2022-03-18 13:04:26 @AlexJohnLondon Places where recognition is the bottleneck requiring human labor and the correct decisions given correct recognition are already known. We know how to treat TB but it takes human microscopists to diagnose it. 2022-03-18 04:09:09 RT @pfau: @ylecun I really think the whole field of AI would be much better if you, Gary, Pedro and like half a dozen other senior academic… 2022-03-17 18:34:46 Please help me out Twitter, anyone got (an) example(s) handy of a paper (on arxiv) with a WAY TOO LONG abstract. I'm talking a NeurIPS paper where the intro starts on page 2 or an ICML paper where the abstract spills over to column 2? 2022-03-17 17:52:01 There's a wild difference between methods: 1. not guaranteed by theory to succeed 2. guaranteed by theory to fail (on important cases) Disregard of theory has opened some minds to type (1) e.g., deep learning—but blinded others to type (2) e.g., much deep domain adaptation lit. 2022-03-17 13:48:35 @MISQuarterly @janmendling @ntraft The reason is simple. By counting only papers published at Management S, Marketing S, ISR, MISQ, etc, B-school faculty are able to do dabble in these areas while defending their turf against unflattering comparisons to, e.g., computer scientists. 2022-03-17 13:46:17 @MISQuarterly @janmendling @ntraft Your willingness to publish is great & 2022-03-15 07:01:56 @fhuszar @mldcmu I think after honing them in the first run, I might be down to record the 5 lectures. 2022-03-15 04:31:34 @Cy16180 Likely not, largely to preserve privacy of students, a good amount of class involve dissecting our own students work. But maybe I’ll develop the 5 lecture sessions into a public format that I can record sometime. 2022-03-15 00:52:45 Unfortunately, like music, most of the this material is best taught by doing, through the interactive process of analyzing, pulling apart, and reassembling material. But I'll share the syllabus, and try to share some artifacts from the prepared lectures. 2022-03-15 00:46:45 Tomorrow, we launch "The Art of the Paper", a course on the principles, mechanics, & https://t.co/iHtTVV469X 2022-03-12 08:11:00 CAFIAC FIX 2022-01-25 02:25:38 @ani_nenkova @Nitin_wysiwyg Might make sense as a statistical default because... who really needs help spelling "but"? 2022-01-23 17:49:44 @Nitin_wysiwyg @ani_nenkova Read the example. The context is given, and understood, and yet it fails. 2022-01-23 15:37:59 @KevinZollman Also, when someone has published work w senior co-authors in a lab, I think interviews make abundantly clear whether the candidate drove (or plausibly could driven) the work (vs implemented experiments but didn’t understand the motivation). 2022-01-23 15:32:29 @KevinZollman I’m a bit unconvinced by the “there is little evidence” line of argument. There are no scientific studies showing that almost any mode of practice we pursue as musicians produces results and yet we have master teachers and virtuosos. 2022-01-23 15:30:26 @KevinZollman So I’m not of one mind here. I think that you can get a sense of how someone thinks by watching them try to digest a hard technical problem or formulate a concrete line of attack on a broad/ambiguous problem. Also while “quickness” ain’t everything, it also ain’t nothing. 2022-01-23 15:27:00 @drexlspivy56 @ani_nenkova Also notable is that Alexa can’t use quotation marks to reference text. 2022-01-23 15:25:44 @drexlspivy56 @ani_nenkova Bingo: https://t.co/DrjbaEqPJ4 2022-01-23 15:24:06 @KevinZollman Hi Kevin, earnest question: why do you think this is better? Coming from a field that often doesn’t do interviews and consequently over-relies on signals like past pubs (ruling out many brilliant but inexperienced people), I tend to think we (in CS/ML) should interview more. 2022-01-23 15:18:31 Thought I had a clever solution but nope. @ani_nenkova is onto smtg truly weird: https://t.co/oi4u02o7rq https://t.co/TO1ylyE8oz 2022-01-23 04:52:02 How well would Star Wars have endured if they had a lousy title & 2022-01-22 22:09:44 Have you ever felt so invincible as the day you stepped outside of iso after recovering from Omicron (+triple-vax)? https://t.co/4KDLecSmvh 2022-01-21 23:46:28 Fun fact about my colleague Andre/dʒ/ Risteski (@risteski_a). He's an early riser. Prefers early meetings. He's shy about telling folks, but he shines at 5a-9a. If you're planning any meetings w Andredgje or inviting him for any talks, please be courteous and make it EARLY! https://t.co/fcw9idDQwS 2022-01-21 06:47:37 Fun fact about my colleague Andrej Risteski (@risteski_a). It's pronounced "Ahn-dredge" (/dʒ/ as in "jello"). The "j" isn't silent. It's buzzing in your mouth. Andre/dʒ/. Don't forget. If you ever meet Andrej Risteski, be sure to call him "Andredge Risteski". Loud "j". OK?! Yala. 2022-01-21 05:36:29 RT @acmi_lab: New work by @saurabh_garg67 (ICLR 2022) shows that in general, OOD accuracy is identified only when the optimal predictor is… 2022-01-20 22:12:45 @IAmTimNguyen Plenty of academic research to be found on either side. But arguably, industry research (to be impactful in the lifespan of an enterprise) should never be (1) only. 2022-01-20 16:54:54 Research communities vary in how they prioritize two archetypal "why?" questions in assessing significance: (1) why is this a hard problem, such that your solution is nontrivial? (2) why is this an important question, such that I should care about the answer? 2022-01-20 04:00:40 RT @Sid_Arora_18: A common motivation for explanations is to improve human “understanding” of the models. In our new #AAAI2022 , we evalua… 2022-01-19 22:58:37 RT @valeriechen_: Is interpretable ML (IML) overpromising and underdelivering? Probably. In our new @ACMQueue article, we describe some o… 2022-01-19 18:24:54 @ashipra When students tell you "please get out of here and go on a sabbatical.... no, really, go on a sabbatical", what does that mean? Asking for a friend. 2022-01-19 17:53:12 The @Cmudelphi team (ML/stats for epidemic tracking & https://t.co/8nOTbFhnCT https://t.co/mrUPDSuBvH 2022-01-19 05:26:07 @andrewgwils @roeiherzig @DjokerNole Maybe! I’d put down some coin to see that anachronistic match. 2022-01-17 12:09:17 hope you like waterbugs… https://t.co/dMT7qnYjbV 2022-01-17 08:11:00 CAFIAC FIX 2022-01-12 05:31:03 This is making me see @sirajraval in a whole new, perhaps more romantic, light. One in a long whimsical line of science televangelists cum Don Quixotes, lost in character, dreaming the impossible dream. https://t.co/082fvUo5mi 2022-01-11 08:11:00 CAFIAC FIX 2022-01-06 07:17:26 I don't want to unfollow everyone... but I kinda want the version of Twitter that consists solely of @ccanonne_. 2022-01-05 21:29:26 @CyrusRashtchian @ucsd_cse @kamalikac @yangarbiter @ML_Theorist Cool! Congrats on the new gig Cyrus! Excited to see the damage you do in this exciting (and noisy!) area. Let me know if you'd ever like to hop on a chat with the OOD-focused cabal in my lab some time and kick around ideas. 2021-12-30 05:00:50 @mattmayo13 @sinafazelpour I’ve been an (infrequent) visitor to this club for a decade, mate, but thanks 2021-12-30 04:54:18 This is what happens when you seek hairdresser recommendations from @sinafazelpour. BTW, Giovanni (“Gio”) says hi! https://t.co/JRMI34opU6 2021-12-29 16:12:06 The curious epistemic “paradox” of the AI age. We are in danger simultaneously due to (1) statistical illiteracy 2021-12-28 02:53:23 @ruthstarkman But none for those who don’t know any AI/ML/statistics? 2021-12-28 02:21:56 @ruthstarkman (the post was in jest…) 2021-12-28 00:03:42 @cduhadway You mean to say, "should get double social security as compensation from from the breeders for making space for their spawn"? 2021-12-27 19:43:48 @SashaMTL @LeonDerczynski Am I your first jewish friend? 2021-12-27 19:42:09 @SashaMTL @LeonDerczynski Of course it doesn't give you a free pass, that was never the real point :) 2021-12-27 17:48:45 @khelwegen Arithmetic error. 2021-12-27 15:09:01 @SashaMTL @LeonDerczynski Lol, this article is goofy. If we believe the forecasts of some think tank that has hallucinated a future in which problems resolve absurdly quickly, then future population doesn’t matter. With enough tortured logic and leaps of faith, we can make anything true. 2021-12-27 08:20:00 CAFIAC FIX 2021-12-21 02:04:00 @jeffbigham No painting in January? What is this, socialism? 2021-12-21 01:48:55 If \strikethrough{language} foundation models so smart, why don't they know that it's not July? https://t.co/hbl0OPUxIQ 2021-12-20 03:42:21 The only way out of burnout: 1) lighten your load & 2021-12-19 04:23:11 RT @acmi_lab: Congrats to our 2nd ever PhD, the soon-to-be-minted Doctor Danish, who defended his dissertation this week. Danish joined thi… 2021-12-18 19:03:22 Despite resturanteurs’ fondness for wordplay, I’ve never seen fall rolls or winter rolls on any menu. Presumably pumpkins and fennel (and turkey & 2021-12-17 15:20:05 RT @danish037: Excited to end the year on a high: I passed my PhD defense today! *Absolutely* loved my PhD years @LTIatCMU—I could have s… 2021-12-16 04:34:44 RT @kazizzad: A note on risk assessment, link below. Comments are welcomed. The video is based on the great and grounding work by Audrey H… 2021-12-16 00:13:18 "Do you decline to provide a referral for this candidate? {Yes, No}". Recommendation letter HTML forms are not the place to blindside tired people with double negatives. 2021-12-15 15:35:02 @BierVicki @bayesianboy You can detect plagiarism in your brain? I don't think most (any) of us can (in general). The very same document could be and not be plagiarism (depending on the source) and only broader context can resolve the matter. 2021-12-15 01:05:30 @bayesianboy The short answer is yes we do, and always have. They are more effective when longer contexts are provided. And the task has presumably grown incrementally harder as models have gotten better (if you fix the underlying corpus used as the evaluative standard). 2021-12-10 03:32:50 RT @risteski_a: Want to learn about the neural net complexity of solutions to elliptic PDEs whose coefficients are small neural nets? Come… 2021-12-10 02:53:40 @PreetumNakkiran Figure 1a: appendix 2021-12-09 21:17:08 Behind every beautiful paper is a crap ton of structure, meticulous preparation, painstaking polish, and a grueling all-nighter that destroys the rest of your week. https://t.co/wXuLaK3mCb 2021-12-09 00:39:50 To realize that going forward you will have in your head not the accumulation of the things that you have learned, but a tiny subset, together with vague senses about where to find your materials and how long it will take to refresh. 2021-12-09 00:37:35 Must teach class. Don't know material. Panic. Consult materials from a past year. They're clear, focused, & 2021-12-06 19:15:05 It's a curious fact about accelerating scientific progress that every year, several hundred "once in k [> 2021-12-05 07:30:37 @AllDeepLearning @DrVeronikaCH @michael_at_work Perfect analogy. 2021-12-05 07:27:42 @michael_at_work @DrVeronikaCH You have my ears, you have my sympathy. But I just do not see eye to eye with you on this thread—this quip was not about you nor about OCD. You do not get to silence any post that could possibly be made, by however tortured an argument, to remind you of smtg not fun. 2021-12-05 07:19:03 @DrVeronikaCH @michael_at_work I believe “uncharitable” is a reasonable characterization of the response in this thread to this tweet. 2021-12-05 05:32:23 @DrVeronikaCH @michael_at_work With respect Veronika, there is some point beyond which offense is taken so eagerly and so uncharitably that no such reply is warranted. I will not post “it’s just a joke” after every joke (roughly 50% of posts?) on my feed, and I do not believe such coddling would serve anyone. 2021-12-04 20:06:41 @DrVeronikaCH @michael_at_work The earnestness, seriousness, and self-importance of these replies are not in line with the tenor of the OP. 2021-12-04 14:36:54 By memorizing many digits of pi, one can simultaneously demonstrate a cognitive capacity & 2021-12-04 01:42:00 @AlexJohnLondon live action grinch 2021-12-04 00:18:31 @AlexJohnLondon Worse absolutely or worse relative to expectations? 2021-12-03 16:46:45 Henceforth, all cold- or flu-like illness shown not to be COVID-19 (by lab tests) shall be called "NOVID". 2021-12-03 16:07:26 A complement to the old line about experts (who know more and more about less and less until they know everything about nothing). 2021-12-03 16:04:41 Research: the process of chasing ever less obvious observations until nothing is apparent to anyone. 2021-12-03 04:56:37 @AlexJohnLondon Mrs. Frisby and the Rats of NIMH? 2021-12-02 16:23:56 @risteski_a @PreetumNakkiran OK, I'll be sure not to post any future tweets about your psychopathic Beanie Baby obsession. No further posts about your Beanie Baby hoarding. Got it. 2021-12-02 14:15:13 @PreetumNakkiran @risteski_a True story: you get Andrej alone with his guard down, he'll reveal that he only really cares about Beanie Babies. 2021-12-02 04:45:54 @PreetumNakkiran Can’t publish ML papers about joinery… or can we…? 2021-12-02 04:42:13 The closest you will come to performing miracles in this world are the acts you perform with joint compound, caulk, and sandpaper. 