The Flexibility of Abstract Concepts — AI Alignment Forum