OpenAI Tells Evaluators: The Harness Is Part of the Result
OpenAI's new evaluation playbook says harness and budget choices can swing frontier model scores by up to 59%, and reports must disclose what claim they actually support.
Topic
Metr
Topic
Topic
Topic
OpenAI's new evaluation playbook says harness and budget choices can swing frontier model scores by up to 59%, and reports must disclose what claim they actually support.