27B Model Predicts Mathematical Interestingness via Proof-to-Statement Ratio
September 25, 2026
A 27B parameter model predicts mathematical interestingness by calculating the ratio of proof length to statement length, a metric that correlates with downstream utility. Optimizing for this difficulty primitive enables models to generate more useful theorems than frontier general-purpose models.
HOW THIS AFFECTS YOU
●
researcherYou can use proof-to-statement ratios as a proxy for evaluating the utility of automated theorem generation.