[Paper Presentation] Enhancing Large Language Models by Integrating Human Preferences and Conditional Reinforcement Learning