[Intro to brain-like-AGI safety] 5. The “long-term predictor”, and TD learning — AI Alignment Forum