1 result for: open-data

A position paper argues AI evaluation must publish item-level benchmark responses and ships OpenEval - 10M model responses across 155k items - to prove the point.