Calibration

GDP.pdf → HLE w/ tools

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

HLE w/ tools is estimated from GDP.pdf with a offset logistic curve fitted on 8 models measured on both: y = 0.0000 + (0.6121 − 0.0000) / (1 + exp(−200.00·(x − 0.0477))), R² = 0.80, cross-validated error 8.9 pp. It is used for 0 estimates.