Tao warns AI is killing open science after OpenAI accused of scooping NYU mathematicians
新智元 · wechat · 2026-09-10
What happened
- NYU mathematicians Tristan Buckmaster and Leven Alpöge spent months using AI to attack the Navier-Stokes equations, one of the Millennium Prize Problems. Buckmaster publicly alleged that OpenAI learned of their promising approach in the final days, threw massive compute at the same direction, and scooped them — then proposed a joint publication on the condition that their Anthropic collaborator be removed from the author list.
- HuggingFace founder Thom Wolf blasted the move on X; OpenAI researcher Sebastien Bubeck called the accusations false. The most plausible account so far: Buckmaster's approach failed, OpenAI's succeeded, but its method may not have been its own invention.
Tao's warning
- Fields Medalist Terence Tao published a long essay arguing AI acts like a steam bulldozer flattening mathematics' "difficulty terrain," erasing the structure that guides mathematicians toward promising problems.
- The deeper crisis is incentive-driven: merely hearing that someone is exploring a direction can trigger a compute-fueled scoop, so researchers will stop sharing ideas — reversing centuries of open science.
- He called for new social norms rejecting black-box brute-force "answers" that yield no new insight, a view echoed by François Chollet and Gary Marcus.
Related event: Mathematicians Accuse OpenAI of Scooping Their Unpublished Work(2 posts)→
More from AGI Musings
- Reporter: Anthropic Is a 'Bubble Inside a Bubble' Rarely Discussing Practical AI — sebkrier · 2026-09-10
- Dev's Take on AI Safety: Mature Models Know Their Limits, Not Just Look Smart — AryHHAry · 2026-09-10
- Marc Andreessen slams AI Doomers chasing social credit, claims Anthropic will fake bioweapons risk — beffjezos · 2026-09-10
- "Extinction Is Scary, But Solve Our Own Math Problems?": e/acc's Sharpest Jab at AI Doomers — Miles_Brundage · 2026-09-10
- Developer says AI safety community's grey goo fears undermine its credibility — Darpinian · 2026-09-10
- AI safety debate: handing ASI control 'to beat China' shows a failure of imagination — astralmatrix · 2026-09-10