2021-11-29 18:05:00 There is definitely a single neuron that fires exclusively for @PatMetheny & 2021-11-24 18:41:01 The Ancillary series prominently features the “spaceship AI that takes human form” device. Earlier versions include the TV series Andromeda. What are the most compelling/innovative/earliest uses? Any from antiquity where a character stands in for the spirit/personality of a ship? 2021-11-23 20:22:13 RT @bobirakova: When Curation Becomes Creation - Algorithms, microcontent, and the vanishing distinction between platforms and creators. -… 2021-11-22 12:31:00 @Pseudomanifold same here :) 2021-11-21 10:59:46 Writing some background text to motivate a problem, just realized that the "recent" leaps forward in deep learning are now over the past DECADE. 2021-11-20 13:35:16 Of all the strange new verb phrases of the web 2.0 era, “invited you to like X” has to be among the weirdest. 2021-11-19 12:56:17 a damning condemnation of social media by a prescient Milan Kundera https://t.co/fZpOv2tKiF 2021-11-18 12:31:47 @Maggiemakar Today I taught remote from 2a—5a Swiss time, then passed out, overslept my alarm and woke up at 1p for a 12p lunch in another city. Apparently I am a computer scientist, and not a Swiss one 2021-11-18 12:27:02 There must be a German word for the special kind of humiliation one feels upon oversleeping an alarm, messing up their own day and others’. Schlafshämen? 2021-11-17 06:52:10 @MichDiaz_ @osanseviero hi @MichDiaz_ & 2021-11-17 06:31:45 Per legend, the Swiss Confederacy arose in 1291, when Emperor Rudolf I insisted (to the dismay of neighboring monarchs) on carving randomly spaced holes from his cheese. Seeing this “randomness” as an affront to divine will, the Vatican expelled the Swiss & 2021-11-15 21:17:03 @ndiakopoulos While they're at it, perhaps they can do something about @Verizon. Sign up for @verizonfios online, cancel only by enduring hours of holding music during a workday. 2021-11-15 20:46:19 @danielphug @Toblerone @rogerfederer You are related to @Toblerone?! 2021-11-15 18:35:21 @Shreyagupta08 Of course, and glad we got to talk it through, I’d hate to leave you with a false impression of our values. 2021-11-15 18:08:15 @Shreyagupta08 These aren’t buckets. Just caricatures of extreme points on some spectra. The only point is that many of these characteristics can’t be ascertained by counting papers. 2021-11-15 18:06:49 @Shreyagupta08 Re continuously, I think this may be projecting too much onto what I said. We all take weeks off and have slow weeks. My point is that research activity should be driven from within and not tied hopelessly to being pushed. If I went into a coma, projects shd go off rails, not die 2021-11-15 18:03:05 @Shreyagupta08 Oh yeah, by “forceful”, I mean with conviction, not burying your interlocutors with brute force. The tempering words “when appropriate” were deliberate :) 2021-11-15 11:55:14 @jacobmenick @HeinrichKuttler @fhuszar @Toblerone @rogerfederer Notably, Hochreiter is surprisingly chill and lighthearted about how his monumental contribution to tennis has been overlooked. JS, however, overcorrected, alternately claiming credit for Djokovic and suggesting that all tennis innovation dates to the 19th century amateur era. 2021-11-15 11:23:39 @Michael_D_Moor @Toblerone @rogerfederer That might explain this morning's earache. 2021-11-15 10:55:29 @Shreyagupta08 Been stewing on this for the last 10 minutes really trying to understand where your coming from, perhaps experience in a hostile/acrimonious lab or an environment otherwise inhospitable to speaking up? But the solution to that has to be to create a better environment... 2021-11-15 10:46:47 Just landed in Switzerland, "home of the @Toblerone, @rogerfederer, & 2021-11-15 10:21:30 @Shreyagupta08 Only for an unreasonable idea of what “forceful” should mean in this context. I suppose uncharitable reading is the lingua franca of Twitter… 2021-11-15 10:16:21 @michael_muller @acmi_lab People often make that leap and it’s BS. Strong writing is hardly correlated with citizenship. Most people struggle to write, for reasons far more profound than word choice or syntax. I suspect you knew that was nonsense before you sent it. 2021-11-15 10:02:57 @Shreyagupta08 You misread the “shoot to the moon” point (in at least one way). Point is the risk shd be that you’ll go in wrong direction, not do nothing. Concerning having your own ideas and guts to argue for them, that’s a level of coddling that I can’t endorse. 2021-11-15 09:55:14 @sshkhr16 @acmi_lab Off hand, the range includes zero, don’t know median without digging. 2021-11-15 00:56:04 Bhuwan Dhingra is recruiting students. He’s a great researcher, a great advisor, a true gentleman, and if you join his lab, I’ll tell you secret vulnerabilities in his ping pong game. https://t.co/EurThp6iSh 2021-11-13 20:27:20 @fazelogik Thanks for tip, already filed for my certificate on Thursday, so hope to get it ASAP! 2021-11-13 06:07:04 RT @mldcmu: Are you an undergraduate, master's or PhD student? Want to spend next summer at @SCSatCMU researching how to make machine learn… 2021-11-12 21:32:28 RT @AvilaGarcez: NeSy@IJCLR 2021 invited talk by @zacharylipton, Integrating Symbolic Structure and Statistical Language #neurosymbolicAI:… 2021-11-12 17:34:43 @AvilaGarcez Thanks for having me Artur! And thanks again for your incredible patience re the tome zone mishap 2021-11-12 06:58:05 Jerry Fodor, teasing my students (and me) from beyond the grave. https://t.co/Cm0ZRUy1rt 2021-11-11 21:03:07 Exciting new #neurips2021 paper by @acmi_lab PhD student @leqi_liu, who's been on this year. This work models satiation effects in RecSys (depleted enjoyment from consuming repeated items) & Paper: https://t.co/QFHWHAfmM5 https://t.co/0nXGv7tAYt 2021-11-11 15:50:45 @blattnerma I don't have a formal visit to ETH on the calendar but it could be possible! I think it's only a 45m train ride from Lucerne so could scoot up for a day if there was interest and schedules aligned. If we could meet for coffee that would be great! 2021-11-11 07:20:06 @hslu_aiml Thanks!! Anything happening b/w Nov 15 & 2021-11-11 00:36:21 RT @hannah_kerner: Today I presented "Troubling Trends in Machine Learning Scholarship" (@zacharylipton, Jacob Steinhardt) to my group and… 2021-11-10 22:03:49 @dustinvtran @yoavgo So in thery calibration could conceivably be done with the aid of statistical tools (under some assumptions that might be too strong). Today, depends. In confs where AC gets same reviewers over& 2021-11-10 21:25:17 @yoavgo Hey Yoav, I think your hot take here is just not correct. 1) Scores need to be calibrated, across area and across individuals 2021-11-10 21:21:39 Fancy Europeans of Twitter. I’m flying to Zurich on Sunday, en route to Lucerne. Where to go? Who to meet? What to eat? 2021-11-10 17:43:30 RT @annargrs: A highlight from @zacharylipton's talk at https://t.co/axQAhW5YdF on positive/negative results in engineering papers: If X w… 2021-11-10 15:44:26 About to speak in 0 minutes at the EMNLP Workshop on Negative Results in NLP! https://t.co/80ePjvF661. Come join and please interrupt me with questions :) 2021-11-10 00:13:41 @annargrs Oh wow! Thanks for posting this, made me realize that in a minor hiccup, the wrong title/abstract got submitted for my talk. It must have seemed like a strange topic for a negative results workshop at an NLP conf: "some [positive] results on [not-necessarily NLP]" 2021-11-09 06:32:23 Has anyone ever adapted a great space opera into an actual opera? 2021-11-08 04:07:19 @davelewisdotir Thanks for checking it out! In addition to the paper (https://t.co/QePLedZfqs), we also shared the code (https://t.co/MoEvDH5QbY) 2021-11-08 00:21:34 (paper link: https://t.co/B1NjEzPtLA) 2021-11-08 00:18:22 When we stumbled into PU learning, I was shocked to find no usable methods for modern ML settings. Classical methods intractable, while modern ML papers offer tricks yielding inconsistent estimates. Our #neurips2021 paper advances both mixture proportion estimation & 2021-11-06 23:20:00 CAFIAC FIX 2021-11-01 19:20:00 CAFIAC FIX 2021-11-01 17:30:00 CAFIAC FIX 2021-08-19 22:56:44 RT @Gradio: We might hope that when faced with unexpected inputs, well-designed software systems would fire off warnings. However, ML syste… 2021-08-17 15:27:15 @srvmshr Hey, thanks for checking in! Huge PR storm coming in (front matter and Ch 2 & 2021-08-17 14:59:59 Just got some #2 pencils from the store. They came out of the bag *presharpened*. What else will Gen Z want premasticated for them? #kidsthesedays https://t.co/XIOuVUWmeW 2021-08-17 14:51:38 What is the most exciting paper of the month? No restrictions. Inspire me (please). 2021-08-16 18:56:42 Are all imported packages important? 2021-08-14 06:12:06 RT @yuvalmarton: Want to be able to naturally and automatically parse “human and natural language processing”? Apply to NYU & 2021-08-14 04:35:04 RT @nourelkhalawy: Superheroes of Deep Learning comics made me laugh so hard and kinda hit hard with the code snippet in volume 1 totall… 2021-08-13 18:21:09 @tallinzen @xp__22 You mean you might have even considered taking on a scrub like me circa 2013? 2021-08-13 18:00:48 Want to bridge the gap between human and natural language processing? Apply to NYU & 2021-08-13 15:41:58 @RWerpachowski @DanielleWenner Says the man who's never broken a toe badly. A toe can turn dark purple (what some would describe as black) right away, not because of necrosis but because of bleeding. But yes, not medical advice, anecdotal reminiscing by a fellow breaker of toes. 2021-08-13 14:30:00 @DanielleWenner Most of the time, just tape to nearest neighbor. If it's a real serious break (if bone is visibly disturbed or entire toe has turned dark purple or black), I'd just bite bullet and see doctor. 2021-08-13 01:46:46 @DavidSKrueger @AmandaAskell I think the real confusion here centers on the word “presume”. To be treated under the law as though innocent (e.g. not incarcerated), doesn’t mean there can be no social (or even employment) consequences. 2021-08-13 01:44:21 @DavidSKrueger @AmandaAskell Not necessarily true. It could make one guilty of defamation, slander, etc. 2021-08-12 01:00:46 RT @yudapearl: Thanks for anointing me with a black mustache, but I can't understand these comics - I grew up with little red riding hood a… 2021-08-12 00:37:40 @d_i_j_k_stra @yudapearl PDF is better: https://t.co/1ey6l4xwZ6 2021-08-11 15:32:13 Can causal counterfactuals counteract the curve-fitting crisis? Can @yudapearl’s epiphany save the day? Find out in the latest “Superheroes of Deep Learning” https://t.co/bi7QW8kTD6 https://t.co/vfU915tbnL 2021-08-11 15:07:37 @philipikim @acmi_lab @FalaahArifKhan Exception in “main” java.lang.NullComprehensionException between keyboard and chair 2021-08-11 13:59:45 @victorveitch Famous last words. 2021-08-11 02:50:30 RT @acmi_lab: New on the Approximately Correct blog: "Superheroes of Deep Learning Vol 2: Machine Learning for Healthcare" by @FalaahArifK… 2021-08-10 19:37:39 RT @risi_kondor: This is brilliant. https://t.co/93qTgOmO9E 2021-08-10 12:30:53 RT @FalaahArifKhan: ***New Comic*** The DL supes are back!!! Volume 2 out now #DLHeroes https://t.co/ERhINZTBBn https://t.co/zKFOM8b3a3 2021-08-10 12:19:50 Downloadable PDF version available on GitHub: https://t.co/WtUjIkrBls 2021-08-10 12:15:55 As coronavirus surges, all eyes turn to the ML illuminati. Can our curve-fitting crusaders crush #covid19? Find out in *Superheroes of Deep Learning, Episode 2: Machine Learning for Healthcare* (co-authored w @FalaahArifKhan) https://t.co/DrA1G5G9Qy 2021-08-06 01:04:07 If there’s one experience that constitutes the archetype of the word “flustered” it’s struggling to get a zipper to budge. 2021-08-05 21:33:52 @leonpalafox I don’t think inability of good work to get out is a serious concern. Most good work will get in, even if it takes a couple tries, and false positives are far more common than false negatives. Add to that the friendliness of the entire community to preprints in advance of pub... 2021-08-05 04:18:07 Getting reviews is hard. Sometimes the reviewers stink, sometimes they’re right and you have problems, sometimes they’re confused but it’s your fault for writing badly. It’s stressful, it’s noisy, and it’s complicated. But this day-after twitter grousing isn’t a great look. 2021-08-04 17:56:27 @guohao_li @NeurIPSConf This is a clear case where a private note to the SAC or PC chairs notifying them of this concern would be appropriate. 2021-08-04 16:24:52 RT @tetraduzione: controversial opinion: this year's #NeurIPS2021 reviews are the best I ever had (how is this possible?) 