Collect Human Feedback For Fine-Tuning Chatgpt Models