Shaping safer goals — AI Alignment Forum