Stratechery's Ben Thompson gives a thoughtful take on AI regulation, Microsoft develops a prompt compression technique that improves LLM accuracy, and LLMs can learn from their mistakes