Against empathy-by-default — AI Alignment Forum