Learning with catastrophes — AI Alignment Forum