Claude 2.1
Anthropic
partial detail 14 method fields published
Evaluation
Factual accuracy recall evaluation (Claude 2.1)
whether the model gets facts right from memory on a large set of complex factual questions probing known weaknesses; rubric distinguishes incorrect claims from admissions of uncertainty
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.