-
Tweets42
-
Followers360
-
Following701
-
Likes536
🧵Can we understand vision language models by interpreting linear directions in their latents? Yes! In our new paper, Line of Sight, we use probing, steering, and SAEs as useful tools to interpret image representations within VLLMs.
I'm at ICLR to present Switch SAEs. Come by 3pm - 5:30pm today at Hall 3 + Hall 2B #272.
I'm at ICLR to present Switch SAEs. Come by 3pm - 5:30pm today at Hall 3 + Hall 2B #272.
(1/11) New paper! “Low-rank adapting models for Sparse Autoencoders.” While SAEs find interpretable latents, they hurt downstream behavior—e.g. using TopK SAE activations on GPT-4 mimics a model trained w/ 10% compute. Our fix? Adapt the model for the SAE, not just vice versa.👇
the only thing worse than bad evals is realizing there are no bugs to blame
Our paper on Switch Sparse Autoencoders has been accepted to ICLR 2025 – see you in 🇸🇬!
Our paper on Switch Sparse Autoencoders has been accepted to ICLR 2025 – see you in 🇸🇬!
🦜Introducing the Stochastic Parrot 🦜: An AI-powered motivational companion! The Stochastic Parrot sits on your shoulder while it listens, looks, and talks to help you crush your goals 🎯(1/6)
Since the internal structure of neural networks, through training, comes to reflect the structure of the external world, advances in interpretability and our understanding of neural computation more generally could have a huge impact across science over the coming years...
Since the internal structure of neural networks, through training, comes to reflect the structure of the external world, advances in interpretability and our understanding of neural computation more generally could have a huge impact across science over the coming years...
Announcing Transluce, a nonprofit research lab building open source, scalable technology for understanding AI systems and steering them in the public interest. Read a letter from the co-founders Jacob Steinhardt and Sarah Schwettmann: transluce.org/introducing-tr…
1/11: New paper! "Decomposing the Dark Matter of Sparse Autoencoders." We find that SAE errors and error norms are linearly predictable using model activations. Why is this, and what does it imply for SAE scaling and the structure of language model representations? Answers in 🧵

Swarnim Jain @swar_ja
632 Followers 949 Following @nuance_labs // cs @ cambridge, yc summer fellow // prev. iit bombay
Max Chuang @maxxchuang
0 Followers 2 Following
Lillian @Eteagauj14445
11 Followers 921 Following Stay strong. Make them wonder how you’re still smiling.
Weena @Weena2337
17 Followers 1K Following
Florence @Couabu422800
15 Followers 947 Following Poetry is when an emotion has found its thought and the thought has found words.
Munmun Bhattacharjee @Munmun0690
22K Followers 7K Following A women of 21st Century. Founder of a NGO which works for Helpless Street Children,we also do other community services. #RadheKrishna #BharatFirst #JaiShriRam
GD @ilovesjca
0 Followers 229 Following
Michelle Li @michellezli
120 Followers 172 Following bio and econ @ harvard, 20 y/o, curious @ https://t.co/89WwwlkGaJ, filled with random thoughts
Linda Xue @xuelinda7
2K Followers 329 Following chief executive pookie @yourworkpookie squatting @joinstationf cs, neuro, design @mit
Awhafker @Awhafker8595
76 Followers 2K Following
Shashank Kirtania @5hv5hvnk
466 Followers 2K Following pre doc research fellow @prosemsft | interested in AI & Formal Methods
ella schlaghecke @ella_schlags
1K Followers 853 Following i run a policy hackathon and i love people 💃🏻 🍓🌟 @scholarsHQ @cansbridgeproj
Aunaoku @Aunaoku9348818
12 Followers 629 Following
Phonic @phonic_co
400 Followers 7 Following Reliable Voice AI | The next-generation voice AI platform to build, observe, and evaluate your voice AI agents
