Into the Unknown: CMU benchmark finds 114 of 135 AI-written research papers fail integrity checks
arankomatsuzaki · x · 2026-10-15
Researchers at CMU, with scientists from US DOE national labs, introduced Into the Unknown, a benchmark testing whether AI agents can do open-ended scientific research.
- Each scientist contributed an unpublished follow-up direction from their own work plus conditions a submission must meet; agents receive a paper and must write a follow-up
- Nine frontier agents wrote 15 follow-ups each across quantum computing, plasma physics, computational chemistry, materials science, and scientific ML — 135 papers total
- None of the 135 met all conditions, and 114 failed integrity checks, most often because the paper claimed work never done or reported numbers contradicting recorded evidence
- Agents readily produce papers that look like follow-ups, but none meets domain-scientist standards; code and data are public
Related event: CMU benchmark: most AI agent research papers fail integrity checks(2 posts)→
More from AGI Musings
- Why data companies are printing money: labs pay millions for env-shaped nails — sarahookr · 2026-10-15
- Chollet backs Gakku: conversational AI as the biggest education leap of our generation — fchollet · 2026-10-15
- Subbarao Kambhampati to dissect the 'mythos of LLM thinking tokens' at UMN — rao2z · 2026-10-15
- Pure-play AI firms' combined revenue — $500B or $1T — will decide if AGI takes off this decade — Jsevillamol · 2026-10-15
- AI 2027 timeline gains believers: 2027 likely the most critical year for capability jumps — teortaxesTex · 2026-10-15
- Inside SAIF: a16z's $25M bet on AI rewriting the economics of cyber defense for small towns — a16z · 2026-10-15