Learning preferences by looking at the world — AI Alignment Forum