mei @multiply_matrix
training foundation models, intelligence to its limit. AGI timeline 2027. use AI to solve bio & cure diseases ❤️ @MIT dropout linkedin.com/in/matrixmulti… Joined February 2021-
Tweets38
-
Followers1K
-
Following62
-
Likes150
wow - only 1 hour of training data to reconstruct images from fMRI !🧠📷❤️ @iScienceLuvr @humanscotti
wow - only 1 hour of training data to reconstruct images from fMRI !🧠📷❤️ @iScienceLuvr @humanscotti
not my examples, taken from @trishume: - moe => cs aesthetic - 2nd order optimizers => math aesthetic - scaling laws => physics aesthetic
@joshua_saxe @fchollet One on math: super-human mathematician AI by June 26. The other: 95% at the end of 2025 on the current test set of of ARC. We can work out the details on the first. I feel much less confident about the second, esp. given the current challanges with evaluation on proprietary…
Claude 3 being lit is a big W for synthetic data. All the rumors I've dropped about Anthropic synthetic data on the blog are obviously confirmed in their thorough technical report. (for real tho, huge congrats, awesome model so far)
essential is all u need
Ilya Sutskever personally inspired me to enter the field of AI, he's moral, empathizes deeply, and is the bravest human I know. Before the thousands of wannabe genAI startups existed.... long ago, AI was unpopular and underfunded, so CS researchers poured attention into other…
Real progress in AI can only be achieved through a very intense work ethic
@VikramVoleti and I integrated and optimized a "2D⇒3D" feature in ThreeStudio -- an open-source text-to-3D and 2D-to-3D project. Here are recent results! (more results at the end of the thread)
In December, we discussed Med-PaLM, at that time a SOTA medical LLM that achieved a 67.6% score on the USMLE MedQA evaluation (passing is 60%). Today, we're describing Med-PaLM2, which improves on this by +18% with a score of 85.4% ("expert performance")! Kudos to all involved!
In December, we discussed Med-PaLM, at that time a SOTA medical LLM that achieved a 67.6% score on the USMLE MedQA evaluation (passing is 60%). Today, we're describing Med-PaLM2, which improves on this by +18% with a score of 85.4% ("expert performance")! Kudos to all involved!
Our paper got accepted at #CVPR2023! (w/ @yu_takagi) We modeled the relationship between human brain activity (early/semantic areas) and Stable Diffusion's latent representations and decoded perceptual contents from brain activity ("brain2image"). sites.google.com/view/stabledif…
Announcing the launch of the MedARC! MedARC is a novel, open, & collaborative approach to medical AI research. It was created to develop large-scale AI models for medicine & build interdisciplinary teams to address clinical needs. (1/10) medarc.ai/blog/announcem…
my colleague Lucidrains built open-source RLHF on PaLM (similar to chatGPT). now someone with a supercomputer needs to train it 😉💪team @StabilityAI github.com/lucidrains/PaL…
Phil Wang aka `lucidrains` wrote yet another awesome PyTorch implementation, this time of Enformer, Deepmind's model for predicting gene expression. Thanks to @NielsRogge we can now load the pretrained models from the huggingface hub directly from the Enformer codebase 🔥
award for outstanding paper won by @laion_ai at @NeurIPSConf ! team @StabilityAI 💪 laion.ai/blog/laion-5b/
by retraining Alphafold, @open_fold can predict protein-ligand structures, and understand unseen regions of fold space 🧬 supported by @StabilityAI biorxiv.org/content/10.110…
by retraining Alphafold, @open_fold can predict protein-ligand structures, and understand unseen regions of fold space 🧬 supported by @StabilityAI biorxiv.org/content/10.110…
🎉Introducing RoentGen, a generative vision-language foundation model based on #StableDiffusion, fine-tuned on a large chest x-ray and radiology report dataset, and controllable through text prompts! @PierreChambon6 @Dr_ASChaudhari @curtlanglotz 🧵#Radiology #AI #StanfordAIMI
Today we're releasing the structures of over 600M metagenomic proteins—the least understood proteins on earth. A view into the dark matter of the protein universe. Explore the ESM Metagenomic Atlas here: esmatlas.com
zero to 1 million users in 27 days. on track to generate 1 billion images per day. stability.ai is the fastest growing AI company in history.
zero to 1 million users in 27 days. on track to generate 1 billion images per day. stability.ai is the fastest growing AI company in history.
