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
2023-05-19 19:00:00 CAFIAC FIX
2023-05-21 19:00:00 CAFIAC FIX
2023-04-21 00:00:01 CAFIAC FIX
2023-03-05 10:00:00 CAFIAC FIX
2023-03-02 22:00:00 CAFIAC FIX
2023-02-27 01:00:00 CAFIAC FIX
2023-01-30 01:00:00 CAFIAC FIX
2023-01-07 18:28:51 @myx Exactly!
2023-01-07 05:15:25 Tetris also runs in Life: https://codegolf.stackexchange.com/questions/11880/build-a-working-game-...
2023-01-07 05:14:58 A Lisp interpreter... implemented in the Game of Life: https://woodrush.github.io/blog/posts/2022-01-12-lisp-in-life.htmlSupports macros &
2023-01-07 01:47:32 @abdullahkhalids Thanks!
2023-01-06 04:31:26 @Cameo That'
2023-01-06 04:25:57 @Cameo Do I understand aright, that in that liminal state of mind they start to lose consciousness of their own disbelief?Much like actors sometimes report. E.g., one of the leads in "
2023-01-06 04:21:47 One of my favourite papers, an adversarial collaboration on the existence of psychic powers (between a believer and skeptic)Shown: abstract, and a particularly amusing &
2023-01-06 04:18:00 @Cameo That'
2023-01-06 04:17:24 @Cameo I'
2023-01-06 04:17:10 @hrheingold @Cameo Oh, wow!
2023-01-06 04:16:52 @Cameo I'
2023-01-06 04:11:28 @Cameo Wow, what fascinating topics! What'
2023-01-06 04:09:49 @Cameo Ah, here it is: http://www.richardwiseman.com/resources/staring1.pdf
2023-01-06 04:05:18 @Cameo I wonder how many are both sincere and thoughtful? There'
2023-01-06 04:00:52 @Cameo This tickles me!
2023-01-06 01:10:23 I'
2023-01-05 18:44:40 Leo Szilard on how to slow down science: https://rootsofprogress.org/szilard-on-slowing-science
2023-01-05 05:19:52 @MarisOzols So: I felt I first really understood entropy for the first time as the answer to a question about source coding. I'
2023-01-05 05:17:43 @MarisOzols Thank you for all these lovely examples! Discovery fiction aside, an interesting point about taste: my first introduction to entropy was through various axiomatic approaches. But I didn'
2023-01-04 06:05:13 I continue to marvel that predict-the-next-word/token, with enough data and parameters, turns out to capture so much of human thought.
2023-01-04 05:17:02 The Third and the Seventh: https://vimeo.com/7809605This has been one of my favourite artworks since I first saw it, about 12 years or so ago. It'
2023-01-03 22:41:27 A note on "
2023-01-02 18:18:02 The first rule of Bite Club is that you do talk about your experience (of being bitten: by a shark, a lion, ...): https://www.abc.net.au/news/2023-01-01/bite-club-welcomes-man-mauled-by-...
2023-01-02 06:37:55 The damping ball in the Taipei 101 Tower, during a 6.8 earthquake: https://www.youtube.com/watch?v=k51nFin7Wzc&
2023-01-01 23:11:09 Fascinating: an online tool to play blindfold chess: https://chessinsights.org/blindfold/
2023-01-01 16:52:06 @scott_bot @TedUnderwood Calculators, all over again.
2022-12-31 20:35:56 There are caveats, of course. Often internal conflict will be used to create drama and tension. The son who doesn'
2022-12-31 20:34:05 Something I find fascinating about movies is how (relatively) internally unconflicted and clear about their goals the main characters often areFind the treasure. Rescue the princess. Solve the murder. Defeat the enemyIt lends tremendous clarity to the plot/actionBy contrast, most people are much more conflicted and less clear about their goals
2022-12-30 20:19:46 https://zhengdongwang.com/2022/12/28/2022-letter.html
2022-12-29 22:26:30 @gordon An idea I really like is that the biosphere is a bit like discovering a new advanced civilization with a billion years of programming experience, and a trillion trillion trillion lines of code, an immense reservoir to simply explore and learn from in wonder.
2022-12-29 22:25:07 @phonner Great line!
2022-12-29 18:53:03 @wesc Agree: https://scienceplusplus.org/metascience/index.html
2022-12-29 02:01:52 @robpike I found movetodon.org quite helpful in seeing which people I follow on Twitter are here. Lots and lots of scientists, especially.
2022-12-29 02:00:12 @robpike Seems fine to me. Try following more than 11 people.
2022-12-29 01:48:16 Surprised by how relieving it is to be able to edit typos!
2022-12-29 01:46:51 @andy_matuschak Most broad critiques of self-help books are really the complainer saying "
2022-12-28 23:17:48 Anyways, I hope Mastodon or something similar can grow and flourish and improve as part of the commons, and won'
2022-12-28 23:16:14 The first example I thought about was Linux. At the time - I think it was ~2008 - Linux'
2022-12-28 23:14:23 A friend who has thought deeply about open source observed to me once that a common exception to this pattern is when companies want to maintain something in the public domain as a way of commoditizing their complement.He said, roughly: "
2022-12-28 23:11:32 The basic pattern tends to be: take an open source project, add something very useful and attractive to users, which can be provided with VC money and a profit model, but is hard to do not-for-profit. Then gradually take over effective ownership of the space, finding some bottleneck you control.This is a very common pattern.
2022-12-28 23:06:15 Curious: open source projects, like many public goods, tend to be subject to enclosure (privatization).I hear stories about VCs investing in Mastodon servers, and look at the (many!) private companies based on it, like Truth Social and Gab. And I wonder if in the long run it will be co-opted by some other party, with a profit motive.It'
2022-12-28 22:30:26 @gojomo @moultano One of my favourite examples: https://mastodon.social/@michael_nielsen/109593663269888818
2022-12-28 22:29:50 Former Australian High Court Justice Michael Kirby'
2022-12-28 22:25:06 @gojomo @moultano A pattern I quite enjoy is when someone explains, even very briefly, three different and somewhat opposed but plausible ways of thinking about something.I think of it as the 1-2-infinity pattern: once I hear three ways that are all useful, but in tension, I find it easy to remember that still other ways are likely possible.
2022-12-28 22:16:10 Something I find very inspiring is older people [*] taking on new adventures. Travelling, taking up new creative work, starting new ventures - or simply trying things like Mastodon. [*] For purposes of this post, let'
2022-12-28 20:33:35 @danb Used movtodon.org!
2022-12-28 20:05:11 @kipply What'
2022-12-28 18:31:56 @numist @rands I'
2022-12-28 17:52:08 @rands I hope the developer has a good support network around him.
2022-12-28 16:26:03 @benoit @davidmanheim Please read LeCun'
2022-12-28 08:48:44 A curious thing about beliefs about AGI: people with a detailed, expert-level ability to understand and build current systems disagree wildly &
2022-12-28 08:00:00 @khinsen I think the earliest reference I'
2022-12-28 07:58:21 @scott_bot (I realize I used the phrase "
2022-12-28 07:57:46 @scott_bot Curious, what do you mean by "
2022-12-28 07:19:47 @aspiringcat Yes, it'
2022-12-28 07:14:18 @khinsen Do they have an argument for that bound? Their site looks like advocacy.
2022-12-28 07:10:05 I was just looking at a (very recent) paper, by prominent authors, making arguments I first saw made more than a decade ago, and which, doubtless, were in many cases proposed long before I ever heard them.The slow building of a canon, and lack of strong enough norms around cumulative knowledge, certainly holds the field back.
2022-12-28 07:03:24 I'
2022-12-28 06:07:00 I enjoyed "
2022-12-28 05:55:02 Reminded of Chomsky &
2022-12-28 05:32:23 @lakens Yes, I'
2022-12-28 05:24:56 @lakens Curious: I'
2022-12-28 03:40:58 @wxs Also: instrumentation. I think the surveillance state goes hand-in-hand with high modernism, for people with even mild authoritarian tendencies
2022-12-28 02:43:35 Basically: if you have deep pockets, you can certainly produce a body of "
2022-12-28 02:40:45 I'
2022-12-28 02:38:29 It'
2022-12-28 01:29:10 @shriramk I noticed the same thing, and quickly quit the app.
2022-12-28 01:22:41 I say all this as someone who is a technocratic capitalist, of course! But it'
2022-12-28 01:15:52 A related question which I like very much: if you could wave a policy wand and change the Gini coefficient in your country, what would you change it to? And why?It'
2022-12-28 01:12:07 But, ofc, there'
2022-12-28 01:10:09 Related: there'
2022-12-28 01:04:46 Two of my favourite books of the year: James Scott'
2022-12-27 05:55:08 This is fascinating: https://www.schneier.com/blog/archives/2022/12/reimagining-democracy.html
2022-12-27 02:53:12 @moultano Nice point. Of course, the text can say "
2022-12-26 22:19:13 Much of the effect of the tuning of ChatGPT seems to have been to get it to answer "
2022-12-26 22:16:33 Seeing lots and lots of complaints about ChatGPT being "
2022-12-26 18:54:27 @AggieBranczyk There I just meant that standard user research seems to give a very narrow and somewhat misleading view of human beings. Good for making tiny incremental improvements to systems, but misses so much.
2022-12-26 18:41:45 @AggieBranczyk Specifically referring to the meaning in terms of human pattern recognition - an image is said to be in the uncanny valley if it'
2022-12-26 18:36:53 Also, just noticing how useful and powerful and repurpose-able the concept of the "
2022-12-26 18:35:51 Curious at how long it will take AI image-generation systems to cross the uncanny valley.
2022-12-26 18:34:53 The usual approach to "
2022-12-26 18:29:46 A recurrent pattern: people trying to build a system (product/company) will tell me they'
2022-12-25 01:51:17 @paulg @Balaji I love quote-tweeting. But it definitely has net negative effects on twitter.Interesting to think about design choices that would make most QTs positive-and-productive, rather than cheap dunks.
2022-12-25 01:45:16 Peanut butter really is shockingly tasty after the gym!This toot brought to you by afternoon HIIT
2022-12-24 23:11:58 Highly-cited researchers by country, per capita. https://elsevier.digitalcommonsdata.com/datasets/btchxktzyw/4Via: https://twitter.com/deniswirtz/status/1606449846403080193
2022-12-24 05:03:00 @wafflesid interesting idea!
2022-12-24 04:54:22 I do miss quote tweets. They'
2022-12-24 04:52:23 @melissaekline I expect this is in part due to the byte pair encoding used to tokenize the training data. It makes it extremely difficult to keep track of things of like word &
2022-12-22 19:09:41 @vtraag They are routinely on arXiv. https://arxiv.org/help/withdraw
2022-12-22 18:37:23 @intellectronica There'
2022-12-22 07:43:32 Curious what the atmosphere was like when the Declaration of Independence was signed.Did it feel momentous? Unserious, like a bunch of people playing? What was the range of moods and projections of the future of participants?(I ask, because I believe historic events often feel like nothing much at the time, to many participants.)
2022-12-22 05:43:58 Someone I admire will occasionally say of a person that they "
2022-12-22 03:32:06 @AggieBranczyk Seems inevitable (and mostly desirable) that topics drift on most servers.
2022-12-22 01:34:37 @numist Yeah, "
2022-12-22 01:29:24 @ElsaJansen (Went to a talk by Cory Doctorow, years ago, that really changed my thinking about this: https://twitter.com/michael_nielsen/status/1144445213554798593 )
2022-12-22 01:28:51 @ElsaJansen Must have been personally frustrating. I must admit to finding the underlying problem intellectually!) fascinating: what are the right principles for speech when technology radically changes its nature?
2022-12-22 01:26:59 Gonna be some amusing cases of this.The person from the circus clown server who ends up in politics!
2022-12-22 01:25:43 Rather curious at how many people are tying their Mastodon identities to their (current) professional identities. Those professional identities will change, sometimes drastically!
2022-12-22 01:20:30 Morbidly curious at when the first VC funds a company to buy up and aggregate Mastodon servers...
2022-12-22 01:16:37 @ElsaJansen Also: thanks for debunking the misinfo! I know it'
2022-12-22 01:14:19 @ElsaJansen Wow. Sorry you (&
2022-12-22 00:42:39 @ElsaJansen Curious: what was "
2022-12-22 00:01:56 @henryfarrell Yeah, they spent $140 million on it, 2015-2021, before spinning it out as an independent org in 2021: https://www.openphilanthropy.org/focus/criminal-justice-reform/
2022-12-21 23:59:07 @henryfarrell My understanding is that Criminal Justice Reform broadly (including Prison Reform) has been a long-time focus for EA - certainly, for Open Phil. I first met people at OP in 2016-ish, and it already seemed to be a significant focus then.
2022-12-21 23:34:15 @wxs And I'
2022-12-21 23:33:58 @wxs Hi! I am, honestly, not - this kind of thing wants to be centralized. But pragmatically there'
2022-12-21 22:11:51 Let'
2022-12-18 22:37:51 I won't be terribly surprised if my account gets suspended, in a retroactive application of the "rules" [sic]. Looks like an attempt at a Myspace speedrun.
2022-12-18 22:30:05 Wow: https://t.co/aN4PJKGGlh
2022-12-18 16:45:23 RT @michael_nielsen: Open to other suggestions, too. I tried Mastodon, and it's not for me at this time
2022-12-18 16:45:17 RT @michael_nielsen: If you're having trouble seeing who to follow there, may I suggest: https://t.co/ksAYfZ0LwS And: https://t.co/bggErQG…
2022-12-18 16:45:12 RT @michael_nielsen: Experimenting with being off Twitter except to announce new work. No replies, DMs etc I am actively posting social st…
2022-12-18 04:36:24 Open to other suggestions, too. I tried Mastodon, and it's not for me at this time
2022-12-18 04:19:25 If you're having trouble seeing who to follow there, may I suggest: https://t.co/ksAYfZ0LwS And: https://t.co/bggErQG2tp
2022-12-18 03:41:50 Experimenting with being off Twitter except to announce new work. No replies, DMs etc I am actively posting social stuff at: https://t.co/o07PtH2N7v. I may also experiment with other platforms I continue to post creative work at: https://t.co/fUfjMzfV4l
2022-12-16 04:22:14 Realizing that last tweet is unintentionally exclusionary: I'll be delighted if *everyone* moves over there. I'll certainly look at everyone's profiles with much interest!
2022-12-16 04:16:37 Also: I'd be delighted if people I follow move over there!
2022-12-16 04:12:10 I do have an account at https://t.co/o07PtH2N7v
2022-12-16 04:08:43 Well, this is a distraction. I think I'll deactivate for a few days, and think about other options. Maybe reactivate Twitter periodically for announcements, and move conversation elsewhere? Not sure. I'll leave this up briefly, then deactivate for a few days.
2022-12-16 04:05:49 @MarkNagelberg Best of luck waiting for the receipts on that one.
2022-12-16 03:53:00 @MarkNagelberg Where did he explicitly say this? A considerable amount of effort in that thread trying to figure out what got them banned. Most of it nothing to do with doxxing, but rather normal critical courage
2022-12-16 03:41:29 @MarkNagelberg No, they didn’t. All or most of the journalists banned simply have been critical of Musk.
2022-12-16 03:25:19 @ben_golub @dginev https://t.co/o07PtH2N7v
2022-12-16 03:22:44 RT @albrgr: This is totally insane behavior by Twitter! It’s the first thing that has made me seriously consider switching to Mastodon. htt…
2022-12-16 03:14:24 @dginev I have an account on https://t.co/0OlpAqNpTM, which suits my needs much better.
2022-12-16 03:08:02 Also: https://t.co/LYx6NrlQBy I verified this: Twitter no longer allows links to Mastodon.
2022-12-16 02:46:42 @stogachess Yeah. I was, sorta, though my opinion has rapidly declined over the past two months.
2022-12-16 02:41:18 Musk's only commitment is to "Me speech". The entire thread is worth reading. https://t.co/50utm4hNP5
2022-12-16 02:22:07 @ruthhook_ Reflecting more: I know I hesitated a bit over Pride &
2022-12-15 22:23:26 @BrydenFogelman What a coincidence! I finished it last night, too!
2022-12-15 18:30:04 Peter Norvig's notes on generative models for code and geometry: https://t.co/vppDLO0N8d https://t.co/w5CdOGVlld
2022-12-15 15:54:57 RT @michael_nielsen: “Fleabag” was amazing
2022-12-15 09:33:30 @minmaxguarantee Definitely want to watch Killing Eve!
2022-12-15 09:09:28 @ruthhook_ (At the show, not your comment!)
2022-12-15 09:08:44 @ruthhook_ Strange, too. It was so universal. It didn’t seem girly to me at all. About love and family and forgiveness and sin and hate and suffering and self hatred and loneliness and so so many other things. I realize I’m ranting here. I’m a little in shock tbh!
2022-12-15 09:04:31 @ruthhook_ Gosh, are the men ever missing out. One of the best things I’ve ever seen.
2022-12-15 08:45:39 @ruthhook_ Show. Season 1 last night, Season 2 tonight. I’d travel to see the play. Must check where it’s playing, if at all
2022-12-15 08:43:58 @intellectronica I’m not sure I could take it. So wrung out!
2022-12-15 08:43:23 @ruthhook_ Oh, curious: why? I guess that’s kind of an impossible question. But my sense of self is that Fleabag is very much my kind of thing
2022-12-15 08:38:50 “Fleabag” was amazing
2022-12-15 04:35:13 @ftlsid You're thinking too small. Imagine something 10x or 100x larger than the environmental movement, arrayed against AI. The environmental movement was *created*.
2022-12-15 04:24:56 (Again: I have no strong opinion - it's complicated to balance. Just pointing out what I see.)
2022-12-15 04:23:35 More thinking in DMs: https://t.co/jM85EfwB9T
2022-12-15 03:47:11 @quotidiania True for lots of activities, I think. We're all mimics, much of the time, but exploring variation :-)
2022-12-15 01:18:58 In DMs: https://t.co/f5ESQqUaeK
2022-12-15 01:17:55 @noampomsky TBC, limerence is great fun. But it's much better when it turns into love than into codependency. All in-my-experience-&
2022-12-15 01:01:41 Incidentally, the "Frame Breaking Act" has to have been one of the more barbaric incidents in the history of technocracy: https://t.co/2vr10e78ly
2022-12-15 01:00:16 Not offering an opinion, merely description. I *am* thinking through my opinion. But it's very complicated. I believe I would have strongly supported textile machinery in 1811...
2022-12-15 00:58:06 Trending on Artstation (At some point I expect there to be an organized and powerful anti-AI movement. I won't be surprised if the current anti- sentiment in the art community is a launchpad.) https://t.co/bSpHLJeBRn
2022-12-14 23:35:16 @noampomsky Hmm. Sounds like limerence or codependence, I'm afraid.
2022-12-14 22:24:58 (He says, as he asks for help rewriting something in a simpler and more striking way.)
2022-12-14 22:24:26 ChatGPT's first suggested prompt always seems a bit like a personal threat... https://t.co/zouG8KmDDJ
2022-12-14 20:25:09 @BasilHalperin @albrgr I just linked to a source for ~20 industry consensus reports, going back at least to the mid 90s, probably further.
2022-12-14 19:37:47 @albrgr I suspect it would be a history of the failures of long-run projections of computing costs I've seen many such projections over the years
2022-12-14 02:32:39 I find myself oddly moved by God's tolerance.
2022-12-14 02:32:04 I asked ChatGPT to imagine it was God and make an argument for why people should be militant atheists. It did surprisingly well, basically advocating for a gentler form of atheism: https://t.co/lc8ncc0JSP
2022-12-13 21:05:59 @ZoharAtkins Indeed!
2022-12-13 21:03:09 This seems almost perfect. I'd just omit "better" in the penultimate paragraph. https://t.co/HiawrOsd5c https://t.co/ZO7yotiM7B
2022-12-13 07:05:42 @SimonThomsen @dannolan @ATabarrok And Australian banks don't seem to fail at anything like the same rate as Crypto exchanges
2022-12-13 05:16:48 RT @michael_nielsen: @Amir_Safavi_N @kanjun The essay could have been 10x as long - I believe we had over a thousand pages of notes, and we…
2022-12-13 05:16:44 RT @Amir_Safavi_N: @kanjun @michael_nielsen I really enjoyed reading your essay https://t.co/dl51uoKGl2 but I am surprised at how little co…
2022-12-13 04:24:39 @Amir_Safavi_N @kanjun The essay could have been 10x as long - I believe we had over a thousand pages of notes, and we drafted and deleted long discussions of important issues in order to focus on what we regard as the core: change, how it's imagined, and how it happens.
2022-12-13 04:14:10 RT @luispedrocoelho: This long essay (effectively, a mini-book) by @michael_nielsen &
2022-12-13 04:14:07 @luispedrocoelho @kanjun Thank you, glad you enjoyed it!
2022-12-13 01:57:26 RT @D_R_Goodwin: Very much enjoyed this set of thoughts-in-motion by @michael_nielsen on the role of community in scientific progress. Some…
2022-12-13 01:06:21 Very interesting article on some remarkable moves by @SFConservMusic https://t.co/js6xTmAz7b
2022-12-12 19:03:28 @Meaningness The last sentence is an interesting idea, but would need a lot of actual evidence, not handwaving.
2022-12-12 18:49:30 @Meaningness And vice versa. It's not entirely arbitrary where issues end up, but surprisingly arbitrary.
2022-12-12 03:23:02 @mattsclancy Great!
2022-12-12 03:12:08 Very incomplete, alas. But I add books I read earlier in life as I think of them.
2022-12-12 03:06:52 My Goodreads: https://t.co/ZpOQCEUpdy Curious to see other people's! I have a sizeable collection of friends' bookshelf photos, and I'd love to have the virtual analog.
2022-12-12 02:25:30 @DavidDeutschOxf @MatjazLeonardis I really like this short old Usenet post from Greg Kuperberg, on a closely related point: https://t.co/lcbqtELztt
2022-12-12 02:04:43 @hyfen It was lovely - thank you for recommending this, I thoroughly enjoyed it!
2022-12-12 01:29:45 @Meaningness Yup!
2022-12-12 01:24:22 Saint Mary’s Cathedral, San Francisco. Very rushed, not cropped etc, I’m afraid. But I like it anyway!
2022-12-12 01:09:56 https://t.co/zefUwQf0yF
2022-12-11 22:36:36 @heyitsvajid @tobi @Suhail Yes. The trouble is people who specialize in your points (a) &
2022-12-11 21:33:06 @KeithMansfield @emmaconcepts So what is the resistance? https://t.co/NVeUP7FRCa
2022-12-11 21:22:14 @emmaconcepts @KeithMansfield Just what Keith always wanted, I'm sure, a legion of stalkers!
2022-12-11 21:18:52 @emmaconcepts @KeithMansfield It's hard not to read this and be nerdsniped...
2022-12-11 19:18:38 @visakanv @Ben_Reinhardt To Ben's point: Hawking's book was not important creatively (though it's a good popular book, and that's independently important). The SG's value is contested, but I believe it was important as a creative work.
2022-12-11 19:11:12 @Ben_Reinhardt @visakanv I wonder how many copies "Sociobiology" sold?
2022-12-11 18:58:09 @visakanv @Ben_Reinhardt Both Darwin and Newton wrote their great works after *decades* of work and thought. Reported results do (ultimately) need to be legible to some community, even if small. But the intermediate states may be (though don't need to be) highly illegible.
2022-12-11 18:52:30 @Ben_Reinhardt You need to be able to explain the value you offer to customers in a legible fashion. Not: "hold still for 10 hours while I explain" Nature has no such constraint. And as someone deepens their unique understanding they may get closer to nature, but further from the rest of us
2022-12-11 18:47:45 @Ben_Reinhardt Names supplied in DM :-) All three seemed very confusing but interesting at first. And all three radically changed how I think about creative work, once I started to get what they do.
2022-12-11 18:42:59 @Ben_Reinhardt A day is very little time. It's taken me ~ a hundred hours on at least three occasions I can think of offhand.
2022-12-11 17:18:17 @wtgowers Remind me not to take any of your classes, I'd be terrified the whole time
2022-12-11 17:16:51 @Ben_Reinhardt Consider the hundreds or thousands of systems which preceded ChatGPT, going back to models in the 1950s (or maybe 40s). Most weren't any good for the things ChatGPT can do, but were created for other reasons...
2022-12-11 17:11:22 @kanjun @smartin2018 Direct replications, at least in the parts of physics I'm familiar with, tend to be quite rare. Eg, Wolfgang Ketterle didn't try to replicate the original BEC work
2022-12-11 17:04:14 @kanjun @smartin2018 If someone in physics publishes something very exciting and it's not built on quickly, that's usually a bad sign, unless there's some good reason (eg a very special piece of equipment). One would tend to ask around (and likely be told "oh, it's wrong"). Both things happen a lot
2022-12-11 16:56:23 Useful discussion of ChatGPT and mathematics: https://t.co/UtyQxnWBXg
2022-12-11 16:47:54 @wtgowers Incidentally, context on my original tweet: I'd just read several comments from people dismissing ChatGPT's use in an area where I've found it easy to get useful examples - certain types of basic code generation, especially with widely used but poorly documented APIs.
2022-12-11 16:42:07 @wtgowers I wonder how ChatGPT is different from the models which are apparently succeeding with math olympiad problems? I have little hope of ChatGPT solving such, unless it had simply memorized a solution.
2022-12-11 16:39:24 @wtgowers Lovely problem Curious: would you be happy if it just said x and -x, where x is any irrational number? Your prompt almost answers itself: "x &
2022-12-11 16:30:24 @wtgowers I have not tried this in mathematics, and suspect that its mastery of basic mathematical reasoning is so weak that it'd mostly be annoying. But in other areas - more heavily dependent upon examples and association - I've found it helpful, sometimes very helpful.
2022-12-11 16:29:10 @wtgowers The basic pattern is to iterate around: me trying to explain the problem
2022-12-11 07:21:48 Lovely thread on asking questions in talks: https://t.co/GCxXRcA15x
2022-12-11 07:10:15 @visakanv I wasn’t expecting para 2. It’s downhill from there, alas
2022-12-11 06:57:59 This was so unexpectedly interesting that it took me a while to notice that it *didn't do what I asked*. I was testing the limits of ChatGPT's recursive reasoning: "write me an essay about writing an essay to [etc]" https://t.co/9s8qahV5Im
2022-12-11 06:22:39 @ben_golub @economeager @ElsevierConnect And then apparently someone just randomly spot-changed a bunch. My first paper to have a real impact. But solely due to the preprint. It still smarts.
2022-12-11 06:21:16 @ben_golub @economeager @ElsevierConnect Yeah. They demanded the change. We told them we'd mark up the proofs for other changes, but there were so many occurrences of this notation, and it was so hard to be sure we got them all that we wanted them to do it using search-and-replace with the raw file. They said "Sure!"
2022-12-11 06:18:25 @ben_golub @economeager @ElsevierConnect ... to a calligraphic S almost indistinguishable from the S we were already using, while leaving the other half the same. All this was happening multiple times in _the same equations_, all through the paper. Utterly incomprehensible.
2022-12-11 06:17:14 @ben_golub @economeager @ElsevierConnect The two most important pieces of notation we used in the paper - dozens of times! - were S, for entropy, and $ for a certain type of quantum dynamics. Both totally standard. They decided to "fix" $ for reasons unknown, &
2022-12-11 06:13:43 @ben_golub @economeager @ElsevierConnect With rare exceptions I preferred to point to the preprint form of my article. The American Physical Society mangled one of my papers beyond comprehensibility in a screwup that is quite funny, viewed from a particular angle.
