evhub comments on Robustness of Model-Graded Evaluations and Automated Interpretability

[evhub]

(Moderation note: added to the Alignment Forum from LessWrong.)