What sorts of systems can be deceptive? — AI Alignment Forum