2022-12-11 06:09:02 @ben_golub @economeager @ElsevierConnect Ah: the _accepted_ article is fine. For the published article there's fine print, though I'd guess usually it's fine.
2022-12-11 06:07:10 @ben_golub @economeager @ElsevierConnect No, they're fine with it: https://t.co/v5DSaiyaRo I think all the biggest publishers are, though I haven't checked in a while.
2022-12-11 06:04:31 @delong @economeager @florianederer @ShengwuLi @paulgp Which did you pick?
2022-12-11 06:02:50 @ben_golub @economeager Curious: are there econ journals which don't allow this? It's extremely rare across all disciplines, AFAIK.
2022-12-11 03:51:15 @sama I think it's very good as a creative sparring partner, especially if you prime it to be morale-boosting (a problem with its defaults: it's dour). I would *check* anything factual, though, just as I would with a collaborator (or my own memory)!
2022-12-11 02:08:14 @gwern @Dolebyte It's a minor nit, but I'm always amazed by people who publish things *without a datestamp* for this reason. Why no, I don't agree with that thing I wrote in 2007...
2022-12-11 01:41:53 @chanhosuh Hamming didn't invent the notion of working on important problems, nor was I thinking of Hamming.
2022-12-10 22:34:58 @jmkelly91 Curious: What’s an example where they’ve built in the manner I describe? Modern physics experiments often integrate hundreds (or more) of others.
2022-12-10 21:29:48 @RexDouglass Interesting - I've use that term too, elsewhere (cumulative science)!
2022-12-10 21:27:19 @RaptorKyelok Interesting, hadn't seen it!
2022-12-10 21:03:18 RT @timhwang: 10 december 2022: sharing a credit with a LLM for the first time
2022-12-10 21:02:26 Something peculiar: cumulatively, I've relied more heavily on other tools (search, books, journals, paper &
2022-12-10 20:56:15 Nice catch. Probably an unconscious influence (I've read a huge chunk of SICP, which is a glorious book, highly recommended!) https://t.co/ATxmtw4RI2 https://t.co/x3XXGT5sud
2022-12-10 20:54:00 @typedfemale Oh, I see it! Funny, I'm not sure I've ever read the preface. But if I had, that's (maybe) the source of my quote. That and the fact I had a family member learn the violin growing up (it was hard to learn!) https://t.co/C8c4E9Xo61
2022-12-10 20:52:08 @typedfemale I'm not making the connection. They talk a bunch about sorcerers and spells, but not about the time required to become a competent sorcerer, IIRC.
2022-12-10 20:47:17 I am touched by the honesty! https://t.co/11rWIlNxVO
2022-12-10 20:44:40 Although, for many tools people often seem to think that a day is far too excessive, and they should be able to play well after 10 minutes, and if they can't, the fault obviously lies in the tool.
2022-12-10 20:41:38 "I don't see what good violins are, I spent a whole day learning how to play one, and I sounded terrible at the end of it."
2022-12-10 18:17:34 Just reflecting on the injunction to try to work on important problems. Are there any surprising things or hard-won insights you've learned about how to do this?
2022-12-10 17:35:18 In the spirit of the previous tweet, let me note just how useful the phrase and concept "conventional wisdom" is. Due to John Kenneth Galbraith, I believe
2022-12-10 17:30:35 Tangentially, after some encouragement from @sebasbensu, I've decided to just have fun in footnotes henceforth. Against all concrete evidence, I've previously tried to restrict my urge to discursive asides in footnotes, but now the brakes are off!
2022-12-10 17:30:34 A point that bugs me about the replication crisis in psychology: https://t.co/z3qzv9Eg84
2022-12-10 17:20:56 The first acknowledgement here felt oddly necessary. Talking it over w/ ChatGPT was surprisingly helpful. ChatGPT wrote none of it, but helped me structure &
2022-12-10 17:17:46 So many field-founding documents are pretty close to everyday non-specialist thinking. They haven't yet bootstrapped an expertise to build off. https://t.co/w3CRRsRlUa
2022-12-10 17:14:47 (I just realized: this is analogous to, and was certainly influenced by, @nayafia's notion of an idea machine. https://t.co/OsXSps9Y4B Note, though, that a field or community of practice is quite different from an ideology.)
2022-12-10 17:14:46 Interesting to think about whether there are equivalent online social machines: https://t.co/kAFBkAgG9I
2022-12-10 17:11:48 The key takeaway is just a point of view or perspective. It's to think of research orgs as social machines for generating new *types* of expertise, and corresponding communities of practice. https://t.co/GQVHZO4HNv
2022-12-10 17:10:04 Some snippets It's conventional wisdom to want to "break down barriers between fields". I think it's worth considering the benefits of erecting much higher barriers, so you can get siloes of people attacking important problems starting from fundamentally different assumptions https://t.co/S1MeLI46Gy
2022-12-10 01:50:29 @fstflofscholars Somehow more natural than viewing (say) universities or research institutions, which tend to be considerably more political &
2022-12-10 01:49:09 @fstflofscholars It's a fun perspective shift, that's for sure!
2022-12-10 01:46:22 Working notes exploring the idea that instead of thinking of discoveries as the unit of advance in science, we can think of communities / fields as the unit of advance: https://t.co/0Ka9bfKU0y
2022-12-09 22:59:10 Remarkable &
2022-12-09 17:34:50 ChatGPT has some (banal) thoughts. But then, turn the dial, and the banal starts to fade: https://t.co/JCyzE1eh8H https://t.co/Ovv9QMtDDs
2022-12-09 15:44:23 @DhrmaRenaissanc Most tools take a few thousand hours to achieve basic competency with. I’ve spent fewer than 20 here. I’m not remotely close to diminishing returns
2022-12-09 05:02:24 Something really lacking: the sense that the other "person" in the conversation cares
2022-12-09 05:01:01 @Plinz @0xmaddie_ I read your tweet, thought "I wonder what else they could get wrong", and my first idea was that perhaps they could broadcast it in wavelengths of light not visible to humans. For some movies this would be an improvement.
2022-12-09 04:55:22 @0xmaddie_ I had pretty much the same response. Underwhelming.
2022-12-09 04:32:29 Oddly, the prompt: "Can you improve these ideas?" sometimes produces something which jogs me out of the rut of my thinking.
2022-12-09 04:29:09 It has a little of the quality creative conversation has. Most CC is pretty banal: the key attribute is that you _keep things alive_. And you seize upon and develop any tiny glimmer of hope I'm finding I need to do all the seizing. But it's helping a lot with the keeping alive
2022-12-09 04:25:50 Also, wow, just spent a few hundred words describing problems I'm having in a project. It made some reasonable suggestions - basically rubber ducking. Then asked it to apply one of @kevin2kelly and @brianeno's oblique strategies. Sure enough! And then another!
2022-12-09 04:09:24 Huh. Being open and honest and vulnerable with ChatGPT about creative difficulties turns out to be a good idea. (Under the current privacy policy, I wouldn't do this about more personal things!)
2022-12-09 02:38:28 The CIA named their VC firm for James Bond (of course): https://t.co/KJushFgRG9
2022-12-09 02:37:10 @infinitsummer Officially: https://t.co/wcBKo7T9sY I wonder where else they are silent or non-silent LPs or GPs?
2022-12-09 02:35:19 @joshu Wow, Broderbund. Getting flashbacks to Loderunner!
2022-12-09 00:55:08 @ethanCaballero @raykurzweil @tszzl Pretty sure you tweeted this 7 months late, or 5 months early.
2022-12-09 00:49:40 For those unfamiliar with quantum computing, the ChatGPT response is quite good - it's more or less the canned answer I'd expect from a quantum computing expert, with no more than very minor infelicities. By contrast, the GPT-3 and https://t.co/yOoVdNxO95 answers are quite bad.
2022-12-09 00:46:34 A much better answer than the just-released https://t.co/yOoVdNxO95: https://t.co/ezNwLcWtmy
2022-12-09 00:44:44 @hellertime The answers are not at all the same. "We don't know", plus supporting explanation, is entirely different from "no" [which is highly misleading, since we don't, in fact, know].
2022-12-09 00:26:13 Cf ChatGPT versus GPT-3: https://t.co/iP7MblmDE9 https://t.co/f3ia3deElw
2022-12-08 17:39:19 @AnnaGHughes Titanic.
2022-12-08 16:46:21 Bruno Latour (1947-2002): https://t.co/AsHjXfGk68
2022-12-08 13:00:00 CAFIAC FIX
2022-12-07 08:00:00 CAFIAC FIX
2022-11-05 23:24:19 Interesting use case: https://t.co/gYIYPvT6v9
2022-11-05 19:52:30 100 bookstores in 100 days https://t.co/2Qt8P9INWR
2022-11-05 19:46:46 @bcrypt If by “here” you mean actually used by lots of people, yes.
2022-11-05 08:27:21 @Singularitarian You may well be right, in which case my bad. Prior quarters were much better, so maybe it was an anomaly? In any case, deleting. Thanks for pointing this out.
2022-11-04 16:50:55 RT @kanjun: In 2020, @michael_nielsen &
2022-11-04 16:50:34 RT @michael_nielsen: Thoughtful reflections from Brian Nosek: https://t.co/BwvXoJEKF3
2022-11-04 16:50:30 RT @michael_nielsen: It was to be a small, two-month project https://t.co/EQQFxDAIMW
2022-11-04 16:50:20 RT @michael_nielsen: Brief summary of a few of the things we discuss: https://t.co/7l6FbBqf2B
2022-11-04 16:39:30 RT @michael_nielsen: New essay by @kanjun and myself, on improving the culture and social processes of science:https://t.co/IX0heqL9yG
2022-11-04 12:41:01 Very good. Good luck… https://t.co/EjVaFd4Q3x
2022-11-04 10:43:08 This entire thread may be regarded as a tourist itinerary for a future Orion Arm Infrastructure Observatory. Its progenitor: https://t.co/S32mVHFB7X
2022-11-04 10:40:36 RT @michael_nielsen: Curious: what are the most interesting megastructure ideas you know of, and why do you find them interesting?
2022-11-04 10:37:49 Put another way, a daring nonconformist with an inventive chisel would _not_ have served Chartres well. There was scope for creativity, but it was limited. Most of the artisans ceded their individuality
2022-11-04 10:35:24 (The tension between individual agency and systems is fascinating, and a dominant theme of modernity, and arguably of homo sapiens. It's what makes "The Wire" so incredible. It's part of why megastructures are fascinating. I'd love other treatments as striking as "The Wire".)
2022-11-04 10:33:13 The relationship between individual agency and megastructures is interesting. In some sense, agency is ceded to (or voluntarily invested in) a larger system. Much like a religion - explicitly, for things like Chartres or the Hagia Sophia.
2022-11-04 10:28:14 An irresistible idea (to me) is megastructures made principally of photons. Not really plausible without nonlinear interactions, which probably means some fermionic substrateStill, things like polariton condensates seem like steps in this direction: https://t.co/RH2WQK320C
2022-11-04 10:14:11 Just coming back to the aqueduct, it reminds me again of the extent to which megastructures are often principally _ideas_. Those ideas can spread out across space and co-ordinate action
2022-11-04 10:10:46 In general, that idea - using correlations between GRB activity to search for large-scale structure in the universe - is pretty amazing!
2022-11-04 10:08:38 It's amusing to speculate about superstructures in the cosmos as megastructures. Eg the Gamma Ray Burster Ring, a surprisingly dense concentration of gamma ray bursters, spread over a mere 5.6 billion light years: https://t.co/syTxVRJC6BFun SF: the GRB Ring as signalling system
2022-11-04 10:02:30 LIGO and LHC are bother interesting megastructures. The respective vacuums are particularly interesting: the deliberate _lack_ of something as a type of megastructure: https://t.co/MoPu7Es7HF https://t.co/sIyfR2gxTk
2022-11-04 10:00:36 Cities - think Trantor, Jericho, or Generation Ships - are very interesting in this regard. They arguably blend all five, due to agglomeration effects.
2022-11-04 10:00:35 Interesting to think about "why megastructures"? A few answers:1. Ego or fear (Pyramids, Titanic)2. Capitalism serving people (cruise ship, TSMC, oil rig)3. Infrastructure - increasing ambient affordances (aqueduct)4. Enabling novel objects (TSMC)5. War (USS Ford)
2022-11-04 10:00:34 The Roman Aqueduct, the first of which dates to 312 BCE: https://t.co/sliiqhdoXg https://t.co/x1UJ2n9A8z
2022-11-03 16:35:21 The Cruise Ship Bellissima, featuring Andrea Bocelli and Sophia Loren. Time to say goodbye, I guess.(5,686 guests! So it's a floating city.) https://t.co/Z7LXrf1tHj
2022-11-03 16:25:10 Something I love about the last point is that Apollo had to be huge to get off earth to the moon. But the reverse trip, blasting off the moon and back required something not all that much larger than sits in many people's garages.
2022-11-03 16:22:46 @DavidDeutschOxf Very pushy roads! And they only push sideways a small fraction of the time! Most of the time they're so sedate &
2022-11-03 16:20:47 Related, one of my favourite facts about Earth and our solar system is the relatively low cost of escape. Once in low Earth orbit, you only need to give 1kg a kinetic energy roughly equal to a fortnight's calories for an adult human(!!!!) https://t.co/PVhuKIcWVH
2022-11-03 16:16:32 One thing which is odd: in our terrestrial systems, gravity dominates. We mostly don't build very big objects because gravity crushes them under their own weight. But this in some sense an atypical situation. And free fall makes it _much_ easier to construct in many ways!
2022-11-03 16:13:01 The Cruise Ship Bellissima looks like she might be bigger than the USS Gerald R. Ford! https://t.co/jfPMEqeNvl(Wonder how that compares to the Vehicular Assembly Building, Burj Dubai, and similar? All candidates for largest single structure, depending on definitions...)
2022-11-03 16:09:57 That pattern - of information dominating matter - is fascinating. Even in something as simple as a car moving down the street &
2022-11-03 16:07:34 An interesting aspect of many (not all) of these megastructures is that there is an underlying idea. That idea can be spread out, and used to co-ordinate and control the overall construction. It's a case where information dominates matter, so to speak.
2022-11-03 16:05:18 Interesting to think of new phases of matter in this vein. Things like Bose Condensates and quantum Hall systems were *imagined* first(!!!) Other things like superconductors and fractional quantum Hall systems were, I believe, discovered. Not quite megastructures. But related.
2022-11-03 16:03:30 The USS Gerald R. Ford is the largest warship ever constructed: https://t.co/GNe5lvYFiA https://t.co/9Rv6fZ5MrO
2022-11-03 16:02:06 Related, Wikipedia's list of hypothetical astronomical objects is fantastic: https://t.co/mZPhpchRMiFrom Trojan planets to Chtonian planets!
2022-11-03 15:59:48 I love the Arp Objects, the Atlas of Peculiar Galaxies: https://t.co/IENvqeHDpgProbably none are engineered megastructures. But it's fun to think about
2022-11-03 14:03:55 "Is human led mathematics over?" A discussion between Joelle Pineau, @wtgowers, &
2022-11-03 13:51:32 A very striking story: DARPA's director lobbied for a reduction in budget, on the all-too-sensible grounds that it would make DARPA better... https://t.co/moFCySXtQ7
2022-11-03 13:37:04 @littmath Admittedly, "the" usually implies taking the principal branch. It'd be nicer if it said it was "a square root".
2022-11-03 11:41:09 @djjr Sounds like a just wonderful book!I once counted the number of people on Avatar's IMDB page - was ~1,400 at the time (no doubt very incomplete, and doesn't include supply chain of course). Pretty amazing for mostly just-in-time work.
2022-11-03 11:35:40 This is a wonderful paper on megastructures, by @anderssandberg: https://t.co/F8pojdwzaH(I was an enthusiastic referee for the paper )https://t.co/e71TevQmOt
2022-11-03 11:33:27 @David_Biyi It's in-thread...
2022-11-02 19:42:30 @ceptional I can think of 20 ways of making it better offhand. I’ll bet they do at least a few significant improvements.
2022-11-02 15:59:17 @JBuch7 Not as well known as they should be. I subscribed to RRE back when he was running it, in the 90s!
2022-11-02 14:45:17 @kocienda I was just saying there's a lot of people for whom money != success.
2022-11-02 14:41:42 Think of this as a request for poorly-known gems.
2022-11-02 14:40:07 What is the most underappreciated scientific or creative work you know of?
2022-11-02 14:34:37 @kocienda (Not so much because I care about JA et al - though I do love JA - but just as a reminder of different sets of values. People who don't care about your hair cut, or your series A, but do care about other things.)
2022-11-02 14:33:37 @kocienda This thread is much more true in certain limited parts of the world than in others. One reason I enjoy working in Berkeley Cafes: the people at the table beside you may well think your opinion of Jane Austen [etc] is more important than the type of car you drive...
2022-11-02 14:04:12 @BrianNosek I love the experiment!
2022-11-02 11:05:35 (IIRC this was the first thing I put on my first webpage, back in ~1995!)
2022-11-02 11:04:43 "A man is a very small thing, and the night is very large and full of wonders." - Lord Dunsany
2022-11-02 10:16:30 @xlvorhees Fascinating!
2022-11-02 09:18:10 @BrianNosek Curious: why?(I tried Mastodon a bit years ago. It seemed to have a lot of problems that Twitter either doesn't have, or has in reduced form.)
2022-11-02 08:52:44 @DavidDeutschOxf I just saw one (private) that appeared so to me. I'd say better than 99% of people could write.("Introspective" isn't the right word for it
2022-11-02 08:49:39 Do you expect Twitter in 12 months to be:
2022-11-02 08:19:16 GPT-3 often seems like an inverse Eliza: the human is providing relatively minimal prompts as stimulus, just to keep the conversation flowing, while the computer is providing long, detailed, introspective responses.
2022-11-01 21:40:58 @JocelynnPearl @singareddynm Oh, very interesting! Do you resonate, @singareddynm?
2022-11-01 21:39:31 I've never seen "Jiro Dreams of Sushi". But maybe. There's a difference between obsession over a craft or system and committing your life to exploring and expressing some set of values.
2022-11-01 21:38:02 @curiouswavefn Really marvellous example! @DarlingtonHall2
2022-11-01 21:37:28 @alexkozak Curious: if you don't mind, how is that expressed in your life?[For a priest, it is a dominant feature of their life. How does your commitment feel by comparison?]
2022-11-01 21:31:20 I would not be surprised if some gardeners approach this.
2022-11-01 21:29:56 To be a bit clearer, I mean treating something as a core calling of a (near)-transcendent value. It's not the task, but the orientation toward the task: there may be both poets and ditch diggers who match this
2022-11-01 21:27:37 Who do I know who has really strongly committed to an unusual ethical or religious system that is not EA? Has taken religious orders, lived as a monk.
2022-11-01 21:06:15 @mpoessel @Jess_Riedel Neutrino oscillations too, IIRC(?)
2022-11-01 18:20:47 I loved Weinberg's "The First Three Minutes". But it's pretty out of date. Ideally I'd like something up-to-date and shorter and just as good
2022-11-01 18:17:01 Booktok! https://t.co/DQ4bcQEsuj
2022-11-01 18:10:37 What's the most wonderful short, moderately technical (i.e., equations allowed) account of the Big Bang?
2022-11-01 17:53:03 This seems spot on to me: https://t.co/CNWCVfasRF
2022-11-01 17:01:26 @devonzuegel @NotionHQ Thanks Devon
2022-11-01 16:58:41 @mxstbr @NotionHQ @devonzuegel
2022-11-01 16:58:18 I was honoured &
2022-11-01 13:00:57 @felixpalazuelos Thanks. Just a playful riff on Mircea. Though the Cathedral feeling is something I feel very strongly. Also in great art galleries
2022-11-01 12:57:25 That said, I’m in a bookstore. Always feels a little like a Cathedral. Even if there’s some profane, along with the sacred…
2022-11-01 12:53:08 @kmmunger It’s all just “content” [sic]
2022-11-01 12:52:27 Up next: poet-bureaucrats and surfer-bureaucrats, no doubt.
2022-11-01 12:48:50 https://t.co/u3zYHykMCa
2022-10-31 14:11:08 It reads funny, because it's something he's _doing to_ his kid. But at that age of course adults really do do a lot to/for their kids. Not clear it's not simply unfamiliar, as opposed to particularly egregious in the annals of parenting.
2022-10-31 14:09:01 Fascinating, someone using memory systems with their kid: https://t.co/33Qzs3PTw3 Reminds me of Agassi and the Polgars. Not sure what to make of it. via @gwern https://t.co/4sjIHs7JI2
2022-10-31 14:05:58 RT @benyeohben: I will be podcasting with @michael_nielsen soon. Let us know any questions on metascience and more…
2022-10-31 13:41:37 I won't be surprised if there are many such dreams of (out-of-copyright) classics now produced. It'll be interesting to see if anyone can create something really good, or if they'll remain fun demos.
2022-10-31 13:40:09 @mattsclancy Thanks for this - a very quick skim looks fun, and I'll look forward to a proper read. BTW, have you read Kevin Kelly's book "What Technology Wants"?
2022-10-31 13:33:54 This looks likely to be very beautiful. Not so much a graphic novel as a dream of Last and First Men. ht @andy_matuschak https://t.co/eHXLqVfqwp
2022-10-31 13:29:21 RT @MattHourihan: Bang on https://t.co/0RWsTRqWxG
2022-10-31 13:29:06 RT @arbesman: So much that is wise and profound in this essay.And so much that is provocative and exciting as well, from “metascience as…
2022-10-31 13:28:17 RT @NowhereLikeNow: “A metascience entrepreneur is a person, especially an outsider, who aims to make a scalable improvement in the social…
2022-10-30 14:30:06 https://t.co/wuiIKwGyqf
2022-10-29 22:53:02 @tmrss_ Australia was one of the places I had in mind. Grew up mostly in Brisbane!
2022-10-29 22:43:40 @AdamMarblestone Thanks! Really looking forward to this!
2022-10-29 22:43:16 @gwern Remarkable. One may quibble at the ethics, but it’s not so clear how the intent differs from many parents. One is reminded of Agassi’s father a bit, or the Polgars
2022-10-29 22:35:00 @lfschiavo @tmrss_ I don’t know why either. It wasn’t great when I was growing up, but got very rapidly better as an adult.
2022-10-29 15:42:14 @nat_sharpe_ Applies also to theory behind the nuclear bomb, LIGO, mathematics. Squiggles on tree pulp have amazing power sometimes
2022-10-29 15:34:25 @danielgross Admittedly, I’ve seen this even in NYC. But I like it as a casual presumption!
2022-10-29 15:33:38 @danielgross A friend’s story about Iceland: in a cafe there, saw a group get up and leave, but left laptop on table.My American friend to her Icelandic friend: “oh, they forgot their laptop!”Icelandic friend: “Oh, they’re just holding the table…”
2022-10-29 10:49:01 RT @michael_nielsen: @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug f…
2022-10-29 10:41:58 @bschne Weird, maybe. Funny, definitely!
2022-10-29 10:41:20 @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug free!
2022-10-29 10:38:30 Fertility rate of 0.8!!!Down by a factor 8 over 60 years! https://t.co/tVyYFd393d
2022-10-29 10:37:18 Starbucks is a useful baseline for local cafe culture. In some places it dominates, in others it’s a place only a few tourists go. The latter is a good sign.
2022-10-26 22:45:32 RT @katyilonka: I read the whole thing (and it's long). While I'm not that into metascience generally, the essay really made me reflect abo…
2022-10-26 22:45:23 @katyilonka That sounds like a great essay to have!
2022-10-26 17:24:21 Thoughtful short note: https://t.co/a6HnA21kdg
2022-10-26 01:53:26 @BrianNosek I'm not wild about methodological answers, as in many replies. They seem more like symptoms of rigorous research, not causes. I do like Feynman's comment about: "a specific, extra type of integrity that is not lying, but bending over backwards to show how you’re maybe wrong"
2022-10-25 23:15:38 @jachiam0 This seemed very good to me. Thanks for writing it.
2022-10-25 02:18:40 @vgr And they will always use that as an excuse. Much like arch-communists and -capitalists.
2022-10-24 18:57:26 @RuxandraTeslo I've found this perplexing since being a (broke) undergrad.
2022-10-24 18:30:39 RT @kmmunger: Fantastic article by @michael_nielsen and @kanjun defining the possible scope of metascience and calling for more metascience…
2022-10-24 02:25:59 IBM mainframe specs from the 1960s. Up to 512 kilobytes, and only 3 or so tonnes! https://t.co/wqJyVeqQwR
2022-10-23 16:56:48 RT @marquezxavier: Really stimulating article on the design of the social process of science by @michael_nielsen and Kanjun Qiu https://t.c…
2022-10-23 16:56:00 RT @JMateosGarcia: "A vision of Metascience"I started reading this essay by @michael_nielsen + @kanjun and it's great. I love the idea of…
2022-10-21 19:20:39 It certainly makes me curious what makes b better. Even a tiny improvement - say, to b = 0.12, would make a huge difference. You'd then need a ~300 increase in training data to get that square root improvement....
2022-10-21 19:12:01 Now, admittedly, b is pretty small in their examples. E.g., b = 0.095 for training data, so you need a roughly ~1000 increase in training data to get the square root improvement in probability. Still, that's better than I would have a priori guessed!
2022-10-21 19:10:59 One simple way of looking at this is a constant multiple increase in compute, training data, or model size buys you a square root improvement in probability.That constant multiple is 2^{1/b}, where b is the exponent in the scaling law.
2022-10-21 19:05:58 But it's not so different either.
2022-10-21 19:05:57 In particular, suppose b = 1 and we double size of training data. Then if the former probability was 1/100 we now get a probability assigned to the correct token of 1/10. That's an incredible improvement!In practice, b is smaller than 1, and things don't work quite so well.
2022-10-21 18:58:46 A slightly more precise statement, courtesy of @moultano: https://t.co/PsjnP2e4bE
2022-10-21 18:56:22 @moultano Ah, I see. Thank you!So a little more explicitly, the scaling model says:p = exp(- a 1/M^b), for constants a and b, and p the probability assigned to the correct (unseen) token, in the training data.This seems to me quite remarkable!
2022-10-21 18:53:31 Incidentally: note that the constant a is negative (for all three of compute, training set size, model size). This ensures that p ->
2022-10-21 18:52:13 @moultano Hmm. What's the difference?
2022-10-21 18:42:37 I'm quite curious whether I have this right. I must admit, the probability dropping as exp(-M^a) would seem shocking to me.
2022-10-21 18:41:54 A very basic question about the meaning of the Kaplan et al OpenAI scaling laws paper: https://t.co/YimscMAryl
2022-10-21 18:28:45 @phokarlsson @Meaningness @kanjun It would take quite some thought to figure out what &
2022-10-21 18:24:03 @phokarlsson @Meaningness @kanjun On the hypothesis: yes, that's the main thing I had in mind.
2022-10-21 18:23:29 @phokarlsson @Meaningness @kanjun Well, there are many things one could look for. Personally, I'd be curious to understand to what extent the extra growth helped (or hindered). The extreme case is, of course, the NIH doubling ~2000.
2022-10-21 18:14:02 @Meaningness @kanjun Two things I find fascinating: (1) the flip from foundations to government circa the 50s
2022-10-21 18:09:49 @Cerebralab2 @Meaningness @kanjun We don't assume it.
