Yu Deng: Mathematicians in the time of AI

AI-generated proofs are beginning to extend existing research and challenge mathematicians’ intuition. How can the mathematical community benefit while preserving transparency, fair credit, and trust? Speaking at SAIR Foundation’s Science x AI Summit 2026 at UCLA on October 16, Yu Deng discusses recent AI-generated results, including an extension of his work on the Boltzmann equation and a claimed proof that would disprove one of his conjectures. He welcomes their mathematical potential while questioning the practices surrounding their release. Deng argues that AI research should meet the standards expected of human researchers: explaining the ideas behind proofs, acknowledging contributions, and carefully verifying results before publication. He also calls for meaningful user control over training data and explores how open models could better serve the mathematical community. Although these models currently lag behind leading closed models, he believes mathematicians can help strengthen them through shared expertise and focused training. His vision is one of humans and AI learning from each other, with AI enabling faster exploration and human curiosity guiding the search for new mathematical ideas. Outline of this video: 00:00 Welcoming AI as a tool for mathematics 03:30 The promise of recent AI-generated proofs 05:56 Explaining the ideas behind AI discoveries 08:04 Giving credit and verifying mathematical results 10:47 Commercial AI and the interests of mathematicians 13:33 Principles for responsible AI use in research 17:05 The case for open models in mathematics 20:54 Humans and AI exploring mathematics together