x
In (highly contingent!) defense of interpretability-in-the-loop ML training — AI Alignment Forum