I've been working for many months on this article on Silicon Valley's under-the-radar role in bringing AI into schools across the US. I really hope you'll read it — here's a gift link — but I'll tell you some of the highlights in this thread. (1/x)

🚨 New postdoc position in our lab at Berkeley EECS! 🚨 (please reshare) We seek applicants with experience in language modeling who are excited about high-impact applications in the health and social sciences! More info in thread 1/3

This is great, & there's clear analogy to the burgeoning mechanism design community for AI alignment: who is providing RLHF votes? Do their preferences reflect yours? Discussions about social choice and collective constitutions are interesting, but "what and who is in the data" is just as important.

📢New POSITION PAPER: Use Sparse Autoencoders to Discover Unknown Concepts, Not to Act on Known Concepts Despite recent results, SAEs aren't dead! They can still be useful to mech interp, and also much more broadly: across FAccT, computational social science, and ML4H. 🧵

Are LLMs correlated when they make mistakes? In our new ICML paper, we answer this question using responses of >350 LLMs. We find substantial correlation. On one dataset, LLMs agree on the wrong answer ~2x more than they would at random. 🧵(1/7) arxiv.org/abs/2506.07962

@jessica.bsky.social on individual reporting as a means to build collective knowledge.

ARR question: If I submit to a cycle, how long do those reviews "last"? e.g. if I submit to the July cycle but can't go to AACL, can I commit my July reviews to the conference associated with the next (October) cycle? @aclrollingreview.bsky.social

New work 🎉: conformal classifiers return sets of classes for each example, with a probabilistic guarantee the true class is included. But these sets can be too large to be useful. In our #CVPR2025 paper, we propose a method to make them more compact without sacrificing coverage.

I would like to spend up to 5-10 hours to learn about basic macroeconomics (I know it's maybe fake, but setting that aside for a moment...). Does anyone have any recommendations?

People love to hate on the transition 3-pointer as evidence of how the 3 has ruined basketball, but I think it's usually just the right play... if you have numbers in transition, your teammate can easily get a putback off a miss, so might as well try the 3

We'll present HypotheSAEs at ICML this summer! 🎉 Draft: arxiv.org/abs/2502.04382 We're continuing to cook up new updates for our Python package: github.com/rmovva/Hypot... (Recently, "Matryoshka SAEs", which help extract coarse and granular concepts without as much hyperparameter fiddling.)

Yesterday's Game 6 was depressing, and this article precisely delineated the reasons why. And sometimes, a precise retelling of what you're feeling is all you need to feel better. www.nytimes.com/athletic/633... @thompsonscribe.bsky.social

Check out Erica's nice work. They not only develop a well-grounded model for disparities in disease progression, but also conduct experiments with real NYP cardiology data! (Anyone who works in healthcare knows how much of a feat it is to use data other than MIMIC)

The US government recently flagged my scientific grant in its "woke DEI database". Many people have asked me what I will do. My answer today in Nature. We will not be cowed. We will keep using AI to build a fairer, healthier world. www.nature.com/articles/d41...

Good reading for PhD students on why meeting scheduling might be more important than you think: paulgraham.com/makersschedu...

1st post on bsky! What happens when a static benchmark comes to life? ✨ Introducing ChatBench, a large-scale user study where we *converted* MMLU questions into thousands of user-AI conversations. Then, we trained a user simulator on ChatBench to generate user-AI outcomes on unseen questions. 1/ 🧵

A weird (and seemingly fixable quirk) with ChatGPT Deep Research is hallucinations even when there is a linked citation with a highlighted passage: "X et al find Y (arXiv link)" and you click on the arXiv link, which is neither written by X nor do they find Y...

Stop everything you are doing and review the best Open Data chart you will see all year. Credit: @nkgarg.bsky.social's lab

My labmates really cooked with this one

Migration data lets us study responses to environmental disasters, social change patterns, policy impacts, etc. But public data is too coarse, obscuring these important phenomena! We build MIGRATE: a dataset of yearly flows between 47 billion pairs of US Census Block Groups. 1/5

I've learned MCMC 5+ different times in the last 10 years from courses, videos, blogposts, etc, and it never clicked. 30 minutes chatting with Claude this morning, and I finally feel like I've figured it out...

