AI use in math papers jumped from 1.39% to 14.09% in six months, arXiv study finds
rohanpaul_ai · x · 2026-09-23
The arXiv paper "The Gold Rush in AI4Math: Where Are We Now?" by Jiashun Jin, Zheng Tracy Ke and Bingcheng Sui scanned 32,944 math-related arXiv submissions from March 1 to August 20, 2026.
Key findings:
- 3,575 submissions disclosed AI use; 1,712 involved substantive mathematical contributions.
- Substantive AI use grew from 1.39% of submissions in March to 14.09% by August 20 — a 10x jump in under six months.
- Use is highly uneven across fields: Combinatorics has the most AI-assisted papers, Metric Geometry the highest substantive-use rate; proof construction was the largest category (1,225 papers).
- Geographically concentrated: the US and China account for about two-thirds of weighted country counts.
- AI is already tackling open problems: of 717 named open-problem records with substantive use, 71% are described as fully resolved, with proofs more common than counterexamples.
More from AGI Musings
- Viral take: if AI takes math, Terence Tao outshines you at any white-collar job — ctjlewis · 2026-09-23
- Debate: do entrepreneurial opportunities live in human situations or AI-only problems? — louisvarge · 2026-09-23
- Musk on AI-era education: broad knowledge, and robots eager for your prompts — XFreeze · 2026-09-23
- A new Fermi twist: no AI-filled universe, yet AI smarter than its builders exists — sqcai · 2026-09-23
- Model release gap shrinks from 73 days to 18, industry bets on one a day by 2027 — JRIngallinera · 2026-09-23
- "Bitter Lesson" goes mainstream: a reading list on why SF AI circles believe what they believe — Afinetheorem · 2026-09-23