创建

登录 ReadmeX

或

了解我们如何处理你的信息: 隐私政策

ReadmeX

陪你读懂 AI 动态

今天,想先了解什么?

选一份动态总结,或直接聊聊你的问题。

社区和关注动态总结只属于你自己。

资讯

如何评估 AI 证明的 Lean 验证

AI 总结

一篇分析指出,Lean 代码成功编译只是评估 AI 生成证明的一部分。其四项检查框架包括:Lean 陈述是否与论文相符、代码能否编译、证明是否使用了非预期公理,以及是否利用了 Lean 内核漏洞。即使所有检查均通过,Lean 证明也可能与论文 PDF 中的证明不同。该分析称,OpenAI 的论文合集包含 722 篇论文,约 45% 附有部分 Lean 代码;290 个相关条目中,约 38 个没有陈述、更没有证明论文的主定理。作者还表示,这些检查尚未系统应用于合集中的许多论文,但没有理由认为检查会失败。

深度见解

该框架为判断“AI 生成的正式证明已经正确”提供了具体检查方法,也说明通过 Lean 验证并不能单独证明论文所载证明正确或与 Lean 证明一致。

This post is an attempt to explain in simple terms how to interpret claims of the form “this theorem is correct because it was checked in Lean”. The summary is: there is an important signal that can be extracted from Lean code, but if this code is AI-generated then you need to do some tests first. We explain the tests, and note that in many cases (such as the OpenAI theorem dump) these tests have not yet been systematically done.

Introduction

Last month’s theorem dump by OpenAI consisted of 722 mathematics papers, around 45% of which came with some Lean code. Three of the (non-Leanified) papers have since been retracted because of a sign error. Some people might say “well, the papers that come with Lean code are definitely correct because that’s how Lean works”. Others might say “well Lean had a soundness bug and could well have more, and AIs will reward hack, so Lean verification doesn’t tell us much at all”. The truth, unsurprisingly, is more subtle, and this post is an attempt to explain it.

#OpenAI#Lean#Mathlib#Anthropic

原文约 8 分钟读完阅读原文
群聊

还没有评论,来说说你的看法。