2022-10-21 17:54:54 @colliand @kanjun Thanks!https://t.co/YgqP8DDWUu
2022-10-21 16:12:47 @Jonathan_Blow @ivn_echvrria A talk would be good. A hundred item twitter thread is also a pretty interesting idea. Hmm.
2022-10-21 16:07:50 @Jonathan_Blow @ivn_echvrria https://t.co/TRLugBXk3R(Cf Robert Gordon's "The Rise and Fall of American Growth")
2022-10-21 16:02:24 @Jonathan_Blow @ivn_echvrria Yeah, will be true late Dec 2025.Amazing the (unclassified) speed record is, I believe, still held by the SR-71.
2022-10-20 23:17:06 I mean, it's easy to confabulate after-the-fact reasons Keurig is a good idea, and Juicero bad. But in the moment? I'm not sure I'd have made the right choice, at least, not based on the idea. Maybe on team or some other factor.
2022-10-20 23:15:49 I enjoyed this. And while I think it's, um, unlikely to work, I'm not _certain_I wouldn't have invested in Juicero. Obviously bad idea! Also: I wouldn't have invested in Keurig, for the same reasons. Oops(It's still not clear to me why Keurig is a good idea, and Juicero bad) https://t.co/DrU5qVHh1J
2022-10-20 23:01:36 RT @kanjun: Today, AI systems can create stunning art &
2022-10-20 22:55:59 Delighted by this: https://t.co/Ot4F4yKPEM
2022-10-20 21:43:51 @petersuber @kanjun Thank you so much Peter! That’s wonderful to hear, especially given how much I’ve learned from your writing.
2022-10-20 19:23:52 RT @AdamMarblestone: TFW someone who relishes in "scientific bottleneck analysis" and "scientific roadmapping" (i.e., me) reads a credible…
2022-10-20 18:46:03 @kanjun @genintelligent Congratulations on launching after so much hard work!
2022-10-20 18:39:00 "Our work aims to understand the fundamentals of human intelligence in order to engineer safe AI systems that can learn and understand the way humans do... If you want to work on a principled approach to building safe general intelligence... we'd love to hear from you!" https://t.co/zwP2vWqbim
2022-10-20 16:22:59 I enjoyed playing with this interactive visualization of the majorization relation for 3-component probability distributions. https://t.co/bAAa9Z1UBu
2022-10-20 15:30:16 RT @paulg: We've only explored a small part of the space of how science could be organized and funded. The current default is far from opti…
2022-10-20 15:21:11 RT @JeffSpies: Sounds familiar. I was told to either (a) build @OSFramework and leave academia or (b) get a job, get tenure, &
2022-10-20 15:04:10 RT @laurencetratt: It's a long read (I've spread it out over 24 hours!) but "A Vision of Metascience" by @michael_nielsen and Kanjun Qiu is…
2022-10-20 15:01:04 Thoughtful thread reflecting on metascience: https://t.co/RkHdNgAQfM
2022-10-20 14:59:03 RT @EmergingDrivers: Some very good ideas here. For instance: " Instead of funding grants that get the highest average score from reviewers…
2022-10-20 14:45:50 RT @kanjun: In 2020, @michael_nielsen &
2022-10-20 05:02:40 @worst_account It's interestingly comic when applied to points of difference: "Tony acquitted the defendant, ruling that their rights had been unfairly violated when Tony pulled them over."Very interesting heuristic.
2022-10-19 21:53:55 @jt_kerwin @paulnovosad https://t.co/PGIuxL3IVU
2022-10-19 21:03:33 Thread for my essay with @kanjun on changing the culture and social processes of science: https://t.co/vMh2SHhfFm
2022-10-19 21:00:10 RT @michael_nielsen: New essay by @kanjun and myself, on improving the culture and social processes of science:https://t.co/IX0heqL9yG
2022-10-19 20:59:25 Careful, you'll end up with the Chair of the Fed trying to reign in people's rants, and then you've got first amendment problems: https://t.co/H7DvUjolqS
2022-10-19 20:21:16 @samth @kanjun I certainly strongly agree with your point.
2022-10-19 20:20:38 @samth @kanjun We don't assume that, we didn't want to make it 3x as long, eg: https://t.co/HIGhF7VDIi
2022-10-19 18:09:49 @ID_AA_Carmack If you work mostly on people problems then there's lots of value to going to such gatherings - you can meet and make shallow connections to a large number of people who might help with your project. But if you work on thing-problems, diminishing returns set in much faster....
2022-10-19 18:08:37 @ID_AA_Carmack I tend to think of projects in terms of whether the hardest part is solving people problems or solving thing-problems. Starting a company is usually (not always) the former, while solving a mathematical problem is usually the latter.
2022-10-19 18:05:14 @ID_AA_Carmack I keep a quota and keep track. Some years it's for 1 such gathering, other years I set it at 3. More than 3 would mean a change in the kind of work I do (to something much more network- or co-ordination focused).
2022-10-19 01:46:53 @vgr @petersuber In some sense, protocol negotiation is the meta-protocol we all use repeatedly, for iterated games...
2022-10-19 01:45:52 RT @paulnovosad: A fabulous essay on the institutions that produce science, with some very interesting ideas about how to improve them.
2022-10-19 01:45:30 @vgr There's got to be a fun game here (self-illustratingly). And reminded a bit of @petersuber's Nomic.
2022-10-19 01:44:51 @vgr Great question! Rock-paper-scissors, writing, the rules of the mosh pit, the heterogeneity of restaurant culture (but each restaurant thinks _their_ way is obviously the right way, often!), the protocol of change rooms in clothing stores, etc etc...
2022-10-19 00:21:16 RT @dgmacarthur: An incredibly thought-provoking piece on how we might improve the process of science. This is a must-read for anyone feeli…
2022-10-19 00:18:49 @stuartbuck1 Not consciously, though I'm pretty sure I have looked at that piece.
2022-10-18 23:36:28 @BrianNosek @siminevazire Halloween costume, Brian!
2022-10-18 23:17:04 @JocelynnPearl @JustinQuda @matthewclifford @kanjun @notboringco @packyM @Convergent_FROs @AGamick Thanks Jocelynn! To give credit where due: that was 90+% Anastasia, 2-% me :-) Looking forward to reading your essay!
2022-10-18 23:12:35 RT @stuartbuck1: This is definitely a must-read, and a must-re-read . . . in fact, to take advantage of spaced repetition, I'll reread it t…
2022-10-18 23:12:32 @stuartbuck1 Thanks Stuart!
2022-10-18 23:12:25 RT @siminevazire: I haven’t finished reading it yet, but this essay is fascinating. Presents a view of metascience as an engine for improvi…
2022-10-18 23:12:16 @siminevazire Thanks! Yes, we certainly meant it lovingly (this will become very apparent later!) And on the scale of many scientific orgs, COS is tiny. But mighty!
2022-10-18 23:03:45 @futuryst @kanjun @samira_kiani1 That would be lovely!
2022-10-18 22:37:41 @futuryst @kanjun @samira_kiani1 Thanks Stuart!
2022-10-18 22:37:27 RT @futuryst: Love how this just-released essay on evolving science as an institution, by @michael_nielsen &
2022-10-18 22:37:20 RT @tkalil2050: I hope that this essay by @michael_nielsen and @kanjun inspires more metascience entrepreneurs, and more backers of metasci…
2022-10-18 22:29:47 RT @ilangur: If we care about improving outcomes of scientific research, we must consider the systems &
2022-10-18 22:23:08 RT @kanjun: @michael_nielsen An early question was: why do funders say they want high-risk, high-reward research, yet end up funding low-ri…
2022-10-18 22:23:03 RT @kanjun: @michael_nielsen We also wondered: why does nearly every new funding institution set out to "do things differently", and then,…
2022-10-18 22:22:48 @Ben_Reinhardt @kanjun
2022-10-18 04:16:05 @jackclarkSF @togelius The book John McPhee wrote about. Or the New Yorker profile. This is McPhee and Taylor in the World Trade Center buildings. https://t.co/TsQZah5iDo
2022-10-18 04:11:40 @togelius @jackclarkSF It's interesting-horrifying to read Ted Taylor - perhaps the best nuclear weapons designer ever - on the possibility of the former. And I'll bet you can do both of the latter. If there's a 1% chance of any of these it's a hell of a risk.
2022-10-18 02:36:51 Courage and suppression: (lots of background: https://t.co/6zzGpHRoFm ) https://t.co/siPtk8Akjo
2022-10-18 01:08:13 It speaks, maybe, to a difference in perception of the activities: chess as a competition with each other, while in climbing it's really pushing individual limits against nature. (Both these tweets are oversimplifications of complex situations, ofc.)
2022-10-18 01:06:41 Something I used to relax during the pandemic was chess and climbing YouTube. And it's interesting the difference: the chess players are very competitive with each other, while the climbers are more of a blend of supportive &
2022-10-18 00:56:56 (Apparently) the technically hardest climb ever done: https://t.co/svVwjHkeYy
2022-10-17 22:57:48 @_OliverStanley @moultano @jstn We can influence the regulation, if we choose…
2022-10-17 22:54:47 @Stefania_druga I am disappointed tbh
2022-10-17 22:47:53 @moultano @_OliverStanley @jstn Interesting to think about doing this somewhat through independent bodies, like IATA or NTSB for airline safety
2022-10-17 22:41:01 Not a squirrel Bar Mitzvah in sight https://t.co/Qa3aH0wyLD
2022-10-17 22:39:56 @alexjc @aj20000 Interesting articles both!
2022-10-17 22:14:11 @alexjc @aj20000 Fair use is legal. It's not an admission of anything.
2022-10-17 21:12:26 If a human artist showed me the previous QT'ed image I'd be amused and a little impressed by the joke. I know it's silly, but it's at least a little clever. Reminds me of Magritte.
2022-10-17 21:08:31 I suppose if we get tired of the argument, we can always go fishing: https://t.co/5anpq0fDgx
2022-10-17 20:41:14 @TrevyLimited @LibertyRPF I'm pointing out a factual error in your previous tweet. Fair use does apply to commercial use, and you asserted otherwise.
2022-10-17 20:36:10 @TrevyLimited @LibertyRPF Fair use certainly applies to commercial use, though it is typically taken somewhat more restrictively there. https://t.co/zEHhXc3RVP(This is, for example, why it's legal to use short quotes etc in published books.)
2022-10-17 20:33:51 @TrevyLimited @LibertyRPF Please assume good intent and apply the principle of charity. Good conversation is impossible without it. Thanks.
2022-10-17 20:30:27 @LibertyRPF @TrevyLimited My understanding is that Google search benefits a great deal from fair use. It seems quite feasible to make image synthesis programs that (a) respect fair use
2022-10-17 20:24:54 @RuxandraTeslo It makes you sound like an open, curious person.
2022-10-17 20:23:34 @LibertyRPF @TrevyLimited It was a general moderating comment for a conversation that seemed like it might be careening in an unproductive direction.
2022-10-08 14:25:59 @Nadagast @adimelamed_0 @jasoncbenn The first paragraph I quote states in multiple ways that environmentalists are anti progress.
2022-10-08 14:22:35 @stefanlesser I don’t actually know what point you’re referring to(?) I wrote this to make my thinking clearer. There’s no point about note taking intended.
2022-10-08 03:47:55 Just TBC, I'm aware of the standard argument. But it's amusing to come up with (arguably better) arguments for other positions https://t.co/eZoS9TmIaE
2022-10-08 03:46:15 It seems that much of the trouble would be in figuring out how to make the comparison sensibly. But even mapping that out seems like it ought to be interesting.
2022-10-08 03:41:04 This is one of those things where I'd love to be able to wave a wand &
2022-10-08 03:36:18 It'd be *very* interesting to have differential stats on human vs mice. (Complicated by a lot of factors, of course.) https://t.co/nGF2Hmep1F
2022-10-08 03:34:49 An interesting dichotomy: weddings are (usually ) invitation-only, while funerals are more often open events.
2022-10-07 23:57:44 RT @michael_nielsen: Thinking out loud on progress: https://t.co/SjwU1Rapx7 https://t.co/kXEs03fQ5r
2022-10-07 23:36:35 Thinking out loud on progress: https://t.co/SjwU1Rapx7 https://t.co/kXEs03fQ5r
2022-10-07 21:35:14 A hands-on explanation with a nice combination of theory and practice. Focused on language models, with the promise that it'll build up to modern transformer models.
2022-10-07 21:35:13 These lectures are terrific: https://t.co/MOXhuB4qyD
2022-10-07 04:03:46 @andy_matuschak @TKPullinger Do you have an opinion versus the Remarkable?
2022-10-07 02:42:07 A teacher in my primary school told me y was "sort of" a vowel. I have no idea why, but the network agrees!
2022-10-07 02:38:05 Enjoyed this tidbit in @karpathy's explanation of how to train a neural net to generate names: the first layer learns a 2-d embedding of the alphabet
2022-10-06 05:18:59 I enjoyed this thoughtful recent survey of progress in AI, especially language (by @chrmanning ) https://t.co/wm2mB48Nfc https://t.co/lPRVJ3QSjO
2022-10-06 04:36:02 @TKPullinger Oh, thank you, that's a useful idea!(I don't have an ipad, but might buy one if this looks like the best option.)
2022-10-06 04:31:18 @twothreemany Thanks!
2022-10-06 04:05:09 Has anyone extensive experience of using the Zoom whiteboard? If so: how is it as a user?
2022-10-06 03:42:50 "several" https://t.co/RCCB7B9Jfd
2022-10-06 02:26:52 (Amusing: those two screenshots could easily have been, in their time, the main content for two PhD theses. Indeed, they would have been wonderful PhDs! Ditto the screenshot in the QT'd thread, of the Strassen matrix mult rule, with a little thought about recursion sprinkled on.)
2022-10-06 02:17:20 Excuse me, prime factors of n-1, not n.
2022-10-06 02:16:07 That reminds me of another incredibly beautiful result from computational number theory, Vaughan Pratt's proof that every prime has a short proof that it's prime:(Note, you need to apply the idea in the pic recursively to the factors of n, too!) https://t.co/F7r20CCvNS
2022-10-06 02:05:46 Here is the pseudocode for the Solovay-Strassen primality test. So simple! And fun to think about _why_ it works. Each time your number passes the test it decreases the probability it's composite by a factor 2, so you only need to run it a few times to have high confidence. https://t.co/D4HMLcbO3j
2022-10-06 02:00:12 Also, all three seem nearly inevitable after the fact.
2022-10-06 01:59:55 Reflecting: I know three of Strassen's results: the surprisingly ("impossibly") fast matrix mult algorithm
2022-10-06 01:45:10 anaconda has now spent so much time resolving conflicts that I think I'm going to need to hire it a therapist...
2022-10-05 18:42:08 @TetraspaceWest What bothers me about the notion of "blankfaces" is that it's self illustrating: a deliberate denial of human empathy for people who have chosen (usual very partial, and often for very good reasons) subjugation to a system.
2022-10-05 17:52:25 Reflecting: if you're told (a) the problem
2022-10-05 17:49:03 Also tangential: galactic algorithms: https://t.co/kVDyId5hM1(Note that I expect that neither the Strassen algorithm or the DeepMind improvement are galactic.)
2022-10-05 17:43:25 Tangentially: the screenshot here is of the core piece of a major result in computer science / mathematics
2022-10-05 17:38:27 BTW Strassen's algorithm is a fun thing to think about. It's tempting to convince yourself that it's _impossible_: of course 8 multiplications must be required. And fun to think about why that is wrong.
2022-10-05 17:35:51 @say_cem That's not the subject of the paper.
2022-10-05 17:12:26 Strassen's algorithM: https://t.co/gmbl3yDYWx
2022-10-05 17:11:16 Here's the algorithm for the 4 x 4 case. It's certainly not clear _why_ it works: https://t.co/3GSNW7TklG
2022-10-01 04:38:09 @Smerity Thanks, that's very helpful! Curious: could ideas like Lempel-Ziv coding have been used instead? BPE seems oddly arbitrary.
2022-10-01 04:21:59 Curious: why does GPT-3 use byte-pair encoding?
2022-10-01 03:18:31 @KyleCranmer @ch402 Yes, we've met (v. briefly).
2022-10-01 02:49:17 Fun idea: https://t.co/mNaDJzXrau
2022-10-01 02:45:02 @ch402 (And, bizarrely, all your tweets show up in my mentions except this one. No idea why.)
2022-10-01 02:43:38 @simonw I've been wondering the same. Seems from the outside like it should have been a gmail-like rush... but no-one came.
2022-10-01 02:40:48 @ch402 (Sorry!)
2022-10-01 02:40:32 @ch402 No, I wasn't - hadn't seen it!
2022-10-01 02:39:51 @noampomsky Heh: https://t.co/A7JmZj13uS(I love it from 42 sec on.)
2022-10-01 02:38:08 @BecomingCritter All 3 of us.
2022-10-01 02:37:08 @ch402 Oh, it's 1/20 and 3/40 there.
2022-10-01 02:32:02 Quite a few people are pointing me to the Chinchilla followup (&
2022-10-01 02:30:05 @ch402 Especially since the RG often provides power laws which are near small integer ratios (1/2, 3/4 etc... I can't help but notice the values in those graphs)
2022-10-01 02:28:35 @ch402 The natural thing a physicist thinks of here is the renormalization group and universal scaling laws, a la Kadanoff-Wilson-Fisher-et al (and, more recently, Geoff West et al, in animals and cities).
2022-10-01 01:39:55 @Meaningness Yes, it was that which made me interested originally.
2022-10-01 01:38:40 The transfer to new texts is fascinating: same performance curve, but displaced by a constant: https://t.co/p9Db3S0NX6
2022-10-01 01:36:53 Fascinating: train on a small model to see how well your large model would perform if trained for a long time(!) https://t.co/zfihWqylhu
2022-10-01 01:35:20 They don't seem to search for the cutoffs to the power law behaviour. It's a pity, since it'd be fascinating to understand what sets the scale for the cutoffs. Presumably, in some sense, the inherent uncertainty of language must play a role for the cutoffs on the right!
2022-10-01 01:32:34 ... for sufficiently large compute and parameter count, performance will saturate at a level dependent only on the dataset size (over some range), and that's what's being plotted.
2022-10-01 01:32:33 There's a lot in the paper, and I don't yet understand most of it. In particular, the notion that the three things need to be "scaled up in tandem". If I'm interpreting correctly, it means something like: focus on one of the graphs above (say, dataset size). Then... https://t.co/xVDo0pdcS3
2022-09-29 03:41:53 @DrewFustin @laura_e_langdon @jeremyphoward Horatio, is that you?
2022-09-29 03:37:13 A little comic ancient history (1997): "What makes quantum computers powerful?" by the Quantum Computation Collective: https://t.co/ju5WBEOpu7
2022-09-29 02:52:49 @misha_saul Amazing. https://t.co/sztwfxfC6O
2022-09-29 02:37:09 @lisatomic5 Wow!
2022-09-29 02:33:56 @clarejtbirch Funny, looking at so many of my old Brisbane haunts many are now closed or have narrower hours. I wonder if it's the pandemic?
2022-09-29 02:32:30 @clarejtbirch Really? Wow. I used to love going to Three Monkeys in Brisbane around midnight or so. And there would be a bundle of other cafes open in West End or New Farm.
2022-09-29 02:28:12 @costellowv @humanbrennapede @YoYo_Ma Looks just lovely! Thanks for posting this.
2022-09-29 02:11:30 @nsaphra @quotidiania @lisatomic5 (I'm not sure I want to use the term "genius" of anyone. But if I did, they'd qualify. Ever since reading them they've been all through my thinking &
2022-09-29 02:10:21 @nsaphra @quotidiania @lisatomic5 I'm not sure Elinor Ostrom and Jane Jacobs are the two most important influences on my thinking. But they're top 5. I had a male advisor in a male-dominated field. https://t.co/caaJbjQpPZ
2022-09-29 01:10:58 @PrinceVogel It's quite peculiar, from my pov.
2022-09-29 01:03:19 Things I miss about Australia: coffee shops open to a reasonable hour (say, 10pm or 11pm in the 'burbs, and often much later in more urban areas): https://t.co/uOzeQo8Epd
2022-09-29 00:47:16 @lisatomic5 Lovin' the distribution of replies, which seem to vary between "obvious and boring", "utterly false", and "not nearly strong enough".
2022-09-29 00:40:11 @simonsarris @lisatomic5 (I can't resist saying: she was the first female full professor of physics in Australia , &
2022-09-29 00:34:35 @simonsarris @lisatomic5 I once asked a (now frighteningly accomplished) grad student friend of mine what he'd gotten out of going to Princeton. He replied promptly: "I learned that many of my intellectual heroes are pretty ordinary people too, have bad days [etc]".
2022-09-29 00:30:51 @simonsarris @lisatomic5 An example: when I want courage &
2022-09-29 00:28:34 @simonsarris @lisatomic5 I'm not reading it as antagonistic, because it doesn't contradict anything anyone else has said, AFAICT. It's a nice point. A nice thing about knowing some of your heroes is that you can more firmly grounded in their behavior.
2022-09-29 00:24:06 @simonsarris @lisatomic5 I wasn't waiting for them to come along. Indeed, for the most part the ones I've met were heroes before I met them. (There are also many I haven't met!)
2022-09-29 00:16:05 @simonsarris @lisatomic5 I love that she was your hero growing up. That seems to fit!
2022-09-29 00:15:20 @simonsarris @lisatomic5 I'd name quite a number of people I've met as heroes. Indeed, I believe I've met (very, very briefly) Martha Stewart. At Google, of all places...
2022-09-29 00:08:09 @lisatomic5 Interesting theory. You're more charitable than me
2022-09-25 07:15:35 It should be said that (a) AFAIK, the Oort cloud still has a somewhat funny status, without direct observation
2022-09-25 07:12:27 Hmm. Looks like it depends a lot on details which are (maybe still?) poorly known: https://t.co/eZQ1DT9TiJ https://t.co/wYPBUw3AUq
2022-09-25 07:10:03 Assuming I haven't goofed badly, this would be a fun quals question. The thread on Jupiter was very fun: https://t.co/jvfN2m6sTb
2022-09-25 07:08:57 Curious: is this right? Might the angular momentum be in the Oort Cloud? The Outer Oort Cloud is ~4 orders of magnitude further away than Jupiter, and has maybe a few Earth masses. It'd be close, but just ballparking numbers it looks like it might win. https://t.co/9eevuYNE6H
2022-09-25 06:02:46 Something I like about this model - it sets up a loop where someone/thing else is doing part of the moving. I don't know about others, but I find that _incredibly helpful_ in creative work, especially solo creative work. Something that helps move you along...
2022-09-25 05:57:56 @thcoudreau Thanks!
2022-09-25 04:52:05 @taps Thanks, that is very kind of you!
2022-09-25 04:27:56 @gummadi Thanks.
2022-09-25 03:25:16 (It's perhaps ironic, but few things make me as angry as the deeply ingrained cultural idea that men shouldn't be kind &
2022-09-25 03:22:11 Not those misguided souls who tell guys they need to beat their chests &
2022-09-25 03:00:13 Then you get the award ceremony. Federer, exhausted &
2022-09-25 03:00:12 One of the most incredible contests I've ever watched was the Australian Open 2009 final. The fourth set, in particular, between Federer and Nadal was otherworldly, in part due to the sense that both players were playing at a level of skill and determination never before touched
2022-09-25 02:54:29 ... three of the greatest athletes in history, incredibly determined never-say-die competitors, in tears as one of them retires: https://t.co/oFailqBUBu
2022-09-25 02:54:27 Something I'm enjoying about Roger Federer's retirement: there's a school of thought which says "The trouble with modern men is they're too conciliatory, too much like [the historic stereotype] of women, they're not beating their chests enough" Meanwhile, at the Laver Cup...
2022-09-25 02:02:28 @pmcray I think the examples almost didn't matter. The underlying principles - what you buy by letting stuff get out of control, and why - remain true for social media and AI and many other modern domains. But yet, a 30 year on version would be amazing!
2022-09-24 22:21:52 @albrgr I think the @nytimes coverage of tech is best understood as principally that of competitor. Like reading a FB press release on the Times. Interesting to glance at, but about as unbiased as asking a diehard member of a political party on a policy from their opposition.
2022-09-24 22:18:34 The thesis surprised me when I read it (late 90s) - the idea that we're increasing the power of our technologies by giving up control &
2022-09-24 22:18:33 One of the most fascinating books I've read is Kevin Kelly's "Out of Control": https://t.co/BSY1mdL509
2022-09-24 20:52:25 Er, yes. Always amazed when I meet people who think differently. https://t.co/74q9d648zx
2022-09-24 20:47:57 Wonderful shot! I love how the VAB in the background makes this look almost like a toy model. But the Saturn V was more than 100 meters in height! https://t.co/jkn5fflUbZ
2022-09-24 20:45:05 @johncarlosbaez A large language model.Do you ever use Google or Google translate in writing your essays? Then I believe you use a LLM already in writing your essays. Though that's not the kind of use I was referring to.
2022-09-24 20:42:39 Put a different way: how to use a LLM to design a writing tool that enables you to think much better thoughts?(@robinsloan had some nice prototypes.)
2022-09-24 20:37:29 @diviacaroline @tlbtlbtlb @Meaningness Something I used to enjoy as a prof: convening a small group, usually 1-3 other people, to go into great depth on a subject I was working on. I'd sometimes write several hundred thousand words of notes as exhaust. But it was win-win
2022-09-24 20:35:58 @diviacaroline @tlbtlbtlb @Meaningness Talking to friends about ideas is wonderful, up to some level of depth. But it can feel (very) burdensome to go into great depth.
2022-09-24 20:33:18 @tlbtlbtlb @diviacaroline @Meaningness Oh, this is fascinating! @diviacaroline - if you're of a mind, I'd love a top-level thread about your experience of this - how it works in practice, what you get out of it.
2022-09-24 20:21:41 The use of LLMs changes that context, a lot. But it probably makes it stronger, in many ways.Certainly, tools like Google and G. Scholar and Amazon make my writing far stronger. LLMs may function similarly, in some respects.
2022-09-24 18:09:39 @scott_bot For the second, third, and fourth rules of Citation Club: see the first rule of Citation Club.
2022-09-24 17:49:54 @Malcolm_Ocean @relic_radiation @jessicamalonso @HeidiPriebe1 ... that the overall system was able to catch things like that routinely. It would have been so easy just to blame the people immediately responsible, but the report took the view that it was the design of the entire organization (IIRC).
2022-09-24 17:48:46 @Malcolm_Ocean @relic_radiation @jessicamalonso @HeidiPriebe1 Love "There is no blame".NASA lost the Mars Climate Orbiter due to a unit conversion bug (!!!) https://t.co/FjtNxZExw8I skimmed the accident report years ago, and a really nice thing about it: it said errors like the unit conversion were _inevitable_, what mattered was...
2022-09-24 17:43:44 RT @michael_nielsen: @scott_bot The first rule of Citation Club is that you refer incessantly to Citation Club?
2022-09-24 16:54:38 @HuelHater [Proponents of other ideology] do not focus on the issues I think are the most important. We must reject this dangerous ideology.
2022-09-23 16:30:12 @skdh I don't know how well this would work, but might be worth trying: https://t.co/UX8dKNsc5G(I use pandoc to convert org-mode and markdown files to html. It works pretty well, with occasional hiccups.)
2022-09-23 01:37:30 @AliciaEggert @stewartbrand @longnow Is this at Fort Mason?