AI research + bio @ludo_antonov @volokuleshov @YevgenChebotar @yaroslavvb @ibab_ml @_MichaelAntonov
Joscha Bach @Plinz
129K Followers 755 Following FOLLOWS YOU. Artificial Intelligence, Cognitive Architectures, Computation. The goal is integrity, not conformity. https://t.co/rFUNzdYXuKEric Jang @ericjang11
69K Followers 3K Following physical AGI at 1X. Author of "AI is Good for You" https://t.co/eFg4WXhg0pRobert Scoble @Scobleizer
504K Followers 68K Following Follow me on my new podcast with AI startups, Unaligned. Tech industry color commentator since 1993. Author/Blogger. Former strategist @Microsoft.Irwan Bello @IrwanBello
6K Followers 2K Following Supercomputers & Friends AGI research & products ex @OpenAI, founding team @character_aiAndrew Van Beek @AndrewVanBeek1
78 Followers 178 Following Developer turned founder: https://t.co/KFOnPOykFm…Zhengyao Jiang @zhengyaojiang
1K Followers 268 Following Cofounder and CTO @WecoAI, building AutoML Agents. Final year PhD student at UCL @UCL_DARK @ai_ucl. (Zheng=j-uhng, j as in job; yao=y-aoww)Taylor @Taylor37148900
10 Followers 93 FollowingSteve Zhong @stevestevezhong
256 Followers 984 Following Startup founder. Previously @ycombinator W23. Mentoring startups @techstars and @AlchemistAcc. CS+Math grad from @harveymudd.Kinjal Nandy @itsKinjalNandy
1K Followers 2K Following i wanna talk to chatGPT like i talk to a humanHamza Mahmood @admiral_caramel
34 Followers 194 Following Nanofab | Hardware | Bio // mover of current // ECE @uoftengineeringDiane Green @DianeGreen91775
13 Followers 747 FollowingGerry Fernando Patia @GerryFPatia
1 Followers 79 FollowingJames Hill-Khurana @jtvhk
4K Followers 5K Following Eclectic. Curious about machine learning, tech history, design, HCI and biomimicry. Prev, philosophy + cogsci, @uwaterloo.Anna Monaco @annarmonaco
2K Followers 821 Following building @tryparadigm, backed by @ycombinator | prev @penn cs, swe & pm @google @microsoftBucky Moore @buckymoore
5K Followers 1K Following Partner @KleinerPerkins to founders advancing the frontiers of computingBrandon White @bwhite5290
966 Followers 2K Following Best Seven Foot ML Engineer in the World! AI + Bio + Design.Kevin Peterson @kevinmpeterson1
104 Followers 282 Following Rocket ships, autonomous cars, and bikesEric Frank @IHaveSweaters
853 Followers 3K Following research scientist, formerly founding team @ uber ai labs & @ml_collective & geometric intelligence, ex toy designer @KiteandRocket 🌈✡️Fu*#ingTwoFaced @BadClutchwork
51 Followers 574 Following I can do whatever I want when I want...I love being Catfished especially by blonde bitches with big tits. So lie to me I like it.Louis de Benoist @louisdebenoist
65 Followers 66 Following CEO of Cube - building AI agents for logistics operations. Previously at @Cambridge_Uni, @HECParis, and @PolytechniquePablo @PabloAIChat
31 Followers 201 Following Pablo is a voice-first application that bridges the gap between complex AI services and the everyday user.stechert @stechert
867 Followers 5K Following Wrote search @Splunk, planner/schedulers/vision/datamining @NASAJPL, robotics @UCLA. Opinions mine.e.bstan @ebstan
6 Followers 1K Followingclaire night skies �.. @clairebookworm1
2K Followers 993 Following professional yapper, (increasingly) amateur sleeper 🥐🥐 || @hackclub @angelhacks_la @mcgovernmit @contrary @nucleatedojoJean-Baptiste Mouret @jb_mouret
