Learning the prior — AI Alignment Forum