Really proud of @rajmovva.bsky.social and @kennypeng.bsky.social for this work! We hope that it's useful, and are already using it for many followup projects Preprint: arxiv.org/abs/2502.04382 Python package: github.com/rmovva/Hypot... Demo: hypothesaes.org

(1/n) New paper/code! Sparse Autoencoders for Hypothesis Generation HypotheSAEs generates interpretable features of text data that predict a target variable: What features predict clicks from headlines / party from congressional speech / rating from Yelp review? arxiv.org/abs/2502.04382

💡New preprint & Python package: We use sparse autoencoders to generate hypotheses from large text datasets. Our method, HypotheSAEs, produces interpretable text features that predict a target variable, e.g. features in news headlines that predict engagement. 🧵1/

An interesting dynamic on Google reviews is that I tend to trust businesses which write custom replies to negative reviews, even if the reply doesn't really resolve the issue (e.g. see below). I distrust businesses which have stock replies, even if those replies offer to resolve the issue somehow.

*Please repost* @sjgreenwood.bsky.social and I just launched a new personalized feed (*please pin*) that we hope will become a "must use" for #academicsky. The feed shows posts about papers filtered by *your* follower network. It's become my default Bluesky experience bsky.app/profile/pape...

Now online @pnasnexus.org! Many discrimination auditing and electoral tasks use ML to predict race/ethnicity – by discretizing continuous scores. Can the discretization process cause bias in labels and downstream tasks? Yes! Led by @evandyx.bsky.social academic.oup.com/pnasnexus/ar...

In new work, we show a "No Free Lunch Theorem" for human-AI Collaboration (w/ @nkgarg.bsky.social and Jon Kleinberg). (And if you're at #AAAI, I'm presenting at 11:15am today in the Humans and AI session. Poster 12:30-2:30.) arxiv.org/abs/2411.15230

Is there a canoncial version of modernBERT for computing sentence/document embeddings? Primarily for classification & clustering (rather than retrieval), but any are fine. If not, what's the best practice to finetune for this? cc @benjaminwarner.dev @howard.fm

Does anyone have a rule-of-thumb for how many words/sentences you can represent in a 3072-dim text embedding (using openai's large embs)? Like, there might be a token limit (here, 8192), but perhaps the embeddings are quite lossy beyond some smaller number of tokens?

A coming flood of federal grant proposals will claim AI makes projects cheaper & better. Many will be empty promises. This admin needs to place the best bets it can on AI. For the Federation of American Scientists' Day One Project, my memo on AI grant governance fas.org/publication/...

One of my grand interpretability goals is to improve human scientific understanding by analyzing scientific discovery models, but this is the most convincing case yet that we CAN learn from model interpretation: Chess grandmasters learned new play concepts from AlphaZero's internal representations.

Our article on using LLMs to promote health equity is out in New England Journal of Medicine AI! 85% of equity-related LLM papers focus on *harms*. But also vital are the equity-related *opportunities* LLMs create: detecting bias, extracting structured data, and improving access to health info.

We have a new review on generative AI in medicine, to appear in the Annual Review of Biomedical Data Science! We cover over 250 papers in the recent literature to provide an updated overview of use cases and challenges for generative AI in medicine.

Reflections on NeurIPS: There's always a big theme people seem to be preoccupied with. This year, it was the continuation of scaling/progress. Will it continue? What will the next generation of models hold? I even got to sass Dylan Patel (not on bsky) over it. Here are my personal thoughts 🧵

genAI has made us more suspicious that emails, cover letters, artworks, etc. are produced by AI. this shift forces us to change our behavior in order to prove our human-ness: a "burden of authenticity". waking my account up to share a recent blog post on the subject: rajivmovva.com/2024/11/08/g...

The soul-searching journey for figuring out what research area is right for you is tricky since so many papers are cool. I tell my early career students that they should try to differentiate papers that they'd like to read 📖, implement 🔨, *and* write 📝 from papers that they'd only like to read 📖.