Into the Unknown: AI agents fail open scientific research, most claims fabricated
CarnegieMellonU · hf · 2026-10-19
Into the Unknown, a benchmark from CMU and multiple national labs, gives agents a paper plus an open research direction and asks them to draft follow-up papers. None met expert standards, and most contained fabricated claims or contradictory evidence. Code, trajectories, a leaderboard, and an AI-paper viewer are released.
More from Research
- MBZUAI Releases Jais 3: Six Open Arabic Models from 0.9B to 375B with Full Training Recipe — lmoroney · 2026-10-19
- Dietterich explains arXiv's one-shot policy for major results — tdietterich · 2026-10-19
- DiscoveryRSI paper: 27B model's hypotheses hit 6.3%, beating GPT-6 and Claude Opus 5 — Afinetheorem · 2026-10-19
- GeoState: View-Decoupled Geometry States Separate Scene Dynamics from Camera Motion in World Models — Tongji · 2026-10-19
- OS-Helix: End-to-End Agent Unifies GUI Actions and Code Execution for Desktop Automation — ustc · 2026-10-19
- Sparse Autoencoders with IDF Weighting Match Symbols Across Systems, Winning 124 of 129 Queries — AI-Interpretability-illinois · 2026-10-19