OpenAI Publishes 700+ AI-Generated Mathematics Preprints
OpenAI published more than 700 mathematics preprints generated entirely by AI on GitHub on October 6, according to 科技新報. Frank Calegari described the release as a milestone and said the papers show breakthrough contributions in several mathematical fields, while also criticizing their writing, citation practices, attribution norms, and some errors. Ursula Martin said OpenAI left messy drafts for the academic community to clean up, and Michael Harris said he would not read them; OpenAI said it uses difficult problems across scientific fields to evaluate and develop models.
The unusually large batch of AI-generated mathematical preprints provides a prominent test of AI-assisted research, but the reported expert criticism shows unresolved questions about rigor, attribution, and academic standards.


數學家弗蘭克·卡萊加里(Frank Calegari)在陶哲軒(Terence Tao)的網誌發表客座文章,指 OpenAI 在 10 月 6 日發布超過 700 篇 AI 數學論文是數學史上的里程碑,但他也批評這些論文寫得極差、違反引用慣例,也存在部分錯誤。他呼籲學界調整聘請、出版論文及培訓學生的方式,應對 AI 對數學研究的改變。
OpenAI 早前在 GitHub 發布 700 多篇完全由 AI 生成的數學預印本,即未經同行評審的論文初稿。數學界對這批論文意見不一:牛津大學數學家娥蘇拉·馬丁(Ursula Martin)批評 OpenAI 把凌亂初稿丟給學界收拾,哥倫比亞大學數論學家 Michael Harris 則表示不會閱讀這些論文。