human evaluation

Type: Concept

Expert annotators deeply assess model outputs (fact-checking, code correctness, physics equations) as a preferred measure of progress over benchmarks/leaderboards.

Field: AI evaluation