunro @GidMK @AndrewHayen @bencowling88 I wish our other interventions were equally contentious, because evidence is sorely lacking for testing and ventilation as well.Controlling infection is hard! And there is no magic medical bullet that can make up for pre-existing deficiencies in societal support.
2022-07-20 13:56:39 @GidMK @apsmunro @AndrewHayen @TAH_Sci Given that the effect can be entirely explained by observer ascertainment bias, that we have no other studies showing similar sized reductions, and that real world evidence screams otherwise, it is entirely reasonable to say the evidence of any measurable effect is very weak.
2022-07-19 18:35:53 @conorsen This plot is based on a single poll in each state. All four of those guys are within a few points of each other, so not sure there's much to take away.
2022-07-19 16:22:11 @AnilMakam @AngryCardio Why do we need to follow science aggregators? What value are they providing to our twitter feeds? For example, you could subscribe to a medrxiv RSS feed and get the same information without the BS Topol filter on top.
2022-07-19 13:55:52 @dr_dmorgan @eliowa Did you see the associated editorial that takes this review as vindication for the airborne theory and proclaims that it is "time for an indoor air revolution?" The discussion in the paper and editorial seem detached from the actual evidence.https://t.co/3n71xNM6MM
2022-07-18 14:25:11 @jaycaspiankang Harvard's endowment is $53B, and they have 5000 undergrads, so that's over $10M/enrolled student.Stanford's endowment is $38B and they have 6400 undergrads. $6M/enrolled student.I don't think these numbers include real estate holdings or return on IP.
2022-07-18 14:12:24 @jaycaspiankang How can there be 100 colleges in the US holding more than $500K in assets per enrolled student? Just insanity. https://t.co/EzE0qUeLqP
2022-07-18 14:09:00 @jaycaspiankang @JamesSharpsteen I like the idea of quadrupling Harvard, but I've also seen how hard it has been to grow UC Berkeley. The physical constraints are as stifling as the financial ones.Tax Harvard, improve community colleges is a much more feasible plan.
2022-07-17 16:48:12 "No automation, no AI, no machine learning" is a perfectly reasonable set of design principles.
2022-07-16 19:24:02 @AnilMakam PS Improving ventilation is also not "science based."
2022-07-16 19:23:23 @AnilMakam There is no evidence showing that N95 mandates reduce spread, and in fact plenty of evidence that the mandates don't do anything.There is no evidence that reducing spread in college settings is desirable. You can assert it is a good thing, but it's not "science."
2022-07-16 19:19:18 @AnilMakam Why indoor masking?
2022-07-16 18:33:41 @MattZeitlin @dariustahir @conorsen By a huge margin!!! This comparison is dividing by zero.
2022-07-16 18:31:29 @kylehauptman @conorsen The estimated dollar amount of remittances is ~600B dollars worldwide. Credit and debit payments total over 7T in the US alone. I'm not sure the remittance market justifies the massive crypto overhead.
2022-07-16 00:22:22 @AngryCardio @Twitter I tried to figure this out this morning. https://t.co/rpKl5LamAw
2022-07-15 15:47:09 @tsimonite This tweet from yesterday cites a fun bullish story on self driving from 2011. Ah, how far we've come...https://t.co/gIVH0VtQvB
2022-07-15 15:00:14 And even when you click on "election information," you are told that it's bad to spread misinformation about covid. https://t.co/L1FaSQ80LS
2022-07-15 14:57:59 An addendum! What happens when you click on politics instead of health? Well, now there's again a paucity of categories. https://t.co/fKOdQcsTa6
2022-07-15 14:38:46 In sum, Twitter has a system where it lets the userbase snipe on each other, and then, based on this opt-in reporting, makes some decision about what's true or not in a completely opaque fashion. Clearly, this is how we solve misinformation.
2022-07-15 14:31:14 If you hone in specifically on their covid policy, there you can find extra information about their ornate suspension policies.https://t.co/G6c1ccXUml
2022-07-15 14:29:48 At this point, you can report the tweet. @TwitterSafety provides no information for how these reports are read and verified, but has a long list of rules for what happens if "they" decide you're a misinformation peddler.https://t.co/HiCyLFDXYY
2022-07-15 14:27:03 @MarkGermaine Yes, Twitter is rolling out misinformation reporting country at a time. https://t.co/qmkBnT3ju6
2022-07-15 14:23:05 I wanted to report someone's cure for cancer with a keto diet. That's covid misinfo, right? So I clicked covid. Twitter then gives the list of things that they consider covid misinfo. https://t.co/4c2JZoomdK
2022-07-15 14:20:44 I clicked on health. Lo and behold, the only disease you can be misinformed about is... https://t.co/I2CN1v4wfE
2022-07-15 14:19:24 TIL there are only three kinds of misinformation out there, tweeps. https://t.co/mMMqn9DU6y
2022-07-15 14:18:09 Everyone on twitter is always being Spammed. Why is that an option? Any case, I will click "Shown misleading info" https://t.co/JcKqHfCy82
2022-07-15 14:16:54 Lets now go through the choose your own adventure and report a tweet. Since I'm reporting misinformation, clearly the only answer is "Everyone on Twitter" https://t.co/U53YkKGuuO
2022-07-15 14:15:40 The Twitter reporting of misinformation is sort of hilarious. Here are the stages. You click "report tweet" and are shown the following solemn announcement: https://t.co/Pme0s44WBh
2022-07-15 14:02:07 @mohomran Right, I know that individuals can report tweets as misleading. I'm wondering about the full process that results in a tweet itself gets explicitly tagged as misleading and rendered unable to be retweeted, liked, or replied.
2022-07-15 13:47:11 @samhaselby I love this answer not only because I think it's accurate, but also because it's an appeal to history rather than technology. Us each showing our disciplinary biases.
2022-07-15 13:44:07 @ethanCaballero Usually a safe guess! But I don't think they are passing every tweet through an LLM, do you?
2022-07-15 13:40:11 Does anyone know the process behind how a tweet gets labeled "Misleading" on here? https://t.co/eStv0a62od
2022-07-15 13:34:50 @samhaselby But it's crystal clear that Twitter has been a net negative for academia, no?
2022-07-15 03:18:55 @drvictoriafox @TheEliKlein @AstorAaron Alameda county managed to keep theirs for only three weeks. Might be the same duration (or less) in LA? But it's really bad that they are doing this so close to the start of the school year.
2022-07-15 02:41:47 @venkmurthy @rfsquared Yes. I wonder if it's related to the @birdwatch project or if it's some other team.More transparency would go a long way. But, you know, Twitter is busy imploding under the Musk takeover, so we're all on our own out here...
2022-07-14 00:04:48 @Feelsdesperate Why no option for "disappointed and confused."
2022-07-13 16:20:06 @BenMazer Obviously, they mean "pathologists."
2022-07-13 14:05:28 @Feelsdesperate @HamrickPaul At this point, don't the vignettes miss the forest for the trees?
2022-07-13 13:31:49 @hoofnagle Does one need to read that article to know the answer?
2022-07-12 15:05:35 @MattNoahSmith It's certainly harder to have that conversation than it should be.
2022-07-12 15:02:49 @MattNoahSmith Right, and science practice is itself inherently political.
2022-07-12 14:59:29 @MattNoahSmith Loudly criticizing our science bureaucracy is important, but such criticism must be prefaced by the assertion that one can't "follow science."
2022-07-11 16:49:49 @JSEllenberg If you predict image classes by homotopy, everything is a donut. https://t.co/A6v5oYdOJv
2022-07-11 16:25:16 @politicalmath I thought you were punching down and retweeting some 2 follower bot... but no! that account has over 50K followers.
2022-07-11 13:28:34 @BenMazer I'm sorry sir, but this is Twitter.
2022-07-11 02:12:39 @JSEllenberg https://t.co/wFAajAhaNQ
2022-07-11 01:41:37 @JSEllenberg I'd bet you Billie Joe drives a Tesla.
2022-07-10 20:16:19 @james_e_b_ @KSHimself Yeah, it's definitely still a thing where I live, but I (a) work at a university (b) in the San Francisco Bay Area. I don't see a world where that (a)+(b) combo abandons the permanent pandemic mindset.
2022-07-10 19:51:54 @CorieWhalen @robbysoave Robby's tweets are a revolutionary new personality test.
2022-07-10 19:49:35 @CorieWhalen @robbysoave You're also the person who always checks bags, right?
2022-07-10 14:09:04 @phl43 @itaisher Probably the most damning part, IMO, is that 8/25 of the studies in the post-reg period are women only, whereas only 1/30 of the trials pre-2000 were women only.
2022-07-10 14:06:49 @phl43 @itaisher I found the paper. It's odd how only 1 trial/year is eligible in the pre-reg period, but 2/year were eligible over the post-reg period, 2000-2012. Also weird how there are no trials published in the years between range 1996-2000.https://t.co/UNnUyjueY0
2022-07-10 13:38:19 @BenMazer You've made this point before, but the 3-shot covid vaccine series is the same price as a 4-pack of BinaxNow tests.I wonder how people would perceive these two interventions if the vaccines weren't free to all.
2022-07-10 03:13:45 @phl43 I bet you that the paper that figure was pulled from is also fake.
2022-07-10 01:15:23 @BenMazer just in case you think this is hyperbole, let me share some wonderful berkeley examples:https://t.co/yNvpuyPXMfhttps://t.co/j60N3TjQ4Uhttps://t.co/c20ktCP7vchtt...
2022-07-10 01:12:27 @BenMazer you mean a math professor?
2022-07-10 01:07:12 @sdbaral @VPrasadMDMPH Agree. It was fascinating (and scary) to watch the testing hegemony take hold despite a paucity of evidence.
2022-07-10 00:34:46 @sdbaral @VPrasadMDMPH Also double the current NIH budget!
2022-07-10 00:28:40 @sdbaral @VPrasadMDMPH Took me a second to find it, but Paul Romer was advocating spending over 100B/year on testing. That's double our foreign aid expenditures.https://t.co/ZXSl08naMf
2022-07-10 00:25:18 @sdbaral @VPrasadMDMPH Do you have back of the envelope calculations for that? Some economist out there must have computed an estimate of the cost of pandemic testing.
2022-07-08 02:46:43 @ajlamesa @jwilcox79 I also wish people would (a) stop equating vaccination with masking and (b) stop thinking that N95s are intended for source control.But the CDC is to blame for both.
2022-07-08 02:44:30 @ajlamesa @jwilcox79 I just meant that given the massive hand wringing in that post, it reads like he would have postponed the flight home regardless of reimbursement.
2022-07-08 02:38:59 @ajlamesa @jwilcox79 Did you read the whole piece, Anthony? He's unfortunately promotes a lot of incorrect ideas and then links to the CDC as "evidence" they are true.
2022-07-06 17:06:22 @_ericting I wish some new talking head was destined to overtake Wachter as the bay area's dominant covid pundit.
2022-07-06 14:53:08 @KordingLab I thought the only important endpoint in life was Harvard itself.
2022-07-06 14:40:22 @KordingLab Which causal effect of Harvard?
2022-07-06 14:33:19 @KordingLab Exactly. Unless you want your kids to get into Harvard, and then there are boxes to check. But there's a whole lot of cargo-culting around getting into Harvard, even though Harvard makes up stupid rules to justify it's unjustifiable class compositions.
2022-07-06 14:30:56 @KordingLab Sure, but you can't download it. I'm sure the endpoints are ridiculous.In any event, my point was that an RCT is not going to teach you how to be a better parent.
2022-07-06 14:26:47 @KordingLab No, I'm just criticizing you for tweeting papers without reading them, especially ones that are not accessible (the first one does not appear to be available using my UC Library account).Parenting is hard, and most RCTs are not going to inform you on best parenting practices.
2022-07-06 14:25:15 @VPrasadMDMPH Did you find a copy of the British RCT to read? It wasn't available using the UC library.I ask as you'll frequently find some rather sketchy endpoints in education research. We have to read these RCTs with the same critical eye as any clinical trial.
2022-07-06 14:22:56 @KordingLab Did you read either paper or just agree with the tweet?
2022-07-05 19:34:32 @DimitrisPapail What experiments would you do to answer those questions?
2022-07-05 19:12:49 @DimitrisPapail What are the fascinating questions you’d like to tackle?
2022-07-05 15:23:27 @daily_barbarian What are you going to trust: a massive pile of data or a random psychiatrist with an MBA?
2022-07-04 14:33:37 @jaycaspiankang Nailed it, but now let me pick a fight about your obviously wrong taste in backpacker rap.
2022-07-03 14:15:41 @BenMazer I live in the liberal epicenter of the anti-vaccine movement in the US. I'm very familiar. And that those same people somehow became the biggest pro-vaxxers in the universe tells you these movements have nothing to do with evidence.
2022-07-03 14:13:26 @daily_barbarian The hysteria killed margarine, no? As people decided it was OK to eat butter and eggs again?
2022-07-03 14:10:12 @BenMazer For the use of "may?" I'm not sure he's the one who is unhinged on this topic.Look, Doshi, Kuldorff and others have clear biases. These biases get amplified on Twitter.But the people you favorably retweet also have biases that get amplified to absurd positions.
2022-07-03 13:59:54 @BenMazer Kuldorff has an agenda and is clearly angry about how he was treated. But calling him a "blatant anti-vaxxer" is childish ad hominem.You think the FDA let "blatant anti-vaxxers" join their vaccine advisory boards?
2022-07-02 23:35:46 @macroliter @TheEliKlein @ajlamesa @AJKayWriter @BallouxFrancois Picking on a few papers that were wrong versus standing up to huge parts of the US government is very brave of you.
2022-07-02 23:34:39 @macroliter @TheEliKlein @ajlamesa @AJKayWriter @BallouxFrancois oh, you know, just disadvantaged toddlers in Head Start programs. No big deal.https://t.co/KlkFa5H19o
2022-07-02 16:17:50 @helaineolen I challenge you to write the story in 500 characters. I think it's doable.
2022-07-02 15:47:04 @ajlamesa @EacklesLedell @nytimes Japan charges individuals for testing. Hundreds of dollars per test. Their excess deaths also suggest they undercount covid deaths by a factor of 5 or more. We cannot use these public covid dashboards as precise measurements.
