Code-mixed LLM: Improve Large Language Models' Capability to Handle Code-Mixing through Reinforcement Learning from AI Feedback