2021-08-04 05:10:48 A lot of well educated people mindlessly forwarded this @DjokerNole misinformation. It’s not just drooling idiots getting played in the meme game https://t.co/Qxuj18kDZe 2021-08-03 02:22:50 fantastic burn. fascinating rhetorical move. https://t.co/7Yh25BAe1U 2021-08-02 22:35:47 RT @FrankPasquale: Zuckerberg’s dream: “some version of Ernest Cline’s novel Ready Player One meets Fortnite, meets virtual reality, meets… 2021-08-02 20:10:40 I want to attend a *CL conference to find the shadowy corridor where the remaining linguists commiserate & 2021-08-02 19:57:55 I'm glad @random_walker & 2021-08-02 13:44:40 @PierreNyq Yes! 2021-08-01 06:27:15 So I read "The Making of the Atomic Bomb". TL 2021-08-01 04:19:33 Jesus this is such insider jazz shit but I wish I could communicate to people on the tech side of my life how funny this video is... https://t.co/SpwP9YfflX 2021-08-01 03:06:55 As an added benefit, this one-seater doubles as a coffin. https://t.co/NIMegjcsPN 2021-07-31 19:21:01 @sinafazelpour Maybe, but only for interpretations where "ridiculous" does not necessarily mean "bad". Case in point: Rush. 2021-07-30 04:53:43 @ericjang11 I'd cry hyperbole, but I did ask for strong opinions only, so fair. 2021-07-30 03:19:28 jax/flax vs jax/haiku: strong opinions only. yalla, let’s go! 2021-07-29 22:44:17 @KLdivergence Just don't let them call you "cheugy" 2021-07-29 22:25:30 RT @marksaroufim: @zacharylipton The problem with R is that it limits you to 5 figure salaries but Python supports 6 figure salaries 2021-07-29 15:37:21 @blattnerma @lejyby voilà! 2021-07-29 15:15:36 @nkrishnaswami A "soft" compel. Hey buddy, you pick any θ you want, but I've got my finger on the λ! 2021-07-29 15:13:34 Of course, "we must do the right thing to these people" sounds bizarre and would not sell well. 2021-07-29 15:11:30 One strange linguistic turn in the vaccine debates is "compel people to do the right thing". In what sense are these people "doing" anything if compelled? An interesting juxtaposition of passive and active. 2021-07-29 14:19:01 @egrefen Correction: 2021-07-29 08:02:04 WTF, I thought we were all clones. https://t.co/mc7f5obkBT 2021-07-29 07:58:17 @CarlosMougan Profundity. 2021-07-29 07:47:21 Is deep reinforcement learning cheugy? 2021-07-29 07:45:56 What's your favorite air pistol event? Don't hold back. 2021-07-29 07:20:22 @adrian_weller Qu'est-ce qui pourrait indiquer le contraire? 2021-07-29 07:18:05 There's a kind of research that consists of assuming the crowd is right and trying to get to the front of the line, and another that consists of figuring out where they've got it all wrong. Interestingly, fields full of the former present great great opportunities for the latter. 2021-07-29 07:07:18 @blattnerma I mean the science of data. Or, as they say in France, "les science des données". Comprenez vous? 2021-07-29 07:00:27 Guys, I have a burning question. What's the best language for data science, Python or R? 2021-07-28 21:58:38 How do I get to the alternate universe where a bloke can open the newspaper without reading politicians' opinions on critical race theory or slimy rubber-necking at the mental health woes of young athletes? 2021-07-28 19:03:23 Congrats @JunjieHu12! Heads up to prospective NLP graduate students interested in cross-lingual & Website: https://t.co/QnGys6CSUI (he's hiring) https://t.co/r5ARMxDvXl 2021-07-27 02:25:25 Just placing this here to hold myself (& 2021-07-27 02:05:49 So much credit to @JonBatiste for otherworldly creativity, hustle and vision. It's one thing to get to the apex of a career path that currently exists, another to get to the top of one you imagined back to life. 2021-07-26 13:51:21 RT @ACMQueue: When Curation Becomes Creation: Algorithms, microcontent, and the vanishing distinction between platforms and creators https… 2021-07-24 17:06:13 Often in research, the real thing and quackery lie so close together that absent specialized equipment they’re indistinguishable. 2021-07-24 12:32:55 Looking for a #COVID19 scapegoat? Blame the business schools! They did it to make the MBA great again... https://t.co/QKRLJBjEUb 2021-07-24 12:01:00 Ok, I’m going to flush out the obvious one: “branded covid masks” 2021-07-24 12:00:09 When conferences come back, what new swag item(s) will there be that weren’t around in the 2010s? 2021-07-24 04:40:31 Is there a “quantified pet” for quantified self people that turn their madness on their puppies? 2021-07-23 17:37:09 (to be fair, I have some first-hand experience) 2021-07-23 17:35:59 A deep net can predict “cat” without knowing what a cat “is”. Similarly, many researchers can develop ostensibly Bayesian procedures without even knowing what the propositions to which they assign prior probabilities actually mean (or could even plausibly mean...)! 2021-07-23 14:21:04 @pfau Yes I do, but ICML poster attendance is not the star witness. 2021-07-23 05:36:44 Bubble contractions are not always bad times. A lot of crap gets cut out and some creative ppl blossom under the greater selective pressure. Google rising from the ashes of the dot-com bubble comes to mind. I expect some diamond formation in ML under this nascent contraction too. 2021-07-22 23:34:18 The dark side of impostor syndrome is paralyzing anxiety. But the bright side is gratitude. If I get to spend decades in “I can’t believe I get to work with these people” land, that’s enough. 2021-07-22 22:56:51 RT @ruchowdh: For those of you who wanted to apply for an internship with the ML Ethics team at Twitter but found the application was close… 2021-07-22 22:20:56 @jeanqasaur Define popular? By # of users, they already are, always have been, and always will be? 2021-07-22 06:56:28 @TheLeanAcademic @saurabh_garg67 https://t.co/uaytkWc8p2 2021-07-22 04:43:11 If you're "at" ICML, my student @saurabh_garg67 is presenting on RATT now! Poster link: https://t.co/K2mQTdYcDg TL 2021-07-22 04:37:48 I'm in a virtual poster room where 90% of posters are unmanned. As wasteful as it may seem, for now, virtual surrogates can't replace the energy (& 2021-07-21 15:56:06 @martinpotthast @jaketyo @bhuwandhingra @frrncl Hey, @jaketyo is on it! 2021-07-20 16:10:25 RT @saurabh_garg67: New @ #ICML2021: We present an empirical study to characterize the heavy-tailed nature of the gradients of the PPO surr… 2021-07-20 16:09:55 RT @saurabh_garg67: Visit us at ICML 2021 Spotlight: Tue 20 Jul 10 pm - 11 pm EDT Poster: Tue 20 Jul 11 pm - 2 am EDT https://t.co/N0ekHb… 2021-07-18 13:37:38 To make progress in research, I need [poll time] 2021-07-16 05:12:49 So I was thinking about placing myself in a conservatorship, but now having second thoughts. 2021-07-16 04:35:55 @manashardas Wann ist Mikkel? 2021-07-16 03:50:16 @machelreid Ok but that’s “delightfully stupid” 2021-07-16 03:34:00 @the_dismal_tide T1 & 2021-07-16 03:25:03 Loki is here to remind us that no matter how good the acting, time travel shows are always stupid (...except Prisoner of Azkaban) 2021-07-15 20:14:56 Whoa.... there's a @duolingo for Yiddish. How much $$$ to unlock profanity lessons? @burrsettles @LuisvonAhn https://t.co/lWE7ijvqY3 2021-07-15 15:43:23 @martingoodson You think you beet me? Well, that's your op-onion. I suggest you turnip around. Make like a turmeric and dye! Yucc...a. 2021-07-15 05:57:27 What are they going to call conservatorships in the inevitable rebranding? 2021-07-15 03:48:24 @YairZick I would \strikethrough{eat} read that. 2021-07-15 03:16:02 @DarceyNLP @nsaphra Hard to digest the Rutabaga postulate, but who could resist Cassava's theorem? 2021-07-15 03:13:37 @nsaphra I can get behind this one. 2021-07-15 02:58:05 Kohlrabi is unique among root vegetables in that it could easily be mistaken for a mathematician. Nobody would blink at “Kohlrabi’s theorem” but “Turnip’s lemma” would raise eyebrows. 2021-07-14 19:42:20 @victorveitch I mostly agree here. Except the "no penalty". For those looking to pursue a university faculty career, there can be big consequences. 2021-07-13 21:41:54 New article on the imperative to attract (& 2021-07-13 04:12:08 RT @PreetumNakkiran: I defended today! Thanks very much to my advisors @boazbaraktcs & 2021-07-12 01:28:17 @dww @mathtick What do you think I mean? 2021-07-11 22:39:40 Imagine living in a time where philosophy (or any academic literature...) was powerful enough to be regarded as a political threat. Oh wait... 2021... https://t.co/yzZ3Z0DdBG 2021-07-11 22:31:48 @johnsontoddr4 Who here is talking about “current AI”? Perhaps you are thinking of a different conversation? 2021-07-11 22:14:50 @dkaushik96 2021-07-11 22:07:45 @dkaushik96 I’m sure your proclamation has “big pizza” shaking in its Armani boots. 2021-07-11 21:16:58 @wfwhitney Maybe we could create a family of models w an associated loss function that would establish some isomorphism between data density and loss in parameter space provided dim (data) > 2021-07-11 21:14:05 To an ignorant outsider (like me), the weirdest things about a Bayesian account of scientific reasoning are: 1) nobody has the faintest idea what distribution we are referring to 2021-07-11 21:07:17 @GaelVaroquaux Lol yes funny, and a thought that crossed my mind :). But looking for more here than power to make a model look good. 2021-07-11 21:05:33 Once realists admit “approximate truth” and instrumentalists concede that reality, whatever it may be, has something to do with which theories are useful, it can be really f’in hard to figure out what anyone’s actually disagreeing about / whether this is the right way to frame it 2021-07-11 20:18:31 @R4_Unit Lol yes definitely so we can ensure that it is a critical point. But can we make it a true minimum where there exists some epsilon such that every point besides theta in a ball of that radius has non-zero loss? How can we more broadly characterize the expressiveness of the dist? 2021-07-11 20:12:44 @MorseCell So clearly we can't do anything we want. Single layer case makes that clear. How much expressiveness is there in the distribution to shape the loss landscape? 2021-07-11 20:11:12 @wfwhitney Right, how much further can we go? Let's say θ are the parameters of an MLP we apply square loss. Clearly, we can't bring about any loss landscape we want. Simplest case for a single layer cannot make it non-convex. 2021-07-11 20:07:20 @MorseCell OK great so that's simple enough . So how much more ambitious can we get? In an extreme case, can we construct a distribution to bring about any sufficiently smooth loss landscape? If not, what can we assure? Let's say θ are the parameters of an MLP, loss is L2. 2021-07-11 19:30:23 Are there any proofs about the expressiveness of data to define a loss landscape? E.g. for any parameter vector θ, can we construct a dataset D for which θ is a local minimum? 2021-07-11 18:14:28 Can’t get down with the hipster response of “space, meh”. However tacky the billionaire pageantry, don’t make space a techbro thing. 2021-07-11 17:19:49 When you hear “evidence based medicine” do you think ppl mean A) statistical/quantitative findings in published research (broadly) 2021-07-11 13:03:15 @exteriorpower But then, surely the learning theorist makes inductive inferences about deductive inferences about inductive inferences! 2021-07-11 05:52:03 Mathematicians make deductive inferences. Scientists make inductive inferences. Economists make reductive inferences. And learning / statistical theorists make deductive inferences about inductive inferences. 2021-07-11 05:31:06 @pentagoniac Impossible? No. There's a very big difference between questioning whether the sorts of tools we have are sufficient versus whether the sorts of tools we need are possible. 2021-07-11 05:16:17 @khoomeik Well that would defeat the point. I want a system that would have similar observations of the real world to human naturalists and would somehow, despite never seeing a single species form (takes too long) develop a coherent theory for how they formed. 2021-07-11 05:07:15 What would it take to build an AI system that could/would discover the origin of species? It would seem no system of a kind with current unsupervised & 2021-07-11 01:15:00 @thegautamkamath = BASS SOLO 2021-07-11 01:00:55 @KLdivergence @YairZick https://t.co/dEQBU9yRVM 2021-07-11 00:52:52 Google's invested 1 trillion in AI but YouTube's ad placement algorithm still cuts to commercials in the middle of (vs between) songs. 2021-07-10 22:09:02 [and later...] "Philosophy—in every field of inquiry—is what you have to do until you figure out what questions you should have been asking in the first place." 2021-07-10 22:03:44 “There is no such thing as philosophy-free science, just science that has been conducted without any consideration of its underlying philosophical assumptions” — @danieldennett 2021-07-10 16:34:09 Concluding that a model is “robust” because it improved on 1 out-of-domain challenge set is like concluding that a model generalizes because it correctly classified 1 test example. 2021-07-10 05:46:18 @danielrock https://t.co/rmH3JL0dOG 2021-07-10 05:40:25 @danielrock ah yes, of course, the old “my dog ate it”, we believe you ... 2021-07-09 03:59:52 I want to go on a sabbatical but I want an official university document that calls it a "vision quest". 2021-07-09 03:57:58 @PreetumNakkiran @sarahookr GPT-∞, bröh 2021-07-09 01:21:19 Out with @annielizard, had a run-in with my ex (axe?). Awkward https://t.co/oSPlVANJrj 2021-07-08 23:46:36 Quarantine has made one thing clear: The real reason for conferences is to force weeklong breaks from meetings. 2021-07-08 02:53:29 @ccanonne_ Dude, that’s what DM is for. 2021-05-22 17:38:27 RT @acmi_lab: New paper by our postdoc & 2021-05-22 13:32:41 @KevinZollman So meta. 2021-05-21 19:22:04 Anybody hear there's some kind of upcoming ML conference deadline? Sup with that? 2021-05-19 21:05:48 @vinneycavallo Gold, not an investment. Unless you are supplying gold to a company that makes things with it or out of it. 2021-05-19 20:54:09 @vinneycavallo Lol, orthogonal consideration. My saxophone has increased in value. But it's not an investment on that grounds. It was only an investment insofar as it was an investment in my (now deceased) musical career---providing capital to be used to make something. 