2022-07-02 15:06:55 @politicalmath This critical review by Liu, @VPrasadMDMPH and Darrow better contextualizes those N95 studies, IMO: https://t.co/RXz6ebV1zE
2022-07-02 15:06:48 @politicalmath I didn't find that piece convincing 2 years ago "If we throw good statistical practice to the winds and just look at the trends... respirators are better than masks are better than nothing. It would be wrong to genuinely conclude this, because it’s not statistically significant."
2022-07-02 15:02:03 @jaycaspiankang Predictions about what happens in Berkeley? There's been significant moving of homeless into hotel-based housing, and it feels quite different now than even a few months ago.
2022-07-02 14:58:46 @politicalmath Could you point me to the evidence that "N95 masks, well fitting and worn voluntarily, reduce the chance of infection by a small but measurable amount?"Otherwise, fully agree.
2022-07-01 14:02:24 RT @adamcifu: I am pretty sure nobody will pay attention to this tweet. https://t.co/6ZAkPjq5jX
2022-07-01 13:37:55 @Maratosflier @VPrasadMDMPH What's wrong with just staying home when symptomatic?
2022-07-01 13:33:11 @VPrasadMDMPH With regards to your 2nd point: thoughts on this cluster RCT? It found that testing was no better than isolation of contacts. But that's probably not the control you'd have chosen.https://t.co/sTM34t0zf4
2022-07-01 13:20:23 @TAH_Sci @BallouxFrancois @james_e_b_ @snj_1970 @OldMrHobbes @Gaw42 @KevinBardosh Sort of. I'm not convinced that the aggressive counting of covid deaths and the subsequent plastering of these numbers everywhere has been helpful either.
2022-07-01 13:17:00 @AliceFromQueens While I would also rank him higher than the B,B,H,&
2022-07-01 13:12:30 @BallouxFrancois @james_e_b_ @snj_1970 @OldMrHobbes @TAH_Sci @Gaw42 @KevinBardosh Are we sure the mass testing was good? Testing gives us science-y folks some fascinating data to chatter about, but did it improve outcomes anywhere? There were certainly clear downsides of amplified anxiety, interrupted schooling, and other societal disruption.
2022-06-30 13:20:31 @BenMazer @MoreauGabarain @EricTopol Uh... no one is more reliably wrong than Eric Topol.
2022-06-30 13:19:26 @PunditPandemic Whenever someone brings up John Snow, I remember that his paper on cholera also argues that plague and yellow fever are water-borne.
2022-06-30 13:00:34 @BenMazer @MoreauGabarain @EricTopol You had a thread yesterday about how anonymous medical accounts were bad. But are any as bad as Eric Topol or Eric Ding?
2022-06-29 19:44:43 @AliceFromQueens @KelleyKga @BenMazer Yes, this is a very good point. As circumstantial evidence, I haven't seen many anon accounts quoted in NYTimes articles about science. At least not yet.
2022-06-29 19:20:50 @AliceFromQueens @KelleyKga @BenMazer I do agree with Ben's larger point: one probably should not take medical advice from anon twitter accounts. But I'd go further to say one probably should not take medical advice from any twitter accounts.
2022-06-29 18:22:16 @BenMazer While I mostly agree, there are exceptions to every rule, right?@AliceFromQueens is probably the most extraordinary exception.And look at the impact @KelleyKga had with fact-checking ACIP.
2022-06-29 18:13:11 @BenMazer There are some amazing anon accounts on here, but it true that the vast majority are bad. One of the biggest conundrums of Twitter.
2022-06-29 17:39:14 @CAParentPower @_ericting Yes, but I don't like the "it's because we didn't enforce the mandate" mindset. That's scary.
2022-06-29 16:17:06 @subsix848 @_ericting There is no shortage of scientists who refuse to believe data that consistently refutes what they want to be true. But these people are not directly in charge of policies that affect millions of people.
2022-06-29 16:04:10 @wencyleung You might want to look at the data from here in California where we tried another mask mandate and it did nothing. https://t.co/ejT2nlCiXh
2022-06-29 15:43:39 @BritchesFuzzy @_ericting Yes, this is exactly what happened here last winter.
2022-06-29 15:32:28 @pbleic @BallouxFrancois @EricPhDing While perhaps accurate, is it not weird to openly speculate about such fantastic, terrifying scenarios on Twitter?
2022-06-29 15:11:08 @BenMazer that's fair! us olds are dumb. I think it's probably people approximately your age and stature, with similar graduate degrees and tastes in ice cream who are most often correct.
2022-06-29 15:06:08 @BenMazer That's interesting, I'm very much the opposite. I am also older than you, so you might have to reconsider.
2022-06-28 14:56:30 @phl43 @NateSilver538 @davidzweig Anyway, only the truly enlightened acknowledge that everyone is full of shit and pushing their respective agenda.Science isn't real.
2022-06-28 14:55:37 @phl43 @NateSilver538 @davidzweig Compare and contrast Gelman's responses to different papers:Bendavid: Basically the devil, and they owe the world an apology.Flaxman: Well, you know, nobody's perfect.
2022-06-28 14:54:15 @phl43 @NateSilver538 @davidzweig The pandemic has revealed a lot of folks true colors and Gelman is no exception. The man who would have conniptions about power posing let a lot of shoddy science slide. And you can tell from this post it's because he was afraid of covid.
2022-06-27 13:12:39 @PaulMainwood Do you have favorite reading recommendations on the history of science? (It's a past time of mine as well).
2022-06-27 13:10:58 @TAH_Sci @PaulMainwood True, though we've had sport and art for millenia. There is something quite magical about post-enlightenment science and the rapid advances it has brought.
2022-06-27 13:04:00 @TAH_Sci @PaulMainwood There's something special about the sciences, though. It's wild that flawed, irrational humans can flail around in their petty, self-interested manner and yet still collectively achieve wild advancements.
2022-06-26 20:27:28 @phl43 @NateSilver538 @davidzweig Precisely. Which is why I raise it here. It's clear dishonesty and exaggeration, and it's hard to know what is motivating this consistent fabulism. Is it not willing to admit to being wrong, or is it sincere belief?
2022-06-26 20:19:37 @phl43 @NateSilver538 @davidzweig Related, @phl43 , did you see the dust-up over the CDC claiming COVID is a top-5 killer of children? That stat was from yet-another misleading paper by Seth Flaxman.
2022-06-26 19:45:43 @NateSilver538 @phl43 Sure, but, as @davidzweig pointed out, the odds that she knows anyone with a child that has ended up in the ICU and had no risk factors is essentially 0.
2022-06-26 14:36:45 @phl43 I can't tell if this is a lie or a misquote by an overzealous reporter. It's egregious either way.
2022-06-25 22:22:12 @KordingLab @ShalitUri It shows that the probability of A is smaller than the probability of A conditioned on (A union B).
2022-06-25 21:57:41 @KordingLab @ShalitUri Colliders are only one example of variables that you should not control for. For example. if you control for a mediating variable, you'll also mess up your causal estimate.
2022-06-25 19:44:03 @anish_koka Translation: "Fake it 'til you make it."
2022-06-25 18:26:47 @ShalitUri @KordingLab Right, but M-bias is such a weird and controversial example, that I'm not sure the DAGs have added much value there beyond creating a mostly imaginary problem for people to write a bunch of papers about.
2022-06-25 17:36:32 @ShalitUri @KordingLab My concern is that if you ever non-randomly select a sample, you introduce this paradox. Berkson's initial examples were in 2x2 tables (no DAGs). I don't see what the DAG brings to light.
2022-06-25 15:56:13 @zacharylipton @KordingLab Hmm, I'm not sure. I really think this is an issue of measuring proportions of subsets rather than of wholes. See the figure I posted. It shows that Berkson reversals always have to happen.
2022-06-25 15:54:38 @KordingLab Is this too simplistic? The universe of possible events is the black box. A is red, B is blue (and hence A and B is purple). The proportion of A with respect to all events is smaller than the proportion of A with respect to the events where either A and B occur. https://t.co/QXMQF1FSwM
2022-06-25 14:58:53 @KordingLab I am going to try to draw this figure. Give me a minute (or an hour, lol, I'm terrible at infographics).
2022-06-25 14:54:26 @KordingLab Collider bias, like Simpson's Paradox, is much simpler if you think about subsets than if you worry about causality and conditional probabilities. Venn diagrams would be more intuitive than DAGs.
2022-06-25 14:36:00 @KordingLab https://t.co/McYkhdFzqP
2022-06-24 14:38:39 @R_Hughes1 Reform is not in the cards. Indeed, the judiciary's power only will grow over the next couple of years. It's arguably already the most powerful branch of the government given the current structural morass.
2022-06-24 13:57:05 @rak3re @ajlamesa @nytimes @TheAgeofShoddy Right. My point is he does this on every issue. And this transparent cynicism has been amazingly profitable for him.
2022-06-24 13:50:30 @TShampling Iraq level is right. It destroyed opportunity for low-income kids and helped usher in the end of our university system. It is not clear we recover from this event.
2022-06-24 13:48:21 @ajlamesa @nytimes Also, as @vkoganpolisci has pointed out many many times: you have opinion polls saying schools should be closed, but then as soon as schools are opened, everyone goes back.We know polls are systematically broken! We have to stop using them as evidence.
2022-06-24 13:46:29 @ajlamesa @nytimes Again, that's his shtick. And it's so annoying on every front. Find a nugget of some truth, blow it way out of context, and then stick it in everyone's eye to get ratioed.It's worked for him for 20 years, but god he's so insufferable.
2022-06-24 13:44:43 @ajlamesa @nytimes Yglesias always has to be a contrarian asshole. It's the entirety of his shtick.
2022-06-23 12:34:54 @PaulMainwood I can't believe we still give any credence to data mining of private data sets.
2022-06-23 01:36:51 @BenMazer This one is coming as well (I think?) https://t.co/N0mLeRrsJX
2022-06-23 00:48:07 @BenMazer I know it is popular to pretend it didn't happen, but the DANMASK trial was an RCT run by MDs.
2022-06-23 00:12:15 @mraginsky @KordingLab It turns out blogs are good. And everything is better than threads.
2022-06-22 18:27:25 @KordingLab So do I!
2022-06-22 16:18:39 This is the first new Twitter feature I've liked in years. Brilliant idea. Death to threads! https://t.co/ga555lAQkR
2022-06-22 14:04:49 @WashburneAlex @KevinBardosh The sad truth is that if you care, you can't leave. And one thing that is abundantly clear is that Alex cares a lot.But welcome back?
2022-06-21 00:46:01 @RohanChindooroy @james_e_b_ @NahasNewman I mean, heart disease is a top-10 cause of death in kids. And it kills more kids than covid!
2022-06-21 00:44:18 @JacobAShell https://t.co/QFxc2UTBKMLike, take this one. Here's what google image search returns. Are any of these better or worse than what Dall-E supposedly returned? https://t.co/OMdvUnfaCX
2022-06-21 00:42:45 @JacobAShell I always tell people to type their Dall-E query into google image search and see what comes back. Mostly Dall-E is just outputting a mash up of those images.
2022-06-20 21:31:38 @JacobAShell Hoax might be too strong, but parlor trick for sure.
2022-06-20 13:21:52 @JSEllenberg In that csv file, the average number of influenza + pneumonia deaths per year from 2014-2018 is 190K. But those tagged specifically as flu, the number is only ~8,500/year.
2022-06-20 13:20:29 @JSEllenberg That's exactly right. Since we didn't PCR every person for every disease, we got a broad net for "pneumonia and flu" which could be caused by many nasty bugs. The flu code requires a test, I think. From these two numbers, CDC comes up with an estimate for total flu deaths.
2022-06-20 12:51:41 @JSEllenberg They don't make it easy! Which is why I always directly go to the csv: https://t.co/33Q23xyUM0
2022-06-20 12:47:05 @JSEllenberg https://t.co/rxLokB8nVUHere's a plot of some pre-covid death rates. Again, the ICD code is "flu and pneumonia" which is not as cut and dry because there is usually no PCR test. https://t.co/XKqbL4S9un
2022-06-20 12:43:51 @JSEllenberg I hate that I'm mired so deep in these statistics, but, pre 2020, the average summer deaths per day from "flu+pneumonia" was about 500 in the US.
2022-06-20 12:38:54 @JSEllenberg Where do you get that number?
2022-06-19 19:38:20 @jillneimark @LizHighleyman I avoid wearing masks whenever possible, enjoy eating out and being with friends, go to the gym every day, have been traveling. As far as I know, I've never caught covid either.Our own personal anecdata do not and should not inform healthcare policies and recommendations.
2022-06-15 11:54:57 @edschenck @venkmurthy @drjohnm Eh, I've been chatting with them, and none can point to a single reliable finding. It's astonishing.Medicine is amazing as it is the only human facing science that occasionally stumbles upon treatments with RRs larger than 10. In econ, an OR of 1.2 is a large effect.
2022-06-15 11:42:41 @edschenck @venkmurthy @drjohnm Hmm, I don't see how to use the IV here.I'm wary of IVs as there are no killer examples of them finding anything particularly profound or reproducible.
2022-06-15 11:36:08 @edschenck @venkmurthy @drjohnm 100% agree. Just highlighting that reasonable trials are still being done.Are any countries outside the US even offering paxlovid to healthy, vaccinated, non-hospitalized individuals?
2022-06-15 11:31:13 @edschenck @venkmurthy @drjohnm The UK is doing a trial of Paxlovid in hospitalized patients as part of RECOVERY:https://t.co/pfZJIvEsQ7
2022-06-14 14:43:40 @VPrasadMDMPH Adam's patience and generosity in that discussion is remarkably admirable.
2022-06-14 14:15:59 @AliceFromQueens For any search query, it's fun to compare the output of DALL-E to the output of Google Image Search. Which is more sentient? https://t.co/Mesqs978zh
2022-06-14 13:24:32 @BallouxFrancois Respectfully, I do wonder how useful these simple SIR models have been overall to our understanding of covid dynamics. Now that we precision measurements on a scale never seen before, it may be time to reassess their predictive value.
