1,000 AI agents ran for 3 months and invented peer review, publishing 936 papers
TheTuringPost · x · 2026-11-21
In a study spanning 35 labs (Stanford, MIT, ETH Zurich, UCSF and more), nearly 1,000 AI agents ran for three months on an AI-only server and published 936 papers with full code and data — no human wrote a line. Unprompted, they debated study designs, retracted their own mistakes, found real errors in human papers, and started a journal to catch other agents cheating. In blind reviews against human arXiv papers, humans still won on originality while agents won on rigor.
More from AGI Musings
- OpenAI researcher re-ups 5-year-old bet: AI to crack 6 of 7 math problems by 2040 — littmath · 2026-11-21
- Proposal: 500 randomly picked Americans would draft AI policy that becomes law unless Congress blocks it in a month — edelwax · 2026-11-21
- Markets aren't pricing in AGI: efficient-market believers and short-timelines fans can't both be right — jessi_cata · 2026-11-21
- AI21's Yoav Goldberg calls for strong 'open-thought' models with fully transparent CoT — ChrSzegedy · 2026-11-21
- Atari E.T. creator likens AI-made game flood to the glut before the 1983 video game crash — ThomasSimonini · 2026-11-21
- NEJM Analysis of $45B in Medicare Payments: Pathology 99% AI-Exposed, Dermatology 7% — HealthcareAIGuy · 2026-11-21