IN DEFENSE OF AI EVALS, FOR EVERYONE

Summary from the TLDR AI newsletter

Evals aren't about following a rigid philosophy - sometimes it's fine to be less rigorous, and sometimes the task demands more rigor.