“Constitutional AI uses Reinforcement Learning from AI Feedback (RLAIF), where a model ranks responses based on a principle to generate preference data, replacing the need for human labelers for that task.”