You're building a system where a language model needs to answer questions using a specific knowledge base instead of relying only on its pretraining. You are considering Retrieval-Augmented Generation as the approach.
What is Retrieval-Augmented Generation (RAG), and when would you use it?