Our first protein 🥩 design paper! 🤠 Check out Minkyu's tour de force on T-REX 🦖, an agentic protein design controller for de novo binder design! TLDR we treat protein design as an online allocation problem over heterogeneous tools (Proteina-complexa, Boltzgen, Bindcraft, ✨ etc), use 🤖 agents to reason over what tools + settings to run, designed a granular classification system to manage the agents (e.g. Rescue🛟/Explore🔎/eXploit🚀), that is all orchestrated with a deterministic controller to efficiently brrr the GPUs. 💸 T-REX 🦖 is fully open source: github.com/ml-struct-bio/T-R… Preprint: biorxiv.org/content/10.64898… Please let us know if you have any feedback!

🎙️ I spoke with the @BBC Newshour about the current AI governance debates. You’re hearing a lot right now about AI doomsday scenarios and existential risks. We are developing increasingly capable systems, and there is genuine uncertainty about where those capabilities will lead, which poses important scientific questions to investigate. But the current AI existential risk narrative is blinding us to the actual crisis: the concentration of power in the hands of a few companies in a handful of countries. I fear the current doomsday narrative will contribute to deepening that power imbalance as it may shut down open-source innovation out of fear. AI is an enormous opportunity for development for countries in Africa and the Global South, but models developed elsewhere often fail in local contexts. Those models need to be adapted to local data. Developing nations cannot afford to train billion-dollar frontier models from scratch. There are too many resources needed to do that. Open-source is the only bridge that allows local innovators to adapt frontier AI systems to their own realities. If we crush open-source out of fear, developing nations lose the main mechanism they have to build their own sovereign tech ecosystems. #AI #Africa #OpenSource #AIRisk #UNGeneralAssembly @Vertaix_ @PrincetonCS 📌 Listen to the full interview where I touch on a bit of this and other things (starting at 37:00) here: bbc.com/audio/play/w1730bn70…

Wrote down my own personal story of a theory I love, online convex optimization, and the people along the way: minregret.com/2026/09/15/two…

looking forward to this launch event, AI Alignment & Safety, Oct 19th @ Princeton!!

The @SiebelScholars Foundation has awarded fellowships to 5 grad students in @princetoncs: Linrong Cai, Peter Halmos, Nicolaas Kaashoek, Stephen Newman and Anchengcheng Zhou. The $35,000 fellowships are based on academic achievement and leadership. ⬇️ cs.princeton.edu/news/five-c…

Congrats to all, including our own @CaiLinrong !

Congrats to graduate students Linrong Cai, Peter Halmos, Nicolaas Kaashoek, Stephen Newman and Anchengchen Zhou on being named 2027 @SiebelScholars! 🎉 The fellowships are awarded to students for outstanding academic achievement and leadership. bit.ly/4xW81Jv

AI Alignment & Safety is perhaps the most urgent&important problem we face. Join us as a postdoctoral fellow to work on technical solutions!! NEW POSITIONS at AI Alignment & Safety @Princeton: Apply here: apply.interfolio.com/192161

Congrats to Princeton alum @_pgokhale on winning the 2026 Distinguished Early Career Award at @IEEEQuantumWeek! Gokhale graduated from @Princeton in 2015 with a BSE in computer science. He is now the CTO of @infleqtion. He has a PhD from @UChicago. thequantuminsider.com/2026/0…

"The progress of AI in biology and biomedicine is enormous. But it’s also critical for us to understand how challenging this field is. We really need these purpose-built AI models that can identify and decode the unknown." Professor Olga Troyanskaya recently spoke at @SimonsFdn's 2026 Simons Science Summit about how her lab uses AI to untangle the mysteries of the genome: bit.ly/4h2G9N7

Some thoughts on AI and Theory. 1. To a first approximation, theoretical computer science has been organized around a few major open questions. Much of our work has been motivated by developing approaches to answer these questions. 2. Such “problem-motivated” work has often led to theory-building focused on identifying a general principle that unifies a class of theorems. But much of that theory-building also involved proving new, difficult theorems. 3. Thus, while it’s true that problem-solving was strongly correlated with building understanding, drawing connections, and eventually developing general theories, it would be disingenuous not to admit that our community, perhaps disproportionately in retrospect, focused on and celebrated problem-solving. This was not arbitrary, and was quite defensible. Being able to make progress on central technical questions usually correlated with taste, creativity, persistence, and depth of understanding. Much of our reward structure therefore implicitly relied on the fact that producing an important proof was good evidence that someone possessed these harder-to-observe qualities. 4. It seems likely that we will soon have AI tools available to us that can prove many such theorems in a short time. The cost of obtaining proofs for well-posed mathematical questions will likely fall dramatically. The “scarce” intellectual work will likely shift both upstream: to questions, models and theories, definitions, and conjectures, and downstream: to interpretation, synthesis, explanation, and theory-building. 5. But as long as we believe in humans being meaningfully in charge of our collective decisions and fate, building human understanding of our science (and of science more generally) will remain an essential goal. I plan to expand on this important aspect soon. 6. Historically, finding a solution to an important problem and understanding its significance, implications, and connections were entangled. Finding a proof usually required researchers to discover the right concepts along the way. A dramatic reduction in the time and effort required to prove theorems could break that coupling. We could end up with many more true statements and proofs without a commensurate increase in understanding. Converting an abundance of proofs into human understanding may become one of the central challenges of our field. 7. As a result, I expect the high-level goals of theoretical computer scientists to change. In fact, the advent of powerful theorem provers might help us construct new theories and explore new models far more easily and rapidly, and significantly expand the domains where our models and theories apply. In that sense, the space for theoretical work may significantly expand rather than contract. 8. There’s a high human cost to the disruption that we are likely heading into. Many in our field, and in mathematical communities more broadly, are coming to terms with it. The range of opinions and reactions among mathematicians and theoretical computer scientists is a natural part of this evolution in our thinking as we collectively work through it. Some concrete efforts (including one at @SimonsInstitute) are already underway to think through the immediate scientific and institutional questions arising during this transition.

Petabytes of bodycam video are collected by police depts — most footage is discarded after 90 days. AI is now used to review the footage instead, but Olga Russakovsky and her team found that most AI models aren't able to analyze the footage accurately. bit.ly/4xIWL3P

VERY EXCITED to host a launch event for AI Alignment & Safety at @Princeton on Oct 19th, with v. distinguished speakers and new&important research. C u there!!

Congrats to the 11 COS faculty members recognized for their outstanding teaching! 🎉 🏅 🏅Robert Dondero 🏅Aarti Gupta 🏅@manoelribeiro 🏅Zachary Kincaid 🏅@praveshkkothari 🏅Lydia Liu 🏅@margmartonosi 🏅Pedro Paredes 🏅@OlgaTroyanskaya 🏅David Walker 🏅@ZhongingAlong

#AI still struggles to accurately analyze police bodycam footage, according to new research by @orussakovsky and her team. AI models don’t perform this task reliably because they are trained on video that is staged, well-lit and clear, she said. 👇engineering.princeton.edu/ne…

AI is increasingly being used by police depts to analyze and review bodycam footage. New research from @Princeton shows that in 1 out of 4 cases the AI models can’t identify basic details, like whether an officer handcuffed someone or drew a gun. bit.ly/4xIWL3P