You are the human annotator. Click the response you prefer — watch the reward model's training loss decrease as you add labels.