Meta's IdeaScientist trains RL agents for scientific ideation, beating open baselines by 14%
Jeande_d · x · 2026-10-09
Researchers from a Meta internship present IdeaScientist, a framework that decomposes scientific ideation into gap finding, innovation, and report writing, training each role with reinforcement learning. It introduces the Svalbard Idea Vault, a 2.77M-idea corpus for retrieval and time-controlled evaluation. On Qwen3.6-27B it outperforms the strongest open autoresearch baseline by 14%, with a 25% novelty gain and 75–92% blind human preference win rates against six baselines; cross-domain retrieval substantially improves idea transfer across fields.
More from Research
- LangChain Founder: Evals Work for Narrow Tasks but Break Down for Autonomous Agents — hwchase17 · 2026-10-09
- Hugging Face launches Robotic Episodes Viewer for 24k+ LeRobot datasets — mishig25 · 2026-10-09
- Blind humanoid walks, plays soccer and lifts suitcases with joint encoders only — accepted at Humanoids 2026 — Jan_R_Peters · 2026-10-09
- Delete object info from observations and PPO learns to search anyway — TU Darmstadt on its Humanoids 2026 paper — Jan_R_Peters · 2026-10-09
- U-Space finds an interpretable subspace for LLM uncertainty, no training needed — Tobias Braun · 2026-10-09
- CARE certifies VLA inference speedups up to 10.8x with statistical guarantees — UMCP · 2026-10-09