Mathematicians question AI proof transparency and academic value
A report published by the WeChat account Qin Shuo Pengyou, authored by Irene Zhou, examines mathematicians’ criticism of AI companies that use public scholarly material for model training while often withholding the reasoning behind AI-generated proofs. Mathematician Tao Zhixuan, Fields Medalist Deng Yu, and Rutgers statistician Han Qiyang are quoted on transparency, attribution, research norms, and the declining value of conventional mathematical problem-solving. The report also quotes OpenAI’s announcement that it used about 10,000 parallel AI agents and an unpublished model to address the Navier–Stokes existence and smoothness problem in 88 hours, alongside a 166-page paper and Lean verification code.
The report documents tensions over transparency, attribution, training data, peer review, academic evaluation, and the changing role of human mathematicians as AI systems increasingly participate in mathematical research.

继知名数学家陶哲轩公开批评OpenAI后,近期另一位新晋菲尔茨奖得主邓煜也炮轰OpenAI——不能你蒸馏了全世界,不让别人来蒸馏你。随着顶级模型公司不断破解世界级数学难题,数学家们集体破防。邓煜近期提出的观点大意是,AI公司在构建大模型时,通过“知识蒸馏”或抓取全人类的网络文献、学术论文与知识库来训练自己。然......
本文来自微信公众号: 秦朔朋友圈 ,作者:Irene Zhou
继知名数学家陶哲轩公开批评OpenAI后,近期另一位新晋菲尔茨奖得主邓煜也炮轰OpenAI——不能你蒸馏了全世界,不让别人来蒸馏你。
随着顶级模型公司不断破解世界级数学难题,数学家们集体破防。