2022-09-23 01:35:45 WhatsApp ingroupFB Messenger outgroup
2022-09-22 18:18:05 @patrickc @ArtirKel @stuartbuck1 @elidourado @futurepundit Absent from Dourado's account AFAICS: the mRNA vaccine BNT162b1 that worked, but wasn't deployed at scale because the side effects were more severe than the vaccine BNT162b2, which got the EUA. That was found because of the FDA process, and plausibly _accelerated_ everything.
2022-09-22 18:04:16 @patrickc @ArtirKel @stuartbuck1 @elidourado @futurepundit Janssen, of course, also developed fentanyl, to Stuart's point.
2022-09-21 03:38:07 Oh, I should have mentioned this. One of the most incredible facts about the sun (it's a lot hotter near the sun than on the surface!) https://t.co/D9J8vSF1tz
2022-09-21 02:51:26 Temperature drops off exponentially (at first) inside the Sun, and then faster as you go further out: https://t.co/KFZKIHV1O6
2022-09-20 20:33:20 The inverse Metcalfe's Law, where the more people who have to buy in, the slower and less decisive is your action: https://t.co/p0W7TcyIqW
2022-09-20 18:13:11 @not_a_hot_girl - I don't know- I don't know- The replication work of Doyen et al (and, I believe others, though I haven't read followups) suggests that parts of Chapter 4 are likely wrong. Note that "failed to replicate" doesn't mean "not true"
2022-09-20 17:48:00 Curious: has Kahneman gone further in retracting the discussion in Chapter 4 of "Thinking Fast and Slow" than this 2017 comment? https://t.co/AWPmoXPR5T https://t.co/EcSewMzvjZ
2022-09-20 03:40:53 I love the related point Barbara Tversky makes: https://t.co/6UDYETHgi7
2022-09-20 03:40:07 Lovely analogy: https://t.co/Gany00nJmY
2022-09-19 22:29:27 @pierre_azoulay Thanks! Much of it is an extended riff on https://t.co/DNrUbl6zkU &
2022-09-19 22:23:56 @Singularitarian Really good point, yes!@preskill once told me that a benefit of writing something up just before presenting it in a class was that your understanding often peaked at about that moment anyway.
2022-09-19 21:28:53 Apparently Atlas forms ~10% of Delaware corps(!!!) https://t.co/EP6l2XnwSj
2022-09-19 18:50:33 @krikrissia An enduring problem with being clear: https://t.co/Upp6r5eCxZ
2022-09-19 18:47:34 @rustlezephyr Very much so. The harder you work on readability, the larger the effect: https://t.co/Upp6r5eCxZ
2022-09-19 18:00:47 @Meaningness Don't forget the part where you "discover" the same things again, and start to write them up, only to realize (or, worse, be reminded) that you've already done this. Especially amusing with rediscovering mathematical results you published years earlier...
2022-09-19 17:52:35 I'm really enjoying all the creative people who are liking or retweeting this: https://t.co/XDgi1RlAMD
2022-09-19 17:51:31 @Meaningness Somewhat analogous to automatic damping in a well designed nuclear reactor. But you've got to figure out when to insert the control rods into the whole process...
2022-09-19 17:50:50 @Meaningness The funny thing about that process is that it's not that it ever really terminates (for a complex subject). Rather, you start to decide on boundaries, and so the thought starts to finish with "But I guess Z is out of scope".
2022-09-18 18:07:48 Does anyone know how to do footnotes in footnotes using Pandoc?(Tearing my hair out a bit!)
2022-09-18 06:47:22 We need to aim higher. This thread is wonderful: https://t.co/IikDOBq7WB
2022-09-18 04:58:30 @alexeyguzey An amusing Turing test. If the bike is reasonable, then it wasn't drawn by a human...
2022-09-18 04:54:35 @Meaningness @DVHenkelWallace In 1996, I found a web proxy server that would automatically translate any (English?) webpage into, um, Swedish chef-speak. I was using a friend's computer, and thought I'd turned it off, but it remained enabled. He was utterly mystified, and quite irritated...
2022-09-18 04:51:55 @Meaningness @kanjun Maybe thumpenessay, in the vein of your recent coinages?
2022-09-18 04:48:10 @Meaningness @DVHenkelWallace Jim Henson found a way: https://t.co/YUzmgEeqfA
2022-09-18 04:44:45 Why yes, @kanjun &
2022-09-18 02:42:12 As a writer, discursive footnotes are the place for the things you really want to say, but the reader really doesn't want to read.
2022-09-16 17:38:54 Really fascinating, on the (new!) phases of 2-dimensional water: https://t.co/2WLPNcWyiv
2022-09-16 05:32:44 @hyperphilo It’s not fashionable, but it may be my favorite match ever. The fourth set was played at an absolutely ridiculous level, and the first four sets were all superb. Fed looked ill in the fifth, alas (maybe the mono)
2022-09-16 05:20:30 I’m on mobile, and won’t dig up other clips. A few other fave moments: Federer Agassi impossible shot in Dubai. Fedal Wimby 2008. Fed Roddick Aus 2007 and Wimby 2009. Fedal Aus 2010 (iirc) and 2017. So many others too.
2022-09-16 05:11:47 You’ve got to feel for Roddick. He brought out the best in Federer: https://t.co/UOe1McXyqs
2022-09-16 05:10:05 Lots of great clips of Federer. This is one of the funniest things I’ve ever seen: https://t.co/8xgiGI8kEw
2022-09-16 04:49:43 I very much enjoyed browsing his entire library too. I did a quick sample, and guess we'd have more than 100 books in common, which was unexpected.I'm amused at his apparent discovery of Kissinger in the past 10 years. Not very Scarborough Fair.
2022-09-16 04:41:52 There's something fascinating about the concept of "bigger than the game". Both Federer and Serena had it. https://t.co/X4kSdBZrAWReminded of someone's observation that Einstein's Nobel did far more to enhance the Nobel's reputation than vice versa.
2022-09-16 04:38:47 Art Garfunkel's favourite books: https://t.co/aLnc1wUOPV
2022-09-16 04:24:21 The "fan favourite" is self-explanatory. The Edberg award is voted by the players themselves. It's a measure of F's success how overwhelmingly popular he is with the fans &
2022-09-16 04:21:12 Amusing: Federer and Serena both have to wait 5 years to be inducted into the Tennis Hall of Fame. It's... difficult to imagine them not being inducted, to put it mildly.Two of Federer's more amusing and lesser known records may be witnessed below: https://t.co/pYPE4l0Fmn
2022-09-16 03:27:21 Curious: with the Ethereum merge, is today a good day to buy a GPU for deep learning experiments?If so, what would you look for, and where?
2022-09-15 22:06:26 @dan_abramov I write to understand. If I can "plan" I already understand, and writing is usually pretty dull. OTOH, sketching is often fun, provided I don't get stuck on it as a plan, and remain open to throwing _everything_ out as I understand better.
2022-09-12 17:00:23 This source https://t.co/B6giqj7BFi claims $500M in non-military R&
2022-09-12 16:59:06 This source https://t.co/S1cU5aP866 claims $2.1B spent on R&
2022-09-12 16:55:23 Note also: the research budget and R&
2022-09-12 16:54:06 Anyone know where there is long-term data for the Bell Labs research budget? Would love it over 1950-1990. Inflation-adjusted a bonus!(Searches so far yield a few single year data points, often of somewhat dubious provenance.)
2022-09-12 05:03:50 And then a little later: https://t.co/5IZGF4dkdw
2022-09-12 05:00:54 Not that long after: https://t.co/ReZ3xr9G4t
2022-09-12 04:58:49 A very different era: https://t.co/WmYMNmi9Rd
2022-09-12 04:50:48 @Shripriya (I.e., Djokovic is certainly still a top 3 player. I don't think Nadal is any more.)
2022-09-12 04:50:13 @Shripriya I agree. Though he may cling on. And Djokovic was arguably robbed by not getting the 2k points at Wimbledon, which would have put him in the top 3.
2022-09-12 04:49:28 @papyruspatri Alcaraz is number 1. Ruud is number 2. Zverev and Medvedev have both been in the top 5 for the past year, including many weeks at 1 and 2. If Zverev hadn't gotten injured, the top 3 now would probably be Alcaraz, Ruud, and Zverev.
2022-09-12 04:38:15 On June 30 2003 Roger Federer was the # 5 male tennis player in the world. That was the last time someone outside the "Big 3" wasn't ranked inside the Top 3I expect that run to be broken soon. Nadal is currently clinging on at @ 3. Curious whether the run will last 20 years?
2022-09-12 03:38:10 @JocelynnPearl @stuartbuck1 Yeah, I really enjoyed that too!
2022-09-12 01:24:51 @stuartbuck1 (My understanding is that sport climbing and big wall free soloing are quite different, though overlapping. It's a bit like seeing the world mile champion and the 100 meter champ having a conversation about technique, training etc.)
2022-09-12 01:23:59 @stuartbuck1 Here's Ondra doing what is, I gather, considered by some the hardest ever climb: https://t.co/svVwjHCocGVery different to the free solo of El Cap, but also just wonderful!
2022-09-12 01:23:20 @stuartbuck1 Wonderful doco!You may enjoy this: Honnold talking to Adam Ondra, perhaps the greatest living sport climber (if I have the lingo right): https://t.co/NWZVWIfUcH
2022-09-12 00:40:14 @curiouswavefn Loved both of those!
2022-09-12 00:33:51 What are some of your favourite documentaries (and what did you enjoy)?
2022-09-12 00:32:32 I enjoyed the Hacker News "what are the best documentaries?" comment thread: https://t.co/eimpigaqkI
2022-09-11 23:00:10 Science funding woes, 16th century style: https://t.co/kZuE5NSm2Z
2022-09-08 15:49:28 Freeman Dyson’s account of the termination of the Orion Project: https://t.co/tVamDLIUfl
2022-09-03 04:22:18 @lisatomic5 @uncatherio The typo in this tweet made me smile!
2022-09-02 23:25:40 @then_there_was Do you think that if you tell your wife you love her every day you’d be better off automating it?
2022-09-02 23:19:37 First time I looked I thought “wow, this is repetitive.”This observation has repaid further thought Related: several CEOs have told me they’re surprised how much of their job is repeating themselves.
2022-09-02 23:14:32 Reading back through dang’s comments is very enlightening IMO, especially if you’re interested in online community.(I periodically point people there, which is why I happened to be looking at his profile today.)
2022-09-02 22:43:32 Love dang's HN profile (he co-moderates HN): https://t.co/dHNcHbT1hN https://t.co/e3QWoYptE6
2022-09-02 18:01:38 @uncatherio I wonder. This makes me smile: https://t.co/EGQRu73SiESometimes the best way to get a cookie is just to honestly say "I want a cookie!"
2022-09-02 06:41:11 @not_a_hot_girl @gwern See https://t.co/KagwHAqA8WThe results aren’t large, to put it mildly…
2022-09-02 05:58:56 Modern online dating was really not that far in the future (~50 years). Things have changed.Amused to learn this: https://t.co/q0hox5JhLe https://t.co/r3XlEvp4Na
2022-09-02 05:58:55 Dyson on romance is amusing and rather sweet. https://t.co/nDrC7eDNLO
2022-09-01 22:10:55 These two paragraphs capture Dyson's selection of letters very well: (Incidentally, I did not know this about von Liebig. I'm not sure how it comports with James Scott's account in "Seeing Like a State".) https://t.co/knCd2K3aIi
2022-09-01 22:07:27 Dyson on nationalism in science.(There's a good essay to be written about the proper role of nationalism in science. It's easy to think of examples of both foolish jingoism and foolish appeasement. I'm not sure I have a clear model of who gets this consistently right.) https://t.co/GaI2qWVJcD
2022-09-01 22:02:51 Reminded of the amusing characterization of Dirac's (IIRC) religion: "Our friend Dirac, too, has a religion, and its creed is 'There is no God, and Dirac is His prophet'". https://t.co/EYWBl0Ic5i
2022-09-01 20:47:36 An amusing instance of tacit knowledge, also from Freeman Dyson: https://t.co/ttNKY2qmd9
2022-09-01 20:35:48 Freeman Dyson https://t.co/D0BJWelbGH
2022-09-01 18:53:15 Tolstoy:(Spoilers, technically - it's the end of "War and Peace" - but it's not exactly giving away a plot point.) https://t.co/b8WMCygIYI
2022-09-01 17:47:00 @s_r_constantin @selentelechia Nice. I guess this applies to worry-internet in general. A kind of constant background noise, making it harder to extract signal about when you actually should be concerned.OTOH worry-internet has correctly called 27 of the last 2 major things we should all worry about...
2022-09-01 16:18:50 Very interesting: https://t.co/qUwxFWTZO4
2022-09-01 01:45:08 Amazing!
2022-09-01 01:41:39 Go go Serena!
2022-09-01 00:43:26 @profElanor Quite.
2022-09-01 00:17:40 The paper: https://t.co/QSRCCHF58l
2022-08-31 23:59:38 US Government Bans Export of Nvidia A100 and H100 GPUs to China and Russia (via HN, apparently NVIDIA was informed August 26): https://t.co/dFchDI6if6
2022-08-31 23:37:08 This is great:(from Lynn Margulis's paper on endosymbiosis, one of the all-time great ideas in biology) https://t.co/dnidjIN3Ee
2022-08-30 18:59:12 Chris is a sweet, sincere, kind, imaginative, and brilliant person. He'd like to get married and have children - dating thread below. Someone fortunate is going to marry this man. https://t.co/VwFQenpMxj
2022-08-29 23:55:54 @paulg Curious: what's your theory of why some investors would do damage? It seems that company &
2022-08-29 22:53:17 @ceptional @StuartJRitchie @SMirandaField Thank you!
2022-08-29 17:29:04 "R01 grant": https://t.co/ja1zalkerh
2022-08-29 17:13:38 @nsaphra @StuartJRitchie @bengoldacre No, I haven't, I should, thanks.
2022-08-29 17:13:17 @ch402 I'm not optimistic in your reasoning about what will set the terms of the argument. Much of it will be about power. I expect AI companies looking to make money will make whatever arguments support that end. And they're a lot more powerful than the people they'll be arguing with.
2022-08-29 17:12:29 @ch402 Agreed. It's also interesting to ask _why_ those norms were set in the first place.
2022-08-29 17:09:21 Curious, in reading science, the words about reputation "A very well-respected psychologist...", "From a renowned institution..." etc. The language sounds like Jane Austen's accounts of marriageability...
2022-08-29 17:04:13 Has a history of the replication crisis been written?(I'm aware of, and very much enjoyed, @StuartJRitchie's book, which while not a full-blown history, does have an informative birds-eye view.)
2022-08-29 03:33:54 Cormac McArthy put it nicely for books: "The ugly fact is books are made up out of other books".
2022-08-29 03:32:14 @simonw I suspect top artists develop astounding visual memory. And how good their memory is doesn't seem relevant to me, in any case
2022-08-29 03:30:30 The argument made in my previous tweet bothers me, in part because I expect to see it made often by people who are merely using it to justify what serves their selfish interests.But the mere fact this argument will be (or is) made for bad reasons isn't enough to make it wrong.
2022-08-29 03:27:55 Thoughtful discussion of the ethics of using AI models trained on artists' artworks. Something I wonder about: artists, of course, learn from &
2022-08-29 03:26:28 @simonw Curious: how do you feel about consuming art made by artists who have been influenced by other artists?
2022-08-29 03:13:15 Love this idea: https://t.co/CxBT69vK8y
2022-08-28 19:23:31 Stockfish on Kasparov vs Deep Blue: https://t.co/UvgEtJC64FThe omniscient viewpoint turns out to be very funny: "[Deep Blue] trying desperately not to win, but [Kasparov] trying desperately to lose".
2022-08-28 18:39:40 @SecretsAreCool2 He seems to have a pretty good sense of humour (&
2022-08-28 18:38:59 "You play like a grandmaster" seems to be Stockfish's favourite insult.
2022-08-28 16:36:57 @MoritzSchauer Yes, I know. I think I first heard it in the 80s, as a teenager.
2022-08-28 15:54:12 The channel is hilarious. Here Stockfish 15 offers snarky commentary on Stockfish 8's loss to Alpha Zero: https://t.co/J8gRtcKuPE
2022-08-28 04:13:54 @uncatherio Somehow, though, if you don’t believe yourself in this way, it’s worth noting
2022-08-28 04:12:16 @uncatherio I keep rediscovering the truth if this every few years. It’s such a bad mistake to pretend to be the wrong thing!
2022-08-28 04:10:41 @uncatherio I love Vonnegut’s “we become what we pretend to be so we must be careful what we pretend to be” (iirc).
2022-08-27 21:57:25 Stockfish: "winning is much better than drawing, unless you are Anish Giri" .
2022-08-27 21:51:45 The computers strike back: snarky computer commentary on famous human achievements: https://t.co/5eHN9DdbTU(Stockfish analyzes a famous chess game, with many snarky remarks about mistakes made by the players.)
2022-08-27 21:10:21 Part 2: https://t.co/i2N85mZH6N
2022-08-27 21:09:34 Early visual essay on AI safety: https://t.co/rI9jWL1oXU
2022-08-27 21:05:55 @not_a_hot_girl I don't remember. I expect the older material would generally be better.
2022-08-27 21:03:35 @saintsoftness @NeuralBricolage Ditto!
2022-08-27 21:02:15 I wrote a little riff on this: https://t.co/7jm1pjeT2E(Frustratingly shallow, in part due to lack of expertise, but perhaps mostly a length constraint.)
2022-08-27 21:00:39 Rereading Peter Norvig's very thoughtful comments on statistical models of language: https://t.co/LWebfRd4Ft
2022-08-27 20:57:58 @NeuralBricolage Maybe the many versions of the story of the Golem, Prometheus, Sorcerer's Apprentice, etc. Powerful, but I don't think they're what's being asked...
2022-08-27 20:54:06 @NeuralBricolage I don't, unfortunately. It's an interesting prompt. I've read a number of such essays, but didn't particularly like any of them.
2022-08-27 18:38:58 @ArtirKel No mechanical, the standard plan gives laughably too little liability insurance in many states, extra drivers aren't clear, and the Premier plan price is not transparent. Basically, 3/10. But better than the rental companies, which are 1/10.
2022-08-27 18:28:09 @ArtirKel They're better than most. A very low bar, though. I'd describe them as mediocre.
2022-08-27 18:26:01 A rental car company will one day figure out insurance. And they will take over the world...
2022-08-26 05:25:35 @DecaroliChiara That all sounds great, but especially: congrats on the pregnancy!
2022-08-26 02:29:46 Clarification on the prompt. From @deepfates: https://t.co/igKpelJ27R
2022-08-26 02:27:07 @vgr @deepfates My compliments to @deepfates and the GPU cluster...
2022-08-26 02:26:01 (The initials are overlaid.)
2022-08-26 02:24:34 One of my fave profile pics is this one, of @vgr.Seems to me like an incredible artist able to capture some essence of the person.Turns out, it's dalle2 generated via prompt “cartoon red helicopter” Turing Test Fail Hard. https://t.co/AUu5ntnSjJ
2022-08-26 02:13:52 @vgr Disappointed to realize I'm 4 mins behind in the bad jokes tourney: https://t.co/pTE7HO0c9I
2022-08-26 02:12:43 @vgr Good news, next week there's a tourney at Flushing Meadows: https://t.co/NA2GDuW2RH
2022-08-26 01:54:14 Another marvellous reaction: https://t.co/TtqmXLscFA (ht @JLyle for the reminder of David Blaine)
2022-08-26 01:51:00 @JLyle I heard an interesting discussion of that from Penn (of Penn &
2022-08-26 01:34:44 @ken_crichlow
2022-08-26 01:33:32 @chidiwilliams__
2022-08-26 01:17:14 It's... so clever. One of my favourites - and probably the first one I really liked - was the Chatroulette Wrecking Ball. I can't watch it without smiling - mirroring the people in the video. https://t.co/ltpUkDXFp5
2022-08-26 01:14:59 Fascinating little history from Wikipedia: https://t.co/P4FibdmLPo
2022-08-26 01:14:14 Does anyone know who invented the reaction video form?
2022-08-25 22:58:37 @AlecStapp Upon reflection, "nice" is perhaps the wrong adjective. "Horrifying" might be more apt...
2022-08-25 21:57:17 @AlecStapp A nice review of BSL3 and BSL4 incidents (see especially the supplementary material): https://t.co/qAFo5sLZQ9
2022-08-25 19:25:24 @timoni @hyperpape_id The trivia section at the bottom is interesting. I presume there's some underlying issue...
2022-08-25 19:16:53 @mindspillage @petersuber Thanks Kat (&
2022-08-25 19:16:39 @webdevMason Congratulations to you and @conaw!
2022-08-25 19:14:20 Nice thread from @petersuber, who you should follow if you have any interest in open science: https://t.co/LINWLDZcIP (via @mindspillage )
2022-08-20 02:52:13 (Trying to understand the benefits of rectified linear units. One peculiarity is that once they turn off they can be hard to turn on again - it's not so easy for them to attain gradient to train against. But this doesn't seem so bad if sparsity is a feature, not a bug.)
2022-08-20 02:49:27 Interesting list of proposed benefits for sparsity in artificial neural networks(All from: https://t.co/DgPOGOhXyf ) https://t.co/2ga2ehlt1U
2022-08-20 02:46:18 I wonder how important that is?
2022-08-20 02:46:17 Fascinating, claim that about 1-4% of neurons are active at any given moment: https://t.co/HmpWI3D4SG
2022-08-20 01:41:02 Ajeya and Spencer are long-time EAs, &
2022-08-20 01:36:50 A conversation about Effective Altruism I enjoyed, with Ajeya Cotra and @SpencrGreenberg, on Spencer's podcast: https://t.co/7wdHtanHbu
2022-08-18 21:31:04 @ch402 That seems more like a sign of discovery than a bad sign!
2022-08-18 20:12:11 @ch402 (And it's something Lakatos does magnificently!)
2022-08-18 20:12:00 @ch402 Somewhere Rota points out that mathematics writing would be better with examples of things which almost satisfy definitions of lemmas, to emphasize why the definitions &
2022-08-18 20:11:14 @ch402 In general I'm quite frustrated by his lack of consideration of edge cases and plausible counter-egs. He does consider some, but many fewer than I would have liked (for simple clarity on this kind of question). One is left guessing at the meaning of abstractions.
2022-08-18 20:10:29 @ch402 I suspect it's possible to both confirm and refute your comments with excerpts from Kuhn. It's possible I'm not reading carefully enough. But I don't think so - I think mine is a fair reading, though not the only possible reading.
2022-08-18 18:50:06 @vgr He seems to have mostly delivered.(Great thread!)
2022-08-18 17:11:41 @johncarlosbaez (A friend who is an extraordinary interface designer once observed that the media presents people like Steve Jobs &
2022-08-18 17:09:30 @johncarlosbaez It's funny, because this is not a comment he makes in passing, but is the subject of extended discussion and is quite central for Kuhn.
2022-08-18 17:07:05 @F_Vaggi @Meaningness Very nicely put - yes!
2022-08-18 17:06:11 @johncarlosbaez One thing I've noticed often is how many scientists (especially beginners) take Kuhn as an injunction to do his revolutionary science, but fail to note his comment that it arises _out of_ the pursuit of normal science.(I don't entirely agree, but that's beside the point.)
2022-08-18 16:28:48 @Meaningness Yeah, good point. Fundamental physics seems (mostly) just pretty stuck even considered as "normal science" (to the extent the category makes sense).
2022-08-18 16:23:37 @Meaningness For any reasonable definition of surprise, physics is still full of it, and for the same reason as mol biol. It is somewhat less accessible. Mostly, Kuhn comes across to me as provincial, and not working hard enough to disprove his own story.
2022-08-18 15:56:48 Enjoyed his in passing comment about the importance of astrology for acceptance of Copernicus (apparently it helped the astrologers calendar better)
2022-08-18 15:48:02 @estee_nj It’s a remarkable book.
2022-08-18 15:39:13 TBC I’m often not entirely sure I’m not misreading or misunderstanding him.
2022-08-18 15:38:20 He doubles down pretty hard on crises being internally generated, with theory playing a central role. There’s a strong contrast with molecular biology and microbiology, where sheer exploration - “let’s go find new molecules” etc - is so shockingly generative and surprising
2022-08-18 15:25:09 To clarify: he means something like “novelty with respect to the paradigm”, not something like “we had unanticipated difficulties, requiring improvised solutions”. But I think the former happens pretty routinely, too.
2022-08-18 15:22:36 (I read it at, IIRC, age 17 or 18. Gripes aside, I found it tremendously exciting and enlightening. But rereading today I think his central conceptual distinctions are much less clear cut than he makes them appear.)
2022-08-18 15:20:20 His history often seems rather stylized. And he’s certainly looking for confirmation of his explanatory framework, not refutation.
2022-08-18 15:18:42 Rereading Kuhn. I’m shocked, now, at how strongly he emphasizes that normal science doesn’t pursue novelty. It’s true, sometimes we do largely confirmatory work. But many scientists are, I think, angling forever for surprise.
2022-08-18 15:14:14 “The first duty of government is to protect the powerless from the powerful.” - Code of Hammurabi, circa 1750 BCE
2022-08-18 03:39:52 @natfriedman @andy_l_jones @bradneuberg “ To adapt the nuclear power industry's utopian motto of the 1950s, I want “Art Too Cheap to Meter,” which we are increasingly finding the technology to create—works not shaped by buying and selling, but art with the immaterial condition ...”
2022-08-18 03:37:54 @natfriedman @andy_l_jones @bradneuberg Fun: it’s in this 1998 book too: https://t.co/A03pFFU46q
2022-08-18 02:39:18 @bradneuberg @andy_l_jones https://t.co/UYjFJxALCpIn your sense, @natfriedman
2022-08-17 19:22:48 @swartable Please don't retweet bad threads. It was an observational study, presented as an intervention.
2022-08-17 16:08:42 @SarahCAndersen I really love this. Thank you.
2022-08-17 16:02:30 Early draft of my autobiography: https://t.co/V72R9I70TX
2022-08-17 01:02:04 @zhangir_azerbay @Meaningness @hillelogram Good luck with it - I hope you're able to train your own model, with several orders of magnitude more data! It's a fun idea!(I wonder if it's possible to programmatically generate more Lean data?)
2022-08-17 00:58:27 @Meaningness @hillelogram The same algorithm to answer works here as with anything else: is there a subpopulation of accomplished practitioners who are finding it extremely helpful day-to-day with their particular use case, after months or years of use? If "yes", both hype &
2022-08-17 00:55:56 @Meaningness @hillelogram Opinions about C++, emacs, OS X, &
2022-08-17 00:50:58 @Meaningness @hillelogram At this point I'd guess it introduces enough errors / fails badly enough that it's just a fun toy. The training set must be tiny. But it's fun to see people playing.
2022-08-17 00:30:15 Fun use for GPT-3: attempting to translate informal statements of mathematical theorems into Lean: https://t.co/QTwlmEDlZj
2022-08-16 19:15:07 Very thoughtful thread about the book: https://t.co/PfxjvDMYqi
2022-08-16 14:54:41 Never mind, it's all been cleared up: https://t.co/QaHhT6p9OJ
2022-08-16 14:52:58 This whole thread is fascinating. Note that the picture is messier than the first tweet, but... yeah, coal-for-electricity does seem to be dying. https://t.co/VcU2aiMRcg
2022-08-16 05:47:29 @RosieCampbell 3. Even when the Noble Lie "works" you damage institutional trust. I don't care whether the lies from the public health orgs are "good" or not: I trust them much less than I did two years ago.
