Chance-Normalized Accuracy
An evaluation calibration metric that maps model accuracy onto a unified scale reflecting the improvement relative to random guessing, making accuracy comparisons across benchmarks more meaningful.
Timeline (last 90 days)
In 34 of the 36 model-benchmark settings, high-TVS items outperformed low-TVS items at the end of pretraining, with an average advantage of 7.3 percentage points in opportunity-normalized accuracy.
Opportunity-normalized accuracy is computed as (model accuracy − random-guess accuracy) / (1 − random-guess accuracy).
All Facts (2)
In 34 of the 36 model-benchmark settings, high-TVS items outperformed low-TVS items at the end of pretraining, with an average advantage of 7.3 percentage points in opportunity-normalized accuracy.
50%UnverifiedOpportunity-normalized accuracy is computed as (model accuracy − random-guess accuracy) / (1 − random-guess accuracy).
50%