KMLinear DigressionsAI researchA Key Concept in AI Alignment: Deep Reinforcement Learning from Human Preferences