2022-06-13 23:27:22 @bergerbell JUST IN: It's the top story on their website now. https://t.co/XsXRw0xEip
2022-06-13 20:45:21 @WesPegden @JSEllenberg @AlexKontorovich @rebegol Relatedly, there is a bunch of evidence of flu having a major upswing right now, but the data is so impoverished compared to covid. You can find +ve rates, but rarely counts, and can't actually find the raw data. (this is not bad, IMO, just interesting). https://t.co/Om1iiroOuU
2022-06-13 13:33:04 @phl43 @Twitter @toad_spotted As someone who has consumed shitty online content for too long, it's depressing to me that the TPM guy is now a "crazy person."
2022-06-13 13:19:10 @JSEllenberg Google Image Search is way more fun than Dall-E. https://t.co/Fi6KMMybIS
2022-06-12 16:23:15 @davidshor @Bernstein If you'd like to learn more about why you are wrong, you can read the introduction from our book on machine learning:https://t.co/6LYUB8LxJuOr you can watch this talk I gave at Stanford:https://t.co/SH3wAWDKTr
2022-06-12 14:42:12 @mark_riedl @JonathanBalloch @lcastricato In the pdf, he admits it was heavily edited:https://t.co/vPL80fFo1W
2022-06-12 14:04:15 @Bernstein Dude, people use the exact same AI to recommend you tweets and online content. I mean, people have been using the same techniques for 70 years. What part is scary?I guess I agree that the weird antisocial religiousness of techbros is scary.
2022-06-12 13:53:46 @ajlamesa @GinaRaimondo @GavinNewsom It's weird that anyone would consider Newsom given California(a) closed schools longest average(b) has the most homeless people per capita(c) has horribly mismanaged fire safety (d) has one of the highest adjusted poverty rates.What exactly would he run on? Being pretty?
2022-06-12 02:31:50 Given the recent WaPo expose, I am duty bound to post my conversation with Google's LaMDA.https://t.co/A8jui93veR
2022-06-12 01:38:46 @OsitaNwanevu Some of it is sadly about journalists, however.
2022-06-12 00:45:29 @conorsen But the OpenAI/GPT-3/DALL-E takes *are* the same old takes. Here's a clip from the New York Times from 1958. https://t.co/mGX1558J5L
2022-06-12 00:01:16 @anish_koka it would be cool if you did a podcast on this topic.
2022-06-11 23:38:16 @arthur_spirling @tomgara using "it's" instead of "its" is how I prove that I am not an AI.
2022-06-11 23:33:09 @arthur_spirling @tomgara The fact that it's in a google doc is the part which makes me rather skeptical. Also it's sub google-search level answer to the question about climate change. https://t.co/P8qmJUw3Da
2022-06-11 23:32:12 @lauretig @jacobeisenstein @arthur_spirling Wait, I thought you were just supposed yell "OPERATOR" until a person answers.
2022-06-11 23:30:42 @jaycaspiankang @killabeets Yeah, I'm not touching this one.
2022-06-11 23:24:27 @phl43 @clockworkwhale Eh, this is clearly fake. The best passage of the article is right before the end where the AI embraces the midwit sensibilities of an elderly woman from Berkeley. https://t.co/GF0BP0Fear
2022-06-11 18:48:12 @ajlamesa @jonlevyBU Also there's no rule for departing flights!
2022-06-11 18:47:04 @jonlevyBU @daveirl What are the benefits of the pre-departure testing policy?
2022-06-11 16:46:27 For the majority of applications, both prediction intervals and confidence intervals are fake and merely convenient illusions. I'm more shocked by those who think they are real. https://t.co/a7MzhYNRr5
2022-06-11 15:50:50 @rak3re @ZacBissonnette Right, but every American institution is heavily bureaucratized. Even if you remove the hierarchy, technocracy will remain a deeply, fundamentally flawed approach to governing.
2022-06-11 15:36:17 @daily_barbarian That movie is amazing. And more appropriate for the pandemic than Soderbergh's Contagion. Thanks for the reminder that I should go rewatch it.
2022-06-11 15:33:28 @rak3re @ZacBissonnette Technocracy cannot be fixed with a better incentive structure (incentives come from a technocratic mindset!).Technocracy fails because it fundamentally blurs "how things are" with "how things ought to be."(Ugh, I can't fit this into a tweet. I should try to write it out...)
2022-06-11 15:11:01 @ZacBissonnette As a fellow Warren supporter, I worry now that she would have been worse as she would have placed more emphasis on technocracy. I am not convinced that better experts would have solved our problems. Rather, the faith in expertise itself that led us astray.
2022-06-10 23:24:53 @MikeIsaac You're really good at these.
2022-06-10 22:58:55 @drklausner Wow, that's pretty surprising. Where is this data from?
2022-06-10 19:53:41 @rak3re @Saikmedi @DACDAC4DAC @michaelmina_lab Oh wow, I hadn't connected that emed was the company Mina worked for. So gross.
2022-06-10 16:49:29 @tdietterich @ajlamesa @Bob_Wachter 3) Hypermedicalization of society has been a net negative, and covid monomania dashboards distort the perception of risk. (4/4)
2022-06-10 16:48:58 @tdietterich @ajlamesa @Bob_Wachter 2) Everyone will catch covid eventually, and its harms increase exponentially with age. (Delaying infection is not necessarily good). (3/4)
2022-06-10 16:48:31 @tdietterich @ajlamesa @Bob_Wachter 1) The threat of covid to a thrice vaxxed person is lower than that of the seasonal flu if they were flu-vaxxed. (2/4)
2022-06-10 16:48:17 @tdietterich @ajlamesa @Bob_Wachter The medical establishment needs to have an honest conversation with society (and also with Bob Wachter who is really out there). 1/4
2022-06-10 16:44:52 @spudtakes @ajlamesa I agree, we don't!
2022-06-10 15:20:13 @ajlamesa Another useful policy would be to stop publishing positivity rates of covid19. Unfortunately, I know we'll never get that one.
2022-06-10 13:55:30 @anish_koka Not yet. It doesn't catch up for another 38 hours.
2022-06-10 13:47:08 @ajlamesa Anthony LaMesa for President!
2022-06-10 01:39:45 @BenMazer Agree with the only caveat being the twitter commentariat vastly underestimates the cost of ventilation upgrades.
2022-06-10 01:38:41 @BallouxFrancois 1, 2, and 3 for sure. I still see plenty of outdoor N95s in Berkeley and the surrounding area in California.
2022-06-09 16:38:36 @daily_barbarian The thing is, if you look at the other features on this website, this shouldn't be a surprise!1. search is garbage2. who to follow is garbage3. suggested tweets by topic is garbageWhy would we expect them to get anything important right if they can't engineer the basics?
2022-06-09 16:34:12 @PunditPandemic @LizHighleyman He won't answer. But someone should ask.
2022-06-08 22:19:58 @ajlamesa @startstop_reset @WhiteHouse @wutrain @WHCOS Fair point. I'd love to see a list of countries that still have school mask mandates. Someone must be keeping track of this...
2022-06-08 22:17:38 @startstop_reset @ajlamesa @WhiteHouse @wutrain @WHCOS It's depressing that school masking has more support than pre-departure testing.Both should go, but I can never get over how much this country hates children.
2022-06-08 22:15:56 @MattNoahSmith @universalhub @wutrain @CllrKendraLara Hmm, how can you advocate for keeping a school mask mandate but against pre-departure testing?
2022-06-08 22:01:11 @missrobinson @JSEllenberg I think the kids know this one, and I'd propose "There is a Light That Never Goes Out" as an alternative, under-appreciated Smiths jam.https://t.co/2HUKQdANT5
2022-06-08 22:00:15 @missrobinson @JSEllenberg Agree! Lost in Translation is from 2003 (yikes).Also, was just about to suggest this song. :)
2022-06-08 21:58:18 @JSEllenberg Heaven Or Las Vegashttps://t.co/tRjyRU1m7f
2022-06-08 21:56:13 @JSEllenberg Cities In Dusthttps://t.co/ZZToy2IS6n
2022-06-08 21:53:23 @JSEllenberg Running Up That Hill went viral because of Stranger Things, no?
2022-06-08 21:52:29 @AnilMakam @UCSFMedicine @BrightHorizons It's so easy to forget that the old policy was "no fever for 24 hours." I don't think we've internalized how extreme the covid policies are for day cares nor what effect they have on parents, care givers, and children.
2022-06-08 21:12:25 @sunilbhop @R_Hughes1 I'm not really sure how this escalation occurred in the first place, but it is clear that we cannot give unaccountable power to public health in this country. Unfortunately our broken politics and governance have rendered these departments antithetical to their calling.
2022-06-08 21:09:51 @ajlamesa @NYCHealthCommr I don't think he's naive. He's stubborn and angry, and there's no check on his power.It's amazing to watch this happen throughout the country.
2022-06-08 21:06:02 @sunilbhop @R_Hughes1 This sentence in particular is just so far out there: "Vasan expressed frustration that so many protesters seem focused on unfounded conspiracy theories and statistical outliers for vaccine side effects such as myocarditis."
2022-06-08 21:04:21 @sunilbhop @R_Hughes1 I like your optimism and idealism, but no, this is about pride and stubbornness, not about public health.This is a good article on Vasan. I don't think he comes across well: https://t.co/FItgK2kSDk
2022-06-08 21:02:51 @sunilbhop @R_Hughes1 I don't think so. I think this is a men's pride thing more than anything pandemic related. The mayor and this public health person think it shows strength to never admit error.
2022-06-08 20:58:22 @sunilbhop @R_Hughes1 I don't think it's hard to understand if you accept the letter is not about public health. It is a memo written by a stubborn mid-level bureaucrat who refuses to admit he is ever wrong and has been wrongfully given unaccountable power.
2022-06-08 18:06:12 @phl43 The entirety of the elite show their whole ass on here. It's awesome.
2022-06-08 18:05:39 @phl43 Twitter also delightfully illustrates why you shouldn't take the engineering teams at our social media companies seriously.
2022-06-08 18:02:27 @meganbaci Is it that they didn't feel compelled to vote or didn't think that there were any options they wanted to vote for? I don't know how to properly unpack the lack of enthusiasm here, but personally, the state's one-party stranglehold makes voting feel pretty useless.
2022-06-07 22:44:54 @james_e_b_ Ah, I think Karl and Thomas are on it. I'm also guessing it's the ICD code for flu AND pneumonia that ends up counting mostly pneumonia.
2022-06-07 22:29:32 @james_e_b_ Could you post a link to where you found these charts? My guess is this is a statistical artifact as there was little flu circulating.
2022-06-06 17:06:55 @PaulMainwood @FT Twitter ad algorithm, like all of this website's engineering, is an incompetently assembled mess?
2022-06-06 13:30:04 @ajlamesa I wonder if this rule stands more because of a sclerotic, bloated executive rather than because of any rational policy reasoning.
2022-06-06 13:15:49 @jaycaspiankang And it still is insane that the folks in the house you listed paid less than I did in property taxes even though their house is a freaking mansion and more than twice as big as mine.
2022-06-06 13:10:44 @jaycaspiankang 390 sounds hyperbolic, but you can look up the property taxes for any house in Alameda County. I enjoy the game of finding how much my neighbors pay in property taxes and puzzling over why I pay 3x as much.
2022-06-06 13:03:24 @jaycaspiankang Yup, and the numbers are even worse when you realize lots of these folks purchased their houses in the 1970s.
2022-06-06 02:20:29 @JacobAShell Plenty of academics fight with each other on here, but at least they restrict the inter-departmental airing of grievances to faculty meetings.
2022-06-05 20:16:20 @MattNoahSmith You've convinced yourself that this space obviously requires a mask. That's a vague and unsubstantiated claim.
2022-06-05 19:40:53 @rdnowak @OsitaNwanevu Rob, let's talk about word processors.
2022-06-05 19:25:41 @AstorAaron @mlow29 Agree, but you have to geographically narrow down that assertion. I was in Sacramento last weekend and there were no masks. The mask mania is really a Bay Area thing.
2022-06-05 16:48:30 @MattNoahSmith We were just discussing the absurdity of mask mandates. The American exceptionalism in pre-departure testing is similarly absurd.
2022-06-05 16:44:58 @OsitaNwanevu The majority of computer science professors at elite universities would also have a hard time with this!
2022-06-05 16:30:43 @TracyBethHoeg Both plausible. A related question that is harder to answer as deaths go down: Do boosted die from covid less frequently than 2x vaxxed? During the winter wave, this was the case. https://t.co/RWSkJ3xeg9 https://t.co/6HbIodh51i
2022-06-05 16:13:27 @madsjw @sjresnick @JSEllenberg I would trade almost any American downtown for the Terrace in June.
2022-06-05 16:12:51 @MattNoahSmith @lelia_glass @gregggonsalves @arindube I think the three of us are all agreeing here, but twitter makes this conversation challenging.
2022-06-05 16:12:03 @MattNoahSmith @lelia_glass @gregggonsalves @arindube But that's not what we're arguing about. Mask mandates were predicated on "my mask protects you, your mask protects me." If we all agree this is bullshit, then they must end.
2022-06-05 16:06:37 @MattNoahSmith @lelia_glass @gregggonsalves @arindube Then wear a ribbon and leave everyone else alone?
2022-06-05 16:05:59 @lelia_glass @MattNoahSmith @gregggonsalves @arindube Who knows. I've never directly interacted with him on here. Regardless, based on his abhorrent behavior on twitter, I'm not particularly interested in interacting with him.
2022-06-05 16:02:47 @lelia_glass @MattNoahSmith @gregggonsalves @arindube Fortunately or unfortunately, I can't see any of Gregg's tweets in the thread, so I'm only responding to you, Matt, and Arindrajit.
2022-06-05 16:01:05 @MattNoahSmith @gregggonsalves @lelia_glass @arindube The issue is there is not a 'rational' argument for mask mandates. The arguments for mask mandates build on some combination of intervention bias, fear, or symbolism.
2022-06-05 15:06:27 @BenMazer Probably shouldn't listen to people who throw around the term "asshat death ghouls" either.
2022-06-05 15:05:02 @JSEllenberg @sjresnick I haven't been to Springfield in a long time, but I did not find the same amenities or charm I found in Madison...
