Had a great time talking at @ycombinator about @leanprover and the future of what I call “verified intelligence.” As AI changes how we do math, science, software, and ML, I’m increasingly convinced verification is going to become a much bigger part of the stack. Thanks @FrancoisChauba1 for inviting me!

Really liked @neelsomani post a month ago on verifiable transformers. The idea is simple but powerful: redesign a small GPT-style Transformer so its circuit explanations can be encoded into SMT solvers and checked over finite symbolic domains, rather than only supported by examples and ablations. I wanted to see if I could also do formally in a theorem prover! I used TorchLean that takes one exported checkpoint run and build a Lean-side evidence layer around it! So checking the model metadata, finite prompt traces, circuit summaries, replayed margins, and theorem endpoints. So the SMT solver checks the bounded circuit claim, while Lean checks that the exported evidence package actually matches what it claims! Lean has tactics that use SMT solvers in the background:)! I also proved some theorems regarding this! This week’s artifact is intentionally small and TorchLean can run large GPT-style models and this is the kind of future work I’m excited about. Blog here for more details: robertj1.com/ai4science/veri…

Lean is software that allows mathematical proofs to be written and checked as computer code. Mathematician Terence Tao uses it as a part of his collaborative approach to solving complex mathematical problems. quantamagazine.org/how-terry…

Software like Lean, which allows mathematical proofs to be written and checked as computer code, could usher in a new, more collaborative era of problem solving. One of its most prominent supporters is mathematician Terence Tao. quantamagazine.org/how-terry…

AI technology is on the rise. So are its applications in industry, science, and society at large. At leidendeclaration.ai we, a group of 16 scientists, including mathematicians, computer scientists, philosophers, and historians, have drafted a declaration calling for action on the challenges posed by the use of artificial intelligence in mathematical research. Over several months, we carefully weighed the promises and risks of introducing AI into research workflows and beyond. The declaration has already been endorsed by international communities, most prominently by the International Mathematical Union, as well as by a large group of mathematicians and other scientists. We invite our colleagues to sign the declaration and support its recommendations on how AI should be adopted in our workflows, and how we should establish and maintain relationships with AI labs and industry. One of its central points is the need to preserve balance, rigor, and sound judgment while acknowledging the extraordinary pace of development in the AI community. We hope this declaration, a deeply human act, will serve as a guiding document for those who recognize the human imperative and the need for governance in scientific discovery, especially mathematical discovery.

Last week we hosted the first ever YC Paper Club in Mountain View. We brought together great AI researchers and founders to discuss both the state of the art and what it actually takes to get it into production. Thanks to the following presenters: 0:12 - Intro from YC Visiting Partner @FrancoisChauba1 3:49 - Tanishq Kumar (@tanishqkumar07) — Speculative Speculative Decoding (arxiv.org/abs/2603.03251) 18:33 - Guangyao (Stannis) Zhou (@zhouguangyao) — Diffusion-MPC (arxiv.org/abs/2410.05364) 30:26 - Isaac Ward — LeWorldModeling (arxiv.org/abs/2603.19312) 43:54 - Akshay Vegesna (@akshayvegesna) — Deep Learning is Not So Mysterious or Different (arxiv.org/abs/2503.02113) 51:24 - Konwoo Kim (@konwookim) — Pretraining Under Infinite Compute (arxiv.org/pdf/2509.14786)

Launching today: Formal Frontier from the Mathlib Initiative, a program of @RenPhilanthropy and funded by @xtxmarkets. Formal Frontier will focus on responsible, scalable, and open-source AI-driven autoformalization of mathematics. Read more and learn how to get involved: mathlib-initiative.org/forma…

In the era of #ArtificialIntelligence, when human dignity is threatened by new forms of dehumanization, ours is the pressing duty to remain profoundly human. We must lovingly safeguard the grandeur of humanity bestowed upon us and revealed in its fullness in Christ, the splendor of which no machine can ever replace. #MagnificaHumanitas vatican.va/content/leo-xiv/e…

Amazing article by my advisor @AnimaAnandkumar on AI4Science, also with formal verification and guarantees this will definitely have a big impact across various scientific and mathematical domains.

So exciting!

Solving unsolved math problems is even more fun now! Check it out at aristotle.harmonic.fun

The Future of Mathematics Symposium keynotes from Fields Medalists: Terence Tao, Maryna Viazovska, and Michael Freedman Now available on Youtube, links below.