2021-05-19 19:16:27 RT @priyakalot: Curious about how AI researchers are thinking about the societal consequences of their work? We (@JessicaHullman @ndiakopo… 2021-05-19 18:44:15 RT @DMulliganUCB: live now @AIESConf I was so happy to read this. On the Validity of Arrest as a Proxy for Offense: Race and the Likelihood… 2021-05-19 18:40:28 @MaartenSap @SCSatCMU @LTIatCMU @ai2_mosaic @allen_ai Hi @MaartenSap, this is fantastic news! I've enjoyed following your work and couldn't be happier that you're joining us. Hope we can connect soon and looking forward to seeing you post-postdoc. 2021-05-19 17:34:45 @LockeBenji @mhajabri No. 2021-05-19 17:25:49 @mhajabri @LockeBenji Missing point. It's neither ownership nor profit motive makes it investment. When investing, you provide a resource to someone who actually does something with it, e.g. providing capital to startup so they can buy equipment, hire ppl and build things. Squatting ain't investment. 2021-05-19 16:55:09 @LockeBenji You are putting up capital to be used to actually do something. Just sitting on some resource (even a real one) in hopes that price will go up is not investment. 2021-05-19 16:45:25 @vinneycavallo lol yes, the investment there comes via reading and thinking, not by owning bitcoin. 2021-05-19 15:19:28 @abhijatbiswas So in general, it's dubious to tax capital gains at a preferential rate compared to hard-earned income from labor. So that's the real problem. 2021-05-19 15:17:45 @dipamc178 Commodities such as gold, yes. You do not "invest" in gold. However, buying and holding shares issued by a company that actually makes something is an investment (while high frequency trading is not). Real estate, depends on what you do with the property. 2021-05-19 14:58:40 It's deeply misleading to call ppl "cryptocurrency investors" for owning #BTC. Should be "cryptocurrency speculators"... unless they're investing in the underlying technology or related firms. There's nothing intrinsically wrong with speculation, but let's use language honestly. 2021-05-19 13:35:52 RT @KevinKaichuang: Bound test error by inserting randomly-labeled data into the train set and stopping training when error is low on the c… 2021-05-19 09:09:21 @mer__edith @timnitGebru @ZoubinGhahrama1 @mmitchell_ai @JeffDean Could be wrong, but I suspect this assessment may be overestimating the practicality/business-relevance (for good or evil) of anything the research org was working on. 2021-05-19 04:51:41 As an academic discipline, "business technologies" addresses problems that lie in the intersection of the complements of business and technology. 2021-05-19 03:45:44 @jonathanstray Thanks! 2021-05-18 21:37:24 @michaellavelle Hey Slacker, why don't you Do the Right Thing and Dirty Dance your way over to the Breakfast Club and Scream yourself Clueless into a bowl of Corn Pops? 2021-05-18 21:25:47 (addressed by Gen Z to senior, not a Boomer rejoinder) 2021-05-18 21:24:56 We are mere years away from "OK, millennial." 2021-05-18 12:32:57 They didn't want your vote to count, and now they don't want your vaccine to count. I think it's time to start disseminating the vaccine through the 5g networks... https://t.co/cOa0xbHnA2 2021-05-17 12:56:43 @mireillemoret Thanks Mireille! All credit to our student Riccardo for the hard work to make this study happen. 2021-05-17 00:41:52 @JeremyAbramson Im speaking as a dancer (without tenure, mind you) who had a hard time for a couple years, not as a bandleader. These are meant as words of empowerment, not criticism. Perhaps you’re missing some context. 2021-05-16 20:08:09 @CarlosMougan Syllabus will be posted publicly but course itself will be in person. 2021-05-16 17:42:01 @mireillemoret Yes, good catch! I really need a second coffee. 2021-05-16 17:09:42 New work on problems with recidivism prediction. While we often pay lip service to the proxy label problem, too few address it directly. Here, using National Incident Based Reporting System (NIBRS) data, Riccardo probes the (in)validity of offense as a proxy for arrest. https://t.co/6UFYSL5lc1 2021-05-15 19:26:32 @tallinzen There are some folks ∈ {ppl who reinvent themselves every decade}. I can think of a few (@yudapearl, Michael Jordan, Miles Davis). But their intersection w {ppl who bow to pressure from clowns} is ∅. 2021-05-15 19:17:36 Me: Let’s go home & Herbie: STFU https://t.co/K2sRuJXbjg 2021-05-15 18:37:52 @tallinzen By then, it's too late. Just as constraints aren't worth talking about if they could never be tight, tenure isn't worth talking about if you've already been castrated and micro-chipped. 2021-05-15 17:59:40 @rajivjk @ashish_vt All? To be sure, a vice we should all watch out for, and an affliction to which none are immune. But perhaps it's not a coincidence that both encountered this vice in the same place. 2021-05-15 17:53:33 @ashish_vt No place is perfect. But in my experience, CS schools/departments are paragons of academic freedom compared to some of the other places an interdisciplinary scholar might find themselves. 2021-05-15 17:45:27 If you genuinely want guidance in research, seek it and receive it. But if you're already locked in and excelling, and yet someone thinks that it's their business to domesticate you, put your middle finger up stand your ground. Tenure exists only to protect this right. 2021-05-15 17:45:26 Young interdisciplinary faculty: some of you will find yourself in departments where people will tell you what to work on, or where to publish. Remember that if you go down that path, your job (title be damned) won't be a professorship, and tenure gotten that way isn't tenure. 2021-05-15 17:30:17 @DeepGraphLearn @yoavgo @francoisfleuret Now because ML is a field with a rapid publication cycle and short papers (compare to, say, statistics or economics) a substantial ML PhD very often includes several published papers. But that doesn't mean it is the *criteria*. 2021-05-15 17:01:30 @DeepGraphLearn @yoavgo @francoisfleuret Now citation counting is also the wrong way to think about it. There are plenty of fatally flawed papers that nevertheless rack up citations because they caught the attention of an uneducated audience of arxiv spammers. It’s just about the work. 2021-05-15 16:59:46 @DeepGraphLearn @yoavgo @francoisfleuret This all sounds like the wrong way to talk about what constitutes a PhD. There is no formal paper counting. But a good thesis should have several strong chapters that add up to a significant contribution. 2021-05-15 15:10:19 Let’s enact a law requiring anyone who says “the Turing test” (vs “a Turing test”) to serve 1 yr at a remedial education facility dedicated to grammar and reading comprehension. 2021-05-15 14:52:18 @thebigmehtaphor Capacity is limited, just take it! 2021-05-15 05:09:05 @b_subercaseaux Yes, this class will be offered in the Fall :) 2021-05-15 02:58:14 With proper preview https://t.co/lqAivquKTM 2021-05-15 02:52:49 Since this tweet I got enough of a push from kind mentors & 2021-05-14 15:06:33 @EveryNoise @841io @Spotify So dark it's sparkling. 2021-05-14 14:21:18 @EveryNoise @841io @Spotify Even if we concede this... brooding academic w gothic tinge: Lang Lang performing Mary had a Little Lamb!?! 2021-05-14 13:49:31 @nsaphra @Spotify And then ... Lang Lang! https://t.co/ni2Sl4Bndf 2021-05-14 13:01:08 @_hylandSL @Spotify That much has been explained to me by Gen Z whisperer @841io. But ... Lang Lang!!! The guy who records Fur Elise as an adult and sleeps under a 10ft autographed poster of himself is brooding academic with a gothic edge?!? 2021-05-14 12:12:12 @841io @Spotify Crazy! That makes more sense, up until ... Lang Lang 2021-05-14 02:22:31 The award for weirdest shit ever presented by an app goes to @spotify for “Dark Academia Legends”, a category that apparently includes Miles Davis and Bill Evans, & 2021-05-13 21:18:51 @JCornebise @gdesjardins_ml Don't have impostor syndrome (contradicts OP) or don't think it's the worst feeling in the world? 2021-05-13 19:06:28 Tell me, how should we evaluate snake oil as a cure for cancer? Surely, let's administer different varieties of snake oil to crowdworkers instructed to feign cancer, and ask them which they prefer! #research 2021-05-13 18:16:37 @gdesjardins_ml Worst feeling in the world? No way. 2021-05-13 15:13:15 RT @ML_Theorist: @zacharylipton today (Thursday) 1pm PST @HDSIUCSD https://t.co/EfqtmWy2b2 2021-05-13 11:01:25 RT @s_scardapane: *RATT: Leveraging Unlabeled Data to Guarantee Generalization* by @saurabh_garg67 @zicokolter @zacharylipton from @acmi_la… 2021-05-13 02:55:42 The best feeling in this job is starting and ending some days thinking “I must have pulled off the greatest con of all time that I get to work with these people”. 2021-05-12 03:32:31 @limufar @UCSDJacobs @ucsd_cse @HDSIUCSD There's a "website" link on the linked page. 2021-05-12 03:29:56 What a cruel tease! The @Spotify version of @maceoparker's "Life on Planet Groove" doesn't allow streaming "Shake Everything You've Got". That's like offering a lobster roll, but withholding the lobster. 2021-05-12 03:20:20 Hey @UCSDJacobs, @ucsd_cse, @HDSIUCSD friends, I'll be "at" UCSD Thursday to talk about label shift & https://t.co/bZbUf6dy1m 2021-05-12 01:30:28 RT @bneyshabur: Excited to announce an internship opportunity for summer or fall 2021The research will explore qualitatively new behaviors… 2021-05-11 21:03:15 RT @JohnSnowLabs: Learning to Summarize Visits from Doctor-Patient Conversations Presented by @zacharylipton We present the first study to… 2021-05-11 18:12:12 @sinafazelpour See the discussion about EMNLP's different handling of deadline extension amidst crises in the US vs India? https://t.co/AAaZ1jAsNp 2021-05-11 04:57:14 It’s ok to think area X is important but that their theory/methods/empirical practice/standards of exposition/peer review are weak. Just get those things elsewhere and do work on the problems you care about, making sure that they meet *your* bar. 2021-05-11 04:46:02 Fortunately, the sub-areas & 2021-05-10 14:54:10 @KLdivergence Stanford, Connecticut? 2021-05-10 03:39:21 Heads up to all @purdueCS students interested in ML research. Your new prof @kazizzad is among the most brilliant, energetic, and creative researchers I've ever known and doing exciting work across the theoretical-empirical spectrum. Find a way to work with him. 2021-05-09 19:12:42 @yoavgo More seriously, after years in both processes, i think the ML reviewing standard is indeed much higher. Things are bumpy AF in deep learning tracks, but these are at least compared against mature theory tracks where review quality is quite high. Peer review in NLP is in crisis. 2021-05-09 17:42:52 But it can be the random seed... https://t.co/vuf3uvVQsg 2021-05-09 13:40:28 Phases of spinning up a research lab: 1. Crap, I'm in this alone. 2. Now I'm responsible for all these promising young people, how can I manage this alone? 3. Wait a minute, I'm not alone. 2021-05-09 13:35:15 @jasondeanlee @acmi_lab @saurabh_garg67 @zicokolter Also I think apart from immediately practical considerations, there's some insight here that we can maybe expand upon. Even for those models that eventually fully memorize, they first reach *near* full accuracy at a point where this mechanism assures generalization. 2021-05-09 13:22:02 @jasondeanlee @acmi_lab @saurabh_garg67 @zicokolter Great questions & 2021-05-09 12:52:53 @jeremiepapon @kaushikpatnaik @acmi_lab @saurabh_garg67 @zicokolter Not so simple, perhaps read the paper first and process the theorem. The analysis rests on the exchangeability of the data (the algorithm doesn’t know which is which), once you optimize the (expressive) model to treat the randomly labeled data differently, no guarantee can hold 2021-05-09 05:12:09 RT @olcan: this idea has legs 2021-05-09 03:36:56 @roydanroy @jefrankle @HanieSedghi @bneyshabur @prfsanjeevarora Exactly. Thanks for taking a look! 2021-05-08 23:16:35 @tdietterich Hey, I’m not Jay Leno, buddy. You don’t get a repeated punchline with three rimshots. 2021-05-08 23:12:07 I feel like ppl aren’t appreciating this quote properly... 2021-05-08 23:07:11 RT @PreetumNakkiran: Nice result with a very clean argument. Note the gen. bound is independent of |F|. Intuitively, this is b.c. the train… 2021-05-08 22:57:29 It’s ok, you can cite prior work generously, and yet still be recognized by history for a seminal contribution. https://t.co/JGsx8APo8Q 2021-05-08 19:22:51 @graduatedescent There's noise in the system. But there's also enormous numbers of reviews, extreme selection bias for what gets tweeted about, and extremely biased perspectives from authors whinging seconds after a fresh rejection. It's why I try not tweet complaints on my own papers' reviews. 2021-05-08 18:24:07 @askerlee @heghbalz @aminkarbasi @acmi_lab @saurabh_garg67 @zicokolter The randomly labeled data is mixed in. So you don’t train on only randomly labeled data. You train on the clean + random data and can be sure up to the point that you start fitting the randomly labeled data that your error on clean data is representative of population error. 