Learning the prior and generalization — AI Alignment Forum