Calibration

BioMysteryBench (human-difficult) → CritPt

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

CritPt is estimated from BioMysteryBench (human-difficult) with a offset logistic curve fitted on 6 models measured on both: y = 0.0421 + (0.2779 − 0.0421) / (1 + exp(−200.00·(x − 0.4350))), R² = 0.77, cross-validated error 11.9 pp. It is used for 0 estimates.