Sanjith Udupa @SanjithUdupa
188 Followers 220 Following 18 // eecs & robotics @mit / co-directing @hackmit / prev @dropbox
Constantin Venhoff @cvenhoff00
232 Followers 106 Following PhD Student at Oxford University @OxfordTVG | Intern @Meta
Jonathan Lei @jonathanlei0
687 Followers 830 Following credit card connoisseur // airline miles collector; and also cloud architect @voltagepark
Om Kulkarni @shadowboxingme
38 Followers 2K Following adaptable, entrepre-neural, and training on non-augmented data
René Heldmaier @ReneHeldmaier
4 Followers 22 Following If you see a good parking lot - look for a better one.
Adam Zweiger @AdamZweiger
942 Followers 415 Following Rethinking how language models learn | Researcher @MIT_CSAIL
Bercee @Bercee804904
37 Followers 974 Following
Gaeul @KE3934399331305
2 Followers 120 Following
Kevin Zhu @kevinbzhu
574 Followers 509 Following @mit, @scale_ai, investing @dormroomfund, organizing @hackmit
const unsigned long l... @idle__eyes
46 Followers 271 Following ml infra @ (redacted) foundation model startup
FanPu Zeng @FanPu_Zeng
640 Followers 1K Following ML research @JaneStreetGroup, previously @mldcmu @SCSatCMU. All views my own. 🇺🇸🇸🇬
Himanshu Maurya @Himanshu_nitrr
449 Followers 6K Following Giving meaning to mine share of star dust. Visiting fellow @WinshipAtEmory. Prev at @oracle, @maddox_ai, @KITKarlsruhe, @_nference, @val_iisc, @iitdelhi.
Diego Taquiri @diego_taquiri
411 Followers 3K Following Research in AI for Antibody Design @UCIrvine | Prev. BSc @CayetanoHeredia
Casey Hanna @CaseyHannaa
13 Followers 122 Following Math @ Stanford, summer @Microsoft | essays sometimes, code always
Seatough @Seatough177194
31 Followers 2K Following
amogh @OfficialAmogh
7K Followers 7K Following co-founder @humanbehaviorai (yc x25) // prev stanford cs
Jennifer Wang @wafercookies1
4 Followers 9 Following
Chaya Koelpin @koelpin72767
2 Followers 69 Following
Krithik Ramesh @KrithikTweets
785 Followers 710 Following AI + Math @MIT, compbio stuff @broadinstitute, prev: research @togethercompute
Andy Arditi @andyarditi
719 Followers 478 Following
Ishaan Sinha @imsinha0
52 Followers 397 Following Math + CS @harvard @amazon too many ideas, too little time
Liu Xiaoping 刘小�... @Liuxiaoping_WHU
234 Followers 1K Following Interested in single cell and spatial omics, digital pathology, cancer Fitness enthusiast,guitar noise maker LOVE CHINA
Brian Zhou @brian1zhou
494 Followers 2K Following Cofounded @a37_ai. On leave @Harvard, was @TJHSST.
Jeremy Fox 🦊 @JeremyDanielFox
2K Followers 724 Following Building Claude @AnthropicAI. Ex @google. My views are my own.
David @DavidSHolz
92K Followers 8K Following founder @midjourney, prev founder leap motion, nasa, max planck - random vibeposting @davidvibesonly
Zach Mueller @TheZachMueller
12K Followers 591 Following Let's make billions of parameters go brr https://t.co/rUxXIfNpwh
Daniel Han @danielhanchen
28K Followers 2K Following Building @UnslothAI. Finetune train LLMs faster. LLMs bug hunter. OSS package https://t.co/aRyAAgKOR7. YC S24. Prev ML at NVIDIA. Hyperlearn used by NASA.
Stuart Sul @stuart_sul
1K Followers 123 Following ml research @cursor_ai, cs @Stanford, mlsys @HazyResearch
Deedy @deedydas
205K Followers 5K Following VC at @MenloVentures. Formerly founding team @glean, @Google Search. @Cornell CS. Tweets about tech, immigration, India, fitness and search.
countersignaling @countersignalpd
256 Followers 10 Following what are young people thinking about? podcast produced by @srachasaucee
Sami Shalabi @shalabi
2K Followers 644 Following Building AI agents for enterprise. Founder @MavenAGI; X-@Google; X-@IBM; 4x Startups; @MIT
Karun Kaushik @karunkaushik_
6K Followers 347 Following Co-founder @GetDelve (AI-native compliance) | YC W24 | @ZFellows | Prod | prev. AI @MIT
Yacine Mahdid @yacinelearning
12K Followers 738 Following (neuro/ai) I make technical deep learning tutorials 👺
Ion Stoica @istoica05
5K Followers 20 Following Professor at UC Berkeley, co-founder of Databricks, Anyscale, LMArena, Conviva.