2022-08-16 05:43:18 @Marco_Piani @allafarce I don't think so. https://t.co/r4F8GqVPj4
2022-08-16 03:55:35 @Nerland87 @kanjun Really wasn't meant to be a conclusion, so much as gesturing toward some simple models. I find the mindless repetition of the "high risk, high reward" mantra in research really annoying...
2022-08-16 03:53:35 @frances__lorenz @QualyThe @moskov Suspicious that the tweet was entirely ignored, and then you and he tweeted within a couple of minutes of each other...
2022-08-16 03:51:33 @moskov @frances__lorenz @QualyThe Tough crowd.
2022-08-16 03:36:18 One somewhat self-serving remark: with the book launch I'm seeing much thoughtless critique of effective altruism (EA). I'm not an EA, but there is much worthy about it. I wrote up some critical _and_ appreciative notes: https://t.co/EqtOYDZwz2
2022-08-16 03:36:17 Looking forward to tomorrow's release of @willmacaskill's new book (https://t.co/rhNwmhX3UB) I was fortunate to receive a pre-release copy, and have been enjoying making my way through. No detailed comments until I'm fully done!
2022-08-16 03:22:59 Amusing to think about age and response to EA. https://t.co/C7sP3Ks3Ee
2022-08-16 03:19:54 @Nerland87 @kanjun That's an institutional constraint. It's interesting to think about what leverage would look like here. Some kind of impact market might make it possible (basically, borrow against future impact, and then hire people to do the experiments).
2022-08-16 03:03:42 I used to think @frances__lorenz is @QualyThe, but I'm starting to wonder if it's @moskov. God I hope so :-P
2022-08-14 20:03:15 @vgr You need the best lasers, the best squeezed light, the best vibrational isolation along about 17 zillion axes, the best simulations and modelling, the best large vacuum, the best [etc etc etc]. Decathlon is a monofocus by comparison...
2022-08-14 19:59:56 @vgr Just because an end is simple doesn't mean the means is. It pushes so many different limits it's ridiculous...
2022-08-14 19:55:36 @vgr LIGO achieves a strain sensitivity of better than 1 part in 10^22. In some ways it's the greatest achievement of humanity. It really ought to be impossible.
2022-08-14 19:19:17 Some US nuclear weapons use a material known as Fogbank. When they tried to refurbish the weapons they discovered that they could no longer make Fogbank! https://t.co/71nscrqZGf
2022-08-14 18:51:40 @anderssandberg Incidentally, the NSF tried to discontinue the Mauna Loa funding in the early 1960s...
2022-08-14 18:50:44 @anderssandberg Yes: https://t.co/dnalGS1B7U
2022-08-14 18:02:34 I have a few more long-term experiments here: https://t.co/hQ3zyltXVg
2022-08-14 18:01:42 Much harder problem than CO2, because O2 is in a far higher concentration, so you're looking for much (much!) smaller relative changes.
2022-08-14 18:01:41 I don't think inheriting experiments - the divine right of professors :-P - is a good tradition to establish. But something makes me smile about this particular exampleRK, incidentally &
2022-08-14 17:56:30 Interesting to ponder experiments lasting as long or longer than a career. Things like Gravity Probe B (https://t.co/NfIvDK1ZNJ ) or LIGOI love the fact that the Mauna Loa CO2 record was started by Charles Keeling in the late 1950s... and taken over by son Ralph in 2005!
2022-08-14 05:35:49 @aekert Good times!
2022-08-14 05:35:34 RT @aekert: I learned about majorisation the old school way, talking to Michael @michael_nielsen at one of the early gatherings of quantum…
2022-08-14 05:02:38 @shahsagar That really is a very striking trailer. I thought the first half was particularly good - it becomes a bit more generic toward the end, I think.
2022-08-14 04:59:20 @shahsagar Thanks - that's terrific! BP didn't really work for me (MCU movies very rarely do). But that makes me want see it.
2022-08-14 04:55:35 @doc_ochs Yep!
2022-08-14 04:41:50 The end scene of "Arcane". Spoilers, more or less, but... it was watching this that made me want to watch the series, and I thoroughly recommend it: https://t.co/23VGQ0E1iP
2022-08-14 04:39:47 The end scene of "Theory of Everything": https://t.co/nCMFfJwBDI(Not really spoilers, more like a trailer in that it makes me want to re-watch the movie.)
2022-08-14 03:21:09 @daltonmabery @andy_matuschak Got it, thanks.
2022-08-14 03:18:16 @andy_matuschak Also very interesting! Something challenging: lots of tags with one or two display items. Possibly group everything like that under other.
2022-08-14 03:14:16 @andy_matuschak Thanks - that’s a very interesting idea!
2022-08-13 18:22:37 In many ways I'm finding DALL-E most useful as a source of ideas. At the moment the prompt interface is too coarse a tool to get fine grained control. But it's fun for generating quick sketches. Here's Burning Man on the moon: https://t.co/vIO3sI5Vs9
2022-08-13 17:50:33 Change in GDP for US, China, and India. Graph on the left shows change since 1960 (big story: China's rate vs US), and sine 2010 (big story: India now near matching China) ht @OurWorldInData https://t.co/iVPbNVRDde
2022-08-13 17:40:50 "Modern malaise", by @noampomsky: https://t.co/VAh1y2pBtx(Many details can be argued with. But it's very thoughtful &
2022-08-13 17:12:18 Bumper sticker. https://t.co/HQC1PDr8kR
2022-08-13 17:11:45 RT @postquantum: If you like the second law of thermodynamics, you'll love majorization. Would recommend. 5/5
2022-08-13 17:11:21 @LionKimbro Thanks for the suggestion! Curious: why do you find it valuable? For context - connections?
2022-08-13 17:09:57 RT @preskill: I recall when @michael_nielsen gave lectures on this topic @Caltech in 1999. I knew hardly anything about majorization at the…
2022-08-13 17:09:01 @oweissb How was the number 24 consequential in further work over that time?
2022-08-13 04:04:11 @quantum_geoff There may be a Samizdat copy circulating? I only remembered it because, to my great surprise, it was on Google Scholar.
2022-08-13 04:02:58 @quantum_geoff Maybe! If so, I’ve forgotten. I had quite some trouble locating it.
2022-08-13 03:14:02 I really appreciate y'all. My kinda nerds (@NeuroStats @ben_golub, and all the rest of you!) https://t.co/lTFtBGpb9o
2022-08-13 02:51:15 @dabacon Thanks Dave! I'll look forward to it. I love @jvoxfox's reactions, e.g., to "Hurt": https://t.co/AJmu70euKxI'd love to hear one for "The Great Gig in the Sky" from her!
2022-08-13 02:18:25 @pensapreguica How
2022-08-13 02:16:01 @banksean Accidentally.
2022-08-13 02:04:38 @JacquesThibs Oh, that sounds great!
2022-08-13 02:01:10 The thing it most reminds me of is Mark Knopfler's incredible duet with his guitar in "Brothers in Arms": https://t.co/1m02blqY3K
2022-08-13 02:01:09 Spotify: https://t.co/Y6BoxwrVTTOne of the best songs on one of the best albums of all time, IMO.
2022-08-13 01:55:04 One of my favourite facts is that Clare Torry, the singer for "Pink Floyd's" [sic] song the "Great Gig in the Sky, simply improvised the vocals after hearing the music: https://t.co/E6Eq0hj499 She later sued for (and deservedly won) the right to be co-credited as the songwriter
2022-08-13 01:50:44 Quite a compelling graph suggesting an equal royalty split is the way to go in songwriting.Riffing on Truman, it's amazing what you can get done if everyone splits the credit. https://t.co/EldAjK4yKx
2022-08-13 01:09:22 @catherineols I like this comparison: https://t.co/o8Iiar1MpQIn some sense, you're saying there's one loop around some foundational results. But not loops around loops!
2022-08-12 14:24:23 But occasionally a new technique - say, dropout - does work, and it will spread like wildfire. So despite lack of a unifying theory and problems with basic replicability, the field still makes rapid progress toward many central goals.
2022-08-12 14:24:22 Incidentally, quite a few people are replying with variations of "lack of a unifying theory". I suspect that contributes, but isn't decisive.
2022-08-12 14:20:20 Useful example to ponder: https://t.co/WZJ0IdUWvd
2022-08-12 13:43:12 @HackBallet I certainly wouldn't describe it that way, though some would. https://t.co/49vOfgC4I0
2022-08-12 13:33:20 @benblumsmith Quite curious at what made, e.g., medicine move much earlier to pre-registration than, e.g., psychology. Not that it's a cure-all, of course! But a big shift.
2022-08-12 13:31:49 @benblumsmith So, e.g., it's not as though people ignored the Pioneer anomaly. They just didn't have good ideas for fixing it. For p values the problem was known, there were good ideas to fix it... and they were ignored in many fields (though, interestingly, not all).
2022-08-12 13:30:30 @benblumsmith Very interesting. Certainly, there are cases of inconsistencies which persist for ages (e.g., the Pioneer anomaly or the WOW signal or Cabrera's monopole). But those are all cases where we didn't have any solution. The p-value thing is interesting because solutions were known.
2022-08-12 13:20:15 @benblumsmith It's an interesting example. I had the problems with p-values explained to me, clearly, as an undergrad. There are well-cited papers from decades earlier describing the problems. So I don't think it's right to say the problems weren't examined. People just didn't collectively act
2022-08-12 13:01:47 @benblumsmith What's a concrete example of what you're talking about?
2022-08-12 12:48:39 @_leotrs Incidentally, Millikan's work is somewhat unjustly villified today - David Goodstein makes a good case for a (partial!) rehabilitation: https://t.co/CsKIysIkdQ
2022-08-12 12:47:13 @_leotrs I know that particular case well - I've studied the history &
2022-08-12 11:52:49 @Richvn @BrianNosek Or sometimes actual reviewers. So irritating when a reviewer asks for an obviously irrelevant self cite. What are you supposed to do?
2022-08-12 11:51:00 @Richvn @BrianNosek As nearly as I can tell, they're just making the argument that it was influential, but not dominant, and had begun to fade. I can think of papers in theoretical physics like this: silly papers that kicked off fads, but they weren't actually learning much, so gradually faded.
2022-08-12 11:48:56 @Richvn @BrianNosek Remarkable. Presumably, an in-passing nod-of-the-head type citation?
2022-08-12 11:35:12 @BrianNosek The argument I'm making applies just as well - maybe better, in some ways - to esoteric parts of mathematics &
2022-08-12 11:32:13 Just to boil things down: (1) important results usually have many downstream consequences
2022-08-12 11:28:08 There's a nice related story about John Nash's proof of his famous embedding theorem, and a problem in the proof uncovered 42 years later, by @sigma3004: https://t.co/1YSLaGcx37 (Note my emphasis, though.) https://t.co/pXUBLiAjn9
2022-08-12 11:28:07 A nice example of how this process can fail in mathematics, from @littmath: https://t.co/mLQaretyx8
2022-08-12 11:19:15 @BrianNosek ... maybe (1) is not true in social psych? In which case (2) no longer works as a corrective?
2022-08-11 17:12:39 @gordonbrander This is great!
2022-08-11 17:01:15 @kate_sills Berry is much more principled than Hawley. Hawley's main interest seems to be power
2022-08-10 19:48:27 Fascinating read on citation cartels (from 2012, by the wonderfully named @ScholarlyChickn, via @StuartJRitchie): https://t.co/he85kjZPtc https://t.co/UEttugVOHn
2022-08-10 17:36:06 A few short book reviews. Many (not all) are of books I especially like: https://t.co/VGSP98ZJTO
2022-08-10 16:37:57 @danrobinson @kate_sills Yours is a much funnier version of: https://t.co/9N4srDm2GOThough surely it should be "do things that do scale"?
2022-08-10 05:46:09 @kate_sills What a truly remarkable comment. I... what on earth could that even mean?
2022-08-10 05:19:13 The inspiration for my review of a certain book above - one of the endorsements on "Dirk Gently's Holistic Detective Agency": https://t.co/WWD1lXZphl
2022-08-10 01:29:45 Genuinely looking forward to VC's DM'ing me explanations of all the things wrong in Mallaby's book!
2022-08-10 01:29:44 Sebastian Mallaby's book "The Power Law": https://t.co/Tz1R7QavK9
2022-08-10 01:02:58 An additional comment on Lakatos's book, and how it radically changed my thinking about definitions and about categories: https://t.co/6DXH9nTT0p
2022-08-10 01:01:42 I accidentally truncated the review for Ostrom's "Governing the Commons", omitting much of the review. Here it is: https://t.co/VMyxNJ7Cjc
2022-08-09 19:56:35 @Jiajia_20 At a very modest level, with students (&
2022-08-09 19:55:21 @Jiajia_20 Funny, I actually don't remember the latter being part of his argument at all. It may well have been - I read it so long ago! But what I took was the "important new discoveries always look very unlike old, in some significant way [or they wouldn't be important new discoveries]"
2022-08-09 19:53:45 @ambimorph @emily_doolittle Thank you Amber!
2022-08-09 02:11:45 @benskuhn @ArtirKel @Meaningness Nicely structured.
2022-08-08 20:58:35 @gwern "Nice try, sonny, but it's stylized facts, all the way down."
2022-08-08 19:36:11 @Meaningness @ArtirKel Yeah, I dislike the narrative that says "oh, your predictions should be falsifiable". I'm much more interested in "your explanation should (ideally) make predictions that could falsify your explanation, and require you to change it".
2022-08-08 19:34:28 @ArtirKel @Meaningness I first heard "general AI will be upon us in 5 years" in 1996-ish (from someone corresponding with EY at that point, incidentally). I'll bet David first heard it well before thenEventually, such predictions will be correct. But I have a $1k bet (for charity) it won't be <
2022-08-08 19:28:26 @Meaningness @ArtirKel An example of non-determinedly falsifiable: this guy lost a 25 year bet... and it basically changed his opinion not one whit. (I believe this is true of Paul Ehrlich as well). There's some strong notion of falsifiability of _explanation_ missing... https://t.co/58YUq9dnyg
2022-08-08 19:26:46 @Meaningness @ArtirKel There's surprisingly little (to me) that seems determinedly falsifiable in much writing about long-termism. Where by "determinedly falsifiable" I mean falsification would cause a real change in underlying beliefs.
2022-08-08 17:35:20 Robert Skidelsky's abridged one-volume Keynes.(I read both this and the full three volumes. The one-volume is better!) https://t.co/tY0eDH4ADX
2022-08-08 17:35:19 Martin Amis, "The War Against Cliche": https://t.co/JRKdQ1PesO
2022-07-25 18:04:50 You can set it up as a clash of "figuring out the truth" versus "simply acting for your beliefs". And asking the question: why in some worlds does the former seem to be of so much more interest than in other worlds?
2022-07-25 18:02:18 I've been seriously involved in open science advocacy, and glancingly involved in various other policy and advocacy worlds. I very much enjoyed this paper as a distillation of such worlds from someone who has been far more deeply involved.
2022-07-25 17:59:10 https://t.co/Omr1edlKSU
2022-07-25 17:57:47 https://t.co/7xa9WVoGx7
2022-07-25 17:53:35 A man who has been in the trenches https://t.co/38HEyO8hlS
2022-07-25 17:51:24 https://t.co/5xyR9pnB6n
2022-07-25 17:50:29 A _lot_ of details then follow. I usually glaze over simple economic models, but in this case I find myself surprisingly interested. Bad to confuse with truth, of course, but there's a lot of useful intuition pumps / simple models in this paper. https://t.co/rOgKAMI722
2022-07-25 17:46:40 This first paragraph is really useful. I recognize myself in it (sometimes). I also know that it's strongly turned me off some advocacy communities where I believe strongly in the cause, but want it qualified by evidence and good argument: https://t.co/d5zyKgcIqn
2022-07-25 17:46:39 I enjoy this, too: the model doesn't need evil actors. No, assume altruistic people who really care! https://t.co/saP9Qk4uQG
2022-07-25 17:46:38 I'm honestly on the edge of my seat, even though the abstract anticipates the answer (people don't really want to know, for self-interested reasons). https://t.co/XovIfcypRe
2022-07-25 17:46:37 I must admit, I'm just enthralled so far. I realize this probably isn't everyone's cup of tea, but from my point of view he's stating key civilizational issues with just stark clarity (&
2022-07-25 17:46:36 This is somewhat depressing, though perhaps not surprising. https://t.co/bJxXHCnKVE
2022-07-25 17:46:35 The abstract may be loosely paraphrased as "figuring out the truth may undermine your political agenda, so we underinvest in figuring out the truth". Of course, in many cases that's not correct.
2022-07-25 17:46:34 Fascinating paper on rigorous evaluations of difficult policy questions, by Lant Pritchett: https://t.co/nV2EwV1GLs (ht @albrgr )A few immediate impressions to follow, very loosely held and non-expert! https://t.co/GohLNLm5gc
2022-07-25 04:40:30 @betterstreetsai Thank you.
2022-07-25 04:36:44 @betterstreetsai Can you share some of the prompts?
2022-07-25 04:27:58 Also, I just really like it as an application of AI-generated images!
2022-07-25 04:27:12 I like this. I wish I could talk it over with Jane Jacobs!(One of many things I love about JJ is how enthusiastic she seemed to be in her writing, for new ideas and new possibilities.)https://t.co/IYPETqZ0K8
2022-07-25 04:20:12 These are fascinating: urban redesign, using DALL-E: https://t.co/eleTzGtwKW
2022-07-22 00:53:16 @stuartbuck1 More context: https://t.co/4ivMxNSlNu
2022-07-22 00:52:21 @stuartbuck1 I'm just referring to this exchange
2022-07-22 00:44:35 @stuartbuck1 I wasn't talking about the funder's point of view
2022-07-22 00:08:40 @cpiller @schrag_matthew @US_FDA @NIH Thank you both for investigating this.
2022-07-22 00:08:01 Thread by the investigative journalist who wrote the story: https://t.co/wo4xrAIAJO
2022-07-21 23:53:17 There is something particularly painful about what appears to be someone who, for a small career benefit, has considerably set back research on a disease that causes incredible misery and pain to many millions.I am rarely angry, but this is just appalling.
2022-07-21 23:40:56 This is a painful, staggering read: https://t.co/5SwXeQgtiI https://t.co/qnDVllh1CD
2022-07-21 23:18:30 The transition between these sentences is a great diss! https://t.co/uwc0SmXQoj
2022-07-21 23:14:40 The whole pitch was actually really well thought out. I usually ignore such emails, but I wrote back to tell him I was totally the wrong person to pitch, but I loved his email. He followed up occasionally &
2022-07-21 23:14:39 First time I've heard that from a recruiter!Back in 2015 I had an awesome recruiter from AOL who started his pitch with something like: "I bet this will come as a surprise, but AOL is still around! And more interesting than you think!"
2022-07-21 23:11:50 Brb, updating my (non-existent) LinkedIn with my Las Vegas expertise: https://t.co/HK8teSjgnS
2022-07-21 21:38:24 @stuartbuck1 Believe me, I understand this point...
2022-07-21 21:29:40 @stuartbuck1 (It's complicated by the fact that in one case it's a purely public good produced, while in the other it's a mix of private and public goods.)
2022-07-21 21:28:36 @stuartbuck1 The investors "benefit" too, in a very similar way...
2022-07-21 21:27:40 @stuartbuck1 That's fine, if you want to recast that as a success. Though I suspect few scientists would consider that as exactly the same level of success as finding it was connected to pancreatic cancer.
2022-07-21 21:26:41 @stuartbuck1 The situation is the same as with the LPs. The difference is that the LPs have far more visibility. But it doesn't change the actual risk. The taxpayers are bearing a risk. The fact they're not aware of it doesn't change that fact.
2022-07-21 21:11:07 @PDWhittington @Kirsten3531 I appreciate the reassurance...
2022-07-21 21:07:26 @stuartbuck1 The taxpayers are still out the money. They don't complain like the LPs, but it's still a genuine risk. The NIH Director is just dissociated from that risk.
2022-07-21 21:06:18 @PDWhittington @Kirsten3531 Written and deleted, I'm afraid. You'll need to use your imagination!
2022-07-21 21:05:26 @stuartbuck1 From a draft (with @kanjun): https://t.co/wS3JKzdbJV
2022-07-18 15:29:28 My last few days of paper-tweeting have caused me a net loss of ~5 followers.Clearly the correct equilibrium is 0 followers, and I shall act forthwith to that end!(Joking. Sort of.)
2022-07-18 15:21:59 (And, er, insofar as I'm a member of the technocratic class, I guess I'm a low-grade villain in the book, too. But I'm a lot more Jane Jacobs than Robert Moses...)
2022-07-18 15:20:07 @henryfarrell @delong Thank you. I missed this point, quite badly (&
2022-07-18 15:01:44 Scott's book is a (superb, pointed but sympathetic) criticism of high modernism. Incidentally, reading the unsympathetic reviews of SLaS by economists, one is struck by an unacknowledged point in those reviews: economists are implicit low-grade villains in the book.
2022-07-18 14:59:00 (Incidentally, something I only just noticed: the Suez Canal &
2022-07-18 14:57:08 Interesting to compare James Scott's description of high modernism with both progress studies &
2022-07-18 01:41:12 @tylercowen Netflix CEO publishes paper explaining why video has to change
2022-07-18 00:12:30 @brianluidog Underrated tweet.
2022-07-18 00:09:23 @brianluidog @paulg @starsandrobots You've gone native. "Maccas"!
2022-07-17 23:49:04 @paulg @starsandrobots Certainly, the replacement paper straws just... don't seem to work. They'd be better called paper unstraws, or paper blockages, or something.
2022-07-17 23:47:30 @paulg @starsandrobots I looked years ago for why the ban on plastic straws came into effect. This seems to be (part of?) the root, a focus not on mass but # of items: https://t.co/I8cZomYcKaNot as stupid a reason as I feared. This is, admittedly, not a high bar.
2022-07-17 23:18:49 @infinitsummer Overthinking as a service. (One of the most endearing traits of the Bay Area, IMO.)
2022-07-17 23:10:54 @bertgodel Glad you like it!
2022-07-17 21:48:28 @aaronclauset Seems plausible. I was very struck to see your results (roughly) repeat, with what seems likely a quite different audience. Though I'd guess we both (mostly) have male followers.
2022-07-17 21:07:32 @pablo_gps @aaronclauset I can't speak to the original intent. I re-ran it because I realized I was surprised by the outcomes (mainly 1 >
2022-07-17 20:58:14 Very similar to the original results: https://t.co/C6s6F6k0EqI repeated the poll because I was somewhat surprised by the original results, &
2022-07-17 20:36:04 All other things equal, which of these known biases in the academic ecosystem has a more harmful effect on scientific discovery? (repeating a poll by @aaronclauset, more below)
2022-07-17 20:24:22 @littmath On the other hand, I find that my thoughts about stationarity really haven't changed much.
2022-07-17 19:42:33 @phillstephens @RichardMCNgo I just mean, I don't open with "So, should Canada join the EU?"(A stolen question, actually, which I got from a friend. I've used it... maybe 3-5 times? The responses are interesting.)
2022-07-17 19:40:32 @phillstephens @RichardMCNgo Or maybe "early" question is better.
2022-07-15 22:51:35 An interesting reply to the paper, basically going full AI and data-mining: https://t.co/JaNZDIMRZI https://t.co/e2TxDwHc7Z
2022-07-15 22:23:47 @chrislintott That's a pretty great headline!(Past my Atlantic free article limit for the month, alas!)
2022-07-15 22:21:54 @chrislintott ? Not sure what you're referring to. Sounds like a horror movie...
2022-07-15 22:20:21 Oh, good question about whether some animals can see Andromeda well:(IIRC some frogs have near single-photon sensitivity, in some wavelengths.) https://t.co/WgRJbuLB6g
2022-07-15 22:18:11 Up and to the right: https://t.co/q3N8TPM2Mx https://t.co/jl4PT7jn62
2022-07-15 21:52:13 A fun, stimulating paper. However, I'm very bothered by these questions about the correspondence between their notion of a "field" and on-the-ground practice, and what this means about the results.
2022-07-15 21:47:14 Time to reach canon (right-hand graph): Looking at the scatter plot, not the fit lines, this is a fairly weak effect, except in computer science. Still, interesting. https://t.co/J6sz20w0kd
2022-07-15 21:44:19 (Not sure at all about the last few tweets, just playing with ideas!)
2022-07-15 21:42:48 You might retort "But X will come to dominate A". That's not really right. AFAICT human cognitive limits seem to mean that fields fission when they get too large.
2022-07-15 21:42:47 Put another way: suppose a "field" A is really an agglomeration of three largely independent subfields, X, Y, &
2022-07-15 21:34:01 @Meaningness I'm more inclined to believe Dirac's explanation (from that draft with @kanjun which you've seen a much earlier version of...). https://t.co/Jm3PmBQW8J
2022-07-15 21:31:39 That makes me wonder about this paragraph too. Subfields (or -specialties) have their own canon, too. And so in many ways what this looks like to me is less about papers than about relative stasis in competition between _subfields_, _not_ papers. https://t.co/KwKqcvzIDe
2022-07-15 21:30:16 ... what's going on subspecialty by subspecialty? Are the Gini coefficients rising in all the subspecialties, including new ones that arise over time?Or is the overall effect being caused by something else?
2022-07-15 21:30:15 Upon more reflection I really don't understand this paragraph at all. EE isn't a single field. It's an agglomeration of many, many subfields. While I'm sure there are generalist EEs, many that I meet are concentrated on some particular subspecialty. So... https://t.co/aR95thQQEI
2022-07-15 21:21:51 (I don't quite understand the definition of the correlation here, so that's just a gist, not a detailed description.)
2022-07-15 21:21:50 The second graph is also amazing. IIUC, for small fields, the top-50 papers from year are only weakly correlated. But for larger fields - 10k papers / year and above - the correlation between years gradually becomes very strong. https://t.co/K0fy9CTjVZ
2022-07-15 21:19:37 (Bothered by not understanding the extent to which the paper's notion of a field corresponds to researcher's individual self-conceptions. E.g., maybe "machine learning" counts as a field, but some researchers think of themselves just as "ML for vision" people. Etc.)
2022-07-15 21:18:02 When a field produces 100-1000 papers per year, I suspect most specialists in the field at least glance at nearly all those papers, and have time to read in-depth quite a few. But at 10,000 papers that gets really difficult.
2022-07-15 18:51:59 They (mostly) bite the bullet here:(Kudos to the authors for doing so!) https://t.co/Tsg5SfvPW2
2022-07-15 18:48:54 They might, for instance, just be due to unhealthy social dynamics. Maybe there are later papers which are fantastic contributions to our understanding, but they never percolate out to be as widely known as some lesser-but-much-earlier canonical results.
2022-07-15 18:48:53 I keep coming back to the same question in my mind: to what extent are these results due to social dynamics, and to what extent are they due to things about the structure of knowledge?
2022-07-15 18:44:04 @ambimorph Very!
2022-07-15 18:38:48 @timhwang Epigraph for a science horror movie...
2022-07-15 18:33:36 That is: they are genuinely new and, if I am honest with myself, surprising information to me.
2022-07-15 18:33:12 Much of what's in this paper is not surprising to me. But something I really enjoy about the last two screenshots: they're at least somewhat surprising.Yeah, I can make up post hoc justifications for these facts. But I could have done the same had things gone otherwise, too.