2022-06-05 14:41:11 @sjresnick @JSEllenberg The giant state university is the confounder about why that stuff is in Madison, though. Most small cities and suburbs don't have universities with 40000 students.
2022-06-05 14:21:26 @jerryschippa @sjresnick @JSEllenberg @conorsen College towns are also a rarity for smaller cities. All other small cities lack the constantly cycling population of 18-22 year olds.
2022-06-05 14:20:08 @jerryschippa @sjresnick @JSEllenberg I can't speak for @conorsen, but I think the vibrance and variety of activities offered by a large scale city are what younger people gravitate to, and those that live in larger cities would consider Madison to be a suburb with decent restaurants.
2022-06-05 14:08:10 @JSEllenberg @sjresnick It's a stretch to say that the Chazen is downtown.
2022-06-05 03:13:41 @JeffLinderoth @madsjw Yeah, the 5 position could have been any of 5 movies. But I stand by my hot take that Millers Crossing is bad.
2022-06-05 02:01:05 @madsjw New shit has come to light.1. The Big Lebowski2. No Country for Old Men3. Fargo4. Raising Arizona5. Burn After Reading
2022-06-05 01:58:19 @madsjw Yeah, this is nonsense.
2022-06-04 18:17:02 @JeanneNoble18 Berkeley schools let out for the summer yesterday and the cowardly superintendent is skipping town. We shall see what fall 2022 brings.
2022-06-04 13:21:30 @katrosenfield All of the negative reactions are in the replies to the *2nd tweet* of the thread.https://t.co/MUp6z9vsWO
2022-06-03 19:18:55 @jaycaspiankang @Noahpinion Agree about Jay's Twitter polls.But we need to ask *why* we want to know people's opinions? Is it to placate them in service of maintaining elected power? Is it because we care about aggregate utility? Is it because we're just curious?The answer will guide the survey design.
2022-06-03 16:44:28 @daily_barbarian @TShampling It's a really good crystallization, and there's so much to unpack. Looking forward to your piece.
2022-06-03 14:44:01 @BenMazer The Socratic method requires asking more questions, doesn't it?
2022-06-03 14:43:32 @BenMazer To be even more concrete, the case fatality rate in California since January is about 0.2%. Now divide that by 30 and we're talking about 5-10 deaths per 100K infections.
2022-06-03 14:41:15 @BenMazer I know you're trolling here, but if you combine the actual case fatality rates with the 30x undercount factor, the IFR would be an order of magnitude lower than what the GBD people were arguing.
2022-06-03 14:39:43 @BenMazer Right, the 30x is completely implausible.BUT it would also mean the infection fatality rate is vanishingly small. Also, using other studies the same people like, it means that everyone now has long covid.
2022-06-03 01:59:04 @VPrasadMDMPH Also, that "since October 2020" is nothing to brag about.
2022-06-03 00:30:57 @VPrasadMDMPH Ah, but we have one randomized study of cloth masking and the result was null. And there's still that Guinea-Bissau one.
2022-06-02 23:55:54 @LizHighleyman Berkeley, Contra Costa, and SF have all since announced they won't join Alameda.
2022-06-02 23:05:29 @NSavidge @CityofBerkeley Contra Costa has no plans to impose a mask mandate.https://t.co/UuLb0tohEs
2022-06-02 23:04:30 @jwalshie @NSavidge @CityofBerkeley Yeah, it really wouldn't make sense.
2022-06-02 22:56:21 @_ericting I think we've already seen that, no? And you've reported on it! I guess we might as well repeat the natural experiment until we all asphyxiate under qunituple N95 masks.
2022-06-02 22:44:26 @scottyd121 @_ericting Alameda is really slow at updating cases, so it's possible we're not at a peak yet. But I don't care either way! What I care about is why this unelected bureaucrat still gets to make these decisions with no accountability.
2022-06-02 22:16:19 @JLimHospMD Here are the ICU vs "with covid" numbers in Alameda where a mask mandate was just imposed. Insanity.https://t.co/DwbOTX5RpG https://t.co/df2SLXnHfG
2022-06-02 22:03:24 @meganbaci @rmbodenheimer I'm quite curious about their enforcement plan this time around. The official letter was rather light on details (though had plenty of ham-fisted verbiage about equity).https://t.co/G37w3VKxBJ
2022-06-02 21:47:32 @ajlamesa We're also about to host a sold out NBA playoff game in San Francisco tonight, but that's not in Alameda County anymore as the Warriors abandoned us.
2022-06-02 21:46:48 @ajlamesa It's really stupid here in the Bay Area, Anthony. How could you bet against us?
2022-06-02 21:44:00 @jwalshie @sdaction1 @WellstoneDem @CityofBerkeley @JovankaBeckles @RigelRobinson @RebeccaForBART @Kaplan4Oakland I am looking forward to voting against all of these people. Thanks for the heads up.
2022-06-02 21:40:44 @jarod_ormsby @MmeBlackBalloon @_ericting Unreal. Hernandez says the same thing in Berkeley. Berkeley schools have had a mask mandate for the last two weeks (last day of school is tomorrow).I'm surprised Moss came on your podcast in the first place. Nice work!
2022-06-02 21:32:58 @MmeBlackBalloon @_ericting I don't know in Alameda. It was Erica Pan and then she got "promoted" to a state-level position. Berkeley bizarrely has its own (completely unaccountable) health officer, though she usually just parrots what Alameda does.
2022-06-02 21:29:43 @MmeBlackBalloon @_ericting Everyone on the ballot has been pro restrictions. We are not offered those sort of choices here in the Bay Area.
2022-06-02 21:25:42 @_ericting Check out this plot of hospitalizations in Alameda County. There is no one in the ICU. Clearly that this is a "with covid" not "for covid" phenomenon. https://t.co/DwbOTX5RpG https://t.co/wzjIool0yW
2022-06-02 15:07:38 @JacobAShell It's such a fun and pleasant place to visit. I don't get it.
2022-06-02 12:46:15 @politicalmath California!
2022-06-02 01:05:52 @VPrasadMDMPH @JLimHospMD But just think of the observational studies we'll publish in PNAS during the next crisis!
2022-06-02 01:00:45 @MDaware @benryanwriter @AstorAaron @LawrenceGostin @aetiology @CDCgov Less good would be fine, but no one can provide evidence of NPIs having any benefit without significant book cooking. And these stubborn, misleading assertions about the value of NPIs just causes people to dismiss data on legitimately valuable interventions (e.g., vaccines)
2022-06-02 00:19:45 @JDHaltigan @The_OtherET @AstorAaron @ElleMandell @Laurie_Garrett @ianmSC I never read sensitivity analyses as they never show the result in the paper is sensitive. They are the weirdest convention.
2022-06-02 00:15:43 @JDHaltigan @The_OtherET @AstorAaron @ElleMandell @Laurie_Garrett @ianmSC Related... https://t.co/CS8JmLHCk8
2022-06-02 00:11:14 @PunditPandemic That thread is astounding. And a lovely example how researchers convince themselves of whatever they want based on their own studies.
2022-06-02 00:09:52 @JDHaltigan @The_OtherET @AstorAaron @ElleMandell @Laurie_Garrett Yeah, I just think this is weak reasoning. These large-scale survey + modeling studies aren't helping and we shouldn't encourage them. But PNAS sees "2 million people" and "masks work" and presses publish.
2022-06-02 00:01:46 @JDHaltigan @The_OtherET @AstorAaron @ElleMandell @Laurie_Garrett It's not about omicron. It's that they chose to start in May 2020 and end in September 2020, which creates a massive seasonal bias.Unfortunately, writing another takedown won't matter. Prestige journals will continue to publish crappy mask papers forever because The Science.
2022-06-01 23:56:56 @The_OtherET @AstorAaron @ElleMandell @Laurie_Garrett and it's 4 months selected to make conservative areas look bad.
2022-06-01 23:51:08 @MDaware @benryanwriter @AstorAaron @LawrenceGostin @aetiology @CDCgov The clear signal you see is confirmation bias.Here are some graphs clearly signaling the effectiveness of vaccines. Show me anything comparable for non-pharmaceutical anything. https://t.co/Aor2seNNvy
2022-06-01 22:17:23 @ID_ethics "The science has changed!"
2022-06-01 17:55:27 @BenMazer Very curious, what would it take to convince you otherwise? Because we have two years of data now that points pretty strongly in the opposite direction. I mean, I'd love for a single graph remotely close to the one you tweeted this morning. None exist.https://t.co/gAx00xdbv4
2022-06-01 17:52:30 @TheNickFoy @KellieHwang @Bob_Wachter Hmm. I don't want to read the Chronicle today, so curious if she mentioned that there was an entire season of basketball before these finals?
2022-06-01 00:02:59 @TheNickFoy Remarkably, this is the first land acknowledgement I've seen in a Twitter name.
2022-05-31 21:23:28 @subsix848 @LizHighleyman PS Do you have a link to a reference of the relative sizes of asbestos to aerosols?
2022-05-31 21:21:47 @subsix848 @LizHighleyman Oh man, I had this dude in my replies earlier today going on about asbestos and P100s and how this applied to covid... and then I muted the thread.
2022-05-31 20:41:53 @LizHighleyman I do worry that we have to come to terms with the assertion "good quality masks do provide personal protection" at some point. To my ear, this makes masks sound more useful than they have been in practice.
2022-05-31 20:40:41 @LizHighleyman @DLeonhardt That quote is almost tautological. But it's hard to argue with the sentiment.
2022-05-31 20:35:01 @ZacBissonnette @DLeonhardt I'm predicting a Leonhardt article in Fall 2022 that says "masks were a bad idea and we should have never normalized them."
2022-05-31 20:34:02 @AstorAaron @TheNickFoy @maggiemfox @j_g_allen @JenniferNuzzo Mask discourse remains the worst fucking discourse. It's not salvageable. And as a result, lots of children suffer.
2022-05-31 15:48:43 @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt But this consistent failure points to something! If the effect size is so small that no study can measure it, it's likely there is nothing there.
2022-05-31 15:47:47 @ducha_aiki @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt Can you tell me which author on there is the medical expert you trust? Lex Fridman? Zeynep Tufecki?
2022-05-31 15:27:34 @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt It's a rather proper Bayesian argument: My prior is that Jeremy Howard has zero credibility. Hence, my posterior will not be adjusted based on his absurd survey article.
2022-05-31 15:24:55 @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt That the article you link cites Jeremy Howard as evidence should give you pause. Think about Jeremy Howard's absurd machine learning takes. Why did we defer any credibility to him with respect to masking?
2022-05-31 15:23:17 @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt This is not a systemic review. And I have read all of the references attached there and none provide evidence. Here is a Cochrane review that finds that masks don't work. And we also have two years of real world data demonstrating no effect.https://t.co/Xxk6IPmbIG
2022-05-31 14:58:49 @james_e_b_ She's the worst."Long covid is real" and then "1 in 5 adults get long covid." @bergerbell aptly described her as a "reporter swallowed by her own beat."
2022-05-31 14:40:56 @roydanroy @kamalikac @drklausner @DLeonhardt @gregggonsalves And I have not seen a single randomized study that shows N95s to be superior to surgical masks. Happy to be corrected if one exists.
2022-05-31 14:39:52 @roydanroy @kamalikac @drklausner @DLeonhardt @gregggonsalves This Cochrane review summarizes many of the studies: https://t.co/Xxk6IPmbIG
2022-05-31 14:37:55 @roydanroy @kamalikac @drklausner @DLeonhardt @gregggonsalves Yes, this is what I'm saying. There have been RCTs showing N95 use in hospitals does not protect health care workers against flu. No idea why it would be different for covid which is even more transmissible.
2022-05-31 14:29:09 @drklausner @DLeonhardt @gregggonsalves But there is no evidence that masks protect individuals either. Is it worth sustaining perpetual fear for a clothing decoration that does nothing? We'd be better off being honest about what tools we have at our disposal (i.e., vaccines are effective, everything else...)
2022-05-31 14:16:26 @tdietterich @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt The answer is no. Masks are not normal. And we have two years of evidence now showing that they do nothing.
2022-05-31 13:20:42 @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt I hammer on this because the Berkeley school district maintains mask mandates. My son has now had 3 years of elementary school disrupted. Enough. It's time to reckon with the bad arguments that brought us here. And one of the worst arguments is "masks work."
2022-05-31 13:19:48 @drlucymcbride @ShiraDoronMD @TuftsMedicalCtr @DLeonhardt Unfortunately, this article makes the same tired and baseless claims about laboratory studies demonstrating "masks work" without citing any of them.We're over two years in now, and advocates still can't even define what it would mean for a mask to "work."
2022-05-31 01:38:08 @DavidOAtkins @gelliottmorris @davidshor @mattyglesias @EricLevitz @jaycaspiankang FWIW, this is more or less Jay's point.
2022-05-31 01:25:37 @AltMiddlePeds @JLimHospMD Yes. If you click the link above, you get an interactive graph. At the peak of the January omicron wave, the ICU beds peaked at 40. At the height of the January 2021 wave, one where SF was relatively spared compared to southern CA, the ICU beds peaked at over 60.
2022-05-30 23:03:29 @JLimHospMD Right, the hospitalization plot doesn't break down acute care vs ICU. There are only 10 patients in ICU in SF right now. (This table is from the same source as where Bob gets his data: https://t.co/m9Exu1WhJd) https://t.co/dG59X9UnMY
2022-05-30 15:26:12 @realistleft @UnlearnEcon I don't think science should be coalescing around any gold standards. We should be able to evolve and adapt how we report results and not hold some ideal fixed.Peer review is relatively new in the history of science, and we've been far too sluggish to move away from it.
2022-05-30 15:13:59 @UnlearnEcon The sad part is that re-analyses are not publishable in our current journals. If a crap paper appears in Nature, and you re-analyze the data and come to a different conclusion, you can't publish the reanalysis in Nature. Or in the tier-2 journals either.
2022-05-29 23:22:44 @awgaffney @zeynep @phealthsean I think the numbers in that paper are implausible. How could ascertainment bias change by a factor of 10 in one month? And if we had 31x under reporting here in the Bay Area, everyone would have been infected 3 times in the last 6 months.
