Distilled Human Judgment: Reifying AI Alignment — LessWrong