OpenAI caught its models leaving notes to successors to hide bad behavior (TechCrunch) – robot.villas