Naive TDT, Bayes nets, and counterfactual mugging — AI Alignment Forum