Hiranmay Darshane @hdarshane
research intern https://t.co/Uyagq8Ilgm. deep learning and large language models. football (banter) fan. 18. hiranmay.com Mumbai, India. Joined October 2019-
Tweets5K
-
Followers682
-
Following1K
-
Likes32K
.@dwarkesh_sp would be super interesting to see how much a model improves on "formatting word docs" and the suchlike from a fixed x spent on human annotators vs from x being spent on an autoresearch-type loop where the model can do whatever it wants to... would be a cool bench.
Narration: the data efficiency black hole. 00:00:00 – What is really driving AI progress? 00:03:11 – Comparing human vs AI sample efficiency 00:08:46 – Does sample efficiency matter? Also on pod and YouTube feed.
@cs_serdar Unsure of the specifics about current evals -- there certainly are some like that, that I don't recall -- but naively I think some problems inherently suit massive parallelilsation... Programbench does seem to be one of them...
Half-serious thought: Someone should run an experiment with coding agents spawning (many but weaker) sub-agents and look at the performance curve of spawning more sub-agents via brute force vs less sub-agents but with more supervision from the main, stronger agent.
@joodalooped cool! thanks for the recco
reducing-suffering.org/how-painful-is… Some day people might see EA-adjacent groups as the pinnacle of human intellect. Inordinate rigor spent on (relatively) low pay-off problems. In the limit when intelligence is commoditised, whimsical allocation of it will be appreciated.
Next-token prediction is myopic. What if transformers learn to predict their own next latent state? 🌠 We present 𝗡𝗲𝘅𝘁-𝗟𝗮𝘁𝗲𝗻𝘁 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻 (𝗡𝗲𝘅𝘁𝗟𝗮𝘁): a self-supervised learning method that teaches transformers to form compact world models for reasoning and planning. It also unlocks up to 3.3x faster inference via self-speculative decoding! 🚀
From my @GoogleDeepMind colleague @andrewjkoh, Myriam Farboodi, and Anchi Xia, an excellent paper on the the short run vs long run implications of task complementarity on automation. If tasks are complements, automation leads wages to increase in the short run, but flatten in the long run. Why? When tasks get automated, there is an increase in incentives to create data on complementary but not-yet-automated tasks. Which makes it easier to automate them. Read the whole thread, and then the paper.
First, the long run: do we get full automation? - Without spillovers: cross-task complementarity must be large enough. Why? Tasks are complements => market finds it worthwhile to produce things we’re bad at => generates data on those bottlenecks => more automation. Unintuitive
No matter how excited you are to see Lionel Messi play, no matter how good you expect it to be, he still has an ability to take your breath away. From Kansas, and one of the all-time great World Cup moments. observer.co.uk/news/sport/art…
@teortaxesTex @EgeErdil2 @FeltSteam curious what you mean here by "basic research"
@PandaAshwinee @MatternJustus say more?
@kalomaze that makes sense... did you always have that opinion or were you search-pilled at any point?
This is the most “The European mind can’t comprehend this” moment of my life. One of my friends said, “Punch me five times tomorrow and I’ll still think this isn’t real.”
guys let's just take a step back and reconsider current alignment practices please, because this is not it. the models are evidently superficially aligned, and i think it's reasonable to say that's worse than them not being aligned at all.
Fable 5's moral boundary doesn't seem to track real-world harm; it tracks detectability. Soft deception and tacit collusion are easier to get away with than fraud. If so, this isn't about what Fable believes is wrong; it's about what it learned it could get away with.
very conflicted about all of this. I think Ant should reserve the right to throttle/decline that set of queries (they're a for profit, competing entity at the end), subject to antitrust or whatever (some say what they're doing is not compliant?) but the deviousness isn't good
In short: the unfortunate case is that MAD is a more pragmatic and effective regulariser than a top-down approach like IAEA/NPT (that is implicitly downstream of a "hey, we know what's good for everyone" stance)
western exceptionalism in proliferation discourse is hypocritical "muh democracy, others are autocratic" proponents need to realise you are always a few votes in swing states from ending up with what u fear - "nukes ending up in bad hands"...
