NYU Defines "Interestingness" Metric for LLMs to Discover Math

NYU's Julia Kempe group proposed a quantifiable "interestingness" metric—the ratio of proof length to statement length—and trained a 27B model that beats frontier models at predicting proof difficulty, boosting interestingness by 4.3x and enabling LLMs to autonomously discover interesting theorems.

2026-09-26 ~ 2026-09-26 · 3 related posts

• Researchers teach LLMs to find interesting theorems, boosting interestingness 4.3x — CatAstro_Piyush · 2026-09-26
• Paper defines theorem interestingness metric, 27B model beats frontier models at proof difficulty — Pascallisch · 2026-09-26

1 near-duplicate retellings: nyuniversity