2K Followers 3K Following AI researcher. Robotics. Data-efficient learning. Evolution. Neuro-evolution. Senior Researcher (Dir. de recherche) @ Inria / ERC ResiBots.pointed_max @pointed_max
1K Followers 970 Following a pointed set is a set endowed with a distinguished element, called the base point.Chris Albon @chrisalbon
86K Followers 2K Following Director of Machine Learning at the Wikimedia Foundation. We host Wikipedia.Nick Donahue @PrimalNick
3K Followers 968 Following Designing & Building Homes in North Carolina with @buildatmosGuillaume Lample @GuillaumeLample
37K Followers 648 Following Cofounder & Chief Scientist https://t.co/hLfvKLkFHd (@MistralAI). Working on LLMs. Ex @MetaAI | PhD @Sorbonne_Univ_ | MSc @CarnegieMellon | X11 @PolytechniqueTeknium (e/λ) @Teknium1
29K Followers 3K Following Cofounder @NousResearch, prev @StabilityAI Github: https://t.co/LZwHTUFwPq HuggingFace: https://t.co/sN2FFU8PVE Support me on Github SponsorsDavid @DavidSHolz
54K Followers 5K Following founder @midjourney, prev founder leap motion, nasa, max planckRafael Rafailov @rm_rafailov
3K Followers 638 Following Ph.D. Student at @StanfordAILab. I work on Foundation Models and Decision Making. Previously @GoogleDeepMind @UCBerkeleySacha Ichbiah @SachaIchbiah
805 Followers 3K Following Engineering and AI @Polytechnique // Physics PhD // Accelerating logistics ops with AI @getcubeai // [email protected]kun zhu @kunzhu118672
83 Followers 897 FollowingLidiya Terpel🇺🇦 @LTerpel
185 Followers 384 Following Building something new in AI; startup founder @skyworkerai. Ukrainian in SFTiger @regtled
1 Followers 46 Followingpranav ⠕ @_pranavnt
1K Followers 539 Following making things, sipping cold brew, and blasting taylor swift • @uwcse @atlasfellow @JupeInc @kobra_devJoscha Bach @Plinz
129K Followers 755 Following FOLLOWS YOU. Artificial Intelligence, Cognitive Architectures, Computation. The goal is integrity, not conformity. https://t.co/rFUNzdYXuKEric Jang @ericjang11
69K Followers 3K Following physical AGI at 1X. Author of "AI is Good for You" https://t.co/eFg4WXhg0pJeff Dean (@🏡) @JeffDean
297K Followers 6K Following Chief Scientist, Google DeepMind and Google Research. Co-designer/implementor of things like @TensorFlow, MapReduce, Bigtable, Spanner, Gemini .. (he/him)Andrew Ng @AndrewYNg
1.0M Followers 914 Following Co-Founder of Coursera; Stanford CS adjunct faculty. Former head of Baidu AI Group/Google Brain. #ai #machinelearning, #deeplearning #MOOCsIrwan Bello @IrwanBello
6K Followers 2K Following Supercomputers & Friends AGI research & products ex @OpenAI, founding team @character_aiZhengyao Jiang @zhengyaojiang
1K Followers 268 Following Cofounder and CTO @WecoAI, building AutoML Agents. Final year PhD student at UCL @UCL_DARK @ai_ucl. (Zheng=j-uhng, j as in job; yao=y-aoww)Stefano Ermon @StefanoErmon
13K Followers 362 Following Associate Professor of #computerscience @Stanford #AI #ML #SustainabilityChris Albon @chrisalbon
86K Followers 2K Following Director of Machine Learning at the Wikimedia Foundation. We host Wikipedia.Guillaume Lample @GuillaumeLample
37K Followers 648 Following Cofounder & Chief Scientist https://t.co/hLfvKLkFHd (@MistralAI). Working on LLMs. Ex @MetaAI | PhD @Sorbonne_Univ_ | MSc @CarnegieMellon | X11 @PolytechniqueTeknium (e/λ) @Teknium1
