How AI actually does math: Lean verification leads, AI is a blind component
gerardsans · x · 2026-10-31
A debate over AI's role in mathematics. The quoted post noted that until recently "AI does mathematics" meant olympiad problems with known answers—AI hit gold-medal level in July 2025—while this year the action moved to open problems, and OpenAI's September claim about an internal model showing Navier-Stokes equations can break down refuted the "AI lacks creativity" critics.
gerardsans pushes back on misconceptions: in real math research, the initial formulation by human experts is key, with Lean, a formalisation engine, verifying intermediate and final results. AI itself is a "blind component" in this pipeline, not an independent mathematician.
Related event: OpenAI Drops 722 Math Breakthroughs at Once, Sparking Debate(4 posts)→
More from Models
- AI fact-check thread, Error 7: technically correct but misleading phrasing, reworded — SpencrGreenberg · 2026-11-02
- Reddit user suspects Opus 5.5 got worse after circular bug-hunting, benchmarks show no clear decline — hypesfo · 2026-11-02
- OpenAI weekly: GPT-6.1 Terra priced 50% below GPT-5.6, Rosalind Research rolls out, shared ChatGPT-Codex limits — btibor91 · 2026-11-02
- Rumored 'gpt-6.2-astra' briefly surfaced in Codex CLI, unverified — Angaisb_ · 2026-11-02
- Developers chafe at Claude Code's weekly limits: 75% used, 3 days to reset — adamamcbride · 2026-11-02
- Alleged 14-trillion-param "Mythos-class" weights appear on Hugging Face, unverified — teortaxesTex · 2026-11-02