Primary sourcearXiv: OpenAI
View sourceEvent ·Official first-hand
Lean 验证通过不代表自然语言证明正确:从 Navier-Stokes 看自动形式化局限
arXiv 论文 2610.08144 分析了自动形式化(autoformalisation)的可靠性问题,即用 AI 将自然语言数学文本翻译成 Lean 等形式语言再做机械验证的流程。作者指出,解决数学自然语言中的歧义在 Solvability Complexity Index 层级中位置为 SCI = ∞,高于任何可计算问题(停机问题仅为 SCI = 1)。论文给出多个 AI 将自然语言陈述误译入 Lean 的实例,导致自然语言证明与 Lean 验证不匹配。这些实例包括 OpenAI 公布的 Navier-Stokes 方程解爆破证明,作者论证其形式化 Lean 证明与原始自然语言证明并不对应。
Current read
一篇用 SCI 层级理论拆解自动形式化局限的论文,直接指向 OpenAI 的 Navier-Stokes 证明——Lean 验证过了,原文可能还是错的。
11 sources63° AI heatLast updated 2026/10/6 10:58:01
Evidence
11 sourcesRelated source 1Mark Chen (OpenAI 研究)
View sourceRelated source 2宝玉
View sourceRelated source 3rohanpaul_ai
View sourceRelated source 4@koltregaskes
View sourceRelated source 5kimmonismus
View sourceRelated source 6Gary Marcus
View sourceRelated source 7OpenAI Blog
View sourceRelated source 8Thomas Wolf
View sourceRelated source 9theverge
View sourceRelated source 10IT之家
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.