Reward Alignment

Adjusting model behavior via reinforcement learning or preference data to align outputs with human values or specific goals.