[ASoT] Some thoughts about imperfect world modeling — AI Alignment Forum