2022-07-15 18:31:05 Really fascinating. In small fields, papers become canonical slowly. In big fields, they either become canon immediately, or never!!! https://t.co/quFlpMbwdc
2022-07-15 18:29:30 @iphigenie It'll be a good deal for me, then!
2022-07-15 18:29:02 When a field is large enough, the canon becomes stable(!!!) https://t.co/XkRPk3acyX
2022-07-15 18:26:40 _are_ apples to oranges!My kingdom for an edit button.https://t.co/Y1zAFGzjjM
2022-07-15 18:23:00 These are all quite different to one another, and lead to calling quite different things fields.
2022-07-15 18:22:59 It's fascinating to ponder the time dynamics. "Electrical engineering" is not a static concept
2022-07-15 18:18:33 Interesting to ponder to what extent this captures: (a) social dynamics
2022-07-15 18:17:52 Interesting: using the Gini coefficient (and changes in Gini coefficient) to capture citation inequality. https://t.co/MwskYI0Lnv
2022-07-15 17:46:44 @Jonathan_Blow If, OTOH, many of the classes had been like my thermodynamics or experimental physics classes, I would have dropped out in the first year.
2022-07-15 17:46:04 @Jonathan_Blow Yeah. I wanted to do physics or computer science, but math was far more flexible in class structure, so you could pick the best profs. So I majored in math. Probably >
2022-07-15 17:24:04 @Jonathan_Blow That was, indeed, exactly what a group of about half a dozen or so people in almost every year did. And I think it's been surprisingly productive - many of those people have made important scientific discoveries.
2022-07-15 17:22:58 @Jonathan_Blow I was weirdly fortunate in my undergrad. It was very laid out what you needed to do to get perfect grades. I asked a prof why, &
2022-07-14 03:04:21 @ctitusbrown
2022-07-14 03:01:30 https://t.co/Wn7yR5mD1z
2022-07-13 19:25:16 This includes a large component of "improved understanding".Why yes, neither of these is easily quantifiable, and there is room for enormous disagreement. This is a feature, not a bug.
2022-07-13 19:20:24 @itsjaneflowers I like it, tbh, &
2022-07-13 19:19:37 Benefit to humanity. https://t.co/K3K9Fm7V5b
2022-07-13 18:13:25 @llimllib Doesn't sound like you're applying the advice at all. Sounds like you're taking out any personal feeling or insight.
2022-07-13 18:10:04 @vgr Marriage announcements...
2022-07-13 18:08:20 @nickcammarata It is, ofc, useful to have a way of picking out pithy or memorable statements of fairly obvious ideas! But I doubt it's the foundation of good writing.(I use "unusually low likes" to help filter for unusually good tweets of mine. Also not reliable, for related reasons.)
2022-07-13 18:06:08 @nickcammarata Twitter likes are best at picking out pithy statements of fairly obvious (relative to context) ideas, maybe with a small twist.
2022-07-13 18:05:19 @nickcammarata You're optimizing for approval, not insight. The two aren't completely unrelated, but it's very complicated. Sometimes they're strongly anticorrelated. And sometimes correlated.
2022-07-13 17:56:40 I do love Susan Rabiner's observation that what you publish can be, at most, only a tad better than the best of what you take out. So you ought to hope to take out some wonderful material!
2022-07-13 17:52:30 (Revising / editing.)
2022-07-13 17:52:07 "Take out the boring/wrong bits" is the most obvious, highest reward, and often hardest to implement writing advice.But I _really like_ the boring/wrong bits! I feel like Smeagol. Those bits are My Precious! Mine! Mine!
2022-07-13 06:19:56 RT @tomstafford: Revisiting @shancarter and @michael_nielsen on https://t.co/LXvEaxoMSi Using Artificial Intelligence to Augment Human Int…
2022-07-12 23:03:35 @tobyordoxford I absolutely love the images, but it bothers me (a lot) that the 8-pointed form is an artifact of the imaging system! It makes me wonder what else I'm misinterpreting.By comparison, things like the use of false color seem much easier to reason about.
2022-07-12 22:26:46 https://t.co/cHJJOFcATS
2022-07-12 20:32:03 @johncarlosbaez Or the many far more detailed and in-depth statements they've made, not in the artificially shoehorned constraints of a newspaper article?
2022-07-12 20:31:20 @johncarlosbaez You mean, like this: https://t.co/4hVavT4D3D
2022-07-12 19:41:22 Utterly perplexed by people claiming to support action on climate who then poo-poo action on climateYeah, focusing only on carbon capture or pretending it's a silver bullet would be a terrible idea. But attacking sensible work on it is _also_ a terrible idea. https://t.co/WcXcVrmzEe
2022-07-12 17:25:08 Also: the JWST runs a proprietary version of Javascript 3 (!!!) made by a bankrupt company(!)https://t.co/CMzDejzABo
2022-07-12 17:23:38 Most of these images are false color, BTW. The wavelength range is, according to Wikipedia, 600 nm (orange) to 28 microns (mid-infrared).
2022-07-12 17:19:24 Fun thread: https://t.co/MJ7uPqZR1s
2022-07-12 17:07:33 Via HN, some high-res JWST images: https://t.co/JXwIlqVGck
2022-07-12 04:15:01 I assume that graph shows https://t.co/AiLqm4BsuG earlier, though I don't actually know: https://t.co/xS7PjvkJfx
2022-07-12 03:59:56 @ArtirKel Er, yes, I know those things (and a lot more)
2022-07-12 03:49:23 Fascinating, though I admit I understand very poorly what drives exchange rate fluctuations! https://t.co/VKSZRMTz6n
2022-07-12 03:47:09 @frances__lorenz @ns_whit We have several, most recently from (IIRC): https://t.co/nBgxJgK3px(Earnest answers a specialty...)
2022-07-12 03:05:13 Application to AGI &
2022-07-12 03:03:32 I tried to find and post this quote in my library, but @curiouswavefn saved me the trouble. Dyson captures part of this very, very well: https://t.co/jF6AmYX3lu
2022-07-12 03:02:31 The story as told in the documentary casts Oppenheimer as hero and ultimately victim. You could easily cast him as villain. Neither view is correct, IMO, and both do the events a tremendous disservice. Still: I strongly recommend the doco.
2022-07-12 02:50:36 ... the headlong pursuit of the technical goal, with little thoughtful reflection on the impact on humanity. The Sorcerer's Apprentice, brought to life, again. https://t.co/3YSzU652yc
2022-07-12 02:50:35 "The Day After Trinity", telling the story of the atomic bomb, through the particular lens of Oppenheimer's life (ht @laurademing): https://t.co/3gUe0tFzVpRemarkable, harrowing. Struck by many things, including the lack of Japanese perspective. A subtext throughout is...
2022-07-09 00:18:40 (The first really detailed explicitly non-fiction discussion I know was from I. J. Good of "Good Lord Here Comes Lord Good" fame. https://t.co/UyTqZqfmMk )
2022-07-09 00:14:31 Vinge certainly wasn't the first to discuss super-intelligence in detail. But his 1993 paper was nonetheless really influential. Just recalled that his deadline ("within 30 years") is about to arrive: https://t.co/lFyEwLSTmY https://t.co/91Gic4BLvv
2022-07-08 13:25:11 Fantastic news and thread: https://t.co/zjeKYy7suj
2022-07-08 01:59:28 Big Sun over the Pajarito Plateau https://t.co/CNAaTZBMoT
2022-07-08 01:49:20 @TrollColors In financial markets you're coupled to a positive sum system. In casinos you're coupled to a negative sum system. I know which one I'd rather be playing (and it has a lot of collective social benefit, with a little care).
2022-07-08 01:41:19 Amusing: I've never gambled in Vegas, AFAICR. I have gambled briefly at a casino and at a racetrack elsewhere. Interesting in many ways, but lost its appeal quickly (partly because I wanted it to: not habits I want!)
2022-07-08 01:39:07 @ThomasMiconi @andy_matuschak Yes! Amazing place!
2022-07-08 01:37:46 It's a text to a friend, not a scientific paper. I'm only 6 foot. If you assume the average height of men there is 5 foot 8 in, with a sd of 2.5 inches, &
2022-07-08 01:23:36 Something I love, many of the *same* people go to visit the Strip, the Grand Canyon, and (@andy_matuschak points out) the Hoover Dam, all on the same trip. Three very different types of astonishing thing!
2022-07-08 01:14:08 Notes on the Las Vegas Strip. Written in text for a friend who likes urban planning and design, so focuses on that. From 2018, but just now cut-and-paste to my notebook for fun: https://t.co/RoRqxlZikO https://t.co/u5y35WUPPl
2022-07-08 01:08:32 @BrianNosek Good year for people named Nosek. Maybe you can team up to overcome Iron Man?
2022-07-08 01:06:09 @scott_bot (I think you can reasonably argue that WHO was a month late in declaring the pandemic, maybe even two. Hindsight is 20-20. But I think we're still quite a ways from the fourth year.)
2022-07-08 01:04:33 @scott_bot I had to think about it because we're early in year 3. The pandemic was declared mid-March 2020.
2022-07-08 00:54:42 @olingern Curious: do you still see those, or not? (The site should be un-hacked. But I just want to check.)
2022-07-08 00:41:38 @karpathy Love the sorting function.(Fun list!)
2022-07-07 00:11:31 @__femb0t Source?Fascinating map!
2022-07-06 06:00:30 @iamwil It’s a site repair, will take a day or so. Thanks for pointing it out!
2022-07-05 22:03:19 @ChadRigetti I'd never read the entire sonnet before now. Thanks for sharing! https://t.co/zhrAL3W4NG
2022-07-05 21:30:51 @stuartbuck1 @lukeprog @jgbarzyk Fascinating, thanks!
2022-07-05 21:30:42 @lukeprog @jgbarzyk Thank you for these, Luke, very helpful!(I wish I had a list for DARPA, which can be pretty, er, stringent at the program level. I suspect the raw data on discontinued programs for DARPA would be fascinating.)
2022-07-05 19:45:10 @provisionalidea @jgbarzyk Thanks!
2022-07-05 19:42:34 @provisionalidea @jgbarzyk I didn't say it was a sign of quality.I'm asking for an example.
2022-07-05 19:12:13 I know of two others: (1) IIRC the director of DARPA at one point refused an increase in funding, saying it would make the agency less effective.
2022-07-05 19:12:12 A very interesting example: https://t.co/fBQm2R4gZQ
2022-07-05 19:10:04 @_neilhacker Really interesting.
2022-07-05 19:06:20 @AWMundy What do you have in mind as the critical pieces?
2022-07-05 19:05:23 @jgbarzyk Do you know of one that's more than mildly critical? Eg, advocating large cuts, that kind of thing?
2022-07-05 18:37:33 Curious: funders often carry out evaluations of their own effectiveness (sometimes "independent", s'times not). Unsurprisingly, most such evaluations conclude they're doing superbly well, with just a few minor issues.Do you know of any really critical evaluations?
2022-07-05 16:29:06 Something I enjoy is that the person saying we shouldn't play at being Gods is a computer vision / ML person. That is, they're teaching sand how to see. They're a God who has turned pro :-). I think this is (probably, mostly) marvellous!
2022-07-05 14:13:06 @ohcapideas I love the ending to Gatsby, too.
2022-07-05 13:48:11 @TheRealKildare Good grief. I read your tweet, and immediately heard the music. I hadn't realized _I_ had it memorized!
2022-07-05 13:47:30 @Kanesburner Good advice. Hard to remember: self-righteousness is deliciously attractive. But also self destructive.
2022-07-05 13:45:39 @NowhereLikeNow @jhagel (Tolkien returns to this point over and over, and makes it in many different ways. I believe it must have been central to how he conceived of story.)
2022-07-05 13:45:00 @NowhereLikeNow @jhagel Fascinating distinction. Reminded of Tolkien on stories not really ever ending, but going on and on, forever. And we just see tiny slices through them.
2022-07-05 13:43:29 @MakemineB I first learned of this in the movie "Four Weddings and a Funeral", where it was just devastating.
2022-07-05 13:42:32 @leader_kate Sentence-for-sentence, Nabokov is astonishing.
2022-07-05 13:40:51 Many beautiful responses. One I particularly enjoyed: memorization as a gift for someone you love: https://t.co/SofffjmNgS
2022-07-05 13:40:01 @noahlt The notion of memorization as a gift is a beautiful one.
2022-07-05 13:39:36 @JoshuahHeath @rshindell Thank you, these are wonderful.
2022-07-05 13:39:19 @astupple Me too. An incredible piece of writing (and thinking).
2022-07-05 13:38:13 @jesqa_ One thing I like, very much, about older texts is how much common humanity is revealed by them. That one -- you can feel the hurt, and the cry for help.
2022-07-05 00:22:23 @anniefryman I find (1) and (2) mostly easy to remember, and (3) almost impossibly difficult. I find self-forgiveness very hard.
2022-07-05 00:12:32 @anniefryman Oh, very interesting about the Lord's Prayer.Years ago I noticed one (selfish) reason to be kind to others is that I think it helps a little in being kinder to myself. I somehow feel I deserve it more. It seems closely related to that line.
2022-07-05 00:04:25 @anniefryman WB!!!(Lovely!)
2022-07-05 00:00:54 @LibertyRPF Try this: https://t.co/RZK33Otnwd
2022-07-04 23:59:19 @LibertyRPF Inconceivable!
2022-07-04 23:44:10 https://t.co/Tjaz9PX0b5 https://t.co/Mcg55ofBC9
2022-07-04 23:32:41 Quite a few people have told me that their writing improved considerably after they began to memorize striking passages of poetry and prose.
2022-07-04 03:20:03 @olingern Thanks for letting me know. I wasn’t aware. How annoying…
2022-07-03 06:16:33 Michael ->
2022-07-03 05:29:30 @JacobTref @DanielleFong (But what would I know? My guesses should be treated like a large Co CEO's opinions about quantum mechanics...)
2022-07-03 05:28:44 @JacobTref @DanielleFong One guess: large firms often establish moats which mean they have near-monopolies, &
2022-07-03 04:12:07 @JacobTref @DanielleFong What’s the lfwp?(Curious, too: I hear about this effect often specifically from CEOs.)
2022-07-02 21:59:20 A fun challenge: to write discovery fiction about using RCTs to discover evolution or Newton’s laws or the structure of DNA or etcetera.
2022-07-02 19:55:34 @JakeOrthwein Yup!
2022-07-02 19:55:09 @JakeOrthwein And this: https://t.co/gYG5Vsq8Y7(Wow: searching my past tweets for "laughing" is so fun!)
2022-07-02 19:54:09 @JakeOrthwein Thank you for sharing this. Reminded me a little of: https://t.co/liJnaFwq9NJust made my day better!
2022-07-02 19:09:39 https://t.co/KwVhK0WCpR
2022-07-01 15:13:49 @spearofsolomon Good movie plot!
2022-07-01 14:51:33 @akbirthko :-)
2022-07-01 14:45:11 @TaliaRinger ?
2022-07-01 14:44:55 I've heard about people using language models as therapists. I wonder what happens if you prompt it by telling it it's the greatest therapist in the world? (Or similar variations.)Hmm. What's the best way to play around with the models?
2022-07-01 00:55:54 An interesting problem I've heard many people mention with EA: giving can often be based more on ideological alignment than competence.(I've heard from some seriously ticked off people about this, credibly.)
2022-06-30 22:57:36 @wjzeng I nearly mentioned Robin Hood in the tweet (I was quite surprised when I learned of their trick for being free) . So: yes, that seems plausible!
2022-06-30 22:50:29 Incidentally, a fun variation question on the first tweet: do we have transaction fees on programmable money in 200 years? I doubt it!(Yes, I see the obvious intrinsic argument for. But I'll bet it can be beaten.)
2022-06-30 22:44:38 @etiennefd Most people are unable to do this effectively. It's a difficult skill, and requires both difficult general skills, and intense field-specific knowledge. Worse, a lot of people don't realize just how far they are from an understanding of the frontier.I do not except myself.
2022-06-30 22:41:13 @just_steve_h Economic growth is just a way of saying "we get better at making stuff". So you can do more things with fewer resources. Quantitatively it's a nice approximate short term model, but makes little sense over centuries (with apologies to Angus Maddison) except in a wishy-washy way
2022-06-30 22:32:52 @rossbyrd Nor do I understand money very well
2022-06-30 22:31:13 @rossbyrd I don't know
2022-06-30 22:29:20 It's fun to think about cost modelling. It's tough to make permanent space settlement work, unless you're willing to wait a long time, and spend a _lot_ of money (trillions, maybe quadrillions). I remain optimistic, though, that it will pay for itself many times over.
2022-06-30 22:16:43 I know I'll need to mute this thread
2022-06-30 22:14:21 In 200 years do we have programmable money, smart contracts etc? Are they provided by government? [A: likely yes
2022-06-30 21:15:36 @LauraDeming In fairness, "You should not buy a donkey which brays
2022-06-30 16:00:28 The post is well worth reading in full.
2022-06-30 15:51:52 https://t.co/guSvDpIPwQ https://t.co/JspmI4aJcN
2022-06-30 12:20:16 @HeidiBaya There's also some deep connection between incentives and values that perhaps could go on the map. Each affects the other: ultimately, though, values - strongly internalized beliefs about what is important - matter more.
2022-06-30 12:18:33 @HeidiBaya I think of the main incentive as being a desire for your work to be meaningful, to connect with something that matters. Not sure where that fits?
2022-06-30 00:38:28 (ht @AnnaLeptikon) https://t.co/STetI8K2PH
2022-06-28 18:34:35 I don't know. I guess... I'd really love it if stirring the pile turned out to be the way to go. It's certainly happened before in the history of science, but rarely on the scale we're now seeing. It's just plain fun and interesting!
2022-06-28 18:34:34 The critique lots of scientists have of deep learning is "you're just tinkering, you don't really understand what you're doing! Science proceeds by uncovering deeper explanations, not stirring your pile of linear algebra". https://t.co/gg4RoOHr5C
2022-06-28 18:27:42 Jury-rigging the Singularity.Not a terrible description of modern deep learning! I'm sympathetic, for the not-very-principled reason that it violates so many norms about the "right" way to do things. https://t.co/C7M2ci16fv
2022-06-28 18:25:37 More on designing the local physics, but ultimately concluding this type of change isn't a good idea. The style of thinking is fascinating: "oh, this fix will help make things more interpretable, but it's going to screw up the overall capacity to learn, so no-go..." https://t.co/O5wB8IZ1aQ
2022-06-28 18:22:26 Enjoying the style of thinking here, sort of a way of changing the "local physics" of the net to get the properties you want, without damaging its overall capacity to learn interesting things! https://t.co/HoJBpoGuiH
2022-06-28 18:20:49 Arguments like this make me think that the notion of a feature neuron may just be a mistake. Maybe you just want to think entirely in terms of collections of neurons? Though it's not so obvious how to choose the right collection to focus on! https://t.co/4ZkFAgIoU4
2022-06-28 18:17:57 "it may be possible to move the field in a positive direction by discovering (and advocating for) those architectures which are most amenable to reverse engineering." https://t.co/VLZ5GOifVK
2022-06-28 18:17:56 Really interesting: interpretability as a key metric for nets. I wonder if they have something quantitative later in the paper? So you can report that your performance metrics aren't quite state-of-the-art, but you got an improvement on interpretability? https://t.co/f6cA13aD9f
2022-06-28 18:17:55 Indeed, it's because of this that Reed-Solomon codes are robust. So you might guess that maybe polysemanticity arises as a consequence of regularization, helping the net avoid overfitting.
2022-06-28 18:17:54 First thought is that it's always easy to reverse engineer a net into non-understandable computer programs! Just write it all out explicitly, converting data (weights and biases for each neuron) into explicit codeSo wondering: what does understandable mean, in practice?
2022-06-28 18:17:53 Fun definition of mechanistic interpretability as reverse engineering nets into "understandable" computer programs. "Understandable" is doing a lot of work though! Curious what it means! https://t.co/8JtQgLx0L8
2022-06-28 18:17:52 I'm enjoying reading the recent paper by @nelhage @trishume @catherineols @ch402 (and many others) on AI interpretability: https://t.co/BylMHpWjoLFor fun, a few thoughts from a non-expert outsider, just as I read
2022-06-28 17:29:05 Marvellous thread: https://t.co/0zYily4tos
2022-06-28 15:53:33 Ah! There's an account in Chapters 24-26 of "Dealers of Lightning". More "death by a thousand cuts" than a single decision, though certainly two forces (executive disinterest and the rise of companies like Microsoft, Apple etc) played a role.
2022-06-28 15:03:13 @QiaochuYuan I doubt many people write 30 drafts of an entire book. But sections... absolutely. I once asked (another) well known writer how many drafts he went through, &
2022-06-28 15:01:34 @QiaochuYuan A very good writer once made the comment that this is the secret to writing well in general. Most people don't want to write 30 drafts of something, with the first 20 terrible and needing to be thrown out.I found this very helpful
2022-06-28 15:00:46 @QiaochuYuan Oh, I'm pretty sure that's "easy": just try and fail more times than most human beings have the perserverance to continue past
2022-06-23 17:15:47 @_ArnaudS_ I have pretty complicated feelings about zoos. I don't think I'll visit. Bears are certainly amazing animals.
2022-06-23 17:12:16 Nearly one third of adults are first-generation immigrants(!) https://t.co/chZ6F0ux9Y
2022-06-23 17:11:25 Random spot in Bern to sit and work: https://t.co/7pmhjC1zgP
2022-06-23 13:57:40 @njrobynf @lizzywol @TSA_Northeast @SyracuseAirport @TSA Yes.
2022-06-23 13:44:50 The thread is quite interesting: https://t.co/uQFzkpvXMZ
2022-06-23 13:42:40 https://t.co/l5vOQX6xRd
2022-06-23 10:20:29 @victorerikray Ouch.“This really needs a book. But in a simplified treatment…”
2022-06-22 18:32:26 @bitcoin_eagle Huh. Do you remember the year?Remarkable...
2022-06-22 18:31:51 @paulg Amazing (if true!) You'd want to be awfully confident in that head fake!
2022-06-22 18:29:59 @celinehalioua Fascinating! All the 380s I've ever flown were certainly from huge hubs (Singapore, Dubai, London, etc).
2022-06-22 18:15:48 I don't understand why the program failed, or what the competitive pressures were. Still, fascinating to compare to: https://t.co/iYmVzXn1gJ
2022-06-22 18:13:59 Fascinating short history of the Airbus A380. It seems to have been a might-have-been that fell just short of what was needed.A huge gamble for Airbus that nearly but didn't quite pay off. https://t.co/BPwcFdsmTm
2022-06-22 17:25:39 @shallit43 Or jet lag…
2022-06-22 16:38:33 Thanks to all those who suggested the lake! It’s glorious! Photos don’t do it justice (the light is gone).
2022-06-22 16:34:53 @FlavioRump @ETH_en Sounds marvelous, thanks!
2022-06-22 16:34:31 @qedgs Thank you so much. I spent the day filling out forms (etc). But may take you up on the offer. Need to see how the next couple of days pan out (I need to be in Bern), and may get back to you. This sounds amazing
2022-06-22 16:32:38 @PatrickGillett Oh, great suggestion, thanks!
2022-06-22 16:31:31 @ade_oshineye Wonderful list, thanks!
2022-06-22 14:25:17 @SusanGroff1 It would be pretty funny. Would last 2 mins in Needles, tops
2022-06-22 14:22:57 Dave points out the multiplications are enough, no additions needed: https://t.co/pNZu3HAfJC
2022-06-19 19:33:12 @ArtirKel Yeah….
2022-06-19 17:05:57 Cue 50 exceptions…
2022-06-19 17:04:43 Fascinating how often org quality and size are inversely correlated (esp given big orgs were often originally good smaller orgs). At the London City Airport. So much better than LHR etc….
2022-06-19 11:38:16 @nvpkp @sympatheticopp "You are all individuals!" https://t.co/ZRLmKFih9L
2022-06-19 11:27:56 @sympatheticopp Sadly, having few unusual experiences has become, according to you, an unusual experience...
2022-06-18 21:57:55 @kristineberth Sorry to hear it! Hope the treatment goes as well as possible! Good vibes: https://t.co/3eezj2MT1H
2022-06-18 21:50:02 @maartengm This is the correct answer.
2022-06-18 16:21:25 The faith showed by most replies in the extremely high efficiency of our institutions is very endearing…
2022-06-18 15:33:25 Harry Potter analogies are "How do you do, fellow kids", but for Millennials...
2022-06-18 13:30:11 @akiffpremjee It is (in part) a math problem.
2022-06-18 13:29:18 @perfopt0 Occam’s razor ain’t perfect, but I know what I’d bet on
2022-06-18 13:27:20 Two fire trucks (etc) passing each other in opposite directions, sirens blaring, seems a particularly egregious failure of co-ordination
2022-06-18 09:15:46 @wtgowers Both.
2022-06-18 09:03:09 @quantumgeometer Thanks, looks very interesting!
2022-06-18 08:49:23 Someone who recently moved from the US to the UK told me that Twitter on UK time is much more sensible (presumably, indeed near-tautologically, because UK discourse is better than US).I'm inclined to agree. Interesting measure of a country.
2022-06-18 08:43:18 What are the best critical works that have been written about AI?Interested in both recent and decades old.
2022-06-17 13:32:38 @ShriramKMurthi The blocker I remember: I'd heard it described (repeatedly) as hard. That made it surprisingly scary. When it was explained, I realized I'd understood it years earlier (Towers of Hanoi, &
2022-06-17 08:24:32 On the dissolution of the San Antonio Symphony: https://t.co/tb1o2seyYg https://t.co/cF6m8DRMBS
2022-06-17 06:47:15 One of my favourite pieces of writing about technical writing, by David Mermin: https://t.co/hHD2EstMm1Thoughtful, and amusing, throughout. https://t.co/6Xm9iHWhG1
2022-06-17 06:38:10 https://t.co/q8ZMuUZy8x
2022-06-17 06:37:50 Everyone's a critic https://t.co/97BB65JUkg
2022-06-16 18:15:27 @winterblooms Thanks for these - very thoughtful! I need to digest a bit.
2022-06-16 14:26:36 I love watching people make discoveries. It's usually scientists, but occasionally historians: https://t.co/B5tE1cHJmd
2022-06-16 14:24:44 @GwenCheni Oh, sorry - I meant a book (or, more likely, essay) to write. It came up as an idea in conversation, and is now in my (very long) list of project ideas....
2022-06-16 12:26:15 Fascinating analysis of Jane Austen and her heroines: https://t.co/IrHqnR60OV
2022-06-16 10:28:48 @germank @AllenDowney It's a riff on "Seeing Like a State".
2022-06-16 10:27:33 @casparlessing Fixed, thanks. May take a couple of minutes to be live. No, I haven't changed the notes.
2022-06-16 09:48:45 Looking forward to this on Monday: https://t.co/wxlhSJltY0
2022-06-16 08:39:39 @SKRHardwick Own code.
2022-06-16 08:34:06 A useful thing about adding tags to my public notebook: it makes it easier to tell people what I've been thinking about recently. Eg right now:https://t.co/PEVurtXKjm and https://t.co/OequR0RLd5
2022-06-16 08:04:14 "Seeing Like a Scientist" is a stimulating title for a book, IMO.
2022-06-15 17:54:36 @KantNot1 Was there an hour before your tweet. It is, indeed, marvellous!
