56. Preference optimization and model alignment

Shape model behavior using human or AI feedback, reward models, preference optimization, red teaming, and post-training evaluations.