Should we discount extraordinary implications? — AI Alignment Forum