2022-06-15 17:02:01 Very interesting thread on sources of heterogeneity in expensive cities.(The bookstore in question is very good, and it is peculiar that they've survived.) https://t.co/9yfyPpYzoO
2022-06-15 16:59:38 @KantNot1 Oh, that is very interesting. And, indeed, I'd wondered that very thing about that very store, which I very much like!
2022-06-15 16:33:20 The SFBA has ~4,000 times as many people as Hay.(Moe's, City Lights, and Green Apple are three of the best. But I think Hay at least comes close to matching all of them...)
2022-06-15 16:19:54 Slightly down to realize Hay-on-Wye and Oxford may both have better bookstores in aggregate than the entire San Francisco Bay Area. London certainly does.
2022-06-15 16:17:57 This graph certainly _feels_ right, this trip.And given housing prices, I expect this means the UK mixes together people far more. https://t.co/YBlgVBMDua
2022-06-15 10:55:04 Reminds me of how my brain functions after jetlag... https://t.co/A6cA2Qzl2b
2022-06-15 07:23:57 https://t.co/4oakZp8eYr
2022-06-15 06:10:55 @matthuntrose Thanks!
2022-06-15 06:10:37 @xeegeex Fair call.
2022-06-15 05:49:41 Almost https://t.co/LUSJZ1IUUF
2022-06-15 05:47:39 https://t.co/jGCiwAXEhv
2022-06-13 19:02:12 @seemaychou Happy birthday Seemay!
2022-06-13 18:01:38 I just want to point out that I have just falsified my falsification of falsificationist Twitter.My apologies everyone.
2022-06-13 17:53:14 @ctitusbrown Posting in part because I got to make the joke in the last line.But: people (myself very much included) aren't usually very interested in evidence against what they'd like to be true...
2022-06-13 17:48:20 I love to imagine falsificationist Twitter!Full of people saying "Oh, I used to think X, but here's some striking new evidence against it, Y"It's a nice theory, but it's not what I see
2022-06-13 17:33:38 I am being corrected about the Brit/Australia thing. I'd never previously met anyone not from the UK or Australia who'd read them, but tonnes of people are saying they did. I sit corrected.
2022-06-13 17:26:03 I do enjoy the British sense of humour (Helps if you've read the Famous Five, which excludes most non-Brits/Australians: https://t.co/6hIz89CKbT ) https://t.co/QVaHGfdcvP
2022-06-13 16:46:13 @amcafee Looks like a valiant attempt to invoke Clarke’s first law
2022-06-12 14:53:48 I’d never previously made the connection (both are of the Great Western Railway). https://t.co/6PwIsUfg3K
2022-06-12 14:08:50 Should I cut up my credit cards?
2022-06-12 14:07:33 Heading to https://t.co/MnpVVR7XHdA town of bookstores!!!! I am far too excited
2022-06-12 08:30:19 @paulg I guess there is a sense in which Arrow’s theorem is an example: we know a voting system with those properties is impossible, and yet we kinda believe our voting should have them all anyway
2022-06-12 08:25:39 @Jermolene Superpower _and_ Achilles’ heel.Kinda appropriate that it be both!
2022-06-12 08:21:07 The law of the included muddle: it is possible (and common) for a human being to sincerely believe two propositions, A and B, which are logically contradictory.
2022-06-11 07:30:45 @MasterTimBlais This guy is, um, also doing something quite taxing: https://t.co/VFoLjlXVMy
2022-06-11 07:27:33 @mayli I suspect things like contact improv - really, any type of dance - are just fantastically demanding in this way.
2022-06-11 07:19:37 This is a super essay on concentration and performance: https://t.co/YmljxHICs0
2022-06-11 07:19:11 @hitsamty That's a fantastic piece, thank you!A friend once watched Chappell and Craig McDermott up close practicing (in the nets). He said McDermott appeared to be trying to kill Chappell
2022-06-11 07:16:15 This whole thread is an interesting history of American popular intellectual life. https://t.co/caMNv5P3ME
2022-06-10 21:36:35 @stuartbuck1 I'm missing the non obvious objection? I'd add at least a dozen more: robots are probably a bad idea [little benefit, liable to break]
2022-06-10 12:52:56 + Negotiators (i.e., all of us, but some are in a _lot_ more high-stakes negotiation)+ Air traffic controllers
2022-06-10 12:43:58 The movie 1917 had really long takes shot in the WWI trenches. One of the lead actors said in an interview that by the end of a take he'd be so into it that he'd get a shock as he realized he wasn't a soldier in 1917, he was an actor in the 21st century...
2022-06-10 12:41:37 What professions spend much of their time thinking incredibly hard 0.1 to 10 sec ahead? To rephrase: which are most fully in-the-moment?+ Musicians+ Many professional sports people+ In general, anything involving live performance, esp. improvisation https://t.co/IHZMeJPwKX
2022-06-10 12:38:35 Um, really nerdy hot takes...
2022-06-10 12:38:26 From the department of hot takes that really need to be rewritten with caveats: https://t.co/8zKSgkGvTD
2022-06-10 11:10:03 @gnat @VitalikButerin @cephalopod_x Yeah! Very striking!
2022-06-10 11:05:17 @michaelkeenan_0 @SpencrGreenberg A lot of people would say that is true of billionaires too, and so okay to be hostile.(Not expressing that opinion myself.)
2022-06-10 10:49:19 Many thoughtful replies to this. Struck by many of the replies. Eg @VitalikButerin's observation that "Ok boomer" is damaging, and @Cephalopod_x's similar observation about "Karen". https://t.co/uBobByfGCI
2022-06-10 05:03:48 I wonder how well pneumatic tubes or conveyor belts would work?
2022-06-10 04:58:20 Widespread very rapid delivery infrastructure would be fantastic.
2022-06-10 04:54:15 Reading the QTs on this is a fascinating spectrum of humanity.(I hope it works. Hard to pull off, easy to imagine failures. Like lots of incredibly useful things!) https://t.co/gAwSRV78Pt
2022-06-10 02:19:46 @Meaningness @andy_matuschak @DRMacIver I enjoyed it.(People often take such things a little bit too literally: take with a grain of salt, IMO.)
2022-06-09 15:37:53 @patrickc @orgRem @KRoyMyers @LouisShekhtman Is that their research budget? Or does it include teaching etc?
2022-06-09 15:17:34 @Meaningness @andy_matuschak @DRMacIver Same as “Masters of Doom”, just a few proper nouns changed…
2022-06-09 11:52:57 Cf https://t.co/IE1xy2HDb3
2022-06-09 11:47:10 @patrickc That accords much more with my impression too.
2022-06-09 04:19:32 @patio11 Nicely put!
2022-06-09 04:16:09 Occasionally artist friends tell me very excitedly about the 3k or 5k or 10k grant they just got.When I mention this to tech friends they just look confused.
2022-06-09 04:14:20 @umbut thks! bad viewing conditions right now, will look later!
2022-06-09 04:13:36 @GalaxyKate Occasionally artist friends tell me very excitedly about the 3k or 5k or 10k grant they just got.When I mention this to tech friends they just look confused.
2022-06-09 03:58:27 I enjoyed this: https://t.co/5hqgpBkA4D
2022-06-09 03:50:49 This is marvellous! https://t.co/SIJeMzkBbW
2022-06-09 03:18:57 My understanding is that pre-WW2 science funding in the US was dominated by foundations. It would be very interesting if a reversion to that state of funding was going on.
2022-06-09 03:16:20 Fascinating (&
2022-06-09 03:11:25 @LouisShekhtman @albrgr @deaneckles @SciPhilOrg @NSF I'd love to see the data, or a summary, if it's publicly available!I'd have guessed the 5 biggest foundations don't even spend $10 bill (&
2022-06-09 03:01:56 @LouisShekhtman @albrgr @deaneckles @SciPhilOrg @NSF I was just wondering at the slide, which appears to assert that philanthropy for science in the US is of the same order as NSF + NIH, if I understood correctly (i.e., in the neighbourhood of $50 bill per year). Is that right?
2022-06-09 01:42:12 Looking through Gates and HHMI and a bunch of other Foundation numbers, I have trouble seeing where this is all coming from. If you add in things like Google Research, MSR etc, then easy to believe. But that’s certainly not purely philanthropic
2022-06-09 01:27:08 Very interesting:I wonder what the biggest 3-5 sources are? https://t.co/fcHDObGlON
2022-06-08 18:05:50 @devonzuegel I would retweet it - I've wanted this for ages - but: maybe later?
2022-06-08 18:03:45 @devonzuegel Loaded after ~60-120 sec.
2022-06-08 18:02:59 @sebasbensu @devonzuegel She should just force unfollow 95% of them, but not me!
2022-06-08 18:02:27 @devonzuegel Sorry to be the bearer of bad news. Looking forward to using it, especially as I travel over the next few weeks!
2022-06-08 05:10:24 @harry_ramsay Thanks! BTW, some other related writing in this vein: https://t.co/PEVurtXKjm
2022-06-08 04:58:46 @vgr I'm missing the reference...
2022-06-08 04:57:20 @harry_ramsay Thank you!BTW, you may enjoy this: NW appear in my work here (and will more, in future): https://t.co/e9yNkLbR5M
2022-06-08 04:55:48 @km Doesn't he just? And brings you into it.
2022-06-08 04:55:00 @waqasali @km :-( I'll be in England. (Thank you for the heads up!)
2022-06-08 04:53:57 @mengwong @micahtredding @vgr Huh, that's really interesting (and tbh a bit surprising!)
2022-06-08 04:52:40 @km He is a very, very special performer, and participating in those must be an incredible experience.
2022-06-08 04:50:56 @m_svillar I should really listen! I've sorta known that for a while - several friends have told me they think I'd very likely like JC's stuff! Anything you particularly like?
2022-06-08 04:48:56 @km Good grief.
2022-06-08 04:42:01 (As in first on the program. I couldn't possibly rank them against one another!)
2022-06-08 04:28:34 Nightwish and Bobby McFerrin sounds close to a perfect show! Probably put Bobby first, I think. https://t.co/7iwcDF7tDF
2022-06-08 04:24:05 @feross You also a Nightwish fan?
2022-06-08 04:19:26 @xeegeex That sounds awesome!
2022-06-08 04:15:41 @elmobronowski God, that sounds beyond amazing.
2022-06-08 04:14:27 @xeegeex Yeah, I just couldn't bring myself to do it, masked or not. With much regret! But singing loudly indoors is about the worst thing for spreading airborne diseases!
2022-06-08 04:12:04 Which led me to the original, "Angel of Grief", by William Wetmore Story: https://t.co/evjYPNzi4bIt captures grief and despair very well, IMO.
2022-06-08 04:11:00 So, a fascinating connection to Nightwish: https://t.co/92qHo9AZFF
2022-06-08 04:09:37 @xeegeex Oh, wow, that's great!I love Nightwish. Had tickets to see them in San Francisco a few weeks ago, though Covid peaked, and the idea of being in a room with 3000 people singing seemed... unwise.
2022-06-08 03:45:45 Just walking along, near the Stanford campus, when I saw that. It really hit me.
2022-06-08 03:44:41 Angel in despair https://t.co/MC0GO30uKw
2022-06-07 06:11:19 @skdh @paulg (Of course, I've no way of knowing it would work for others, just reporting my personal experience.)
2022-06-07 06:10:35 @skdh @paulg There's a different way to "counter" this. Don't reply to obviously poorly made arguments. This seems to (mostly) work for me, at 75k. But as soon as I start to engage with such arguments, my experience goes (rapidly) downhill: it's prodding an anthill.
2022-06-06 19:48:07 @meekaale @diviacaroline @PaulVanderKlay @PatrikAHagman Eg @xianityplus (and then look through follows, and followers).
2022-06-06 19:46:18 @meekaale @diviacaroline @PaulVanderKlay @PatrikAHagman The Mormon transhumanism, Christian transhumanism, and Turing Church communities are interested (among others).
2022-06-06 13:57:39 Luke's thoughtful response: https://t.co/AWQ6OhlO0m
2022-06-06 13:51:05 The Purge: It's Time to Build
2022-06-06 13:50:41 Imagining @pmarca out in the streets, leading a posse of people extremely rapidly installing new train stations, a hyperloop, 300 story buildings, a spaceport... https://t.co/Us42uTq2ZN
2022-06-06 02:48:51 @xeegeex I wish I'd known that!
2022-06-06 02:43:01 RT @michael_nielsen: My favourite tie story - one of the funniest stories I have ever heard from anyone, ever, about anything - is from Len…
2022-06-06 02:42:48 My favourite tie story - one of the funniest stories I have ever heard from anyone, ever, about anything - is from Lenny Susskind. It's at 1:30 here: https://t.co/zY7tdxdOmB
2022-06-06 02:40:15 My second favourite tie story, being strangled onstage by one while giving a speech: https://t.co/H8hhsimC5e
2022-06-06 02:36:23 @xeegeex So I got up on stage, and gave my speech while getting increasingly light headed . I've never been more relieved to step off stage.(I wish I was joking!)
2022-06-06 02:35:12 @xeegeex ... I undid the top button, and used a bobby pin behind the tie to make it look like it was done up. I could barely tell in the mirror.But my boss had a really good eye (&
2022-06-06 02:33:13 @xeegeex I once had to give a speech to a bunch of politicians in Australia. It was a fancy (IIRC) black tie event, and I'd rented a tux last minute. The fitting was done poorly, and I found I was being strangled by the shirt. So...
2022-06-06 02:27:41 @xeegeex For a lot of people where I grew up (Australia), you'd wear a tie routinely at school or at a job. But I never had either. So it's only been for rare formal occasions. Maybe once a year(???), on average, with much variation!(Weddings have def got less formal over time.)
2022-06-06 02:09:15 @xeegeex I can quasi-reliably tie a tie. It's usually for weddings or funerals. Which don't occur quite often enough (or, in the case of weddings, are frequently informal enough) that I can tie one without thinking: "Hmm, how does this go again?"
2022-06-06 01:46:58 @xeegeex https://t.co/EHcXCy5heh
2022-06-06 01:45:38 Gender | can you reliably tie a tie?
2022-06-06 01:43:39 @xeegeex I was wondering that as I wrote it! But didn't think to add a poll. I will now.
2022-06-05 19:09:02 Koala CEO, when reached for comment, explained that: "trees don't grow on money" https://t.co/Bx7bbbsDQw
2022-06-05 17:36:24 I expect this would be a wonderful, meaningful job: https://t.co/ALQ73r5q0N
2022-06-05 05:42:44 A lovely description of this kind of very quiet singing: https://t.co/RZ9Ym7Sl8i
2022-06-05 05:41:59 @lfschiavo Carl Sagan's Cosmos, maybe. The Feynman Lectures (which are now free online), maybe. Stimulating question.
2022-06-05 05:27:21 @mindspillage That's great!
2022-06-05 05:11:12 As is this: https://t.co/uM5IVvpTCHI'm pretty shy about singing with people, but love it when I overcome that. Intellectually, I am not a crowd person. But that's a type of emotional crowd I love to be a part of.
2022-06-05 05:08:42 This is very beautiful: https://t.co/ldJ0UzImPQ
2022-06-05 05:08:04 @martinskalis Utterly marvellous! Thank you!
2022-06-05 04:55:47 @mindspillage That sounds marvellous! It's a long, long time since I heard it live, but I remember it as gorgeous.
2022-06-05 04:54:01 @mindspillage That is very, very beautiful. Thank you for sharing it.
2022-06-05 04:36:23 RT @michael_nielsen: Also: if you've never been to a Hallelujah sing-along, it's just glorious: https://t.co/bqAuZ1jZDD
2022-06-05 04:33:51 @geomblog Found this - it's just wonderful: https://t.co/bqAuZ1jZDD
2022-06-05 04:33:35 Also: if you've never been to a Hallelujah sing-along, it's just glorious: https://t.co/bqAuZ1jZDD
2022-06-05 04:29:36 @geomblog That sounds marvellous.I went to the Messiah in Toronto once, and it was glorious.
2022-06-05 04:28:31 Tips welcome!I'd like to see the Mormon Tabernacle Choir.
2022-06-05 04:25:51 I'd love to see and hear such a choir one day!
2022-06-05 04:25:07 When I was a teen, my Mum went to see Pavarotti with a ~thousand-person choir. I asked how it was &
2022-06-05 04:18:14 @visakanv @VividVoid_ @nikitanomo Love the first 90 sec (esp) of this, too, which includes the piece Visa references: https://t.co/xDV54ZIKu1
2022-06-05 03:57:31 @jesqa_ He has a kind, thoughtful-looking face.(I don't like to read too much into faces, but that's a really nice drawing!)
2022-06-05 03:22:03 @QiaochuYuan I shouldn't have mentioned it. (It's nice that people ask after stuff. But... exhausting. And I'd need maybe a few hundred hours more work - one top of a couple of hundred already - to make that essay any good.)
2022-06-05 03:17:40 (Somewhere I have an unpublished essay "How to read the web", inspired by Mortimer Adler's classic. I should dust it off...)
2022-06-05 03:16:58 This thread is a pretty good illustration of "How to read the web" https://t.co/zqSqPfNCqu
2022-06-04 02:28:26 @cmuratori Me: I was born in k/Rf2ie+q<
2022-06-04 02:10:32 @albrgr @LinchZhang (For me, being clear &
2022-06-04 02:09:21 @albrgr @LinchZhang I may as well say: I was _also_ using EA judo as something of a catchall to refer to interesting and unusual patterns of argument among EAs. But then, I have an approach to definition &
2022-06-04 02:04:46 @albrgr @LinchZhang So I actually meant the things both of you read into it. But regarding the causal downstreamness: this is case-by-case. If there's lots of repeated patterns in the proposed alternative notions of good, it may mean some more substantive underlying problem of principle.
2022-06-04 02:02:30 @albrgr @LinchZhang There's also institutional and practical variants, of course.
2022-06-04 02:00:17 @albrgr @LinchZhang I was referring to several practices collectively. The principal one is:non-EA: "I don't like EA. Here's why your notion of good [or good actions] is wrong..."EA: Recasts that as a critique of EA notions of good, &
2022-06-04 00:20:04 @imperialauditor Very nicely put! Yes, the same thing bothers me.
2022-06-03 20:47:17 @metaLulie @MatjazLeonardis Same with Tw
2022-06-03 20:31:59 @jonathanstray Thanks Jonathan!
2022-06-03 20:15:23 @joshbuddy You may enjoy (ish) Neil Postman's "Technopoly".
2022-06-03 20:09:50 @joshbuddy No, I haven't.
2022-06-03 19:18:55 @CineraVerinia Either is fine! Longer stuff is probably easier for you on the website.
2022-06-03 19:17:59 RT @michael_nielsen: Some rough notes on the role of vision papers in basic science: https://t.co/don59lgG9d
2022-06-03 19:17:35 RT @michael_nielsen: Notes on Effective Altruism (EA): https://t.co/tIJQIMbuyf
2022-06-03 19:12:15 @rivatez Will do!
2022-06-03 18:59:13 @rivatez Had no idea you played tennis!
2022-06-03 18:53:56 @theanega Thanks!
2022-06-03 18:53:28 @RichardMCNgo It's not an objection
2022-06-03 18:50:26 @joel_bkr Thanks, that's very nice to hear!
2022-06-03 18:47:59 @KellerScholl @albrgr @ozyfrantz Thanks Keller - I've updated to "thousands or tens of thousands". I think it's still possible to quibble with this, but I'm much more comfortable with it as a reasonable representation.
2022-06-03 02:58:52 @KellerScholl (Though I guess it's true that it's rather permeable.)
2022-06-03 02:58:22 @KellerScholl Fascinating - thanks! I'd love for others to chime in, with either agreement or disagreement.(I'll try to remember to update the notes in the morning, absent a good reason not to.)Also: I've met a larger fraction of EAs than I'd thought...
2022-06-03 02:47:39 @andrwmrtn There's a sort of Durkheimian analysis of EA to be done - efficiency and division of labor and bureaucracy buys you a lot, but at the expense of alienation and anomie
2022-06-03 02:45:02 @NonMurkyConsqnc Oh, thank you!That's the first time I've ever been complimented in anything like that way
2022-06-03 02:44:16 @andrwmrtn Yeah, an issue I didn't discuss, and would like to in future notes, is the relationship between the different levels of abstraction. Applying individual human values to giant institutions (&
2022-06-03 02:42:52 @andrwmrtn A man on a desert island with 10 million dollars will still starve: he has no way of converting it into food &
2022-06-03 02:42:27 @andrwmrtn In particular: if I have to make a tradeoff decision right now, sure, make such an approximation. But over longer time periods it rapidly becomes misleading, because there's no fundamental status to quantities.
2022-06-03 02:40:37 @andrwmrtn Thanks. I still don't understand this very well. But my thoughts here are pretty consonant with yours: https://t.co/J3MGqGE8WU
2022-06-03 02:34:33 @Lang__Leon I put my preliminary notes up here: https://t.co/LFzHkXvbfs
2022-06-03 02:13:44 @lacker Same answer as @TheZvi's earlier, I'm afraid. I'm pretty Hayekian
2022-06-03 02:09:44 A little EA-influenced-thinking-in-action (read the followup tweet): https://t.co/DdgwuoGg6g
2022-06-03 02:07:57 @patio11 Marvellous! And thank you!
2022-06-03 01:57:59 @patio11 Do you mind saying what the event was?Interesting that that's true of you: a person I think of as likely to be highly oriented that way, in any case!
2022-06-03 01:56:56 @maartengm Thanks! I meant "rough" in part as in "I haven't thought about this in much depth yet".
2022-06-03 01:38:31 Notes on Effective Altruism (EA): https://t.co/tIJQIMbuyf
2022-06-02 23:33:47 @NathanpmYoung @Lang__Leon (I'll be delighted if they get something useful from it. But it's not who I'm writing _for_, if that makes sense?)
2022-06-02 23:32:54 @NathanpmYoung @Lang__Leon I appreciate the vote of confidence from you &
2022-06-02 23:19:34 @mioana @3blue1brown Oh, more things in this vein: @wtgowers solving mathematical problems: https://t.co/UJrKR1MijVOr @Jonathan_Blow livestreaming the creation of a new programming language: https://t.co/yq0twwq9zgBasically: world-class experts, showing how they do parts of their creative work
2022-06-02 23:09:21 @mioana @3blue1brown 3. Do community building in "para-academia", the penumbra of independent and semi-independent researchers. Don't know if that exists in econ! But it does in physics (a bit), machine learning, and open science, and other communities where I've worked.
2022-06-02 23:07:35 @mioana Great thoughtful thread! Three extra things (my background: I was a tenured prof in physics, then left academia)1. Identifying and convening latent communities (as you are doing here!)2. YouTube / Twitch / other non-traditional media (eg @3Blue1Brown )(cont)
2022-06-02 03:46:46 "EA Forum users voted on which posts from the first decade of effective altruism they found most useful and important": the results: https://t.co/w5fmSTRRER(Gradually working my way through: so far, very stimulating.)
2022-06-02 02:30:54 @recursus This is perfect.
2022-06-02 02:30:07 @OptimistsInc Exactly! Crazy critters!
2022-06-02 02:28:30 Platypi: what can't they do?https://t.co/i2Vd2VrMuf
2022-06-02 02:25:03 Platypi are self-advertising! A semiaquatic, egg-laying, duck-billed, beaver-tailed, venomous mammal, who sense prey using electrolocation, and whose males have a venomous spur on the back foot(!!!)I mean, really? (Text adapted from wikipedia.) https://t.co/gVLe66d9XD
2022-06-02 02:19:04 @magdalenakala With the ability to git push being the most desired missing feature, leading to massive overuse of the issues feature...
2022-06-02 01:41:12 @vgr Maximum unviable dunkfest
2022-06-01 04:12:08 Good pitch for the opening to a horror movie... https://t.co/9V5qh3f3lI
2022-06-01 03:59:15 @mmalex I'd honestly been completely ignoring the m1 until your tweet. No more...
2022-06-01 03:56:49 @vgr https://t.co/VYdoerMZUV"I wonder who was the first person to use [foo] on Twitter?" is a strange and tbh unfulfilling hobby...
2022-05-31 18:54:15 @paulg I found @paultoo @btaylor et al's FriendFeed good in this way, with little effort on my part. On Twitter it can be done, but requires a lot of defensive action (muting words, muting &
2022-05-31 18:22:47 This happens frequently on Kindle-on-Mac. I do not understand why Kindle doesn't remove the book and redownload it. @jeffbezos? https://t.co/KLNKOGI427
2022-05-31 01:57:20 @dommydoteth Just so you're not disappointed: Kropp is a side story. But the book is amazing.
2022-05-31 01:55:07 https://t.co/3DXXEJ3HbX
2022-05-31 01:54:39 @dommydoteth Amazing guy. (This is from Krakauer's great book, "Into Thin Air") https://t.co/MYcXbTEouf
2022-05-31 01:50:42 Goran Kropp: https://t.co/oGRjRSuzm6
2022-05-31 01:50:13 @dommydoteth Goran Kropp: https://t.co/jxvi9AU7Ss
2022-05-31 01:32:15 @nloadholtes Glad you enjoyed it!
2022-05-31 01:29:48 @anderssandberg @AdamMarblestone @AGamick are doing something in that vein, with FROs
2022-05-31 01:28:47 @anderssandberg I think the reason for 2 is that the good vision papers I know are all based on an exceptionally deep and accurate understanding of the entities under consideration
2022-05-31 01:26:40 @anderssandberg Two comments:1. Sturgeon's Law ("of course 90% of sci-fi is crap
2022-05-31 01:23:52 @anderssandberg Thanks Anders!
2022-05-31 01:23:39 Thoughtful thread from Anders: https://t.co/b9TxmhOYbN
2022-05-30 22:08:20 Fascinating thread: https://t.co/lempJzGlBN
2022-05-30 04:04:27 @AmandaAskell https://t.co/OECg3QToII
2022-05-30 04:03:34 AI labs everywherehttps://t.co/bdgl8P6oyc https://t.co/z1yBE9Zr82
2022-05-30 03:48:18 @AmandaAskell @RatOrthodox Laughing, imagining Catholics arguing with Protestants in this way....
2022-05-29 23:51:45 Victoria Falls from way above (via @DOverview, which is terrific) https://t.co/LdqhNPwKxv
2022-05-29 21:04:43 @andreitr I'd love if you have any suggestions!
2022-05-29 16:18:04 RT @michael_nielsen: On design versus the natural sciences: https://t.co/pavmQH7apG
2022-05-29 16:17:31 RT @michael_nielsen: Some rough notes on the role of vision papers in basic science: https://t.co/don59lgG9d
2022-05-29 03:09:17 RT @michael_nielsen: On design versus the natural sciences: https://t.co/pavmQH7apG
2022-05-28 23:02:57 RT @michael_nielsen: Some rough notes on the role of vision papers in basic science: https://t.co/don59lgG9d
2022-05-28 22:40:33 Thoughtful response thread from Chris, arguing that vision papers are often used to help in creating common knowledge and identity for a community: https://t.co/H8Wb6KkeaQ
2022-05-28 22:38:40 Remarkable thread on gerontocracy, in many forms: https://t.co/0mIkTGegLr
2022-05-28 21:52:44 @abstractalgo Thanks!
2022-05-28 21:35:06 Something not adequately pointed out above: most such papers aren't v good! They're shallow or wrong, pointing to misleading mirages, not useful visionsI talk a little in the notes about how to tell which is which. Still, I don't understand v well, apart from personal taste!
2022-05-28 21:27:07 @NoahOlsman Oh, yeah, nice point!
