How evals might (or might not) prevent catastrophic risks from AI

By Akash @ 2023-02-07T20:16 (+28)

This is a crosspost, probably from LessWrong. Try viewing it there.