29K Followers 3K Following Cofounder @NousResearch, prev @StabilityAI Github: https://t.co/LZwHTUFwPq HuggingFace: https://t.co/sN2FFU8PVE Support me on Github SponsorsNick Donahue @PrimalNick
3K Followers 968 Following Designing & Building Homes in North Carolina with @buildatmosDavid @DavidSHolz
54K Followers 5K Following founder @midjourney, prev founder leap motion, nasa, max planckRafael Rafailov @rm_rafailov
3K Followers 638 Following Ph.D. Student at @StanfordAILab. I work on Foundation Models and Decision Making. Previously @GoogleDeepMind @UCBerkeleyAydin Senkut @asenkut
52K Followers 2K Following Founder @Felicis | 10x Midas List | Investor @Adyen @Shopify @notionhq | lucky husband & proud fatherSynthLabs @synth_labs
12K Followers 43 Following AI Aligned with Your Vision. We’re doing cutting edge research for transparent, auditable AI alignment.Katherine Lee @katherine1ee
6K Followers 931 Following understanding ourselves and our models. senior research scientist @GoogleBrain, @genlawcenter and @CornellCIS, formerly @Princeton @[email protected]Ross @rpoo
25K Followers 1K FollowingBeff Jezos — e/acc .. @BasedBeffJezos
102K Followers 2K Following chief accelerator & founder @ e/acc // thermodynamic priest // Kardashev gradient climber // memetic warlord // building @extropic_aijustin waldron @jtwald
173K Followers 2K Following founder @gemzfun, @zynga, @playco. advisor @dapperlabs, @decentraland, @immutableKevin Hartz @kevinhartz
16K Followers 335 Following Co-Founder & General Partner A*, Chairman & Co-Founder of Eventbrite (NYSE: EB), Co-Founder and Board Member of Xoom (IPO 2013, acquired by PayPal)Sid Sijbrandij @sytses
24K Followers 625 Following Co-founder & CEO of GitLab. I love innovative projects, accessible education, all remote work, new cities, and macro economics. he/himChenlin Meng @chenlin_meng
8K Followers 833 Following Co-founder & CTO @pika_labs | ex @StanfordAILab @StanfordStella Biderman @BlancheMinerva
15K Followers 748 Following Open source LLMs and interpretability research at @BoozAllen and @AiEleuther. My employers disown my tweets. She/herniki parmar @nikiparmar09
10K Followers 780 FollowingManuel Kroiss @makro_ai
14K Followers 60 FollowingLouis Castricato @lcastricato
3K Followers 477 Following Math @uwaterloo, RLHF @BrownCSDept, Goosefluencer. x-RS @aieleuther, x-Head of LLMs @stabilityai, x-lead @CarperAI. co-founder @synth_labs. We're hiring.nathan lile @NathanThinks
2K Followers 887 Following ceo/cofounder @ https://t.co/bDd3J4Lmzf (we're hiring!) #GenerativeAI recurrent rabbit hole victim. swims in data lakes & pools. nothing great is easy.David Deutsch @DavidDeutschOxf
119K Followers 63 Following Physicist. Author of The Fabric of Reality and The Beginning of Infinityhazyresearch @HazyResearch
7K Followers 1K Following A research group in @StanfordAILab working on the foundations of machine learning & systems. https://t.co/JHK58TDorG Ostensibly supervised by Chris RéRobin Rombach @robrombach
6K Followers 398 Following Generative enthusiast and long-term PhD Student @LMU_Muenchen. Author of VQGAN, Latent Diffusion, Stable Diffusion.Eric Lander @eric_lander
62K Followers 81 Following Geneticist, mathematician and teacher. Director, Broad Institute of MIT and Harvard. OSTP Director-nominee and Presidential Science Advisor for @JoeBidenTanishq Mathew Abraha.. @iScienceLuvr
