Grokking “Semi-informative priors over AI timelines” — AI Alignment Forum