OpenAI's Liam Fedus responds to Terence Tao's mathematicians' manifesto on AI

量子位 · wechat · 2026-09-14

OpenAI's former post-training lead Liam Fedus responded to a manifesto signed by Terence Tao and 25 Fields Medalists warning that AI companies' obsession with solving famous problems harms mathematical understanding. Fedus agrees understanding matters but argues future AI could both solve problems and create conceptual frameworks mathematicians build on. Context: OpenAI models recently refuted the 80-year-old Erdős unit distance conjecture and found a Navier–Stokes singularity proof with 10,000 agents in 88 hours, verified in Lean. Meanwhile, a priority dispute erupted with NYU's Tristan Buckmaster over the fluid-equations work. AMS, LMS and the Clay Institute have each issued cautious responses on authorship, verification, and AI's role.

Original post →

More from AGI Musings

AGI Musings channel →