“Constitutional AI uses Reinforcement Learning from AI Feedback (RLAIF), where a model ranks two responses to a query based on a given principle, generating preference data for training.”