54K Followers 1K Following PhD at 19 | Founder and CEO at @MedARC_AI | Research Director at @StabilityAI | @kaggle Notebooks GM | Biomed. engineer @ 14 | TEDx talk➡https://t.co/xPxwKTq6QbShelby Heinecke @shelbyh_ai
348 Followers 831 Following AI Leader, Researcher, & Engineer. AI Research Manager @SFResearch. Math PhD @thisisUIC, Math BS @MIT. On a mission! 🚀Nikhil Naik @nikhil_ai
2K Followers 1K Following Generative AI Research and Applications | Worked at: @MIT, @Harvard, @sfresearch, @GoogleDavid Benhaim @dbenny12
111 Followers 172 Following Founder at BackflipAI (Stealth). Cofounder and former CTO @ Markforged (NYSE:$MKFG)Diogo Mónica @diogomonica
16K Followers 526 Following Co-Founder and President @Anchorage, ex-@Docker, ex-@Square Infinite knowledge; Infinite progress.Edward Saatchi @SaatchiEdward
2K Followers 2K Following 2x Primetime Emmy winner, AI + Movies + Games! @fablesimulation @theculturedaoEd Boyden @eboyden3
27K Followers 1K Following Y. Eva Tan Professor in Neurotechnology, MIT. Investigator, HHMI. Leader, Synthetic Neurobiology Group. Inventor. Entrepreneur.Karl Deisseroth @KarlDeisseroth
27K Followers 265 Following https://t.co/NCSpb81NTR https://t.co/vxqoQhBORdRobert Nishihara @robertnishihara
6K Followers 623 Following Co-founder and CEO @anyscalecompute. Co-creator of @raydistributed. Previously PhD ML at Berkeley.Marc Bhargava @marcbhargava
3K Followers 2K FollowingLaura Florescu @florescl
680 Followers 2K Following Something new. @mosaicML @sambanovaAI @GroqInc, @microsoft, PhD at @nyuniversity.Igor Babuschkin @ibab
44K Followers 685 Following Maybe the real AGI was the friends we made along the way. @xAIBrandon Sim @brandonksim
164 Followers 450 Following CEO @astranahealth (NASDAQ:ASTH), building tech & ops infrastructure to empower providers | stats, physics, CS @harvardI hope Elon buys TikTok and simply deletes it.
This was an awesome project - we teach models to follow constitutional principles with self-supervision (no labels). We also show that a weak model can generate principles for a stronger one, which self-aligns (SUPERALIGNENT!) and can beat the instruction-tuned (RLHF-ed) model!
Constitutional AI showed LMs can learn to follow constitutions by labeling their own outputs. But why can't we just tell a base model the principles of desired behavior and rely on it to act appropriately? Introducing SAMI: Self-Supervised Alignment with Mutual Information!
Every generation of model is building upon the data cleaned up by the last generation of models. It’s the recursive self improvement era. You train a model with refined data until it’s good enough to help you refine your data.
At this point I feel like we understand pretty well what's going on with LLMs: - Outputs are roughly equivalent to kernel smoothing over positional embeddings (arxiv.org/pdf/1908.11775…) - The learned computation model is *probably* bounded by RASP-L (arxiv.org/pdf/2310.16028…) -…
I saw this challenge aimoprize.com to develop an AI that can win a gold medal at the IMO. I competed at that level a couple of times (only silver medals though) and have been working on RL and LLMs for a bit. Here is my thoughts on what the challenges are: 1/N
From the LLaMa 3 blogpost - they use a combination of rejection sampling, DPO and PPO for post-training. Really interested to know what tasks/parts of the process each algorithms benefits the most.