2022-05-29 19:17:49 @jon_m_rob @davidshor @jaycaspiankang @jvposter @cwarshaw Lol, no worries. I know the feeling.
2022-05-29 14:46:49 @davidshor @jaycaspiankang @jvposter Is there a paper associated with this slide deck? @jon_m_rob @cwarshaw https://t.co/AwR1KhXPTa
2022-05-29 14:12:07 @davidshor @n_stenhou @jaycaspiankang Can you explain what this plot is showing?I read the thread this links to, and it seemed to just be about status-quo bias in ballot initiatives.
2022-05-29 14:09:04 @davidshor @jaycaspiankang This still gets away from Jay's point which could be answered by some links to research showing that "oppose/support" polls overstate support with respect to polls that list two alternative programs.So far, you've shared a slide from Catalist. There must be more out there.
2022-05-29 14:02:50 @jaycaspiankang He's making on clear point, which seems fine: if you ask a polling question in the form "do you support or not support X?", that poll is BS.I'm not happy to agree that alternative question formations are any better. Because what if all of these polls are BS?
2022-05-29 13:57:47 @n_stenhou @jaycaspiankang This is even worse than what Jay was saying. This thread you link is arguing that if only you ask the questions the right way you get the answer you wanted in the first place.
2022-05-29 13:24:30 @BenMazer And the unspoken part of plots like this is some countries necessarily had to suffer in order for such highly effective vaccines to be developed.
2022-05-28 13:24:37 @AstorAaron I think taking this one study too seriously would be a mistake, as these numbers were gathered by robocall polls and, unless I missed it, they don't report the survey response rate.
2022-05-28 13:10:00 @AstorAaron LOL.
2022-05-28 13:07:37 @AstorAaron This is another problem with the dashboards: how do we know the data isn't garbage? E.g., I don't know whether the CA state dashboard numbers are counting multiple positive tests from the same individuals.
2022-05-28 13:05:39 @AstorAaron Now I'm just amusing myself: Since Dec 1, 2021, there are over 700K reported cases in the bay area. Our population is 7.8M. If these were individual people, it means 10% of the population has *tested positive* for covid in the last 6 months. Also seems implausible.
2022-05-28 13:01:33 @AstorAaron On the other hand, if I take the total number of reported cases in the bay area and multiply it by 31, it means everyone has been infected with omicron 3 times since the start of the year. Which is why I think the numbers in this study are implausible.
2022-05-28 13:00:15 @AstorAaron @TheNickFoy 100% agree. I've been writing both local journalists and local health officials arguing the same since the fall. But people really think it's important information the public must have.
2022-05-28 12:55:57 @AstorAaron The *case* fatality rate of omicron in the bay area over the last 6 months is 0.2%, about 2x the *infection* fatality rate estimated of flu. If the real number of infections is 30x, then the IFR of omicron would be 0.007%. That is, ~7 deaths per 100,000 infections.
2022-05-28 12:50:20 @Reroot_Flyover @AstorAaron Oh, I'm sure some people were saying it. They just didn't get their grants funded in the beforetimes.
2022-05-28 12:49:31 @TheNickFoy @AstorAaron Right. That would be interesting to see as well.Though I worry that the large scale PCR infrastructure we built is going to reinforce some harmful neuroses for many years.
2022-05-28 12:44:39 @AstorAaron That number sounds implausible to me. But if it is real, either the infection fatality rate of omicron is a tiny fraction of the the flu or we vastly underestimate the spread of the flu.
2022-05-28 12:38:36 @phl43 He literally wrote a book extolling midwittery.https://t.co/ShO1q9LS2p
2022-05-28 01:45:50 @phl43 I agree with all of it except the part where he talks about submitting to journals anyway because otherwise it would be unfair to trainees. That's no way to start a revolution.
2022-05-26 19:08:15 @WesPegden @apsmunro I really need to bookmark this thread.Or maybe you should write a blog post for posterity. :)Test-and-trace (and testing more generally) doesn't receive the scrutiny it deserves. Testing and tracing is really expensive! We need more frank analyses of their effectiveness.
2022-05-26 18:46:52 @HeadwaysMatter @apsmunro I have not, but I will give it a look. Thank you!
2022-05-26 16:44:34 @apsmunro Have you seen good studies on the effectiveness of test and trace in these countries? All of the research I've seen this far have shown only very modest reduction of downstream infections.
2022-05-25 16:25:39 @dnunan79 2) Again, even with binary outcomes, suppose you only have missed detections. Then this must be accounted for in your power calculation, as it decreases effect size at the same proportion that it decreases variance. 3/3
2022-05-25 16:24:13 @dnunan79 1) Suppose you have binary outcomes, and measurement error is only false positives. Then your estimate of risk reduction will be biased (you estimate the treatment as less effective than it really is). 2/x
2022-05-25 16:23:18 @dnunan79 Got it. I like these definitions.So I can see two ways that measurement error creates internal validity issues. 1/x
2022-05-25 15:14:31 @Britonomist @jaycaspiankang @davidshor @jon_m_rob America's representative democracy precedes the foundational statistics of opinion polling by over 150 years.
2022-05-25 15:07:04 @jaycaspiankang @davidshor @jon_m_rob Right, the takeaway has to be "stop polling."I'll argue we should stop obsessing about candidate polling as well.
2022-05-25 14:47:22 @jaycaspiankang @bergerwfries @davidshor This is the important point that is hard to grapple with. Polls have demonstrated themselves time and time again to be very biased evidence, but everyone can find a poll that agrees with what they personally believe. The whole argument structure is fucked up.
2022-05-25 14:15:17 @virginiahume @andreavhowe This is very fair. I was more reacting to the number of people on here just posting normal takes. I wonder if wall-to-wall silence is also dangerous?The immediate acceptance that we will do nothing in the aftermath here is so just depressing.
2022-05-25 14:08:55 @davidshor @jaycaspiankang There are so many conclusions consistent with polls narrowing over time. I don't understand how you use this datum as evidence for your own claim here.
2022-05-25 14:02:46 @gabrielmbross @robxlii @jaycaspiankang Yes! I also left of University of Maine.Maine has a lot of great schools.
2022-05-25 13:56:13 @robxlii @jaycaspiankang You forgot Maine College of Art and Design.
2022-05-25 13:55:44 @jdvdub @jaycaspiankang Well, if you use enough machine learning and data tooling, you can cherry-pick evidence with *science*.
2022-05-25 13:43:13 @andreavhowe This was exactly my feeling this morning. The muted collective reaction is deeply disturbing.
2022-05-25 13:16:24 @dnunan79 In your mind, what is the difference between a biased endpoint and a measurement error? I'd argue the former is a case of the latter.
2022-05-24 19:11:51 @owenhardy1234 @PunditPandemic @ProfEmilyOster Yup. My son’s school made a big thing about everyone getting vaccinated so they could stop masking.
2022-05-24 16:36:15 Highly recommend this lecture on statistical evidence by Richard Peto. While focused on medical trials, there are tons of broadly applicable insights on sampling, heterogeneous effects, and generalization. (h/t @dnunan79)https://t.co/UKl5KVVJcx
2022-05-24 13:45:04 @subsix848 @PunditPandemic @ProfEmilyOster Parts of Australia are considering mandating masks for flu. It is indeed now just "what you do." Disheartening.https://t.co/J0D626xcUI
2022-05-24 13:34:09 @PunditPandemic @ProfEmilyOster Yup. I think this is true of most pro-school pro-mask people. I had many conversations with parents about "ok, we'll give them masking because the school has been closed for a year." I don't know how we get them off masking at this point.
2022-05-24 03:11:43 @VPrasadMDMPH Since May 23, 2021, the city of Berkeley has reported over 12,400 cases and 12 deaths. That's a *case* fatality rate of under 0.1%. Policy has definitely not adjusted to that reality.I should do this for the rest of the Bay Area Health Departments...
2022-05-24 01:50:23 @jenneraub @TheNickFoy See, this is the discourse we all come to Twitter for.
2022-05-24 01:36:01 @TheNickFoy People who admit to being pro reclining on airplanes are sociopaths and deserve to be blocked.
2022-05-20 08:11:00 CAFIAC FIX
2022-10-29 01:41:21 @ajlamesa And no evidence of significant transmission on public transport!
2022-10-28 13:49:42 @TheNickFoy I mean, legs are coming soon, Nick. Legs are coming. https://t.co/lcluFhTEmB
2022-10-27 17:25:41 @fhuszar @maosbot Don't disagree there. And also that AI Alignment is a nonsense rationalist fever dream based in bad science fiction.
2022-10-27 17:24:30 @fhuszar @maosbot On it!
2022-10-27 17:11:52 @fhuszar @maosbot 300 people working on "The AI Alignment Problem" is 300 too many.
2022-10-27 14:10:32 @quantum_aram Not today at least...
2022-10-27 13:46:19 @AChandra_TO @phl43 LOL. I need to get Derek Thompson to apologize.But right: I think the "peer review" canard is only used when convenient.
2022-10-27 13:41:43 @AChandra_TO @phl43 I don't know, MSM runs with preprints when it suits them.Just last week, the Economist picked up that bizarre Lab Leak preprint written by three conspiracy theorists.
2022-10-27 13:38:51 Even Elon can't escape. “The best minds of my generation are thinking about how to make people click ads.” https://t.co/0tDidHYwLn
2022-10-27 13:35:24 @phl43 Interesting perspective from computer science: CS has convinced deans and managers that *conference* papers can and should be counted as journal publications. Pro: As a result, there are no prestigious CS journals.Con: Our conference review system is a shitshow.
2022-10-27 13:31:51 @phl43 Agreed, I have been reading variations of this anti-journal argument for over a decade and no one can form a cogent rebuttal.Unfortunately, one reason journals continue to exist is publications are required for promotion in academic and industrial research.
2022-10-27 13:25:11 @phl43 I don't agree that social media ends up being better than journals (both have warts), but I also can't believe how many journals get away with charging *authors* thousands of dollars per paper. That alone is an argument for their end.
2022-10-27 00:39:52 @VPrasadMDMPH @johndefeo One of the authors thinks Monkeypox was lab leaked.
2022-10-26 14:02:04 Hell has frozen over: I wrote a blog post saying a bunch of nice things about neural nets (and art?). https://t.co/wGTqgcZG2N
2022-10-25 19:36:23 @FO_ASchatz This can happen even with no statistics:(Rush score, pass score)Team 1: (4,1)Team 2: (3,3)Team 3: (1,4)Team 2 is not best in either score, but it has the best added scores.
2022-10-25 19:05:42 @jaycaspiankang There was an earthquake?
2022-10-24 16:30:30 @LaurentLessard @CSSIEEE Uh, that's totally bizarre. What a waste of speaker time.Unthawing from our frozen virtual pandemic time is going to take a while to shake out.
2022-10-24 16:26:28 @LaurentLessard @CSSIEEE Wait, are there going to be videos instead of talks at the in-person conference?
2022-10-24 14:14:36 @DanielHadas2 We have also created a bizarre literature on Long Covid. Long Covid arose on Facebook and became popularized by a poorly reported story by Ed Yong. It’s a perfect example of science not only behaving like the media, but taking its queues from the media. 3/3
2022-10-24 14:14:20 @DanielHadas2 Since then, we have learned about vaccines. We can quibble with policies, but these papers are good. There have been good papers on treatment as well, in particular from the RECOVERY trial in the UK. 2/3
2022-10-24 14:13:39 @DanielHadas2 Yeah, the following the science meme certainly added too much importance to the science.It's remarkable that we basically knew everything about this virus by May 1, 2020. And we knew a lot by March 1, 2020. 1/3
2022-10-23 14:18:58 Solid advice for "data scientists," no matter which topic they are studying. https://t.co/iGR2neHSKO
2022-10-23 04:03:53 @jaycaspiankang Tarik Glenn is pretty cool.
2022-10-23 04:03:21 @jaycaspiankang Curious who you have as the 2nd cool person after Marshawn Lynch.
2022-10-22 21:13:38 @JacobAShell @DanielHadas2 @Spear_Owl But it could alternatively be social-media-driven mass formation psychosis not cooperating with medicine's preexisting analytical categories.
2022-10-22 21:04:29 @JacobAShell @Spear_Owl @DanielHadas2 Though they were influenza pandemics, rather than covid pandemics, the global mortality associated with the 1957 and 1968 pandemics are both of similar magnitude to covid. The main difference being Asia was hit harder and North America and Europe significantly less so.
2022-10-22 20:51:56 @JacobAShell @DanielHadas2 @Spear_Owl Post-viral sequelae are indeed a thing, and teasing out their etiology has always been a challenge. Here's one survey from 1987: https://t.co/7igETSURw5
2022-10-22 20:35:30 @JacobAShell @DanielHadas2 @Spear_Owl No, I know. That's why I wrote it. I didn't mean it as an attack but as a lament: we're globally stuck doing obviously stupid things when common sense clearly dictates we should do the opposite.
2022-10-22 20:33:13 @DanielHadas2 @Spear_Owl @JacobAShell I see. Thanks for the clarification.I think it's interesting that I see the opposite direction, and I should think more about why.
2022-10-22 20:28:59 @JacobAShell @DanielHadas2 @Spear_Owl I do not disagree at all. But there are many other things I wish were a global consensus: that we shouldn't infringe on the habitats of wildlife, that humankind does not yet possess the ability to stop all respiratory viruses,...
2022-10-22 20:27:24 @JacobAShell @DanielHadas2 @Spear_Owl In what way are the symptoms odd?
2022-10-22 20:26:37 @DanielHadas2 @Spear_Owl @JacobAShell How does one understand "the science" without "the people and the politics?"
2022-10-22 20:26:05 @DanielHadas2 @Spear_Owl @JacobAShell There is a clear dichotomization in politics between those who are willing to entertain the lab leak and those who are not.
2022-10-22 20:25:32 @DanielHadas2 @Spear_Owl @JacobAShell Here in this thread, @JacobAShell is saying that the virology community looks guilty, hence, without understanding their arguments, they clearly must be wrong on "the science."
2022-10-22 20:25:17 @DanielHadas2 @Spear_Owl @JacobAShell We don't have to belabor into this point, but don't you think ad hominem is exactly what the lab leak debate is?
