AI+Lean theorem proving loses to Bitter Lesson: OpenAI math gains come from non-formalized reasoning
Singularitarian · x · 2026-10-23
johnloeber's new essay reflects on weekends spent on automated theorem proving with AI + Lean, concluding the Bitter Lesson reigned supreme, and examines the OpenAI Math Repo and the short one-summer window of human-AI collaboration in mathematics. Reposter adibaradwaj highlights the key update: recent AI math gains appear driven by non-formalized reasoning rather than Lean — proofs are generated in plaintext with about half formalized afterwards, suggesting models are far smarter than they appear and their intelligence's spikiness may be underrated.
More from AGI Musings
- Pedro Domingos: AI Is an Intelligence Lever — the Smarter You Are, the Smarter It Makes You — pmddomingos · 2026-10-23
- Anil Seth: Consciousness Is a Property of Life, Not Computation — Building Conscious AI Is a Bad Idea — anilkseth · 2026-10-23
- Pedro Domingos predicts 2027 will see the most security vulnerabilities fixed in history — pmddomingos · 2026-10-23
- Beff Jezos calls for a 'cyber immune system' of constantly adaptive agents — beffjezos · 2026-10-23
- Pedro Domingos argues RL solves nothing that better generalization can't — pmddomingos · 2026-10-23
- Academic Writing in the AI Age: Do You Write for Humans or for LLM Reviewers? — JeremyNguyenPhD · 2026-10-23