An approach to the Agent Simulates Predictor problem — AI Alignment Forum