Study Links Share of Once-Seen Facts to LLM Scientific Discovery
A Johns Hopkins arXiv paper finds that the proportion of 'monofacts'—facts appearing only once—in training data determines an LLM's capacity for scientific discovery, validated in domains like materials science.
2026-11-04 ~ 2026-11-06 · 2 related posts
- Monofact rate in training data governs LLM scientific discovery, preprint finds — CatAstro_Piyush · 2026-11-04
- Johns Hopkins paper: LLMs make more scientific discoveries when training facts appear only once — rohanpaul_ai · 2026-11-06