Selfish preferences and self-modification — AI Alignment Forum