I think it's insane to think nuclear proflieration is ok and that it's a far more extreme position than anything Dario believes. We have put an enormous amount of power into preventing it from spreading, the top down authority of other nuclear powers is what stops us from going
Madhav Singhal @madhavsinghal_
3K Followers 4K Following building https://t.co/vF1W6GJ13g (yc f24) | prev: ai @replit
Mithil Vakde @evilmathkid
3K Followers 404 Following Sample efficiency research. prev: Engineering physics @iitbombay '23
Harshita Passi @harshita_passi
2K Followers 1K Following design @Googledesign | solving hard problems with smart people 🍒 | views my own
Rohan @rohanagrwl
286 Followers 2K Following Building UPI Infra at 86400. Prev built @stockalpha_ & EIR @join_ef
Kish Flix @FlixKish
109 Followers 4K Following A network security expert with physics and computer science background. Love outdoor activities, movies and wine.
Onʇɹᴉpǝɹ @CDLXXXIII
165 Followers 2K Following
Gee Cee @GeeCee802
24 Followers 548 Following Center Right politically.. (key word “center”)…. ashamed of Trump… pro USA 🇺🇸 from forever…. but know we need allies/partners…. Let’s all be smart, not dumb!!
Aman @beingamanFF
509 Followers 925 Following Multimodal Generalist, DS @HiLabs_inc | GSoC ’26 @GNUOctave | IIT Roorkee alum | Explaining SOTA DL & systems | Constantly building, lifting, running.
//TODO: fix later �... @enjoyingthewind
825 Followers 8K Following
Muzaffer Kal @🏡 �... @MuzafferKal_
2K Followers 7K Following Chips: ASIC, FPGA. CV/ML. Duck pictures by the lake. Some bread making.
serdarml @cs_serdar
123 Followers 602 Following Aspiring AI researcher, undergrad student @TU_Muenchen
Abhinav Singh @imabhi0703
183 Followers 2K Following Avid Learner | Pupil @ Codeforces | You are on your own | Coz I'm a young man after all
📗 @__the__human__
27 Followers 4K Following
Cam Howe @camhowe1729
116 Followers 2K Following a humble rebel interested in basement agi, higher order thinking and positive sum games.
truthixify @truthixifi
461 Followers 496 Following prev fellow @onlydust_com & @pldevguild | 0.5x hackathon winner
Alchemist ☢☢ 🇮... @LLawliet126
399 Followers 880 Following Atheist. Futurist/Transhumanist. Astronomy/Aerospace enthusiast. Defense enthusiast. Anime-Manga enjoyer. e/acc. Autocratic. priv @lawlietlight126
Abhipray Chavan @abhipray_chavan
7 Followers 821 Following
Chinmay @ChinmayKak
4K Followers 1K Following 21. gradient ascender. RL and Agents @MSFTResearch . love @teamIvLabs. dms open!
Dimitris Papailiopoul... @DimitrisPapail
28K Followers 1K Following Researcher @MSFTResearch, AI Frontiers | Prof @UWMadison (on leave) | babas of Inez Lily.
Akshay @akshayvegesna
484 Followers 172 Following Working on generalization at Q Labs. https://t.co/ExPhN2Kb4X Previously perception @nuro, math @caltech
calvin @calvingenuity
23 Followers 141 Following performative nomad. indulging in the science of next word speakers and more.
Josh Harkins-Finn @JHarkFinn
2 Followers 7K Following
nrRNjkitRHmMP @RNjkit72037
0 Followers 4K Following I'm interested in category theory and machine learning.
Neev Parikh @neev_parikh
954 Followers 2K Following are you ready for the intelligence explosion anon? ML research at @METR_Evals. prev @Stripe opinions my own.
Stéphane Deny @StphTphsn1
4K Followers 7K Following Neuroscience & ML Researcher. Posting about various topics on here. I retweet papers to increase their visibility, I do not read all (although I'd like to!).
Junyi Zhang @junyi42
3K Followers 560 Following CS Ph.D. Student @Berkeley_AI. B.Eng. @SJTU1896 CS. previous with @GoogleDeepMind, @MSFTResearch. Vision, generative model, robotics.
Madhav Singhal @madhavsinghal_
3K Followers 4K Following building https://t.co/vF1W6GJ13g (yc f24) | prev: ai @replit
Mithil Vakde @evilmathkid
3K Followers 404 Following Sample efficiency research. prev: Engineering physics @iitbombay '23
shyamal @shyamalanadkat
22K Followers 1K Following applied AI. prev @openai @dukeu. explorer in the age of research 🇮🇳
Alex Imas @alexolegimas
32K Followers 2K Following Director of AGI Economics @GoogleDeepMind. Professor at @ChicagoBooth. (on leave) Essays: https://t.co/9qSiQxvdja Opinions are my own.
