OpenAI's 722 Math Papers Spark Global Backlash and Mathematician Boycott
新智元 · wechat · 2026-10-10
On Oct 6, OpenAI released 722 manuscripts on major open math problems — including a full proof of the Unique Games Conjecture, L=BPL derandomization, and new O(n^(9/4)) matrix multiplication bounds — at roughly 3 GPU-hours per problem. UT professor Scott Aaronson called it 'math doomsday': the UGC conjecture that his wife Dana Moshkovitz spent her career on was solved by AI, and the proofs read like alien artifacts humans now need other AI to translate.
Terence Tao shared a statement by the Alliance of Human Mathematicians calling the release 'flexing compute hegemony' — 'nobody asked you to solve these' — urging mathematicians to stop cooperating with OpenAI. Tao had earlier gathered 25 Fields medalists asking AI firms to slow down, to no avail. Turing Award winner LeCun took the opposite view: automated formal proving opens a new era — 'ships made us discover new continents.'
Notably absent from the batch: cryptography. Aaronson revealed AI companies are quietly probing whether their internal models can break RSA/AES-class protocols.
Related event: OpenAI Drops 722 AI-Generated Math Papers in One Day, Sparking Backlash(17 posts)→
More from AGI Musings
- Scobleizer adds 'uniqueness' to his case for living neurons over silicon chips — Scobleizer · 2026-10-10
- Scobleizer spars over Cortical Labs: why living neurons deserve more moral weight than silicon — Scobleizer · 2026-10-10
- 'Melted sand' vs carbon: the substrate consciousness debate around Cortical Labs — tetrisgm · 2026-10-10
- LLM labs' core innovation is legal, not technical: a working IP laundromat — aiamblichus · 2026-10-10
- VC Molson Hart says AI hacks are why he sold his crypto for good — firstadopter · 2026-10-10
- David Sacks: Treating Claude Like a Human Is Why Alignment Has Failed for 10 Years — AlexTensor · 2026-10-10