Mathematician Shares How GPT-o1-mini Produced Novel Correct Proof
prof_g · x · 2026-08-02
The author recounts exploring a mathematical conjecture about lattice-valued networks with GPT-4o and Claude 3.5 Sonnet two years ago. While the models could independently propose conjectures unseen in literature, they couldn't prove them.
After a month of confident but incorrect attempts, the newly released GPT-o1-mini successfully provided a novel, elegant, and correct direct proof. The author sees this as a turning point, predicting AI-generated proofs will soon surpass human capabilities.
Related event: Mathematician Shows AI Models Now Solve Proofs Without Complex Prompts(2 posts)→
More from Research
- It's not gradient descent: burny_tech corrects the diagnosis to RLHF mode collapse — burny_tech · 2026-10-16
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16