RAG Meets LLMs: Towards Retrieval-Augmented Large Language Models