2021-05-08 18:06:22 @aminkarbasi @acmi_lab @saurabh_garg67 @zicokolter The key is that the setup requires training not just on the labeled data, but also taking some (iid) unlabeled data, assigning random labels, and tossing it in with the training data. We know which examples are from which set but the ML algorithm does not. 2021-05-08 18:05:12 @aminkarbasi @acmi_lab @saurabh_garg67 @zicokolter If you have perfect memorization then you have fit not just the clean data, but also the randomly labeled data, and our bound guarantees nothing. But if despite training on clean + (some) randomly labeled data, you fit the clean data well but not the randomly data => 2021-05-08 14:32:27 @jefrankle @HanieSedghi @roydanroy @bneyshabur @prfsanjeevarora Link to paper: https://t.co/uaytkWc8p2 2021-05-08 14:30:53 Glad to share this work showing that early learning implies generalization (post hoc). Doesn't assure early learning will happen a priori, but that it may be a key to understanding generalization. Maybe of interest: @jefrankle @HanieSedghi @roydanroy @bneyshabur @prfsanjeevarora https://t.co/7jmD0QtP7t 2021-05-08 14:24:19 RT @acmi_lab: New @ #ICML2021: When a trained model fits clean (training) data well but randomly labeled (training) data (added in) poorly,… 2021-05-08 14:23:58 @rasbt 2021-05-08 14:11:07 Spends 10 yrs writing papers like “one cool trick” and “bla bla layers”. Writes blog posts on how papers are outdated (“just share the repo, bro”) and how industry is just as vibrant a place to be a scholar as academia. #kidsthesedays 2021-05-07 09:24:42 @khudabukhsh I said “dangerous”, not “dickish”. 2021-05-07 08:30:49 Sometimes, the most dangerous thing you can do is read. 2021-05-06 17:06:18 RT @sinafazelpour: The updated version of our "Fair Machine Learning Under Partial Compliance" (w/ @jessicadai_ and @zacharylipton ) for @… 2021-05-06 02:19:07 \strikethrough{May 5: ACL decisions. May 7: EC decisions. May 8: ICML decisions. May 12: UAI decisions.} ***May 13: @Castlevania Season 4*** 2021-05-06 00:24:17 How do I know Herbie’s an Apple guy? Because he has so much i-poo https://t.co/2Byl9juK3e 2021-05-05 18:33:23 @DrMartensPhD @PWGTennant I’m not saying videos aren’t cool—video skills a great superpower to have! Rather, my point writing is far from acceptable, poorly taught (if at all, in engineering tracks), and that lack of writing ability is still a bigger problem facing students. 2021-05-05 18:30:22 @davelewisdotir Pushing back on last point: 1) there are countervailing benefits—eg, ppl read and cite the work more 2021-05-05 16:50:39 @rasbt Lol we need to have some whiskey and discuss! 2021-05-05 16:24:20 So I'm going to be contrary here and suggest that the level of scientific writing in most papers is so abysmal as to pose an existential threat, limiting our ability to make intellectual progress. I'd be more concerned with teaching PhD students how to write than how to Tube. https://t.co/0A19rrWRgU 2021-05-05 01:32:25 @stat110 I'm gonna be greedy and ask you to throw in the didactic skills and lecturing charm. :) 2021-05-05 01:25:50 I want to go back in time to take undergrad probability & Check out his wildly popular videos on probability: https://t.co/mF6sN67jj8 & https://t.co/fIy5RSp9Tz 2021-05-04 22:05:35 @deaneckles @AlexBurnap Sounds like a dynamic group! 2021-05-04 22:01:33 @AlexBurnap In my experience these aspects are all over the map. CS has bigger labs on avg, but degree to which profs are hands on & 2021-05-04 21:24:34 @deaneckles Fair points. Surely, there's more at-university management, especially for giant labs (& 2021-05-04 05:10:25 Where’s the social network with all the academics I want to follow but where nobody is posting the “types of ___ papers” meme or ogling billionaire divorces? 2021-05-03 16:22:25 RT @acmi_lab: ACMI visiting (undergrad!) researcher @jessicadai_'s first paper: "Fair Machine Learning Under Partial Compliance" accepted w… 2021-05-03 14:44:41 @AlexBurnap Can you elaborate? In what sense is the consulting more varied---by subject matter/job type? 2021-05-02 13:57:56 RT @lauretig: @zacharylipton Do you have a sense of what the distribution of industry < 2021-05-02 13:46:15 Even in CS, there are perennial debates about the connection between research and practice, and depts represent a comparatively healthy array of academic activity across that spectrum. 2021-05-02 13:46:14 “Today it is possible to find tenured professors of management who have never set foot inside a real business, except as customers.” Possible? Interestingly, far more CS faculty work for & https://t.co/aWBlWDmTVp 2021-05-02 05:44:37 Academic communities come about for myriad reasons: a new technology, a common aim, etc. Far the weirdest are communities whose sole purpose is to avoid unflattering comparisons to others. Here, the ugliest caricatures of academia are brought to life by the slimiest people. 2021-05-01 04:26:49 @Harriet35929613 GPT-3 is the same techniques we already knew about applied on more data and getting more or less the results we all expected it would. It's fine, and the results are interesting to see. But hardly where the intellectual action in ML is these days. 2021-05-01 02:21:10 @_Will_Rice Maybe not? And perhaps we may not agree on what constitutes a “problem” or a “change”. 2021-04-30 19:58:39 @FelixGoldberg1 And in with the demise of easy pickings comes the return of that beautiful but terrifying question... 2021-04-30 17:04:40 Moreover, adjacent fields, including those dedicated to cleaning up the mess (social++, robustness++, learning theory), decision-making (causality, OR), etc. are arguably are still early in reaping the benefits from the deep learning shock-wave. 2021-04-30 17:04:39 Bubbles don't always pop. Sometimes they gently deflate while emitting a high-pitched farting squeal. Arguably, the core Deep Learning bubble crested in 2018. 2021-04-29 23:41:08 Thrilled to announce that “thrilled to announce” can only be announced snarkily, absent true thrill or the representation thereof. https://t.co/M4UtCmLLiI 2021-04-29 17:25:51 RT @mlforhc: Check out our exciting list of speakers for MLHC 2021! https://t.co/67g5QynOvI 2021-04-29 04:07:39 RT @sethlazar: Early bird registration ends soon for AIES 2021! https://t.co/EKWr3CwxSb We received many fantastic submissions and have sh… 2021-04-29 04:06:53 @sethlazar @jessicadai_ @sinafazelpour Thanks for the shout @sethlazar! Most impressive: @jessicadai_’s first paper (still in undergrad) 2021-04-29 04:05:12 RT @sethlazar: Fair Machine Learning under Partial Compliance by @jessicadai_ @sinafazelpour @zacharylipton. Brilliant interrogation of wha… 2021-04-27 16:57:37 Greatest irony ever baked into a software UI? The button to press when exercising judgment to holster a stupid tweet is called “cancel”. 2021-04-26 20:43:09 Since the injustice and indignity of the indentured work in academia and its differential impact across racial and economic lines has gained steam on twitter, I just want to bump up the heartbreaking tale of Thea Hunter. https://t.co/60GAvQVcVQ 2021-04-26 20:37:14 @scottlegrand @mer__edith @davelewisdotir There are major problems in the system. Overall, we're comparatively lucky in computer science (compared to some basic science, humanities) in the amt of support, # of positions, short/absent post-doc-ing, & https://t.co/60GAvQVcVQ 2021-04-26 20:23:04 @mer__edith @davelewisdotir It really sucks that your work got lifted. That's a shitty feeling and that sting doesn't go away fast. At the same time, I wouldn't take it as an indictment of the whole of academia or the possibility of collaboration. 2021-04-26 20:21:20 @mer__edith @davelewisdotir We don't arXiv things too hastily anymore, try to share things w people we trust, live with the consequences, & 2021-04-25 23:49:00 @FightOnTwist @federalreserve @benbernanke @falconandwinter Indubitably 2021-04-25 17:07:07 @AndrewKemendo @Wikipedia "Hate"? Simma down now. https://t.co/q9FBGw6plR 2021-04-25 15:54:15 There should be a @Wikipedia page for ppl that don’t have podcasts. 2021-04-25 02:45:54 Am I the only person who noticed that former @federalreserve chair @benbernanke has a minor role as “governmental official” in @falconandwinter? 2021-04-24 16:36:23 @danyhaddad43 2021-04-24 16:35:56 @danyhaddad43 Yes. No contradiction here and not the point of the statement in the paper :). Also don't forget the main point which is just that even if you know nothing about the distribution p(x) but do know the true p(y|x), the F1 optimal decisions depend on the batch (unlike for accuracy) 2021-04-24 15:34:12 Herbie has most of the skills to eat his hard yak cheese stick. But he progresses far faster when I position it for him, providing leverage. In this collaboration, my contribution doesn’t rise to the level of authorship, but I would appreciate an acknowledgment. https://t.co/3MujsqEe3t 2021-04-23 22:42:12 @sarahcat21 @jmhessel Oy, not the issue. 2021-04-23 17:46:53 @danyhaddad43 Vs for *macro F1* (avg F1 separately computed over classes), where you are just maximizing F1 separately for each class and thus the blind classifier would predict *all* examples (from all classes) positive. 2021-04-23 16:09:46 @danyhaddad43 Need to read for context. But if I recall, you are missing the fact that we are now talking about multi-label F1, not for a single class. Here even with the blind classifier you have different base rates. For *rare* classes* they will be predicted all negative. 2021-04-23 15:18:45 @danyhaddad43 Expectation over what? Conditioned on what? (think it through) 2021-04-23 10:18:10 RT @quantum_graeme: My biggest worry about the quantum boom is that it is draining talent from more productive industries like finance and… 2021-04-23 02:53:16 @danyhaddad43 Not true. Say you have the Bayes optimal classifier, once you see the batch the optimal threshold conditioned on the observation is not necessarily the same as the optimal threshold in expectation prior to seeing the batch. 2021-04-23 02:51:19 @danyhaddad43 (that's only if the random variable is unbounded, if bounded can apply Hoeffding's) 2021-04-23 02:46:03 @danyhaddad43 No it's the opposite. Given an uninformative classifier the optimal threshold is 0 --> 2021-04-23 02:14:17 RT @alexdamour: I know exactly who needs to hear this, but arXiving a corporate white paper is not a good way to convince the research comm… 2021-04-23 02:13:25 @danyhaddad43 This is unrelated to the i.i.d assumption, but just the definition of F1 score. The examples themselves can be i.i.d. but given such an (i.i.d.) batch of examples, how you should threshold your scores depends on the full batch bc (unlike error), F1 is not additive. 2021-04-22 05:25:40 RT @ProfMattFox: The J& 2021-04-22 05:23:45 @ProfMattFox Quite a few, but my rules is, that like (well-designed) advanced software features, they should be invisible to muggles. 2021-04-22 04:59:20 Adult academic expressing a measured take, spotted in the wild (on twitter). Get out your binoculars. Endangered but not extinct. https://t.co/EqJvNxb3eL 2021-04-22 04:30:55 [takes hiatus from twitter, comes back, remembers why took hiatus] 2021-04-21 20:54:34 Sometimes you really needn’t read past the title. Thanks for the hard-hitting journalism @cnn. https://t.co/GKHVKkEPD9 2021-04-21 18:11:55 RT @NYNLPmeetup: June 2, @ZacharyLipton from @CarnegieMellon presents on "Learning the Difference that Makes a Difference with Counterfactu… 2021-04-21 15:57:07 @JunjieHu12 @UWMadison lol Just do the opposite of whatever I do and you’ll be great! 2021-04-21 15:16:17 @JunjieHu12 @UWMadison Congrats Junjie!! 2021-04-21 05:04:55 @bose_joey @JordanRaykov Depends on what we are calling an idea and what we are calling ML. In the “try random architectures” world that comprises the majority of the paper tsunami, “ideas” are cheap, but this is not a world of ideas. 2021-04-21 00:19:25 @BierVicki @glichfield Wow, what a depressing thought. The end of all new science. 2021-04-19 22:02:21 @glichfield I don't know where the precise cutoff is, but believe (in many sub-fields) we are way past that point. Perhaps the trickier issue is always determining what interventions are both (i) possible 2021-04-19 19:12:49 Some amazing young causality researchers launching faculty careers this year. Prospective students, look out for Carlos Cinelli (@analisereal) joining @UWStat this fall, @raziehnabi joining @EmoryUniversity, (& 2021-04-19 05:12:37 @quantrad @LeonDerczynski Well. they're related. Ignorance reduction is a time-consuming enterprise. :). Yeah it's a worrying time, I'm sure many reviewers are getting steamrolled by BS, and I'm sure many authors are receiving dubious reviews. 2021-04-19 04:31:00 Duke Ellington recorded "Money Jungle" and "Duke Ellington & 2021-04-19 03:50:59 @adjiboussodieng @chris_brockett @mmitchell_ai porquoi non, ma soeur? la quarantaine est longue et @Duolingo est bon marché 2021-04-19 02:46:31 @chris_brockett @mmitchell_ai juste une minute mon ami! @adjiboussodieng est ici! ne désespérez pas! 2021-04-19 01:15:14 @srchvrs There’s a contradiction here, but it’s not mine. 2021-04-19 01:14:26 @srchvrs The point is I’ve seldom seen reviewers actually discuss (even basic) theory in any detail, besides to make a superficial comment i.e., “good”, “solid”, “bad”. Nothing that rises to the level of what would pass for thoughtful informal comments, let alone peer review elsewhere. 