On the estimation and validity of AI time horizons---a statistical look at the METR plot
Paper recomputes METR AI time horizons with splines and item-response theory across 228 tasks.
3 stories, newest first · one card per story
Paper recomputes METR AI time horizons with splines and item-response theory across 228 tasks.
Rubric Response Theory turns rubric verdicts into scalar RL rewards and cuts judge calls while beating GRPO.
SecProbe adaptively tests coding agents on synthesized vulnerability repairs; frontier models peak at 28.33% success.
Today 1K readers · 1.2K views · 106 reads · all time 211K views