We’ve talked about Retrieval-augmented generation (RAG) a lot in this newsletter and on our channel. It has become a popular technique for enhancing large language models with access to external knowledge. However, basic RAG approaches have limitations.