Video narration
How can an AI answer questions about information it never learned during training? Imagine an open-book test. A chatbot works closed-book, answering from its existing knowledge. So private files, recent updates, or changing policies may be missing. Then the setup flips. Before answering, the AI gets relevant pages. First, retrieval finds useful documents. Those pages are added to the prompt as context-the material the model can read now. Then it generates an answer using that context, not memory alone. This is Retrieval-Augmented Generation, or RAG. It isn't retraining. The model stays the same; the reference material changes for each question. RAG gives access to private or recent information and can ground the answer in pages, though poor material can still mislead it. So how can AI answer beyond training? Give it an open book: retrieve relevant documents, add them to the prompt, and generate from that context. That's RAG.