Benchmark-guided refactor trims OpenAI proof 328’s Lean code
The report’s authors say a self-developed data pipeline and human-AI collaboration were used to refactor OpenAI’s publicly available proof 328. The rewrite replaced the step invoking König’s infinity lemma to construct an infinite branch with a compactness-based argument using the finite intersection property. After reconnecting the revision to the original proof project, the authors report that it passed Lean formal verification. Excluding blank lines and block comments, the three affected Lean files fell from 232 to 153 lines, a reduction of 79 lines, or about 34%.
The report presents a concrete example of using mathematical-philosophy labels and theorem dependencies to guide proof restructuring, with Lean checking the revised argument and the authors reporting a substantial reduction in the affected code.
不久前我们介绍了一套在设计中的全新的“数学哲学思想识别 Benchmark”。
当时,在研究 OpenAI “攻克”千禧年大奖难题 Navier–Stokes 方程的时候,我们提出了一个问题:不考具体算术和背公式,大模型到底能不能从天书般的定理里,读懂背后的“数学哲学”?
在那次涵盖 Claude Opus 5.5、GPT-6 Sol、Gemini 3.8 Flash、GLM-5.3、Qwen3.8-Max、Kimi-K3 等十款模型的横评中,我们发现了多个有趣结论,其中一个耐人寻味的现象是,推理 Token 消耗得越多,模型未必答得越好。
有的模型仅用约 50 个推理 Token 就拿下了 92.7% 的准确率;而有的模型疯狂思考直至耗尽 2,048 Token 的输出上限,依然在死胡同里打转交了白卷。
不久之后,OpenAI 又公开了 700 多份由内部模型生成的数学手稿。这再次引发了轰动与争议:AI 确实能推演出极长的结果,但这些动辄数百行的证明到底对不对、有没有把原命题写歪,依然让人类核验者头疼不已。
#OpenAI#Lean#König infinity lemma#Mathematical Philosophy Benchmark