Get verified
All guests

Daniel Litt

Not on Prends yet

University of Toronto mathematics professor

Speaks English5 podcast appearances

AI for research mathematics · frontier model capabilities · Erdos problems

About

Daniel Litt is a mathematician at the University of Toronto whose research sits where algebraic geometry meets number theory. He has become one of the most closely followed academic voices on how frontier AI models are changing mathematical research, writing essays and appearing on podcasts about what the models can and cannot yet do. He was one of the nine mathematicians who wrote the companion paper on OpenAI's disproof of the Erdos unit distance conjecture.

Known for

  • Research on the interplay between algebraic geometry and number theory, including arithmetic structures on fundamental groups of algebraic varieties and Hodge theory
  • Stanford PhD (2015), then NSF Postdoctoral Fellow at Columbia, Member of the Institute for Advanced Study, and faculty at the University of Georgia before moving to Toronto
  • Co-author of the nine-author companion paper on OpenAI's disproof of the Erdos unit distance conjecture
  • A March 2025 bet with Tamay Besiroglu, at 3 to 1 odds, that AI would not autonomously produce papers matching the best of 2025 by 2030, which he wrote in 2026 he now expects to lose

Key points

  • Argues that solving problems is only one part of mathematics: current models are good at grinding calculations, combining known techniques and searching large spaces, but still struggle with intuition, theory building and choosing the right questions
  • Argues there is no mystical part of math research that is inherently out of reach for models, while flagging truth-seeking, creativity, theory-building and long-horizon projects as areas where model ability is still unclear
  • Says AI-generated proofs that humans then check, which he calls vibe proving, could be genuinely useful despite frequent errors, though he had not yet seen a result he would call revolutionary

Recent appearances

References hosts use to judge fit

Researched from public sources, checked 2026-10-08. This person has not claimed the profile yet.