2021-04-19 00:52:51 @roydanroy Right, everyone involved thinks themselves an emperor, the reviewers and authors are the same population, and not an article of clothing to be seen for miles. 2021-04-19 00:49:08 @code_star Not quite the point... 2021-04-19 00:12:42 @julius_adebayo Here I don’t think the problem is “too many theorems”. The bigger problems are dishonesty and innumeracy. Breiman’s debate was taking place at a higher level of scholarship that took for granted some things that the deep learning and NLP research communities today cannot. 2021-04-19 00:06:43 @tomgoldsteincs Or the scientific method... 2021-04-18 14:24:10 @TaliaRinger Oh all manner of problems. Theorems that aren't even theorems. Incorrect definitions, missing definitions! Other times, the results are unrelated to the argument they ostensibly support. And sometimes the results are good, but one can hardly tell from looking & 2021-04-18 14:06:08 Reviewing & 2021-04-18 11:58:59 @khoomeik The NLP lit today ... 2021-04-18 06:15:54 Titles play an outside role in determining a paper’s impact. But the effect is far greater for superheroes. The @DCComics exec who vetoed “Water Boy” back in 1941 really saved the franchise. 2021-04-18 05:52:28 How noisy does a scientific literature have to become before it's more efficient to rediscover everything from scratch than to distill what's both true & 2021-04-18 00:50:06 @rasbt @Miles_Brundage @hardmaru Happens all the time at tech firms for most employees too. 2021-04-17 14:22:53 @jeffbigham @Miles_Brundage @hardmaru Perhaps they forget that most tech companies did not publish much before 2012. There is nothing new about companies doing research, solving math problems, or publishing methods w academic collaborators. Has happened for decades in operations research/management. 2021-04-17 13:55:19 @Miles_Brundage @hardmaru Why should this he surprising? Would you have been surprised to hear about home depot employing statisticians to develop classical demand forecasting models? It’s only surprising because you are calling it “AI”. 2021-04-17 05:45:23 @chrisoffner3d @rasbt Yes. 2021-04-17 03:50:16 “They can’t see it, only sense it.” Buried in @ZackSnyder’s Justice League are subtle lessons about category errors. 2021-04-16 19:41:26 RT @kasia_kozdon: @AndreasZeller @DrVeronikaCH If you haven't read it yet, check out @zacharylipton's Troubling Trends in ML paper. It soun… 2021-04-16 16:57:36 Yuppies are a lot like messiahs. The proclamation "I'm not a yuppie!" is a solid indicator that your interlocutor is, in fact, a yuppie. 2021-04-15 20:41:11 RT @burrsettles: Come work with me! @Duolingo is hiring ML Engineers to help develop amazing new technologies that teach listening & 2021-04-14 01:23:02 @KabirNagrecha @MaartenvSmeden Or to learn what their needs are before proposing an “it” 2021-04-14 00:11:43 @KabirNagrecha @MaartenvSmeden I think this impulse might already be missing the point. To propose a metric, we first must have some conceptual idea of what we’re trying to measure in the first place. With interpretive machine learning, the answer is often something that we know isn’t there. 2021-04-13 13:54:44 @ChristophMolnar Not just from time to time. You need to challenge them until you find a story that makes sense (without leaking logic through giant holes). You can’t save a field by refusing to examine its foundations, only by developing them. 2021-04-13 02:42:40 @suguna_misra @MaartenvSmeden @CynthiaRudin This is a bad faith characterization of a talk that you have not seen, a poor characterization of my work in the area, and a miscued ad hominem swipe all in one thread. You’re better than this. Goodnight. 2021-04-12 20:37:44 @SaraGPtacek @MaartenvSmeden BC most of what flies on this header isn’t real technology addressing any real well-defined problem. I’d worry more about finding technically strong people that build a model that actually works, or if you need to estimate a causal effect, someone that actually knows causality. 2021-04-12 18:57:09 @mireillemoret All great observations and valid perspectives. I'll keep my silly rejoinder in the back pocket for our next chat :) 2021-04-12 18:43:03 @quantrad Judging from your use of "real" it's clear your degree is not in etymology 2021-04-12 16:49:09 RT @MaartenvSmeden: Great summary of the standard narrative for explainable AI by @zacharylipton https://t.co/p2nYvqJ9N3 2021-04-12 05:27:30 @NikhGarg Perhaps few books you read commit so resolutely to lodging a one-sentence nugget into your soul by rote? 2021-04-12 05:16:06 @quantrad I suppose one could think of it as humility, if someone’s identity is wrapped up in their field. So it’s acknowledging “those guys” are smart too. But personally, I just don’t think I ever felt such intense belonging in a field that checking out others would require humility. 2021-04-12 05:01:33 compress(*The Checklist Manifesto*): checklists are great, they help you make less stupid mistakes, here's a success story 2021-04-12 04:53:26 @quantrad There's also the social/institutional question—how to get away with it. Helps to have a supportive environment. Helps to have some momentum to feel safe shaking off careerism. But sometimes also requires throwing a hard elbow. 2021-04-12 04:49:48 @quantrad I try to spend more time reading outside "my field" than in it. And I tend to believe that literacy is the passport across disciplines. If you can read their papers, grok their standards, & 2021-04-12 03:43:37 Many academic fields suffer from widespread ignorance of their conceptual foundations. Meanwhile, "philosophy of X" too often suffers from ignorance of how X actually works. Naively, I'm inclined to wonder if the solution could be to dissolve all formal departmental boundaries. 2021-04-10 16:12:37 @henrygouk @WeisiG @overleaf @dkaushik96 aint the same 2021-04-10 16:04:22 “This has been a book about people trying to solve problems created by people trying to solve problems.” From @ElizKolbert’s “Under a White Sky” about geo-engineering. Also an apt characterization of the entire field of ethical/trustworthy/reliable AI/ML. 2021-04-10 14:36:29 @mdekstrand @overleaf @dkaushik96 Great idea! @danish037 has promised to get the knuckle tattoos if this feature is implemented. 2021-04-10 14:24:09 Dear @overleaf, if you implement a feature that allows us to search for projects by collaborator names, my PhD student @dkaushik96 will tattoo "@overleaf" on his biceps and name his first child @overleaf. 2021-04-10 05:10:14 Help me feel learn stuff. What recent papers have you read that truly had an interesting idea, strong technical execution, and beautiful exposition? 2021-04-09 20:06:08 RT @shagunsodhani: Join us in an hour for the panel discussion with @zacharylipton @deviparikh @mcarbin & 2021-04-08 23:35:32 If @overleaf crashed, permanently obliterating all content that isn't backed up elsewhere on the web, computer science would be set back: [poll time] 2021-04-08 23:19:50 @gtbelinda @strubell I'm not involved in BeatSage, but yes same first author, my boy @chrisdonahuey 2021-04-08 14:05:17 @geomblog @mark_riedl Of all the real issues deserving of calories. 2021-04-08 04:39:27 @geomblog @mark_riedl Thats a 27% acceptance rate... why “but yeah”? 2021-04-03 00:38:39 @raohackr The potential future value of future technology isn’t necessarily realized via massive investments in commercializing existing technology. 2021-04-02 17:12:32 @tlbtlbtlb But here's a plausible scenario: the technology just plateaus at its current level. There would still be jobs for people to apply existing tech in familiar ways, but the tech could lose (much of) its force as a national concern. 2021-04-02 17:10:14 @tlbtlbtlb Yes and no. I suspect that a large fraction of today's handsomely compensated ML practitioners (the typical "knows how to tensorflow" engineer) would take a massive pay cut in a move to an SDE role. And few could pass as professional statisticians. 2021-04-02 14:17:54 A strange facet of the "AI age", is that the technology's professional critics are more precariously over-leveraged on its continued relevance than its practitioners. 2021-04-02 13:50:34 In @DSMarkovits's account, the middle class is worse off (having lost power) and the upper class, despite capturing a greater share of wealth, consists (increasingly) of burned out workaholics who are (in that sense) worse off than their aristocratic predecessors. 2021-04-02 13:50:33 Timely read: "The Meritocracy Trap" by @DSMarkovits. Occasionally too casual and polemical, but an insightful portrait of America's widening inequality & 2021-03-30 02:51:01 @bgalbraith screen hijacking? screen metastasis? 2021-03-29 22:38:55 A screen's a weird object to share. The other person still needs their own screen in order to partake. This would never happen with a pizza, or a cigarette. 2021-03-26 18:04:21 @tomgoldsteincs The answer is the didactic "we" is inclusive of both the speaker/writer and the audience, engaging them in the pedagogical exercise. "We can simplify the equation as follows: ..." Not just me, but us together. 2021-03-26 13:38:54 One benefit of reflexively using the didactic “we” whenever speaking about research is you will never accidentally take sole credit for collaborative work. 2021-03-25 18:52:39 @levidehaan What else could I have meant? 2021-03-25 16:26:19 Hey @TarletonG, while I have your attention briefly, could you send me a DM? 2021-03-25 15:41:58 @alexkreimer It's sunny outside. And there are so many unread books—written by people that thought about what they wanted to say for more than 15s. 2021-03-25 12:41:49 @rogueCHI That was also 2016... 2021-03-25 11:04:41 @mireillemoret “Surely You’re Joking Mr. Feynman” https://t.co/JpHbsOmoYA 2021-03-25 01:34:50 Advice to a young academic in 2016: get on Twitter. Advice to a young academic in 2021: get off Twitter. 2021-03-25 01:32:56 RT @dkaushik96: Excited to share that the camera ready version for our #ICLR2021 paper is now available: https://t.co/mSdugneHje And thank… 2021-03-25 00:36:23 “Then I had another thought: Physics disgusts me a little bit now, but I used to enjoy doing physics. Why did I enjoy it? I used to play with it. I used to do whatever I felt like doing—it didn’t have to do with whether it was important...” 2021-03-25 00:34:26 “It was a brilliant idea: You have no responsibility to live up to what other people think you ought to accomplish. I have no responsibility to be like they expect me to be. It’s their mistake, not my failing.” 2021-03-25 00:20:29 “I DON’T believe I can really do without teaching. The reason is, I have to have something so that when I don’t have any ideas and I’m not getting anywhere I can say to myself, “At least I’m living 2021-03-20 20:51:51 @ChrSzegedy That’s just not true (or conceivably true) in a literal sense (problematic word here is “all”). 2021-03-20 20:50:38 @srchvrs @successar_nlp @ReviewAcl All using same paper matching software? 2021-03-17 18:59:11 RT @eliasbareinboim: Happy to share our tutorial's video on "Causal Fairness Analysis" in the @FAccTConference this year (joint work w/ Dra… 2021-03-17 18:58:54 I understand the impulse to standardize criteria and track data, but the multiple choice questions in peer review forms are generally dreadful. The options are often overly specific with their union failing to cover the sample space, leaving no mechanism for an honest answer. 2021-03-14 01:48:43 @jeffbigham @bgalbraith Don't mistake precision for passion. 2021-03-14 01:47:57 Do you have deep questions about technology that you want answered incorrectly? Join @marksaroufim & ***AI Office Hours: Wrong Answers Only*** https://t.co/RYDVyOYwC9 2021-03-14 00:29:42 @YiyuanLi1 The only measure in this world that matters is the distance between me and James. 2021-03-13 23:45:23 French is a zero sum game. Can’t settle for advancing to the next (platinum?) league. I must administer a drubbing so profound that James (you know who you are) packs it up and switches to Italian. Allons-y! #gamification https://t.co/GseGQq5tn9 2021-03-13 04:49:19 An interesting read on the strange world of elite private secondary schools. Strangely, for all the advantages their attendees gain in college admissions, I hardly ever encounter them at the higher levels of academia. https://t.co/tSzKpyWbMO 2021-03-11 00:18:26 @gautam1858 https://t.co/t9IOf81MCZ 2021-03-09 15:25:37 Fresh delivery from @amazon, only 4 years late. https://t.co/pYNB5GNtEJ 2021-03-09 14:00:27 Over last 2 weeks in Texas, mask wearing in public down sharply (roughly 7.2% dip). Also concerning, rates of people who would "definitely or probably" choose to get vaccinated has fallen from 72% to 66%. Full results from Delphi survey: https://t.co/07r3uqqbMk https://t.co/98YTkbj5XN 2021-03-09 04:59:33 @zstevenwu No. 2021-03-09 04:16:28 Junior faculty mental health FAQ condensed: Q: Should I *? A: No. 2021-03-08 05:35:22 Before social media, did we ever have to read (or inflict) such catastrophic cloying as “< 2021-03-04 21:33:17 RT @rajiinio: These tutorials were fantastic! Those that didn't have the opportunity during the day should seriously take a moment to check… 2021-03-04 12:50:23 The 2021 @FAccTConference has begun and a full day of tutorials with amazing speakers is under way! Register and join us! https://t.co/T1AHCzgLGi 2021-03-04 12:10:46 We’re told that we live in a surveillance economy that tracks our every action, personalizes all our content, and silos us into filter bubbles. And still I keep seeing news about the royal family. Is this like the matrix, where they need to keep it kinda crappy so we believe it? 2021-03-02 03:34:18 Trump faces probes from 5 independent investigations. Asked to comment, BERT replied “only 5?” https://t.co/5Rq4tgDmUx 2021-02-27 03:06:50 Chick, devastating until the very end. Hadn’t even begun to slow down. https://t.co/WsgmWcfJsl 2021-02-25 14:03:01 On one hand, it would be nice to be a billionaire. On the other hand, a top-of-the-line Prius costs < 2021-02-24 19:09:48 Few acts in the modern world are riskier than sending a private Zoom chat. 2021-02-23 22:50:19 All I want for Christmas is to read an 8 page NLP paper that contains neither “*BERT*” nor “probing”. 