Calibration

BioMysteryBench (human-solvable) → CritPt

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

CritPt is estimated from BioMysteryBench (human-solvable) with a linear curve fitted on 5 models measured on both: y = 5.9615·x + -5.0502, R² = 0.67, cross-validated error 5.7 pp. It is used for 0 estimates.