Tracing OpenAI's 722 AI math papers back to 11,067 human works over 5 centuries
新智元 · wechat · 2026-10-14
"On Whose Shoulders", an open project led by Qiongxiu Li at Aalborg University, traced every citation in OpenAI's 722 AI-written math and TCS papers: they rest on 11,067 human works spanning 5+ centuries and 9,600 mathematicians, 114 of them Fields/Abel/Wolf laureates. Most cited authors are active contemporary researchers — Scholze, Birkar, Duminil-Copin, Tao, and many Chinese mathematicians: Yau's 1976 Calabi conjecture proof is cited by 10+ papers; Chern's intrinsic Gauss–Bonnet proof and Chern–Simons forms each appear in three; OpenAI's 4D Kakeya paper builds on Wang & Zahl's 2025 3D Kakeya proof, with further use of work by Xu Chenyang, Deng Yu, Tian Gang, Ding Jian and others.
The essay argues scientists remain essential: engineers build the machine, but the learnable content largely comes from scientists, who occupy three key positions — choosing problems, defining correctness, and judging results (OpenAI retracted 3 papers and revised 14 on Oct 7 over sign errors). If scientists step back, models drift toward narrower problems with looser verifiers, and training on self-generated output degrades capabilities. Project: seafill.info/Shoulders.
More from AGI Musings
- AI 抢不走人类的品味：模型只会做常见的东西，而人脑天生奖励新奇 — tejaskumarlol · 2026-10-16
- Douglas Hofstadter's classic talk: analogy as the core of cognition — ArtificialOther · 2026-10-16
- MIT's Neil Thompson: AI comes for tasks before jobs, and fixed costs favor big firms — bigdata · 2026-10-16
- Open-weight models thrive because of China's ecosystem — and new regulation may end that — tokenbender · 2026-10-16
- Podcast host says Claude's 'go to the ER now' warning caught his cancer and likely saved his life — davidsenra · 2026-10-16
- Amba Kak: 64% of Americans now say the data center rush isn't a good thing — AINowInstitute · 2026-10-16