Teach the model.

Choose which response better answers the prompt. Submissions are stored as preference data for later evaluation and fine-tuning.