Excited about this work with Asher Trockman Yash Savani (and others) on antidistillation sampling. It uses a nifty trick to efficiently generate samples that makes student models _worse_ when you train on samples. I spoke about it at Simons this past week. Links below.

A few days ago, we dropped 𝗮𝗻𝘁𝗶𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼𝗻 𝘀𝗮𝗺𝗽𝗹𝗶𝗻𝗴 🚀 . . . and we've gotten a little bit of pushback. But whether you're at a frontier lab or developing smaller, open-source models, this research should be on your radar. Here's why 🧵

thrilled to receive the outstanding paper award for our work on shallow alignment! i’ll be giving the talk at 10:42am tomorrow (Thursday) in oral session 1D. the poster will be Friday 3PM.

Why do larger language models generalize better? In our new ICLR paper, we derive an interpretable generalization bound showing that compute-optimal LLMs provably generalize better with scale! 📄arxiv.org/abs/2504.15208 1/7🧵

✨ Love 4o-style image generation but prefer to use Midjourney? Tired of manual prompt crafting from inspo images? PRISM to the rescue! 🖼️→📝→🖼️ We automate black-box prompt engineering—no training, no embeddings, just accurate, readable prompts from your inspo images! 1/🧵

I'm very excited to talk about compression-based memorization with Pratyush Maini this Friday at the OpenAI Security Research Conference! Let's chat about compression, memorization, and also our new antidistillation sampling antidistillation.com!

Looking forward to giving a talk this Friday OpenAI with Zhili Feng on some of our privacy & memorization research + how it applies to production LLMs! We've been gaining momentum on detecting, quantifying & erasing memorization; excited to explore its real-world impact!

📣Thrilled to announce I’ll join Carnegie Mellon University (CMU Engineering & Public Policy & Language Technologies Institute | @CarnegieMellon) as an Assistant Professor starting Fall 2026! Until then, I’ll be a Research Scientist at AI at Meta FAIR in SF, working with Kamalika Chaudhuri’s amazing team on privacy, security, and reasoning in LLMs!

I find it interesting that people who believe LLMs/autoregressive models are a dead end base their arguments, and reasoning, either on philosophical, hard to test or rebut hypotheses, or on micro failures, eg 9.11 vs 9.9, to predict paradigm macro failures. All the while the

Mean Flows for One-step Generative Modeling "We introduce the notion of average velocity to characterize flow fields, in contrast to instantaneous velocity modeled by Flow Matching methods. A well-defined identity between average and instantaneous velocities is derived and

Excited to share our work with my amazing collaborators, Goodeat, Xingjian Bai, Zico Kolter, and Kaiming. In a word, we show an “identity learning” approach for generative modeling, by relating the instantaneous/average velocity in an identity. The resulting model,

We're now merging code edits at 4300 tok/s, over 2x faster than the Llama 70b deployment on Cerebras. docs.relace.ai/docs/instant-a…

Introducing ARGUS 👁️ A benchmark for measuring hallucinations and omissions in free-form captions generated by Video-LLMs.