2022-10-22 20:14:16 @DanielHadas2 @Spear_Owl @JacobAShell The only way that scientific paradigm shifts happen is with new experiments that slowly chip away at an infinite landscape of possibility.Every once in a while we find something amazing, but usually, we just muddle through minutiae while we wait for the new revolution.
2022-10-22 20:13:27 @DanielHadas2 @Spear_Owl @JacobAShell My read of the post-Kuhn landscape is that there is no genuine scientific aporia but only politics.
2022-10-22 20:04:20 @DanielHadas2 @Spear_Owl @JacobAShell But there is no way to prove it either way! There will never be new data. We're all stuck scrutinizing mirages and arguing from our political corners.
2022-10-22 20:02:04 @JacobAShell @Spear_Owl @DanielHadas2 Or maybe, just like Latour told us, scientists are a bunch of crazy weirdos, and they shouldn't be out on Twitter showing the world their whole asses?
2022-10-22 20:00:00 @Spear_Owl @DanielHadas2 @JacobAShell But that thread has input from dozens of other scientists. It's not just Andersen and Rassmussen.And you can assert "there's a cartel here suppressing science!."And we will continue to fight because no one can do a real experiment to prove anything about the fucking lab leak.
2022-10-22 19:58:08 @Spear_Owl @DanielHadas2 @JacobAShell Yeah, I don't disagree. He and Angela Rassmussen both behave like children on here, and I don't understand why they can't see how harmful this behavior is.
2022-10-22 19:53:54 @JacobAShell @Spear_Owl @DanielHadas2 I think everyone on this stupid website protests too much.If anything, I'm more Latourian than I was at the start of the pandemic. But the outsider critics fall into the same traps as the insiders. No one is immune.
2022-10-21 20:22:02 @rdnowak @DimitrisPapail @amirvaxman_dgp Barron showed that functions f where the Fourier transform \tilde{f} satisfied the following inequality were well approximated by a small sum of sigmoidal functions. He also showed the number of required functions to achieve an eps-approx in L2 scaled like 4 B^2/eps^2. https://t.co/QUV74jG7Xh
2022-10-21 20:15:51 @rdnowak @DimitrisPapail @amirvaxman_dgp I agree with Rob. Barron's results were describing the sorts of functions that were well approximated by neural nets. (Cybenko's argument is certainly tidier.)
2022-10-21 16:06:14 @BenMazer @drvictoriafox @GYamey He didn't say either of these things. I understand that Ioannidis is a convenient punching bag and you don't want to engage with what he was saying in good faith.
2022-10-21 15:57:11 @BenMazer @drvictoriafox @GYamey Everyone was running around in fear, as you have admitted you were yourself. Ioannidis was asking "do we have the data to justify the fear?" Someone had to ask. Look at all the stupid shit we did because people were unreasonably afraid!
2022-10-21 15:13:58 @KevinBardosh Was the discussion unplanned and arose at another event? In order for this to be a productive conversation, it seems like thoughtful mediation would be necessary.
2022-10-21 15:06:21 @KevinBardosh I wonder how many (if any) of these discussions have occurred.
2022-10-21 14:13:45 @DanielHadas2 Your threads are too good. I look forward to my daily meditation on them.I hope you collect them in one place at some point.
2022-10-29 16:25:31 @DanielHadas2 @JacobAShell Do you two have any comment on how the latest report "bombshell" requires mistranslating Chinese? (I feel like this is Daniel's lane)
2022-10-29 01:41:21 @ajlamesa And no evidence of significant transmission on public transport!
2022-10-28 13:49:42 @TheNickFoy I mean, legs are coming soon, Nick. Legs are coming. https://t.co/lcluFhTEmB
2022-10-27 17:25:41 @fhuszar @maosbot Don't disagree there. And also that AI Alignment is a nonsense rationalist fever dream based in bad science fiction.
2022-10-27 17:24:30 @fhuszar @maosbot On it!
2022-10-27 17:11:52 @fhuszar @maosbot 300 people working on "The AI Alignment Problem" is 300 too many.
2022-10-27 14:10:32 @quantum_aram Not today at least...
2022-10-27 13:46:19 @AChandra_TO @phl43 LOL. I need to get Derek Thompson to apologize.But right: I think the "peer review" canard is only used when convenient.
2022-10-27 13:41:43 @AChandra_TO @phl43 I don't know, MSM runs with preprints when it suits them.Just last week, the Economist picked up that bizarre Lab Leak preprint written by three conspiracy theorists.
2022-10-27 13:38:51 Even Elon can't escape. “The best minds of my generation are thinking about how to make people click ads.” https://t.co/0tDidHYwLn
2022-10-27 13:35:24 @phl43 Interesting perspective from computer science: CS has convinced deans and managers that *conference* papers can and should be counted as journal publications. Pro: As a result, there are no prestigious CS journals.Con: Our conference review system is a shitshow.
2022-10-27 13:31:51 @phl43 Agreed, I have been reading variations of this anti-journal argument for over a decade and no one can form a cogent rebuttal.Unfortunately, one reason journals continue to exist is publications are required for promotion in academic and industrial research.
2022-10-27 13:25:11 @phl43 I don't agree that social media ends up being better than journals (both have warts), but I also can't believe how many journals get away with charging *authors* thousands of dollars per paper. That alone is an argument for their end.
2022-10-27 00:39:52 @VPrasadMDMPH @johndefeo One of the authors thinks Monkeypox was lab leaked.
2022-10-26 14:02:04 Hell has frozen over: I wrote a blog post saying a bunch of nice things about neural nets (and art?). https://t.co/wGTqgcZG2N
2022-10-25 19:36:23 @FO_ASchatz This can happen even with no statistics:(Rush score, pass score)Team 1: (4,1)Team 2: (3,3)Team 3: (1,4)Team 2 is not best in either score, but it has the best added scores.
2022-10-25 19:05:42 @jaycaspiankang There was an earthquake?
2022-10-24 16:30:30 @LaurentLessard @CSSIEEE Uh, that's totally bizarre. What a waste of speaker time.Unthawing from our frozen virtual pandemic time is going to take a while to shake out.
2022-10-24 16:26:28 @LaurentLessard @CSSIEEE Wait, are there going to be videos instead of talks at the in-person conference?
2022-10-24 14:14:36 @DanielHadas2 We have also created a bizarre literature on Long Covid. Long Covid arose on Facebook and became popularized by a poorly reported story by Ed Yong. It’s a perfect example of science not only behaving like the media, but taking its queues from the media. 3/3
2022-10-24 14:14:20 @DanielHadas2 Since then, we have learned about vaccines. We can quibble with policies, but these papers are good. There have been good papers on treatment as well, in particular from the RECOVERY trial in the UK. 2/3
2022-10-24 14:13:39 @DanielHadas2 Yeah, the following the science meme certainly added too much importance to the science.It's remarkable that we basically knew everything about this virus by May 1, 2020. And we knew a lot by March 1, 2020. 1/3
2022-10-23 14:18:58 Solid advice for "data scientists," no matter which topic they are studying. https://t.co/iGR2neHSKO
2022-10-23 04:03:53 @jaycaspiankang Tarik Glenn is pretty cool.
2022-10-23 04:03:21 @jaycaspiankang Curious who you have as the 2nd cool person after Marshawn Lynch.
2022-10-22 21:13:38 @JacobAShell @DanielHadas2 @Spear_Owl But it could alternatively be social-media-driven mass formation psychosis not cooperating with medicine's preexisting analytical categories.
2022-10-22 21:04:29 @JacobAShell @Spear_Owl @DanielHadas2 Though they were influenza pandemics, rather than covid pandemics, the global mortality associated with the 1957 and 1968 pandemics are both of similar magnitude to covid. The main difference being Asia was hit harder and North America and Europe significantly less so.
2022-10-22 20:51:56 @JacobAShell @DanielHadas2 @Spear_Owl Post-viral sequelae are indeed a thing, and teasing out their etiology has always been a challenge. Here's one survey from 1987: https://t.co/7igETSURw5
2022-10-22 20:35:30 @JacobAShell @DanielHadas2 @Spear_Owl No, I know. That's why I wrote it. I didn't mean it as an attack but as a lament: we're globally stuck doing obviously stupid things when common sense clearly dictates we should do the opposite.
2022-10-22 20:33:13 @DanielHadas2 @Spear_Owl @JacobAShell I see. Thanks for the clarification.I think it's interesting that I see the opposite direction, and I should think more about why.
2022-10-22 20:28:59 @JacobAShell @DanielHadas2 @Spear_Owl I do not disagree at all. But there are many other things I wish were a global consensus: that we shouldn't infringe on the habitats of wildlife, that humankind does not yet possess the ability to stop all respiratory viruses,...
2022-10-22 20:27:24 @JacobAShell @DanielHadas2 @Spear_Owl In what way are the symptoms odd?
2022-10-22 20:26:37 @DanielHadas2 @Spear_Owl @JacobAShell How does one understand "the science" without "the people and the politics?"
2022-10-22 20:26:05 @DanielHadas2 @Spear_Owl @JacobAShell There is a clear dichotomization in politics between those who are willing to entertain the lab leak and those who are not.
2022-10-22 20:25:32 @DanielHadas2 @Spear_Owl @JacobAShell Here in this thread, @JacobAShell is saying that the virology community looks guilty, hence, without understanding their arguments, they clearly must be wrong on "the science."
2022-10-22 20:25:17 @DanielHadas2 @Spear_Owl @JacobAShell We don't have to belabor into this point, but don't you think ad hominem is exactly what the lab leak debate is?
2022-10-22 20:14:16 @DanielHadas2 @Spear_Owl @JacobAShell The only way that scientific paradigm shifts happen is with new experiments that slowly chip away at an infinite landscape of possibility.Every once in a while we find something amazing, but usually, we just muddle through minutiae while we wait for the new revolution.
2022-10-22 20:13:27 @DanielHadas2 @Spear_Owl @JacobAShell My read of the post-Kuhn landscape is that there is no genuine scientific aporia but only politics.
2022-10-22 20:04:20 @DanielHadas2 @Spear_Owl @JacobAShell But there is no way to prove it either way! There will never be new data. We're all stuck scrutinizing mirages and arguing from our political corners.
2022-10-22 20:02:04 @JacobAShell @Spear_Owl @DanielHadas2 Or maybe, just like Latour told us, scientists are a bunch of crazy weirdos, and they shouldn't be out on Twitter showing the world their whole asses?
2022-10-22 20:00:00 @Spear_Owl @DanielHadas2 @JacobAShell But that thread has input from dozens of other scientists. It's not just Andersen and Rassmussen.And you can assert "there's a cartel here suppressing science!."And we will continue to fight because no one can do a real experiment to prove anything about the fucking lab leak.
2022-10-22 19:58:08 @Spear_Owl @DanielHadas2 @JacobAShell Yeah, I don't disagree. He and Angela Rassmussen both behave like children on here, and I don't understand why they can't see how harmful this behavior is.
2022-10-22 19:53:54 @JacobAShell @Spear_Owl @DanielHadas2 I think everyone on this stupid website protests too much.If anything, I'm more Latourian than I was at the start of the pandemic. But the outsider critics fall into the same traps as the insiders. No one is immune.
2022-10-21 20:22:02 @rdnowak @DimitrisPapail @amirvaxman_dgp Barron showed that functions f where the Fourier transform \tilde{f} satisfied the following inequality were well approximated by a small sum of sigmoidal functions. He also showed the number of required functions to achieve an eps-approx in L2 scaled like 4 B^2/eps^2. https://t.co/QUV74jG7Xh
2022-10-21 20:15:51 @rdnowak @DimitrisPapail @amirvaxman_dgp I agree with Rob. Barron's results were describing the sorts of functions that were well approximated by neural nets. (Cybenko's argument is certainly tidier.)
2022-10-21 16:06:14 @BenMazer @drvictoriafox @GYamey He didn't say either of these things. I understand that Ioannidis is a convenient punching bag and you don't want to engage with what he was saying in good faith.
2022-10-21 15:57:11 @BenMazer @drvictoriafox @GYamey Everyone was running around in fear, as you have admitted you were yourself. Ioannidis was asking "do we have the data to justify the fear?" Someone had to ask. Look at all the stupid shit we did because people were unreasonably afraid!
2022-10-21 15:13:58 @KevinBardosh Was the discussion unplanned and arose at another event? In order for this to be a productive conversation, it seems like thoughtful mediation would be necessary.
2022-10-21 15:06:21 @KevinBardosh I wonder how many (if any) of these discussions have occurred.
2022-10-21 14:13:45 @DanielHadas2 Your threads are too good. I look forward to my daily meditation on them.I hope you collect them in one place at some point.
2022-10-29 16:25:31 @DanielHadas2 @JacobAShell Do you two have any comment on how the latest report "bombshell" requires mistranslating Chinese? (I feel like this is Daniel's lane)
2022-10-29 01:41:21 @ajlamesa And no evidence of significant transmission on public transport!
2022-10-28 13:49:42 @TheNickFoy I mean, legs are coming soon, Nick. Legs are coming. https://t.co/lcluFhTEmB
2022-10-27 17:25:41 @fhuszar @maosbot Don't disagree there. And also that AI Alignment is a nonsense rationalist fever dream based in bad science fiction.
2022-10-27 17:24:30 @fhuszar @maosbot On it!
2022-10-27 17:11:52 @fhuszar @maosbot 300 people working on "The AI Alignment Problem" is 300 too many.
2022-10-27 14:10:32 @quantum_aram Not today at least...
2022-10-27 13:46:19 @AChandra_TO @phl43 LOL. I need to get Derek Thompson to apologize.But right: I think the "peer review" canard is only used when convenient.
2022-10-27 13:41:43 @AChandra_TO @phl43 I don't know, MSM runs with preprints when it suits them.Just last week, the Economist picked up that bizarre Lab Leak preprint written by three conspiracy theorists.
2022-10-27 13:38:51 Even Elon can't escape. “The best minds of my generation are thinking about how to make people click ads.” https://t.co/0tDidHYwLn
2022-10-27 13:35:24 @phl43 Interesting perspective from computer science: CS has convinced deans and managers that *conference* papers can and should be counted as journal publications. Pro: As a result, there are no prestigious CS journals.Con: Our conference review system is a shitshow.