Andre Saraiva @andresnds
3K Followers 138 Following o1-preview, o1-mini, o1, o3-mini,o4-mini, o3... Reasoning Researcher at OpenAI. Ex-DeepMind.
Dwarkesh Patel @dwarkesh_sp
127K Followers 911 Following Host of @dwarkeshpodcast https://t.co/3SXlu7fy6N https://t.co/4DPAxODFYi https://t.co/hQfIWdM1Un
Sebastien Bubeck @SebastienBubeck
56K Followers 1K Following I work on AI at OpenAI. Former VP AI and Distinguished Scientist at Microsoft.
Together AI @togethercompute
50K Followers 387 Following AI pioneers train, fine-tune, and run frontier models on our GPU cloud platform.
Zhuohan Li @zhuohan123
9K Followers 865 Following mts @ openai | cs phd @ 🌁 uc berkeley | building @vllm_project | machine learning system | the real agi is the friends we made along the way
Ryan Bai @ryanbai1412
461 Followers 29 Following
Jack Lindsey @Jack_W_Lindsey
6K Followers 237 Following Neuroscience of AI brains @AnthropicAI. Previously neuroscience of real brains @cu_neurotheory.
Shane Legg @ShaneLegg
58K Followers 56 Following Co-founder and Chief AGI Scientist, Google DeepMind
Kimi.ai @Kimi_Moonshot
50K Followers 98 Following Built by Moonshot AI to empower everyone to be superhuman.
Elsewhere @elsewhere_today
897 Followers 16 Following The world's first late-night café coworking chain for builders, creatives & nomads.
Prem Qu Nair @premqnair
5K Followers 910 Following @cognition, previously @nuro @princeton. Pursuing 70mm, 225lb, and $0.10/piece
Minh Le @minhxle1
120 Followers 219 Following Research Fellow @AnthropicAI | Prev: @Parafin, @Robinhood
Alex Cloud @cloud_kx
120 Followers 60 Following
Yi Tay @YiTayML
46K Followers 81 Following research scientist @googledeepmind ✨♊, model co-lead/captain of gemini deepthink imo gold medal 🥇, opinions are my own.
Reducto @reductoai
3K Followers 20 Following The most accurate document processing platform + API for AI teams to parse, split, and extract data from unstructured docs. --- https://t.co/hVP38Angy5
Mechanize @MechanizeWork
6K Followers 1 Following We're a software company building RL environments to power the full automation of the economy.
Jack Morris @jxmnop
45K Followers 978 Following research @cornell @meta // language models, information theory, science of AI
Psyho @FakePsyho
25K Followers 366 Following Game Designer; Problem Solver; past: OpenAI (Dota), Pro Competitive Programmer, Poker
Michelle Li @michellezli
120 Followers 172 Following bio and econ @ harvard, 20 y/o, curious @ https://t.co/89WwwlkGaJ, filled with random thoughts
Ndea @ndea
8K Followers 78 Following A new intelligence science lab founded by @fchollet & @mikeknoop.
François Chollet @fchollet
572K Followers 813 Following Co-founder @ndea. Co-founder @arcprize. Creator of Keras and ARC-AGI. Author of 'Deep Learning with Python'.
Linda Xue @xuelinda7
2K Followers 329 Following chief executive pookie @yourworkpookie squatting @joinstationf cs, neuro, design @mit
Zeyuan Allen-Zhu, Sc.... @ZeyuanAllenZhu
20K Followers 452 Following physics of language models @ Meta (FAIR, not GenAI) 🎓:Tsinghua Physics — MIT CSAIL — Princeton/IAS 🏅:IOI x 2 — ACM-ICPC — USACO — Codejam — math MCM
sampada 🎇 @sampadanepal
236 Followers 212 Following
a16z @a16z
874K Followers 52 Following we invest in software eating the world https://t.co/A9eTFq6plZ https://t.co/MXGUBJoesw Watch "The Ben & Marc Show": https://t.co/eRuDhx7kpe
ella schlaghecke @ella_schlags
1K Followers 853 Following i run a policy hackathon and i love people 💃🏻 🍓🌟 @scholarsHQ @cansbridgeproj