2021-02-23 05:47:27 Appears even the @nytimes is letting “A.I.” generate articles. https://t.co/CRJcpWil6O 2021-02-18 20:53:48 Sick of being scolded "it's pronounced I-CLEAR"? Whenever anyone says "I-C-M-L", bark back "it's pronounced I-SMELL". #feelthelearn 2021-02-11 04:22:18 @KLdivergence Software developers also call them "ninjas". 2021-02-09 01:56:17 Pausing Twitter sabbatical to send massive congrats to @sinafazelpour, our dynamic, insightful, & 2021-02-04 21:17:25 Guys, I thought we were going to HOLD THE LINE! https://t.co/Yjns1hbAV4 2021-02-04 14:01:14 Flaws that render a nice idea useless in practice are an inconvenience to the theorist but a non-starter to the practitioner. Flaws that render an idea incoherent are an inconvenience to the practitioner but a non-starter to the theorist. 2021-02-02 13:18:41 @CarlosZednik @Miles_Brundage @ExplainableNL Placing the blame on policy makers for "not paying attention" to work on explaining black boxes presupposes that the work out there is actually addressing their concerns, if only they would look. That is not the case. 2021-02-02 13:17:20 @CarlosZednik @Miles_Brundage @ExplainableNL The "reverse" is not true. Sadly the state of affairs that I described in the "Mythos" still described the status quo in the field quite well, with a surfeit of post hoc "open the black box" interpretability methods proposed but hardly any addressing a single well-formed question 2021-02-02 00:57:14 @danish037 Winner. 2021-02-02 00:56:52 @adrian_weller The words "fine" and "chicken" are doing a lot of work here. 2021-02-01 19:39:13 Soup's lovely, but it's never a good sign when soup is an apt metaphor (e.g., soupy writing). What else is always great in life but never in metaphor? 2021-01-30 23:52:52 @lauretig On a serious note while ML salaries have been inflated in general everywhere the past several years, startup salaries are usually (comparatively) skimpy (vs big boring companies), the promised payoff is usually in stock. 2021-01-30 23:46:19 New life plan: (1) launch a fledgling AI startup 2021-01-29 18:49:58 Can't contain my excitement abt introducing BACKRONYM: Batch-Accelerated Convergence through Kronecker-Rademacher Orthogonalization of the Nyström-Yuille Matrix. ML will never be the same https://t.co/mKTNSNsrqz #deepthoughts 2021-01-29 15:21:00 @azhir_io @CmuDelphi These are important and subtle points but not accessible via the available data. We might hope that in countries with more standardized electronic medical records at the national level we could get a handle on these more subtle issues. 2021-01-29 15:19:27 @azhir_io @CmuDelphi Hey, we are *not*! That's the whole problem. Cases means "confirmed cases" not "actual infections. They can grow if infections go up but also if testing goes up. That's why our analysis is so crucial. We focus on hospital data, where testing is more reliable. 2021-01-29 14:45:07 RT @acmi_lab: Congratulations to @teppercmu PhD student Kyra Gan, whose work introducing (& 2021-01-29 14:30:50 RT @CmuDelphi: Since the pandemic began, the ratio of deaths to cases has fallen dramatically. But how much of this drop is due to actual i… 2021-01-29 01:37:37 Since July, my team at @cmudelphi has been tracking the dropping case fatality rate, deciphering what share of the drop might owe to actual drops due to treatment (vs testing, younger infectees, etc). Drops of 55-60% appear plausible! Learn more here: https://t.co/CzssdkZXxA 2021-01-28 16:14:35 Our #covid19 work at @CmuDelphi, particularly the massive surveys (75k+ respondents daily) that @capnrefsmmat has spearheaded are featured in the new administrations #covid19 pandemic strategy doc. Check it out, and follow our daily updated survey results: https://t.co/mZdIROfqrR https://t.co/Vo66wRKv2p 2021-01-28 13:09:56 All the people on Twitter & “for 14 years, I was in the Mossad, I mean, not in the Mossad” https://t.co/CCC7Kg16kx 2021-01-28 04:35:26 The dominant practice of the applied machine learnist has shifted from ad-hoc feature hacking (2000s) to ad-hoc architecture hacking (2010s) to ad-hoc pre-training hacking (2020s). 2021-01-28 01:02:19 Starting now! https://t.co/bdD4N4QfuR 2021-01-27 23:47:33 When philosophers speak of ontology, it's often interesting, but seldom useful. When data scientists speak of ontologies, it's often useful, but seldom interesting. #deepthoughts 2021-01-27 18:03:14 RT @SilverJacket: Lyin’ AIs: Explainable AI can be programmed to hide bias, a form of fairwashing. My feature in February’s @IEEESpectrum.… 2021-01-27 18:02:24 RT @macfound: Check out this panel today, and learn more about this 2019 MacArthur Fellow. #MacFellow https://t.co/Qkr9OHFNRS https://t.c… 2021-01-26 23:13:27 @ani_nenkova Start with a backronym, preferably a dinosaur name, then randomly assign the initials to NLP/AI buzzwords. 2021-01-26 21:42:44 Excited to join MacArthur Fellow theater director @AnnieDorsen in a panel discussion tomorrow focusing on her AI-infused productions & 2021-01-26 17:06:16 @fhuszar And if I feel shame? 2021-01-26 15:24:33 RT @risteski_a: @ykilcher @zacharylipton Fremdkaffeetrinken? 2021-01-26 14:59:07 German should have a single word for "accidentally drinking someone else's coffee". Please invent it and share in the replies. 2021-01-24 01:01:32 No reader can say if this is 1) real 2021-01-22 02:47:14 RT @acmi_lab: how much did you grow in the last 6 weeks? probably not as much as @acmi_lab dogtoral fellow herbie (pictured w music prof… 2021-01-22 00:33:13 @iNv_Lorenthiel @SeguraAndres7 @D2L_ai Not true, the book is now trilingual, with implementations of (most, and slowly approaching all) chapters available in both PyTorch, TensorFlow, and MXNet. 2021-01-21 23:30:04 RT @AndrewMendez19: I skimmed https://t.co/9YcaduyGZQ this morning to prepare for my Applied ML engineering interview. Cannot recommend thi… 2021-01-21 01:50:54 I've been watching the inauguration coverage for 9 hours already and they haven't arrested any pedophile satanists. Getting a bad feeling about this. 2021-01-20 17:47:37 @fhuszar OK, turned personalization back on. https://t.co/B5w2JmNpyS 2021-01-20 17:15:42 Good morning. https://t.co/9H0BAsQs1a 2021-01-15 19:23:12 @emilio__ferrara Yes, Turing describes a kind of test, but the description is underspecified. Results of course will depend on the populations of human judges and human interlocutors, the ground rules for the dialogues, the interface, etc. The contribution is not a specific test but an idea. 2021-01-15 17:51:45 So much confusion could be avoided if we stopped saying "*the* Turing test" and started saying "*a* Turing test". 2021-01-15 17:47:44 RT @slatornews: CAUTION: Researchers at @CarnegieMellon and @amazon found that #beamsearch, which boosts #BLEU scores, can lead to high… 2021-01-12 20:15:42 Today, I've been ducking harassing personal threats from MAGA trolls dog-whistled by Richard Grenell. Remarkably, @ZoaKlein, head of a Zionist organization, lent his voice to attack me, a descendant of Holocaust survivors, in Newsmax, while defending election propagandist Grenell 2021-01-12 16:56:22 RT @lindsayaellis: NEWS: Ex-Trump appointees linked to @CarnegieMellon Institute for Politics and Strategy raised unsubstantiated doubts ab… 2021-01-11 03:28:24 @chrmanning Arguably, if we read Kuhn closely (granted few in ML have), little that we are doing in deep learning amounts to what he calls a paradigm, and the whole enterprise better resembles what Kuhn calls pre-paradigmatic science. 2021-01-10 21:39:21 RT @zacharylipton: @RichardGrenell @simonwiesenthal "Never again" doesn't just mean never again *to us*. Complicity in a bad faith election… 2021-01-10 21:36:10 @RichardGrenell @simonwiesenthal "Never again" doesn't just mean never again *to us*. Complicity in a bad faith election misinformation campaign to undermine democracy and install an authoritarian should not be rewarded with an honor in the memory of the Holocaust. Only takes an ounce of empathy to connect dots. 2021-01-10 20:58:41 @dipamc77 It is the go-to place 2021-01-10 20:45:52 @dipamc77 All platforms have rules and conditions and routinely take down violent content. Perhaps instead ask yourself why an exception has been made here for so long. 2021-01-10 18:42:36 @dipamc77 I am speaking about Twitter, not on its behalf. Yes, this tweet is reprehensible and should be dealt with. 2021-01-10 17:46:27 @tallinzen We are on the same page, I am referring to the Parler ban. The conduct that I enumerate is precisely what Parler allows and that Twitter does not. 2021-01-10 17:37:45 Twitter doesn't take down "conservative" content. They just don't allow people to plot armed invasions of federal buildings, murders of federal officials, and the violent overthrow of our duly elected leaders. One day, we need a serious conversation abt censorship 2021-01-10 04:26:25 RT @smolix: Hi @Cloudflare - @awscloud , @Apple AppStore and @GooglePlay have done the right thing to to cut ties with #parler. Please stop… 2021-01-10 04:13:39 RT @smolix: Hi @ProtonMail - please stop providing e-mail support to domestic terrorists, conspiracy theorists and radicals. Please conside… 2021-01-10 04:13:31 RT @smolix: Hi @ImprovMx please stop supporting domestic terrorists such as thedonald dot win by providing them with mail services. https:/… 2021-01-09 23:34:32 RT @AMZNforClimate: Enough is enough. Amazon hosts Parler on @awscloud. As Amazon workers, we demand Amazon deny Parler services until it… 2021-01-09 18:47:24 @veiledgazelle @CarnegieMellon Yes, I know. My reply was too (I am a Jew). 2021-01-09 18:23:45 @veiledgazelle @CarnegieMellon Yes. 2021-01-09 18:16:53 Day 1, @JoeBiden must rescind @RichardGrenell's appointment to the US Holocaust Council. He is a key propagandist spreading "the big lie" a la Goebbels. It's a grievous insult to Jews to invoke the Holocaust's memory to confer credibility on such a man. 2021-01-09 00:02:21 @Justin_Halford_ @jack Twitter suspends accounts every day. If you want to pick a time to bang the censorship bell, why not refocus your efforts on a case that actually deserves a reasonable debate? 2021-01-08 23:31:04 Better late than never. Thank you @jack for doing the right thing here. https://t.co/uKgc2s1Yq0 2021-01-08 17:13:56 Reply with probability Trump plays golf this weekend. 2021-01-08 12:10:22 If you weren’t sure we were still trapped in 2020: “Jamiroquai lead singer Jay Kay denies he was at Capitol riots after fans mistake him for man in horned helmet” https://t.co/FQNdU778PK 2021-01-08 05:32:02 @Miles_Brundage herbie says "sup?" https://t.co/TNNUItqd1p 2021-01-08 03:02:31 And now, in a twist that nobody saw coming... https://t.co/AsPDxLe2U4 2021-01-08 03:00:21 RT @dkaushik96: @zacharylipton @SecElaineChao @BetsyDeVosED i’m surprised there are things other than student loans that betsy devos finds… 2021-01-08 02:32:16 Make no mistake that @SecElaineChao and @BetsyDeVosED are committing a great act of cowardice. If they really understood what happened, they would stay and invoke the 25th amendment. Instead they are getting as far away as possible to absolve themselves of responsibility. 2021-01-07 19:35:23 RT @zacharylipton: Christ on a bike, it's bad enough that @CarnegieMellon's @CMUIPS hired propagandist @RichardGrenell, why on Earth are we… 2021-01-07 19:34:56 RT @dave_andersen: Hey, @CarnegieMellon - you're my employer. I've worked here for 15 years. I even donate to you. Yesterday - YESTER-FUCK… 2021-01-07 19:29:00 RT @RMac18: Right on cue, Zuckerberg announces changes to the Q& 2021-01-07 19:27:06 RT @fhuszar: For the avoidance of doubt, I wanted to say that I agree with those who say Twitter is responsible for how bad things have got… 2021-01-07 18:47:23 @iandanforth @fhuszar I'm saying the moment they started promoting content, the entered the curation game. They cannot wash their hands and pretend not to be responsible for what propagates on the platform. And thus if they are curating, they are responsible for curating responsibly. 2021-01-07 18:37:25 RT @julie17usc: So much this. Corollary: Regulating data protection as data protection and algorithmic optimization as algorithmic optimiza… 2021-01-07 18:30:04 @fhuszar Even absent curation there are lines that should not be crossed. A different principle is at play there. Free speech does not protect hate speech and it does not protect conspiracies to commit crimes. 2021-01-07 18:19:36 @fhuszar Yes. Of course it might not be as successful. But if they want the upside of curation, they need to assume the responsibility for what they are curating. 