Thought Experiments Provide a Third Anchor — AI Alignment Forum