Answering questions honestly given world-model mismatches — AI Alignment Forum