x
Better priors as a safety problem — AI Alignment Forum