2022-10-27 13:31:51 @phl43 Agreed, I have been reading variations of this anti-journal argument for over a decade and no one can form a cogent rebuttal.Unfortunately, one reason journals continue to exist is publications are required for promotion in academic and industrial research.
2022-10-27 13:25:11 @phl43 I don't agree that social media ends up being better than journals (both have warts), but I also can't believe how many journals get away with charging *authors* thousands of dollars per paper. That alone is an argument for their end.
2022-10-27 00:39:52 @VPrasadMDMPH @johndefeo One of the authors thinks Monkeypox was lab leaked.
2022-10-26 14:02:04 Hell has frozen over: I wrote a blog post saying a bunch of nice things about neural nets (and art?). https://t.co/wGTqgcZG2N
2022-10-25 19:36:23 @FO_ASchatz This can happen even with no statistics:(Rush score, pass score)Team 1: (4,1)Team 2: (3,3)Team 3: (1,4)Team 2 is not best in either score, but it has the best added scores.
2022-10-25 19:05:42 @jaycaspiankang There was an earthquake?
2022-10-24 16:30:30 @LaurentLessard @CSSIEEE Uh, that's totally bizarre. What a waste of speaker time.Unthawing from our frozen virtual pandemic time is going to take a while to shake out.
2022-10-24 16:26:28 @LaurentLessard @CSSIEEE Wait, are there going to be videos instead of talks at the in-person conference?
2022-10-24 14:14:36 @DanielHadas2 We have also created a bizarre literature on Long Covid. Long Covid arose on Facebook and became popularized by a poorly reported story by Ed Yong. It’s a perfect example of science not only behaving like the media, but taking its queues from the media. 3/3
2022-10-24 14:14:20 @DanielHadas2 Since then, we have learned about vaccines. We can quibble with policies, but these papers are good. There have been good papers on treatment as well, in particular from the RECOVERY trial in the UK. 2/3
2022-10-24 14:13:39 @DanielHadas2 Yeah, the following the science meme certainly added too much importance to the science.It's remarkable that we basically knew everything about this virus by May 1, 2020. And we knew a lot by March 1, 2020. 1/3
2022-10-23 14:18:58 Solid advice for "data scientists," no matter which topic they are studying. https://t.co/iGR2neHSKO
2022-10-23 04:03:53 @jaycaspiankang Tarik Glenn is pretty cool.
2022-10-23 04:03:21 @jaycaspiankang Curious who you have as the 2nd cool person after Marshawn Lynch.
2022-10-22 21:13:38 @JacobAShell @DanielHadas2 @Spear_Owl But it could alternatively be social-media-driven mass formation psychosis not cooperating with medicine's preexisting analytical categories.
2022-10-22 21:04:29 @JacobAShell @Spear_Owl @DanielHadas2 Though they were influenza pandemics, rather than covid pandemics, the global mortality associated with the 1957 and 1968 pandemics are both of similar magnitude to covid. The main difference being Asia was hit harder and North America and Europe significantly less so.
2022-10-22 20:51:56 @JacobAShell @DanielHadas2 @Spear_Owl Post-viral sequelae are indeed a thing, and teasing out their etiology has always been a challenge. Here's one survey from 1987: https://t.co/7igETSURw5
2022-10-22 20:35:30 @JacobAShell @DanielHadas2 @Spear_Owl No, I know. That's why I wrote it. I didn't mean it as an attack but as a lament: we're globally stuck doing obviously stupid things when common sense clearly dictates we should do the opposite.
2022-10-22 20:33:13 @DanielHadas2 @Spear_Owl @JacobAShell I see. Thanks for the clarification.I think it's interesting that I see the opposite direction, and I should think more about why.
2022-10-22 20:28:59 @JacobAShell @DanielHadas2 @Spear_Owl I do not disagree at all. But there are many other things I wish were a global consensus: that we shouldn't infringe on the habitats of wildlife, that humankind does not yet possess the ability to stop all respiratory viruses,...
2022-10-22 20:27:24 @JacobAShell @DanielHadas2 @Spear_Owl In what way are the symptoms odd?
2022-10-22 20:26:37 @DanielHadas2 @Spear_Owl @JacobAShell How does one understand "the science" without "the people and the politics?"
2022-10-22 20:26:05 @DanielHadas2 @Spear_Owl @JacobAShell There is a clear dichotomization in politics between those who are willing to entertain the lab leak and those who are not.
2022-10-22 20:25:32 @DanielHadas2 @Spear_Owl @JacobAShell Here in this thread, @JacobAShell is saying that the virology community looks guilty, hence, without understanding their arguments, they clearly must be wrong on "the science."
2022-10-22 20:25:17 @DanielHadas2 @Spear_Owl @JacobAShell We don't have to belabor into this point, but don't you think ad hominem is exactly what the lab leak debate is?
2022-10-22 20:14:16 @DanielHadas2 @Spear_Owl @JacobAShell The only way that scientific paradigm shifts happen is with new experiments that slowly chip away at an infinite landscape of possibility.Every once in a while we find something amazing, but usually, we just muddle through minutiae while we wait for the new revolution.
2022-10-22 20:13:27 @DanielHadas2 @Spear_Owl @JacobAShell My read of the post-Kuhn landscape is that there is no genuine scientific aporia but only politics.
2022-10-22 20:04:20 @DanielHadas2 @Spear_Owl @JacobAShell But there is no way to prove it either way! There will never be new data. We're all stuck scrutinizing mirages and arguing from our political corners.
2022-10-22 20:02:04 @JacobAShell @Spear_Owl @DanielHadas2 Or maybe, just like Latour told us, scientists are a bunch of crazy weirdos, and they shouldn't be out on Twitter showing the world their whole asses?
2022-10-22 20:00:00 @Spear_Owl @DanielHadas2 @JacobAShell But that thread has input from dozens of other scientists. It's not just Andersen and Rassmussen.And you can assert "there's a cartel here suppressing science!."And we will continue to fight because no one can do a real experiment to prove anything about the fucking lab leak.
2022-10-22 19:58:08 @Spear_Owl @DanielHadas2 @JacobAShell Yeah, I don't disagree. He and Angela Rassmussen both behave like children on here, and I don't understand why they can't see how harmful this behavior is.
2022-10-22 19:53:54 @JacobAShell @Spear_Owl @DanielHadas2 I think everyone on this stupid website protests too much.If anything, I'm more Latourian than I was at the start of the pandemic. But the outsider critics fall into the same traps as the insiders. No one is immune.
2022-10-21 20:22:02 @rdnowak @DimitrisPapail @amirvaxman_dgp Barron showed that functions f where the Fourier transform \tilde{f} satisfied the following inequality were well approximated by a small sum of sigmoidal functions. He also showed the number of required functions to achieve an eps-approx in L2 scaled like 4 B^2/eps^2. https://t.co/QUV74jG7Xh
2022-10-21 20:15:51 @rdnowak @DimitrisPapail @amirvaxman_dgp I agree with Rob. Barron's results were describing the sorts of functions that were well approximated by neural nets. (Cybenko's argument is certainly tidier.)
2022-10-21 16:06:14 @BenMazer @drvictoriafox @GYamey He didn't say either of these things. I understand that Ioannidis is a convenient punching bag and you don't want to engage with what he was saying in good faith.
2022-10-21 15:57:11 @BenMazer @drvictoriafox @GYamey Everyone was running around in fear, as you have admitted you were yourself. Ioannidis was asking "do we have the data to justify the fear?" Someone had to ask. Look at all the stupid shit we did because people were unreasonably afraid!
2022-10-21 15:13:58 @KevinBardosh Was the discussion unplanned and arose at another event? In order for this to be a productive conversation, it seems like thoughtful mediation would be necessary.
2022-10-21 15:06:21 @KevinBardosh I wonder how many (if any) of these discussions have occurred.
2022-10-21 14:13:45 @DanielHadas2 Your threads are too good. I look forward to my daily meditation on them.I hope you collect them in one place at some point.
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-24 02:52:50 @JSEllenberg I swear by this rule, and have suffered no resulting FOMO.
2022-11-24 02:43:37 @JSEllenberg If you wouldn't consider messing up your calendar next week to make time for an event, you shouldn't say yes to no matter when it is.
2022-11-23 14:54:49 @DimitrisPapail In the longer tech report they link to, there is this plot which again shows accepted and rejected papers are likely indistinguishable. https://t.co/CpW50usIVN
2022-11-23 14:54:03 @DimitrisPapail I agree with you. I don't see any other interpretation of the plot they give. https://t.co/CVlyHQMvWS
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-25 16:30:31 @JacobAShell @samhaselby 100%. Effective Altruism is essentially "Trust The Science" applied to everything.
2022-11-25 16:13:18 @JacobAShell @samhaselby And you can scroll through the grants here to see how much money GiveWell, one of the most famous EA organizations, has given to Universities. https://t.co/vnb85mCR8N
2022-11-25 16:08:48 @JacobAShell @samhaselby Stanford, of course, has a "sustainability accelerator" https://t.co/DZAdorQTjZ
2022-11-25 16:04:24 @JacobAShell @samhaselby There's this group at Berkeley that is the "Center for Effective Global Action" https://t.co/CQkPyFt1cn
2022-11-25 16:04:02 @JacobAShell @samhaselby The ideas of being "most effective" in charity are pervasive on campus. This influential group at MIT was founded by two Nobel Prize winners and aims "to reduce poverty by ensuring that policy is informed by scientific evidence." https://t.co/9tSCNmL9Ye
2022-11-25 15:59:22 @JacobAShell @samhaselby I get that. There's the weirdo online version of utilitarianism that is what most people call EA. But they profess the most extreme version of mainstream economic tenets. In econ, utility maximization is taken as "correct." There's none of the associated moral philosophy.
2022-11-25 15:05:43 @JacobAShell @samhaselby All of economics is essentially promoting effective altruism. Moreover, EA is very popular in STEM departments, especially computer science. And STEM + Econ hold far more power at universities than most humanities departments.
2022-11-24 02:52:50 @JSEllenberg I swear by this rule, and have suffered no resulting FOMO.
2022-11-24 02:43:37 @JSEllenberg If you wouldn't consider messing up your calendar next week to make time for an event, you shouldn't say yes to no matter when it is.
2022-11-23 14:54:49 @DimitrisPapail In the longer tech report they link to, there is this plot which again shows accepted and rejected papers are likely indistinguishable. https://t.co/CpW50usIVN
2022-11-23 14:54:03 @DimitrisPapail I agree with you. I don't see any other interpretation of the plot they give. https://t.co/CVlyHQMvWS
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-25 16:30:31 @JacobAShell @samhaselby 100%. Effective Altruism is essentially "Trust The Science" applied to everything.
2022-11-25 16:13:18 @JacobAShell @samhaselby And you can scroll through the grants here to see how much money GiveWell, one of the most famous EA organizations, has given to Universities. https://t.co/vnb85mCR8N
2022-11-25 16:08:48 @JacobAShell @samhaselby Stanford, of course, has a "sustainability accelerator" https://t.co/DZAdorQTjZ
2022-11-25 16:04:24 @JacobAShell @samhaselby There's this group at Berkeley that is the "Center for Effective Global Action" https://t.co/CQkPyFt1cn
2022-11-25 16:04:02 @JacobAShell @samhaselby The ideas of being "most effective" in charity are pervasive on campus. This influential group at MIT was founded by two Nobel Prize winners and aims "to reduce poverty by ensuring that policy is informed by scientific evidence." https://t.co/9tSCNmL9Ye
2022-11-25 15:59:22 @JacobAShell @samhaselby I get that. There's the weirdo online version of utilitarianism that is what most people call EA. But they profess the most extreme version of mainstream economic tenets. In econ, utility maximization is taken as "correct." There's none of the associated moral philosophy.
2022-11-25 15:05:43 @JacobAShell @samhaselby All of economics is essentially promoting effective altruism. Moreover, EA is very popular in STEM departments, especially computer science. And STEM + Econ hold far more power at universities than most humanities departments.
2022-11-24 02:52:50 @JSEllenberg I swear by this rule, and have suffered no resulting FOMO.
2022-11-24 02:43:37 @JSEllenberg If you wouldn't consider messing up your calendar next week to make time for an event, you shouldn't say yes to no matter when it is.
2022-11-23 14:54:49 @DimitrisPapail In the longer tech report they link to, there is this plot which again shows accepted and rejected papers are likely indistinguishable. https://t.co/CpW50usIVN
2022-11-23 14:54:03 @DimitrisPapail I agree with you. I don't see any other interpretation of the plot they give. https://t.co/CVlyHQMvWS
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-28 16:41:01 @phl43 Exactly who I was thinking of. https://t.co/e83zJZaP8w There's also Michael Mina who left academia to join a testing company.
2022-11-28 16:38:27 @phl43 Can we name names?
2022-11-27 15:27:40 @drjohnm Medicine is the only human-facing science that regularly finds interventions with ORs of greater than 10. This, much more than RCTs, is why we can make advances.
2022-11-27 15:12:54 @drjohnm @kaulcsmc @stephensenn Definitely RCT. I have always wondered why the same people who disparage observational studies are proponents of meta-analysis. They share most of the same biases.
2022-11-27 15:10:52 @drjohnm Just off the top of my head: smoking cessation antibiotics vitamins B3, D (A and B1 discovered in 1910s) mass production of vaccines chemotherapy MRI ibuprofen ...
2022-11-25 16:30:31 @JacobAShell @samhaselby 100%. Effective Altruism is essentially "Trust The Science" applied to everything.