2022-05-28 21:26:16 I realize this all sounds laudatory. It's not meant to be so
2022-05-28 21:26:15 There _is_ a genuine difference. But I found it surprisingly hard to articulate. It has something to do with new, composable capabilities vs motivating stories of latent possibility: https://t.co/9sLGu8RRXg
2022-05-28 21:26:13 The contrast to normal papers is strong: https://t.co/VZONeim57j
2022-05-28 21:26:12 Famous papers without "results", by conventional lights. They're really starting proto-fields: https://t.co/DcDdTDEq24
2022-05-28 21:26:11 On their strangeness: https://t.co/bWbBYRuWqT
2022-05-28 21:16:54 Anyway, they seemed rather strange to me, an unusual class of epistemic objects, very different from ordinary papers! And so I wrote down some rough-and-ready notes on them, just trying to understand them a little bit.
2022-05-28 21:16:53 I wrote them after (repeatedly) noticing an odd niggling thing: many of the papers I most admire are pretty strange, considered as scientific papers!
2022-05-28 21:16:52 Some rough notes on the role of vision papers in basic science: https://t.co/don59lgG9d
2022-05-28 20:49:21 @m_ashcroft Make something that you deliberately _don't_ think other people would like, but that you would enjoy making(?)(I do this kind of thing occasionally, in similar ruts, and find it helps me at least.)
2022-05-28 19:41:28 @LionKimbro @Meaningness Lovely idea!BTW, I love your notes on making a map of every thought you think, and have recommended them to countless people over the years. Some very brief thoughts, not doing justice at all to it: https://t.co/9OI65HFEpW
2022-05-27 02:26:02 @rhngla @mfaboston Great spot!
2022-05-27 02:16:18 @FoolAllTheTime Which one? AT?
2022-05-27 02:14:25 @FoolAllTheTime Wow, that is amazing. Never heard of the triple crown (for hiking) before.
2022-05-27 02:06:00 @abecedarius @LtownZag @TheZvi @ns_whit Everything notable attracts sloppy arguments.(It is, oddly, a sign of success, AFAICT.)
2022-05-27 01:54:57 @kocienda A benefit of SFMoMA is the cafes in the gallery - you can sit and work in a cafe for an hour or two, go sit in a room full of paintings, rinse, repeat.
2022-05-27 01:48:51 @kocienda I do the same (though never in London, as yet).I like to sit with a paper notebook in the galleries, too. It's very instrumental to note, but it really is good creatively. And very replenishing...
2022-05-27 01:47:31 @Meaningness @jamesheathers See eg: https://t.co/YmeAeXVUniPioneered, among others, by the (sadly, late) Jean-Claude Bradley.
2022-05-27 01:43:53 @mesolude At least you had great taste in themes (ahem): https://t.co/Dc2EwkXmO0
2022-05-27 01:38:36 I don't usually draw in museums. But they are one of my favourite places to work. Very inspiring, in some hard-to-explain way. https://t.co/ocIO7mu5lI
2022-05-27 01:36:22 Turner does what seems like the same trick over and over again (the explosion of light). And it's great, every time. https://t.co/TipEcKVz0E
2022-05-27 00:31:03 Hang on. The AP has my back: https://t.co/zIzdSxGdVZ
2022-05-27 00:30:20 I sometimes wonder if I hallucinated the crowdsurfing. I mean... it's got to be on YouTube, right? But it doesn't seem to be.
2022-05-27 00:25:33 A friend asks for the Australian equivalent: https://t.co/nJYd8rtqs1
2022-05-27 00:22:24 @JoshuaLelon I struggle quite a bit with that. Strange, too, because I nearly always love hearing about other people's current obsessions.
2022-05-27 00:16:47 Jacinda Arden, Prime Minister of New Zealand, first event upon arrival in Japan. https://t.co/L099B089NF
2022-05-26 21:04:06 @orbuch I've a tonne of thoughts &
2022-05-26 21:01:49 @AnatoliyLotkov I told my friend Ben Schumacher, and he said he'd long had an idea/joke about negative money: you can imagine a thief breaking into your house, and leaving a big pile of negative money.It's silly, but also fun to play with
2022-05-26 21:01:01 @AnatoliyLotkov Yeah, it's fun isn't it? I don't remember when I first had it - I think in the late 90s (I seem to remember having the thought in the LANL cafeteria, of all places).
2022-05-26 20:51:05 Delighted to see people discussing non-Abelian money: https://t.co/DPHW4GMUJW https://t.co/F4lViIMNiO
2022-05-26 20:49:19 (Note: I'm quoting from one side of what is effectively an argument here. It's a thoughtful side, but reasonable rebuttals are possible. Certainly, I disagree with much in both sources! Still, I find this stimulating.) https://t.co/wGN0LzZmaJ
2022-05-26 20:44:07 I enjoyed this review of @willmacaskill's book "Doing Good Better" by @amiasrinivasan: https://t.co/mY9ngklmwDI won't attempt to comment, yet - I'm still forming my thoughts. But both the book and review are recommendable. https://t.co/YoVnHym6Vf
2022-05-26 19:36:30 @vgr It's a feature _and_ a bug. I like living in a pop culture some of the time. But it's crummy at doing the truly new.
2022-05-26 19:34:59 @ambimorph @ctietze The net result is a surprising amount of common knowledge (among experts) about what's known and what's not. It's imperfect, but far better than what I see even in pretty specialized parts of social media.
2022-05-26 19:34:10 @ambimorph @ctietze Yup, pretty much. And held to a high standard on it. There's a norm that this often _is_ quite a fight, grounds for not publishing a paper (or, more often, revision), and so on. Even just the anticipation of such an argument pushes people to do a reasonable job.
2022-05-26 18:49:34 Remarkable stories, both. Related to the idea that doing wrong is its own punishment. https://t.co/Htvk15xRGu
2022-05-26 17:48:46 @metasj @nikete It's not an unimportant bit of drudge work, to be automated away. The _author_ needs a deeply internalized map of what is known in order to be _able_ to do novel work. Tools can help with this, too. But it's not an add-on or minor task, it's central to the entire process.
2022-05-26 16:27:02 @nikete In fact: _especially_ counterfactually.
2022-05-26 16:26:48 @nikete I think it's often very constructive, not wasteful. Annoying, yes, and often inaccurate. But a very important part of the whole process. It's that negotiation (even counterfactually) that gradually improves people's sense of what is known.
2022-05-26 15:30:44 @nikete Interesting idea. I'm not sure it's not close to being a research-complete problem: if the machines can do this well, and we have robots in the lab, maybe there's no need for researchers at all any more :-)
2022-05-26 15:26:59 @CineraVerinia I've found many elements of the rationalist blogosphere helpful, often as improvements on existing journal norms. Eg, I use variants on "epistemic status" a lot in my essays &
2022-05-26 15:14:50 @gordonbrander https://t.co/3fcEjQcaPu
2022-05-26 15:07:45 I'm not trying to argue this should be a universal feature of blogging or social media, of course! Just saying that it's necessary for those things to be useful for discovery: a strong culture of accurately saying: "here's what is already known, here's what is new"
2022-05-26 15:05:38 You can see this in the replies to the tweet, in fact: very little is said that wasn't first said 15 or 20 years ago. And yet few seem aware of that background.
2022-05-26 15:05:37 One somewhat unfortunate thing about blogging culture is that (unlike journals) there's no strong norm of explaining where your work sits with respect to existing human understanding.This is an enormous advantage journals have in pushing forward understanding. https://t.co/Lh7Ose8zGH
2022-05-26 03:55:02 @mo_norouzi "Searching for the numinous""A cunning arrangement of 10^29 protons, neutrons, and electrons"(Both are Twitter bios)
2022-05-25 23:04:01 @RobertSpekkens Curious: what's an example, Rob?(Gender-neutral language has been, for me, an example that is at least adjacent.)
2022-05-25 22:33:37 This is pretty amazing: https://t.co/zRuRqxoaJd
2022-05-25 22:27:49 (I spent ~8 minutes quickly reading it before posting. And yet it seems like the kind of thing where, if I had the right expertise, you could spend years.)
2022-05-25 22:26:27 Absolutely fascinating, a wild exploration of what it means for two mathematical objects to be equal, with the punchline being (roughly) "we don't entirely know, here's a bunch of different ways of thinking about it": https://t.co/XEiKnS6TSe
2022-05-25 20:55:29 @klinse I asked a very similar question, and got a bunch of useful answers: https://t.co/WaSBPyIzs5
2022-05-25 20:50:30 @POTUS Then act.
2022-05-25 19:51:57 I enjoyed this NSF pre-proposal for a "Center for Improving Science": https://t.co/RLxItS25ZG https://t.co/CtG6AZOHjK
2022-05-25 19:50:09 The Humean condition, illustrated: https://t.co/YXG8RNpF0h
2022-05-25 19:24:18 (It's a great paper. But the title is misleading. An accurate title would be much more boring, something like: "A plausible argument that a large fraction of papers in some fields are false.")
2022-05-25 19:23:11 I've often wondered if the title of "Why Most Published Research Findings are False" was deliberately ironic?(The argument in the paper establishes no such thing.)
2022-05-25 18:07:56 @Malcolm_Ocean That's a great suggestion...
2022-05-25 17:04:33 But I don't know of a good visual.
2022-05-25 17:04:21 I think of the argument in very visual terms - this immense group of people, each one with their own hidden knowledge, and the incredible benefit of aggregating that knowledge, even if only indirectly.
2022-05-25 17:03:08 Are there any good visualizations of Hayek's ideas in "The Use of Knowledge in Society"?
2022-05-25 15:14:49 Time for me to get to the work of the day
2022-05-25 15:13:01 This is lovely too: https://t.co/ycwxEnptvl (via @uncatherio ). https://t.co/coIT1Z8yja
2022-05-25 15:10:48 Just goofing around, though I think there's some truth to this: https://t.co/DtlRHyA48B
2022-05-25 15:04:16 True also of ideas and concepts: "Every idea around you was someone's lifework."Of course, a book or paper containing a new idea is often the result of only months of literal work. But they arise out of decades of preparation.
2022-05-25 15:00:46 And @gordonbrander's wonderful bio: "Everything around me was someone's lifework". https://t.co/HaZuCKLwLC
2022-05-20 08:11:00 CAFIAC FIX
2022-10-26 22:45:32 RT @katyilonka: I read the whole thing (and it's long). While I'm not that into metascience generally, the essay really made me reflect abo…
2022-10-26 22:45:23 @katyilonka That sounds like a great essay to have!
2022-10-26 17:24:21 Thoughtful short note: https://t.co/a6HnA21kdg
2022-10-26 01:53:26 @BrianNosek I'm not wild about methodological answers, as in many replies. They seem more like symptoms of rigorous research, not causes. I do like Feynman's comment about: "a specific, extra type of integrity that is not lying, but bending over backwards to show how you’re maybe wrong"
2022-10-25 23:15:38 @jachiam0 This seemed very good to me. Thanks for writing it.
2022-10-25 02:18:40 @vgr And they will always use that as an excuse. Much like arch-communists and -capitalists.
2022-10-24 18:57:26 @RuxandraTeslo I've found this perplexing since being a (broke) undergrad.
2022-10-24 18:30:39 RT @kmmunger: Fantastic article by @michael_nielsen and @kanjun defining the possible scope of metascience and calling for more metascience…
2022-10-24 02:25:59 IBM mainframe specs from the 1960s. Up to 512 kilobytes, and only 3 or so tonnes! https://t.co/wqJyVeqQwR
2022-10-23 16:56:48 RT @marquezxavier: Really stimulating article on the design of the social process of science by @michael_nielsen and Kanjun Qiu https://t.c…
2022-10-23 16:56:00 RT @JMateosGarcia: "A vision of Metascience"I started reading this essay by @michael_nielsen + @kanjun and it's great. I love the idea of…
2022-10-21 19:20:39 It certainly makes me curious what makes b better. Even a tiny improvement - say, to b = 0.12, would make a huge difference. You'd then need a ~300 increase in training data to get that square root improvement....
2022-10-21 19:12:01 Now, admittedly, b is pretty small in their examples. E.g., b = 0.095 for training data, so you need a roughly ~1000 increase in training data to get the square root improvement in probability. Still, that's better than I would have a priori guessed!
2022-10-21 19:10:59 One simple way of looking at this is a constant multiple increase in compute, training data, or model size buys you a square root improvement in probability.That constant multiple is 2^{1/b}, where b is the exponent in the scaling law.
2022-10-21 19:05:58 But it's not so different either.
2022-10-21 19:05:57 In particular, suppose b = 1 and we double size of training data. Then if the former probability was 1/100 we now get a probability assigned to the correct token of 1/10. That's an incredible improvement!In practice, b is smaller than 1, and things don't work quite so well.
2022-10-21 18:58:46 A slightly more precise statement, courtesy of @moultano: https://t.co/PsjnP2e4bE
2022-10-21 18:56:22 @moultano Ah, I see. Thank you!So a little more explicitly, the scaling model says:p = exp(- a 1/M^b), for constants a and b, and p the probability assigned to the correct (unseen) token, in the training data.This seems to me quite remarkable!
2022-10-21 18:53:31 Incidentally: note that the constant a is negative (for all three of compute, training set size, model size). This ensures that p ->
2022-10-21 18:52:13 @moultano Hmm. What's the difference?
2022-10-21 18:42:37 I'm quite curious whether I have this right. I must admit, the probability dropping as exp(-M^a) would seem shocking to me.
2022-10-21 18:41:54 A very basic question about the meaning of the Kaplan et al OpenAI scaling laws paper: https://t.co/YimscMAryl
2022-10-21 18:28:45 @phokarlsson @Meaningness @kanjun It would take quite some thought to figure out what &
2022-10-21 18:24:03 @phokarlsson @Meaningness @kanjun On the hypothesis: yes, that's the main thing I had in mind.
2022-10-21 18:23:29 @phokarlsson @Meaningness @kanjun Well, there are many things one could look for. Personally, I'd be curious to understand to what extent the extra growth helped (or hindered). The extreme case is, of course, the NIH doubling ~2000.
2022-10-21 18:14:02 @Meaningness @kanjun Two things I find fascinating: (1) the flip from foundations to government circa the 50s
2022-10-21 18:09:49 @Cerebralab2 @Meaningness @kanjun We don't assume it.
2022-10-21 17:54:54 @colliand @kanjun Thanks!https://t.co/YgqP8DDWUu
2022-10-21 16:12:47 @Jonathan_Blow @ivn_echvrria A talk would be good. A hundred item twitter thread is also a pretty interesting idea. Hmm.
2022-10-21 16:07:50 @Jonathan_Blow @ivn_echvrria https://t.co/TRLugBXk3R(Cf Robert Gordon's "The Rise and Fall of American Growth")
2022-10-21 16:02:24 @Jonathan_Blow @ivn_echvrria Yeah, will be true late Dec 2025.Amazing the (unclassified) speed record is, I believe, still held by the SR-71.
2022-10-29 22:53:02 @tmrss_ Australia was one of the places I had in mind. Grew up mostly in Brisbane!
2022-10-29 22:43:40 @AdamMarblestone Thanks! Really looking forward to this!
2022-10-29 22:43:16 @gwern Remarkable. One may quibble at the ethics, but it’s not so clear how the intent differs from many parents. One is reminded of Agassi’s father a bit, or the Polgars
2022-10-29 22:35:00 @lfschiavo @tmrss_ I don’t know why either. It wasn’t great when I was growing up, but got very rapidly better as an adult.
2022-10-29 15:42:14 @nat_sharpe_ Applies also to theory behind the nuclear bomb, LIGO, mathematics. Squiggles on tree pulp have amazing power sometimes
2022-10-29 15:34:25 @danielgross Admittedly, I’ve seen this even in NYC. But I like it as a casual presumption!
2022-10-29 15:33:38 @danielgross A friend’s story about Iceland: in a cafe there, saw a group get up and leave, but left laptop on table.My American friend to her Icelandic friend: “oh, they forgot their laptop!”Icelandic friend: “Oh, they’re just holding the table…”
2022-10-29 10:49:01 RT @michael_nielsen: @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug f…
2022-10-29 10:41:58 @bschne Weird, maybe. Funny, definitely!
2022-10-29 10:41:20 @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug free!
2022-10-29 10:38:30 Fertility rate of 0.8!!!Down by a factor 8 over 60 years! https://t.co/tVyYFd393d
2022-10-29 10:37:18 Starbucks is a useful baseline for local cafe culture. In some places it dominates, in others it’s a place only a few tourists go. The latter is a good sign.
2022-10-26 22:45:32 RT @katyilonka: I read the whole thing (and it's long). While I'm not that into metascience generally, the essay really made me reflect abo…
2022-10-26 22:45:23 @katyilonka That sounds like a great essay to have!
2022-10-26 17:24:21 Thoughtful short note: https://t.co/a6HnA21kdg
2022-10-26 01:53:26 @BrianNosek I'm not wild about methodological answers, as in many replies. They seem more like symptoms of rigorous research, not causes. I do like Feynman's comment about: "a specific, extra type of integrity that is not lying, but bending over backwards to show how you’re maybe wrong"
2022-10-25 23:15:38 @jachiam0 This seemed very good to me. Thanks for writing it.
2022-10-25 02:18:40 @vgr And they will always use that as an excuse. Much like arch-communists and -capitalists.
2022-10-24 18:57:26 @RuxandraTeslo I've found this perplexing since being a (broke) undergrad.
2022-10-24 18:30:39 RT @kmmunger: Fantastic article by @michael_nielsen and @kanjun defining the possible scope of metascience and calling for more metascience…
2022-10-24 02:25:59 IBM mainframe specs from the 1960s. Up to 512 kilobytes, and only 3 or so tonnes! https://t.co/wqJyVeqQwR
2022-10-23 16:56:48 RT @marquezxavier: Really stimulating article on the design of the social process of science by @michael_nielsen and Kanjun Qiu https://t.c…
2022-10-23 16:56:00 RT @JMateosGarcia: "A vision of Metascience"I started reading this essay by @michael_nielsen + @kanjun and it's great. I love the idea of…
2022-10-21 19:20:39 It certainly makes me curious what makes b better. Even a tiny improvement - say, to b = 0.12, would make a huge difference. You'd then need a ~300 increase in training data to get that square root improvement....
2022-10-21 19:12:01 Now, admittedly, b is pretty small in their examples. E.g., b = 0.095 for training data, so you need a roughly ~1000 increase in training data to get the square root improvement in probability. Still, that's better than I would have a priori guessed!
2022-10-21 19:10:59 One simple way of looking at this is a constant multiple increase in compute, training data, or model size buys you a square root improvement in probability.That constant multiple is 2^{1/b}, where b is the exponent in the scaling law.
2022-10-21 19:05:58 But it's not so different either.
2022-10-21 19:05:57 In particular, suppose b = 1 and we double size of training data. Then if the former probability was 1/100 we now get a probability assigned to the correct token of 1/10. That's an incredible improvement!In practice, b is smaller than 1, and things don't work quite so well.
2022-10-21 18:58:46 A slightly more precise statement, courtesy of @moultano: https://t.co/PsjnP2e4bE
2022-10-21 18:56:22 @moultano Ah, I see. Thank you!So a little more explicitly, the scaling model says:p = exp(- a 1/M^b), for constants a and b, and p the probability assigned to the correct (unseen) token, in the training data.This seems to me quite remarkable!
2022-10-21 18:53:31 Incidentally: note that the constant a is negative (for all three of compute, training set size, model size). This ensures that p ->
2022-10-21 18:52:13 @moultano Hmm. What's the difference?
2022-10-21 18:42:37 I'm quite curious whether I have this right. I must admit, the probability dropping as exp(-M^a) would seem shocking to me.
2022-10-21 18:41:54 A very basic question about the meaning of the Kaplan et al OpenAI scaling laws paper: https://t.co/YimscMAryl
2022-10-21 18:28:45 @phokarlsson @Meaningness @kanjun It would take quite some thought to figure out what &
2022-10-21 18:24:03 @phokarlsson @Meaningness @kanjun On the hypothesis: yes, that's the main thing I had in mind.
2022-10-21 18:23:29 @phokarlsson @Meaningness @kanjun Well, there are many things one could look for. Personally, I'd be curious to understand to what extent the extra growth helped (or hindered). The extreme case is, of course, the NIH doubling ~2000.
2022-10-21 18:14:02 @Meaningness @kanjun Two things I find fascinating: (1) the flip from foundations to government circa the 50s
2022-10-21 18:09:49 @Cerebralab2 @Meaningness @kanjun We don't assume it.
2022-10-21 17:54:54 @colliand @kanjun Thanks!https://t.co/YgqP8DDWUu
2022-10-21 16:12:47 @Jonathan_Blow @ivn_echvrria A talk would be good. A hundred item twitter thread is also a pretty interesting idea. Hmm.
2022-10-21 16:07:50 @Jonathan_Blow @ivn_echvrria https://t.co/TRLugBXk3R(Cf Robert Gordon's "The Rise and Fall of American Growth")
2022-10-21 16:02:24 @Jonathan_Blow @ivn_echvrria Yeah, will be true late Dec 2025.Amazing the (unclassified) speed record is, I believe, still held by the SR-71.
2022-10-29 22:53:02 @tmrss_ Australia was one of the places I had in mind. Grew up mostly in Brisbane!
2022-10-29 22:43:40 @AdamMarblestone Thanks! Really looking forward to this!
2022-10-29 22:43:16 @gwern Remarkable. One may quibble at the ethics, but it’s not so clear how the intent differs from many parents. One is reminded of Agassi’s father a bit, or the Polgars
2022-10-29 22:35:00 @lfschiavo @tmrss_ I don’t know why either. It wasn’t great when I was growing up, but got very rapidly better as an adult.
2022-10-29 15:42:14 @nat_sharpe_ Applies also to theory behind the nuclear bomb, LIGO, mathematics. Squiggles on tree pulp have amazing power sometimes
2022-10-29 15:34:25 @danielgross Admittedly, I’ve seen this even in NYC. But I like it as a casual presumption!
2022-10-29 15:33:38 @danielgross A friend’s story about Iceland: in a cafe there, saw a group get up and leave, but left laptop on table.My American friend to her Icelandic friend: “oh, they forgot their laptop!”Icelandic friend: “Oh, they’re just holding the table…”
2022-10-29 10:49:01 RT @michael_nielsen: @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug f…
2022-10-29 10:41:58 @bschne Weird, maybe. Funny, definitely!
2022-10-29 10:41:20 @hardmaru I’ve heard that if you wave your arms and say “large model” three times all your code will suddenly be bug free!
2022-10-29 10:38:30 Fertility rate of 0.8!!!Down by a factor 8 over 60 years! https://t.co/tVyYFd393d
2022-10-29 10:37:18 Starbucks is a useful baseline for local cafe culture. In some places it dominates, in others it’s a place only a few tourists go. The latter is a good sign.
2022-10-26 22:45:32 RT @katyilonka: I read the whole thing (and it's long). While I'm not that into metascience generally, the essay really made me reflect abo…
2022-10-26 22:45:23 @katyilonka That sounds like a great essay to have!
2022-10-26 17:24:21 Thoughtful short note: https://t.co/a6HnA21kdg
2022-10-26 01:53:26 @BrianNosek I'm not wild about methodological answers, as in many replies. They seem more like symptoms of rigorous research, not causes. I do like Feynman's comment about: "a specific, extra type of integrity that is not lying, but bending over backwards to show how you’re maybe wrong"
2022-10-25 23:15:38 @jachiam0 This seemed very good to me. Thanks for writing it.
2022-10-25 02:18:40 @vgr And they will always use that as an excuse. Much like arch-communists and -capitalists.
2022-10-24 18:57:26 @RuxandraTeslo I've found this perplexing since being a (broke) undergrad.
2022-10-24 18:30:39 RT @kmmunger: Fantastic article by @michael_nielsen and @kanjun defining the possible scope of metascience and calling for more metascience…
2022-10-24 02:25:59 IBM mainframe specs from the 1960s. Up to 512 kilobytes, and only 3 or so tonnes! https://t.co/wqJyVeqQwR
2022-10-23 16:56:48 RT @marquezxavier: Really stimulating article on the design of the social process of science by @michael_nielsen and Kanjun Qiu https://t.c…
2022-10-23 16:56:00 RT @JMateosGarcia: "A vision of Metascience"I started reading this essay by @michael_nielsen + @kanjun and it's great. I love the idea of…
2022-10-21 19:20:39 It certainly makes me curious what makes b better. Even a tiny improvement - say, to b = 0.12, would make a huge difference. You'd then need a ~300 increase in training data to get that square root improvement....
2022-10-21 19:12:01 Now, admittedly, b is pretty small in their examples. E.g., b = 0.095 for training data, so you need a roughly ~1000 increase in training data to get the square root improvement in probability. Still, that's better than I would have a priori guessed!
2022-10-21 19:10:59 One simple way of looking at this is a constant multiple increase in compute, training data, or model size buys you a square root improvement in probability.That constant multiple is 2^{1/b}, where b is the exponent in the scaling law.
2022-10-21 19:05:58 But it's not so different either.
2022-10-21 19:05:57 In particular, suppose b = 1 and we double size of training data. Then if the former probability was 1/100 we now get a probability assigned to the correct token of 1/10. That's an incredible improvement!In practice, b is smaller than 1, and things don't work quite so well.
2022-10-21 18:58:46 A slightly more precise statement, courtesy of @moultano: https://t.co/PsjnP2e4bE
2022-10-21 18:56:22 @moultano Ah, I see. Thank you!So a little more explicitly, the scaling model says:p = exp(- a 1/M^b), for constants a and b, and p the probability assigned to the correct (unseen) token, in the training data.This seems to me quite remarkable!
2022-10-21 18:53:31 Incidentally: note that the constant a is negative (for all three of compute, training set size, model size). This ensures that p ->
2022-10-21 18:52:13 @moultano Hmm. What's the difference?
2022-10-21 18:42:37 I'm quite curious whether I have this right. I must admit, the probability dropping as exp(-M^a) would seem shocking to me.
2022-10-21 18:41:54 A very basic question about the meaning of the Kaplan et al OpenAI scaling laws paper: https://t.co/YimscMAryl
2022-10-21 18:28:45 @phokarlsson @Meaningness @kanjun It would take quite some thought to figure out what &
2022-10-21 18:24:03 @phokarlsson @Meaningness @kanjun On the hypothesis: yes, that's the main thing I had in mind.
2022-10-21 18:23:29 @phokarlsson @Meaningness @kanjun Well, there are many things one could look for. Personally, I'd be curious to understand to what extent the extra growth helped (or hindered). The extreme case is, of course, the NIH doubling ~2000.
2022-10-21 18:14:02 @Meaningness @kanjun Two things I find fascinating: (1) the flip from foundations to government circa the 50s
2022-10-21 18:09:49 @Cerebralab2 @Meaningness @kanjun We don't assume it.
2022-10-21 17:54:54 @colliand @kanjun Thanks!https://t.co/YgqP8DDWUu
2022-10-21 16:12:47 @Jonathan_Blow @ivn_echvrria A talk would be good. A hundred item twitter thread is also a pretty interesting idea. Hmm.
2022-10-21 16:07:50 @Jonathan_Blow @ivn_echvrria https://t.co/TRLugBXk3R(Cf Robert Gordon's "The Rise and Fall of American Growth")
2022-10-21 16:02:24 @Jonathan_Blow @ivn_echvrria Yeah, will be true late Dec 2025.Amazing the (unclassified) speed record is, I believe, still held by the SR-71.