Skip to main content

Math+AI seminar: Liam Fowl

Friday, November 6, 2026 | 1:00 PM - 2:00 PM CT
Technological Institute, Tech LG52, 2145 Sheridan Road, Evanston, IL 60208 map it

Title : The Rapidly Evolving Landscape of AI for Math: From Natural Language Reasoning to Lean


Abstract : In roughly one year, the community has seen frontier LLMs progress from gold-medal performances on the IMO, to millennium prize contributions, and widespread use at the frontier of mathematical research. In this talk, we give a (brief) overview, from the perspective of those working in AI, of the progress and state of AI and math. We then detail several efforts from the Princeton Language and Intelligence group aimed at building compute-efficient, open-source tools for the broader community. Toward this goal, we examine why extended reasoning often stalls, identifying failure modes where models become trapped in unproductive lines of inquiry and lose the ability to self-correct. We then demonstrate how structured problem decomposition can break these patterns across both natural language and formal domains. Finally, we describe how these decomposition principles guide our recent formal systems in Lean, enabling efficient, structured proof-search architectures as well as scalable methods for translating informal mathematical insights into Lean code.


 

Audience

  • Faculty/Staff
  • Post Docs/Docs
  • Graduate Students

Contact

Antonio Auffinger
Email

Interest

  • Academic (general)

Add Event To My Group

Please sign-in