Discover the Performance Gain of LLMs with Retrieval Augmented Generation (thenewstack.io)

Interesting results: combining Llama-13B with Wikipedia backed by a vector database can reduce hallucination, and significantly improves its performance on MMLU scores.