Anthropic Researchers Make Major Breakthrough In Understanding How an AI Model Thinks