Cool plot from @stanfordnlp's annual report on foundation models, showing who is training models. @AiEleuther's 6 models becomes even more impressive when you consider that that's about 0.5 models/staff member! We are continuing to push the frontier of impact-per-dollar.
The AI Index editors chose “the most notable model releases of 2023”. 9 of the 15 were Large Language Models. 3 more involved language: text to image and speech models. 2 image models and a watermarking model brought up the rear. aiindex.stanford.edu/report/ #NLProc #BiasedTakes
As the natural world's human data becomes increasingly exhausted through LLM training, we believe that: the data carefully created by AI and the model step-by-step supervised by AI will be the sole path towards more powerful AI. Thus, we built a Fully AI powered Synthetic…
New preprint is out on interplay between DPO and verbosity. Some of the first feedback we got on DPO was that training on LARGE scale the model becomes increasingly verbose until it diverges. Verbosity effects have also been observed in the OS community. Credit to @peterjliu
jobs.synthlabs.ai you're looking at two people working on sick open source research. work with us.
Multi-turn interactive RL should be a bigger focus. Current methods are not well-suited for this - i.e. PPO can't train with user in the loop generally and offline Q-learning still does not work at scale. It's interesting to see more work in that direction.
When prompting language models to complete a task, users often leave important things unsaid. Can language models teach themselves to ask clarifying questions? In STaR-GATE, we explore LMs' ability to self-improve by rewarding the model for generating useful questions!
Really excited about this work! We figured out how to apply DPO (Direct Preference Optimization) to diffusion models and got big improvements in image quality
Excited to announce DPO has gone multi-modal! New paper out on RLHF for text-to-image diffusion models! We obtain large-scale state of the art results with 70% win rates against Stable Diffusion XL on human evals! Deep dive below 🧵
🧠👁️Our MindEye2 preprint is out! We reconstruct seen images from fMRI brain activity using only 1 hour of training data. This is possible by first pretraining a shared-subject model using other people's data, then fine-tuning on a held-out subject with only 1 hour of data.🧵
We have now hit 330 *million* cumulative downloads since our first model was uploaded @huggingface 18 months ago. Just overtook @Google (321m all time), on our way to 1 billion thanks to the great efforts of the team @StabilityAI Some announcements to come, big weeks ahead..
Stability AI is the leader in open AI models. Open models trained on our cluster are the most downloaded globally (below @huggingface) & with what we have coming up by us & ecosystem it’s going to be a fun year ahead Open ftw Super proud of team, onwards 🚀 🌟
I'm excited to share our latest fMRI-to-image paper 🥳 MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Data! 🧠👁️ arXiv: arxiv.org/abs/2403.11207 project page: medarc.ai/mindeye2
There’s no such thing as an actual AI company. Just guys with software sales teams competing to buy Nvidia GPUs.
There's no such thing as an AI startup, just guys with models competing to be absorbed by one of the three actual AI companies 😇
Claude 3 Opus just reinvented this quantum algorithm from scratch in just 2 prompts. The paper is not on the internet yet. 🔥🤯
Fun little paper to appear tonight on the arXiv. How to do Hamiltonian Monte Carlo on digital Quantum Computers. As physics-based probabilistic ML accelerators are on the horizon, important to test how QC's could try to compete. Best way to predict future is to invent it.🙂
Our goal is to develop foundation models for robotics 🤖: models that can ultimately control any robot for any task This is really hard. But we have some of the core ingredients: - cross-platform robot policies - transfer from VLMs, LLMs - dextrous skills with imitation learning
Introducing Orca-Math, our Mistral-7B offshoot excelling in math word problems! 🧮🐳 - Impressive 86.81% score on GSM8k - Surpasses models 10x larger or with 10x more training data - No code, verifiers, or ensembling tricks needed
In the first technical RLHF interview I've hosted, with @lcastricato of @synth_labs (+ @AiEleuther), we cover maybe every topic: interconnects.ai/p/rlhf-intervi… DPO, PPO, REINFORCE, KTO, long-context, multi-modal, video vs image, evaluation, license terms, Carper, TRLX, data…