on my way back to NYC, i met wise Leon Bottou in the airport. we talked. then i told him "you should tweet that!" and, he delivered much more than a tweet: a blog post with thoughts and insights on AI research only he can deliver this clearly and succinctly.

While at Meta, I worked on this optimizer-wrapper (outer step lookahead momentum) we're calling Snoo (arxiv.org/abs/2510.15830). You can use it with AdamW or Muon and see really strong scaling. Here's a plot where we ran it against (tuned) AdamW up to 1e23 training flop scales.

Will agents replace systems developers? Yes or No? Upper level 5AB, debating 2:15-3:15pm. Moderated by yours truly. 🔥🔥 #MLforSystems #neurips2025

Was an honor to be back at the ML for Systems workshop at NeurIPS! Huge shoutout to the organizers for an awesome program and audience. Oh, and we’re all hiring all across the board for Codex 🧵

Ty for the salad + steak for the workshop speakers and authors! Incidentally I had to lay down to digest — seeing my potential inebriation, Mangpo ended up writing our super useful debate outline 🔥 So if you loved the debate / moderation, please diffuse the credit assignment.

It was so wonderful to be there for the MLforSystems workshop at NeurIPS this year! Congratulations to all the organizers! Honored to be invited for a talk and panel debate! Lunch was delicious too :)

i gave a keynote talk at NeurIPS'25 just last week. here's the slide deck (link below) i've used to share my thoughts on who we are and what we do.

Ty xiao! Love my close captioning “it’s a very exciting topic” The unsaid script: Agents are not the ml for systems we planned for, but it’s the ml for systems we deserve, when we are definitely not hitting agi yet. 1:10->1:5 audience split for agents replacing system devs.

Not writing is writing, in all senses. You deliberate, ruminate, and painstakingly squeeze out words until their shapes disappear. You research, make mistakes, experience shame, and you cope too hard. You take a walk, you travel, you open up a blank page & feel no difference.

If 2026 becomes the year my thesis on cryptographically secure computation that supports large optimization workloads becomes useful that would be oddly very apt. For context, few cared about my topic until ChatGPT got released. (scale and deployment were all I needed 🥹).

Transformers are super cool. But what if there’s something else…

10 years, 2 PhDs, too much et cetera later, we meet in puffers. Harvey Mudd Alumni 🌴⏩🧣 #computerscience #math #chemistry #statmech

How AI Can Accelerate Systems Performance Research This new research introduces AI-Driven Research for Systems (ADRS), a framework where LLMs iteratively generate, evaluate, and refine algorithms for systems performance problems automatically. The researchers applied three

thank you friends for all the goods and memories ❤️

Do you want “suggestions for hobbies found in your Google Photos”? Curious if I’m just privacy-pilled or is this product decision oddly stacked against all market research on how real humans feel re/AI. Cloud photos = private, passive. Even “with permission”, sketch factor high

Thrilled to announce our $300M Series A at a $4B valuation! Chips are the fuel for AI. At Ricursive Intelligence, we are using AI to design better chips faster, closing the recursive self-improving loop between AI and hardware.

Congrats to the basis set team!