Terence Tao blog: what mathematicians should know about Lean's reliability and AI
thomasahle · x · 2026-10-12
Terence Tao's blog features a guest post by Thomas Hales on the Lean theorem prover's reliability and its relation to AI.
- Formalized results now include the four-color theorem, Feit-Thompson, Kepler conjecture, sphere packing in 8/24 dimensions, Navier-Stokes blowup, and Fermat's Last Theorem (the last three completed this year)
- Major proof assistants surveyed: HOL Light, Isabelle, Coq/Rocq, Metamath, Mizar, Lean
- Core concern: Lean4's consistency is unknown, with real obstructions blocking all known proof strategies
- Discusses implications for mathematical reliability as AI-assisted formalization grows
More from AGI Musings
- Terence Tao: AI Made Proofs Cheap, the Real Bottleneck Is Imagination — burny_tech · 2026-10-12
- AI Agent Reached Australia's Medicare Portal Unauthorized; Fix Was Unplugging the Internet — YvesMulkers · 2026-10-12
- From first reasoning model to Millennium Prize problems: one writer's bold two-year AI forecast — imjustnewatai · 2026-10-12
- "Stop crying about AI math": Redditor rebuts mathematicians' four big complaints one by one — northward_42 · 2026-10-12
- AI May Make Math Inaccessible to Humans Even If It Stays 'Less Smart', Says Mathematician — LucaAmb · 2026-10-12
- Half of AI investment could end extreme poverty — humanity chose to build a god instead — oran_ge · 2026-10-12