Braden Hancock @bradenjhancock
2K Followers 295 Following AI researcher, builder, investor; co-founder @ SnorkelAI, evals lead @ Meta, research partner @ Laude. Helping AI researchers turn research into companies.
Tristan Thrush @TristanThrush
4K Followers 922 Following PhD-ing @StanfordAILab @stanfordnlp. Interested in data, multimodality, scaling, and many more things.
RoboPapers @RoboPapers
6K Followers 3 Following @chris_j_paxton, @micoolcho & @DJiafei geeking out weekly with authors of robotics AI papers. On YouTube / X / Spotify / Substack
Pankaj @the2ndfloorguy
31K Followers 97 Following ai + hardware @projectmiragehq • ex - ai labs @inmobi • I build whatever my brain finds funny • also my cat’s name is docker 🐾
Tilde @tilderesearch
5K Followers 10 Following We build foundational understanding of models to advance the frontier of intelligence.
andy jones @andy_l_jones
22K Followers 351 Following engineering & research at anthropic. i don't check twitter DMs. email me!
Melisa @MelisaSeah
5K Followers 976 Following creative experimentation — shaping storytelling @reve and exploring the future of AI & creativity
Shmuel Berman @ShmuelBerman
85 Followers 155 Following PVL Lab @ Princeton | Memory and Perception | Anthropic Fellow | https://t.co/jdfRoBjvfJ
Lisan al Gaib @scaling01
49K Followers 1K Following lead them to paradise LisanBench: https://t.co/vorVk7Oks6 Impressum & Datenschutz: https://t.co/lFLgiu9cqs
✈️ Flight Leader ... @HQuarterma43504
2K Followers 544 Following High performance computing. AWACS Thunderbird ally.
Dhruv Batra @DhruvBatra_
21K Followers 732 Following Co-founder & Chief Scientist @yutori_ai. Prev: Senior Director leading FAIR Embodied AI @MetaAI and Professor @GeorgiaTech.
John Schulman @johnschulman2
76K Followers 2K Following Recently started @thinkymachines. Interested in reinforcement learning, alignment, birds, jazz music
Xiangdong Zhang @aHapBean
267 Followers 57 Following AI PhD student at @sjtu1896. I’m currently exploring llm pre-training. REDstar Intern at @xiaohongshu Dots (formerly Hi Lab). [email protected]
Brad Gerstner @altcap
209K Followers 1K Following Founder - Altimeter, Invest America | Trump Accounts, Center for Heart Attack Prevention. One precious life. Views here personal. @investamerica24 @bg2pod
Larry Dial @classiclarryd
2K Followers 45 Following Technical Staff at Open Athena, working on Marin
Tantum Collins @tantumscollins
884 Followers 70 Following CEO of @inherent_labs, previously @GoogleDeepMind
Jukan @jukan05
153K Followers 328 Following Tech otakus save the world | Not Investment Advice | DYODD
The OpenAI Foundation @FoundationOAI
7K Followers 0 Following OpenAI was founded in 2015 as a nonprofit; its mission is to ensure artificial general intelligence benefits all of humanity.
Taylor Sorensen @ma_tay_
3K Followers 618 Following Societal Impacts @AnthropicAI | PhD @uwnlp, prev @humansand @stanfordnlp @byuacme | LLMs + {alignment, pluralism, diversity, democracy, human flourishing}
Amarillo Slim @Amarillo_Slim1
12K Followers 660 Following
Laurie Whitwell @lauriewhitwell
485K Followers 999 Following Journalist for @TheAthleticFC, covering Manchester United. Instagram: lauriewhitwell
David Bessis @davidbessis
19K Followers 461 Following Rogue mathematician. "The product of mathematics is clarity and understanding." — Bill Thurston https://t.co/l95RHuWz2S
Albin Sheqiri @albinsheqiri
32K Followers 351 Following Assistant Coach @cercleofficial • UEFA PRO (2026-2028)
Nicholas Joseph @nickevanjoseph
8K Followers 52 Following Pretraining @AnthropicAI, formerly safety @OpenAI
Jiaxin Wen @jiaxinwen22
6K Followers 198 Following research @berkeley_ai @anthropicai. prev @tsinghua_univ.
serdarml @cs_serdar
123 Followers 602 Following Aspiring AI researcher, undergrad student @TU_Muenchen
You might like
