[Intro to brain-like-AGI safety] 4. The “short-term predictor” — LessWrong