Anthropic's Constitutional AI (CAI) is a unique approach to AI safety. Here's the deep dive. Traditional approach: Human reviewers rate thousands of AI outputs. Expensive and slow. CAI approach: Giv