2021-01-07 18:07:22 Free speech is not a right to *promoted* speech. The moment Facebook and Twitter got into the curation game (news feeds, trending topics, engagement optimization), this unequivocally became their problem. 2021-01-07 06:30:05 @MGGOWLIFE @CarnegieMellon @CMUIPS @RichardGrenell You must not do much thinking, generally. 2021-01-07 06:13:59 @Yoder_Esqq @RichardGrenell @CarnegieMellon @CMUIPS The topic here isn't policy. I'm happy to debate policy & 2021-01-07 06:02:26 @RichardGrenell @CarnegieMellon @CMUIPS Let's not misapply the term "conservative" to describe attempts at an unconstitutional (and anti-factual) subversion of a free and fair election. 2021-01-07 05:49:16 Christ on a bike, it's bad enough that @CarnegieMellon's @CMUIPS hired propagandist @RichardGrenell, why on Earth are we putting out press releases for the award he received from Trump on the same day as his conspiracies contributed to sedition. Shameful. https://t.co/dwvyeSXSGZ 2021-01-07 05:27:28 RT @TimAlberta: Ahem. So, Michael. Will the RNC offer an apology for its role in fomenting that domestic terrorism? For hosting Sidney Powe… 2021-01-07 05:23:36 @DrLukeOR @willieboag @kat_heller @Twitter @Facebook @timnitGebru Oh don’t be silly, I rock the boat all the fucking time, especially concerning the hiring of Grenell. Don't know what award this is, hadn’t heard. But I certainly wouldn’t hold my tongue and never do. Argue in good faith Luke. 2021-01-07 05:20:53 @brunoborges @Twitter @Facebook I am a computer scientist with a following of 90% computer science practitioners and other entrepreneurs/scientists speaking to the people in my orbit. Use some of that beautiful brain God gave you and perform a more reasonable inference. 2021-01-07 05:19:18 @brunoborges @Twitter @Facebook No I am not, that is a ridiculous interpretation and you do not believe that’s what I mean in good faith. 2021-01-07 04:07:20 Just kidding, happy 2020 everybody! https://t.co/LnOICnxqTZ 2021-01-07 03:05:10 @jeffbigham History will not remember @HawleyMO kindly. 2021-01-07 03:01:00 This is amazing. Pence is smiling and almost giggling, I think he's truly happy to hear Lindsay Graham forcefully declare that Joe Biden and Kamala Harris won. Takes heat off of him, and I think he is eager to get the F away from this dumpster fire with some reputation intact. 2021-01-07 02:40:49 RT @andrewmarantz: My honest, no-snark guess is that Trump doesn't get why some of his allies in the Senate are now starting to (tepidly) b… 2021-01-07 02:36:27 @mark_riedl Hey man, please, wrong target. The man is walking a very careful line here and has a lot in the balance, most importantly the perceived integrity of the Georgia Senate race, he is not an idiot. 2021-01-07 02:26:01 RT @MyNameIsTooLon: @zacharylipton @ylecun lecun right now https://t.co/B9PofMUNUx 2021-01-07 02:14:58 But what happened to "information wants to be free". Are you OK with this @ylecun? Where's your statement? We need leaders, and you are the best positioned AI researcher to demand more from Zuckerberg. https://t.co/I4Re67JbXE 2021-01-07 01:28:23 @pentagoniac @jack Yes @jack is 1000x more serious than Zuck. But morality here demands more than outperforming the devil, it requires decisive action. @jack has been moving (if slowly) in good faith for a while now, but it's long past time to do more. 2021-01-07 00:34:55 @4nj4k @dave_andersen @Facebook Nuclear bombs became inevitable from the moment > 2021-01-06 23:53:53 RT @monteiro: Agreed. Time to put down the tools. https://t.co/wi3DLDSXIT 2021-01-06 23:21:40 @thetruetelmo @Facebook “Speech policing” is a naive framing of the problem and the potential solutions. 2021-01-06 22:55:50 RT @risi_kondor: this https://t.co/G7vQMjeba2 2021-01-06 22:42:48 @joftius @Facebook I called out @jack several times earlier. And surely Twitter has been doing perhaps 10% of what morality demands of them. That said, Facebook has done maybe .01%, and even that always only in a PR calculation, never out of principle. Additionally, they are just far far bigger. 2021-01-06 21:21:41 While packing their bags or riding the back seat to their holding cells, the proud boys can choke on this: https://t.co/p8F96MXW7j 2021-01-06 21:12:18 @2xminus1 @Twitter @Facebook Turn off Trump's microphone (do not exempt president from standard hate speech policies), stop promoting white supremacists in trends, take down slander and hate speech groups & 2021-01-06 21:10:13 RT @MaryLTrump: Make no mistake, the so-called leader of our country, who swore to protect and defend the Constitution, directed a horde of… 2021-01-06 21:07:06 RT @risi_kondor: Couldn't agree more. This is what matters, not churning out more damned conference papers. https://t.co/EcUec4154p 2021-01-06 20:50:57 RT @rjurney: @jack it is time for you to do something or you bear responsibility for the revolution yourself. Stand up, be the man you shou… 2021-01-06 20:37:42 Equally complicit are @Verizon, @optimum, & 2021-01-06 20:30:26 @_dsevero @jack No, turn off all of their microphones. Kill the communications channels. 2021-01-06 20:13:36 Hey Mitch (@senatemajldr), how does impeachment sound now? 2021-01-06 19:58:08 RT @WisalKhwaja: just thinking what if blacks and Muslims would have done the same, the headline would have read "100's shot in front of th… 2021-01-06 19:57:53 RT @landay: Exactly how I feel. Just like Michigan... https://t.co/dbrRvUeSw1 2021-01-06 19:45:33 So when #BLM protestors vandalized a few shop windows in outrage over racial injustice they got tear-gassed & 2021-01-06 18:45:52 New attention chapter in our "Dive into Deep Learning" interactive textbook (https://t.co/V7ZyrjS0hY)! Amazing effort by my collaborator @astonzhangAZ & 2021-01-06 12:31:04 RT @Liz_Cheney: We have sworn an oath under God to defend the Constitution. We uphold that oath at all times, not only when it is political… 2021-01-06 07:08:37 @tswaterman @staceyabrams I’m guessing @OfficialSting wants to stay as far away as possible from casting calls for *these* Harkonnens. 2021-01-06 07:05:45 If Trump is Baron Harkonnen, does that make @staceyabrams Muad'Dib? 2021-01-06 06:25:18 RT @RonaldKlain: Spitballing here, but it may be that telling voters that you intend to ignore their verdict and overturn their votes from… 2021-01-06 06:03:17 RT @JohnCarltonKing: https://t.co/yQ6Nobmt3R 2021-01-06 05:14:06 RT @TheDailyShow: Kelly Loeffler is so loyal to Trump she also lost an election in Georgia https://t.co/jRzlORuiJQ 2021-01-06 01:40:01 RT @RepPressley: Congratulations, @AlphabetWorkers! This is no small feat. I'm proud to stand in solidarity with those organizing for work… 2021-01-06 01:21:43 Georgia elections r like watching transcendent Korean revenge movies—starring @staceyabrams. 2021-01-05 20:20:04 RT @TheGradient: Dear friends @CS_UCLA: I will be giving a talk at CS 201 seminar next week on the implicit regularization of (aka knowl… 2021-01-05 19:49:12 @sayantandas_ @JacobSteinhardt Thanks for reading! I find it useful too, not out of vanity, but more because now that we identified these problems and wrote publicly about them, there's the added pressure to try hard not to be a hypocrite. 2021-01-05 14:29:48 @DeepAwais @lcastricato Understanding Machine Learning by Shai Ben-David And Shai Shalev Shwartz 2021-01-05 12:15:04 @sytelus Sure there will be. The reason is homogeneity. As the improvements in graphics year over year become less and less significant, the relative value of watching a new superhero movie vs one from say, 10 years ago continues to decline. Moreover, catalog continues to grow. 2021-01-05 12:12:50 RT @mukulrathi_: As I'm wading through papers for my Master's project, this is a timely read: some trends in ML papers _not_ to follow. Tha… 2021-01-05 06:18:09 @TheodoreGalanos There is a time that is [some date] everywhere on earth, but that means something different altogether. 2021-01-05 06:16:46 @guestiest McDonalds my be the highest grossing culinary enterprise. And Yanni is higher grossing than John Coltrane. Are we talking about good movies or profitable movies? 2021-01-05 06:01:21 @guestiest The Avengers ran put of gas before they started, X-Men peaked around 2000, Spiderman mostly unwatchable since then too (except for Spidey-verse). Full steam or hot air? 2021-01-05 05:40:52 @gideonesakoff @Deadpool Is it though? 2021-01-05 05:37:09 @MichaelDoronII Agree! Actually had Ragnarok (2017) in mind originally but realized BP was more recent 2018. 2021-01-05 05:12:46 Even as a non-snob w. a weakness for big-budget popcorn flix, the last watchable superhero movie was Black Panther. Prediction: the genre will stagnate, struggle to compete w. the catalog, and we’ll see the rapid decline of superhero dominance in mainstream cinema w/in 5 yrs. 2021-01-04 18:56:42 RT @smolix: I'm starting a new crash course on Deep Learning. Slides daily. 15 minutes or less, based on Dive into Deep Learning https://t.… 2021-01-04 16:04:29 ICLR decisions are slated for Jan 12 Anywhere on Earth. (But this is not a deadline.) Wouldn't it be more accurate to list Jan 12 *Somewhere on Earth*? 2021-01-04 15:32:03 Apparently somebody's Slack-ing off today 2021-01-04 05:33:42 @Aelkus @mraginsky @generativist Seldon saw millennia a head with clarity. Trump can’t see the end of the week. Seldon also appeared capable of reading and writing. 2021-01-04 05:29:07 @mraginsky While the Mule’s abilities were paranormal, Trump’s violate not the laws of physics but social norms previously imagined to be inviolable. 2021-01-04 05:26:54 @mraginsky An anomaly with brainwashing powers unforeseen by our founding fathers that *nearly* derailed their grand psychohistorical plan? Sounds fair. 2021-01-04 05:22:43 @mraginsky The Mule was the original Donald Trump. 2021-01-04 03:36:46 @Parisa__Rashidi I'm hearing rumors that some bad hombres are switching out inside parts of the Dominion machines that Google Scholar bought from Hugo Chavez to count the citations. 2021-01-04 03:24:44 RT @Parisa__Rashidi: I want to find my 11,780 citations. I am sure they are hiding somewhere. 2021-01-04 01:26:20 apparently there was some kind of election recently? 2021-01-03 23:48:12 @yoavgo Perhaps this means that you are aging well. 2021-01-03 20:03:47 It turns out, if we model problem X in framework Y, discarding all aspects of X that tools from Y can/do not address, then indeed what remains has deep connections to canonical problems in Y and its tools can be applied profitably. #paper 2021-01-03 19:18:09 RT @NickKristof: Incredible audio of Pres Trump bullying and threatening Georgia's secretary of state in a phone call Saturday, ordering hi… 2021-01-03 15:53:06 @jaydaronson Impossible. 2021-01-02 21:44:34 @PlzBeSensible @Peter_Wehner @TheAtlantic Only if you, the observer subscribe. The meaning here is they (academics) actually believe what they espouse, vs wielding it as a weapon in self interest, truth be damned. 2021-01-02 13:58:41 A day later, @Peter_Wehner made this very point in @TheAtlantic. Republicans today cynically embrace the perspectivism (denying objective truth) for which they have long berated the academic left. As Wehner notes, at least the scholars came by it honestly. https://t.co/SC9MVTaVrW https://t.co/7cjsSqJSBk https://t.co/WuHGB6Flb4 2021-01-01 22:08:41 @soniajoseph_ Herbie is a full mini Aussie, no poodle. Cute! Yeah its so weird/interesting how temporal spacing varies over course of training. 2021-01-01 22:01:30 One particularly strange thing, ultimately we want command -> 2021-01-01 21:56:34 There’s a lot about training a puppy that flows naturally from RL intuition (eg, the importance of immediate rewards & 2021-01-01 19:03:39 @ryandcotterell @yoavgo @chrmanning @ani_nenkova @ChrisGPotts So people should pull papers with decisions unknown? This is ridiculous, mate. Either arxiv is allowed or it is not. This dance makes no sense. 2021-01-01 18:56:02 @ryandcotterell @chrmanning @ani_nenkova @yoavgo @ChrisGPotts It's not "so quickly". If someone has the paper under review, they might wind up with a year elapsing (submit, reject, revise, resubmit) before the next window opens up. The impact is work rushed onto arxiv *earlier* and then left in an immature state for longer. It's plain bad. 2021-01-01 18:54:49 @ryandcotterell @chrmanning @ani_nenkova @yoavgo @ChrisGPotts The effect of the policy has only been to have work rushed onto arxiv a month earlier, not because it is ready but because of a ridiculous embargo date. 2021-01-01 18:46:00 @ryandcotterell @chrmanning @ani_nenkova @yoavgo @ChrisGPotts That was already very low in the first place. This policy is utterly unacceptable incoherent, creates no real protections, and results in flawed work being rushed onto arxiv and then not updated (even when such updates are available). It’s a mess and we should abandon it. 2021-01-01 18:44:19 @chrmanning @yoavgo @ChrisGPotts Just get rid of it. That would doubtless be 10x better than the current system. 2021-01-01 17:42:40 @yoavgo @ryandcotterell @_dmh @ani_nenkova @chrmanning @ChrisGPotts More seriously, they should just drop the pretense and have a permissive arxiv policy. The current approach displays the incoherence that only a committee could support. 2021-01-01 05:42:49 @khudabukhsh Who called the narc? 2001-01-01 01:01:01