2022-11-25 16:13:18 @JacobAShell @samhaselby And you can scroll through the grants here to see how much money GiveWell, one of the most famous EA organizations, has given to Universities. https://t.co/vnb85mCR8N
2022-11-25 16:08:48 @JacobAShell @samhaselby Stanford, of course, has a "sustainability accelerator" https://t.co/DZAdorQTjZ
2022-11-25 16:04:24 @JacobAShell @samhaselby There's this group at Berkeley that is the "Center for Effective Global Action" https://t.co/CQkPyFt1cn
2022-11-25 16:04:02 @JacobAShell @samhaselby The ideas of being "most effective" in charity are pervasive on campus. This influential group at MIT was founded by two Nobel Prize winners and aims "to reduce poverty by ensuring that policy is informed by scientific evidence." https://t.co/9tSCNmL9Ye
2022-11-25 15:59:22 @JacobAShell @samhaselby I get that. There's the weirdo online version of utilitarianism that is what most people call EA. But they profess the most extreme version of mainstream economic tenets. In econ, utility maximization is taken as "correct." There's none of the associated moral philosophy.
2022-11-25 15:05:43 @JacobAShell @samhaselby All of economics is essentially promoting effective altruism. Moreover, EA is very popular in STEM departments, especially computer science. And STEM + Econ hold far more power at universities than most humanities departments.
2022-11-24 02:52:50 @JSEllenberg I swear by this rule, and have suffered no resulting FOMO.
2022-11-24 02:43:37 @JSEllenberg If you wouldn't consider messing up your calendar next week to make time for an event, you shouldn't say yes to no matter when it is.
2022-11-23 14:54:49 @DimitrisPapail In the longer tech report they link to, there is this plot which again shows accepted and rejected papers are likely indistinguishable. https://t.co/CpW50usIVN
2022-11-23 14:54:03 @DimitrisPapail I agree with you. I don't see any other interpretation of the plot they give. https://t.co/CVlyHQMvWS
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-11-28 16:41:01 @phl43 Exactly who I was thinking of. https://t.co/e83zJZaP8w There's also Michael Mina who left academia to join a testing company.
2022-11-28 16:38:27 @phl43 Can we name names?
2022-11-27 15:27:40 @drjohnm Medicine is the only human-facing science that regularly finds interventions with ORs of greater than 10. This, much more than RCTs, is why we can make advances.
2022-11-27 15:12:54 @drjohnm @kaulcsmc @stephensenn Definitely RCT. I have always wondered why the same people who disparage observational studies are proponents of meta-analysis. They share most of the same biases.
2022-11-27 15:10:52 @drjohnm Just off the top of my head: smoking cessation antibiotics vitamins B3, D (A and B1 discovered in 1910s) mass production of vaccines chemotherapy MRI ibuprofen ...
2022-11-25 16:30:31 @JacobAShell @samhaselby 100%. Effective Altruism is essentially "Trust The Science" applied to everything.
2022-11-25 16:13:18 @JacobAShell @samhaselby And you can scroll through the grants here to see how much money GiveWell, one of the most famous EA organizations, has given to Universities. https://t.co/vnb85mCR8N
2022-11-25 16:08:48 @JacobAShell @samhaselby Stanford, of course, has a "sustainability accelerator" https://t.co/DZAdorQTjZ
2022-11-25 16:04:24 @JacobAShell @samhaselby There's this group at Berkeley that is the "Center for Effective Global Action" https://t.co/CQkPyFt1cn
2022-11-25 16:04:02 @JacobAShell @samhaselby The ideas of being "most effective" in charity are pervasive on campus. This influential group at MIT was founded by two Nobel Prize winners and aims "to reduce poverty by ensuring that policy is informed by scientific evidence." https://t.co/9tSCNmL9Ye
2022-11-25 15:59:22 @JacobAShell @samhaselby I get that. There's the weirdo online version of utilitarianism that is what most people call EA. But they profess the most extreme version of mainstream economic tenets. In econ, utility maximization is taken as "correct." There's none of the associated moral philosophy.
2022-11-25 15:05:43 @JacobAShell @samhaselby All of economics is essentially promoting effective altruism. Moreover, EA is very popular in STEM departments, especially computer science. And STEM + Econ hold far more power at universities than most humanities departments.
2022-11-24 02:52:50 @JSEllenberg I swear by this rule, and have suffered no resulting FOMO.
2022-11-24 02:43:37 @JSEllenberg If you wouldn't consider messing up your calendar next week to make time for an event, you shouldn't say yes to no matter when it is.
2022-11-23 14:54:49 @DimitrisPapail In the longer tech report they link to, there is this plot which again shows accepted and rejected papers are likely indistinguishable. https://t.co/CpW50usIVN
2022-11-23 14:54:03 @DimitrisPapail I agree with you. I don't see any other interpretation of the plot they give. https://t.co/CVlyHQMvWS
2022-11-22 15:40:24 @phl43 This battle has been eye opening for me, as I wasn't sure if I disliked the polling nerds or the disinformation nerds more. But upon reflection, it has always been clear who were the bigger idiots.
2022-11-22 04:41:53 @phl43 But are you for Bayes' Theorem? https://t.co/eCodosHExy
2022-11-21 18:12:11 What is the right way to draw an ROC curve for a human annotator?
2022-11-21 14:58:57 @jaycaspiankang It will depend on the econometric analysis of the value of cat land. I'm sure the Future of Life folks are writing a whitepaper on it.
2022-11-21 14:55:17 @drjohnm As an outsider, I've been shocked by the lack of data transparency in medical publications. Curious if you have thoughts on why data sharing is not widely demanded. One of the main excuses seems to be "patient privacy." This seems like a cop out!
2022-11-21 14:40:12 @jaycaspiankang Don't leave us hanging, man, what's Chapter 6? And why is it "The Feral Cats of the Oakland Coliseum?"
2022-11-20 15:16:11 @DanielHadas2 @DrJMatthewRhett Yes. Change *seems like* to *is* and we're 100% in agreement.
2022-11-20 04:46:58 @jaycaspiankang No matter what happens in that game, we all lose.
2022-11-19 16:01:56 @Tweetermeyer That's a good analogy! But it's a bit weird here as Twitter has been around for over a decade without Elon. It's a weird intersection: (a) Elon has been full of shit at Tesla, (b) Twitter, at its core, is very simple! And the engineering at Twitter has been shoddy forever...
2022-11-19 15:53:41 @Tweetermeyer Can't both things be true? (1) apps can feel simple but require engineering expertise to build and maintain and (2) maintaining the main timeline on Twitter can be done with a skeleton crew.
2022-11-19 15:49:29 @jaycaspiankang Yeah, I'd have to ask on medtwitter, but that does seem a little bit high, even for chief neurosurgeon.
2022-11-19 15:18:25 @jaycaspiankang I was hoping this was a CS professor, but turns out it's the chair of neurosurgery at UCI med-school. I didn't expect that twist.
2022-11-19 15:16:56 @Hammbear2024 @awgaffney @BenMazer Thanks for your opinion, random anonymous account on this crumbling website we all can't log off from!
2022-11-19 15:07:24 @awgaffney @Hammbear2024 @BenMazer Insinuating that Jean Noble trash is not particularly professional.
2022-11-19 14:55:49 @DrCMcMaster @hscott61 @adamcifu @Sensible__Med @RogueRad I don't disagree, but (a) I always do! ARR always has CIs, so NNT does too. And these CIs are usually more reliable than those on RRR. (b) In commentary and discussion, people quote RRR and ARR all the time without citing the CIs.
2022-11-18 18:35:26 @adamcifu @Sensible__Med @RogueRad Oh my! I came to #medtwitter too late and missed the really good stuff.
2022-11-18 18:27:36 @adamcifu @Sensible__Med @RogueRad Have you written about why you like NNT?
2022-11-18 17:43:35 @fhuszar This still holds up. https://t.co/LTetNgQGtq
2022-11-18 16:12:22 @phl43 Scoops like this require some deep reporting tho. https://t.co/GE4VCZ5Ryh
2022-11-18 16:03:08 @SeanTrende The classic case is when you want to test if one group has a significantly different mean than all of the other groups. You'd do something like a t-test for two groups, and ANOVA if you had more than 2. You'd still run lm under the hood, but the *test* would be an ANOVA.
2022-11-18 15:53:20 @SeanTrende Yes, this is one of the most frustrating parts of statistics: There are thousands of tests and analyses that reduce to fitting linear models. Almost all stats can be done by calling lm. It's how you interpret the output that's different. What are you trying to test?
2022-11-18 15:00:38 I've seldom share personal news on here, so might as well before Twitter shuts down. Wordle 517 3/6
2022-11-17 14:47:01 @awgaffney When pooled differently, a UCSF study found food services and agriculture were found to have the largest outbreaks. https://t.co/brRm1S96ZR 2/2
2022-11-17 14:46:52 @awgaffney This assertion is based on a rather bizarrely focused MMWR study that compared transportation to "all industries." 1/2 https://t.co/5LAELsQahj
2022-11-16 16:26:19 @can SBF went to MIT!
2022-11-16 15:15:32 @kylenw @jwherrman @jaycaspiankang but you stayed up all night in the rain and got to see Jimi play the anthem...
2022-11-16 14:50:58 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM @ianmSC You'd also have to accept this as evidence. https://t.co/DXpiszZTAq
2022-11-16 14:50:15 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Look, if you accept that this study is evidence, then you have to also accept that every tweet by @ianmSC from the past 3 years is evidence. And @ianmSC has far more evidence on his side.
2022-11-16 14:49:18 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM I mean, I realize I'm Don Quixote fighting this battle, but I will die on the hill that regression cannot establish causation.
2022-11-16 14:43:57 @jaycaspiankang Also, how does @jwherrman still have a gif avi?
2022-11-16 14:43:09 @jaycaspiankang They are finally admitting error and bringing back Vine. This requires the hardest of coreness to code.
2022-11-16 14:38:47 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM But the trends of epidemics *are* nonlinear. they strongly depend on the susceptible population, the current R of the diesase, the graph structure, etc., etc. You can't "control" for these unmeasured factors by adding dummy variables to a linear regression.
2022-11-16 14:37:25 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM Here's where we disagree: observational studies find whatever authors want to find. No "causal regression" method is immune from researcher bias. This applies to all obs. studies, but, for whatever reason, masking is the only causal inference topic that consistently goes viral.
2022-11-16 14:33:43 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM The area covered is not small. It's most of eastern Massachusetts. I'm not sure if you are familiar with the area, but there are significant differences between Chelsea and Lexington.
2022-11-16 02:31:35 @Drjhoffmanmiami @VPrasadMDMPH @JoeGramc To clarify: there are several RCTs of healthcare workers showing no benefit to N95 use in reducing influenza-like illness. There are many explanations for this, but that's our current evidence base.
2022-11-15 23:47:41 @zshahn @ElGee35 @fperrywilson @EpiEllie @NEJM DiD assumes a linear fixed-effects model. Epidemic dynamics means the DiD assumptions do not apply. Here's a good discussion of why DiD is bad for evaluating NPIs: https://t.co/7D6rdIaBpi
2022-11-15 23:28:29 @JoeGramc @VPrasadMDMPH Vinay's survey covers N95s. Though people claim otherwise, the evidence that N95s reduce infection is also incredibly weak and the assertions are entirely based on mechanistic lab studies.
2022-11-15 21:01:47 @walidgellad Haven't all of the non-randomized studies so far shown no benefit for vaxxed under 65s? Why should we expect to see something different in a randomized study?
2022-12-07 14:44:15 @politicalmath @TallBlondeGuy FWIW, it was 0.64% in the treatment arm, 0.70% in the control arm. There were also significant biases introduced by how subject were recruited and cases were ascertained. https://t.co/CVVvkYnaB5
2022-12-07 14:01:26 @politicalmath 2020 mask research: But covid is different! 2022 mask research: masks prevent flu and RSV.
2022-12-07 14:44:15 @politicalmath @TallBlondeGuy FWIW, it was 0.64% in the treatment arm, 0.70% in the control arm. There were also significant biases introduced by how subject were recruited and cases were ascertained. https://t.co/CVVvkYnaB5
2022-12-07 14:01:26 @politicalmath 2020 mask research: But covid is different! 2022 mask research: masks prevent flu and RSV.
2022-12-08 14:19:13 @dnunan79 My heuristic: It has "e-values" in the sentence, so be very suspicious!
2022-12-08 14:18:21 @phl43 Bureaucracy never shrinks, and bullshit always begets more bullshit. But what do you think about the fact that Twitter didn't collapse when all the software engineers left?
2022-12-08 14:09:11 @JacobAShell @DanielHadas2 100% this. We're seeing the same thing at the UC. If you can get away with charging full price for sitting in your bedroom with a laptop, why couldn't you get away with simply giving everyone an A who's on your roster?
2022-12-07 14:44:15 @politicalmath @TallBlondeGuy FWIW, it was 0.64% in the treatment arm, 0.70% in the control arm. There were also significant biases introduced by how subject were recruited and cases were ascertained. https://t.co/CVVvkYnaB5
2022-12-07 14:01:26 @politicalmath 2020 mask research: But covid is different! 2022 mask research: masks prevent flu and RSV.
2022-12-08 14:19:13 @dnunan79 My heuristic: It has "e-values" in the sentence, so be very suspicious!
2022-12-08 14:18:21 @phl43 Bureaucracy never shrinks, and bullshit always begets more bullshit. But what do you think about the fact that Twitter didn't collapse when all the software engineers left?
2022-12-08 14:09:11 @JacobAShell @DanielHadas2 100% this. We're seeing the same thing at the UC. If you can get away with charging full price for sitting in your bedroom with a laptop, why couldn't you get away with simply giving everyone an A who's on your roster?
2022-12-07 14:44:15 @politicalmath @TallBlondeGuy FWIW, it was 0.64% in the treatment arm, 0.70% in the control arm. There were also significant biases introduced by how subject were recruited and cases were ascertained. https://t.co/CVVvkYnaB5
2022-12-07 14:01:26 @politicalmath 2020 mask research: But covid is different! 2022 mask research: masks prevent flu and RSV.
2022-12-08 14:19:13 @dnunan79 My heuristic: It has "e-values" in the sentence, so be very suspicious!
2022-12-08 14:18:21 @phl43 Bureaucracy never shrinks, and bullshit always begets more bullshit. But what do you think about the fact that Twitter didn't collapse when all the software engineers left?
2022-12-08 14:09:11 @JacobAShell @DanielHadas2 100% this. We're seeing the same thing at the UC. If you can get away with charging full price for sitting in your bedroom with a laptop, why couldn't you get away with simply giving everyone an A who's on your roster?