RAG (Retrieval-Augmented Generation)

RAG retrieves relevant context before generation, which can improve factual grounding and reduce hallucinations when retrieval quality is high.

What to Watch in Practice

In RAG systems, answer quality is often decided less by the model itself than by how retrieval is designed. If chunking, embeddings, or search criteria are rough, the answer may look polished while the grounding is still weak.

In practice, it is usually more useful to frame this as a retrieval quality problem than as an LLM problem.

Practical Note

RAG (Retrieval-Augmented Generation) usually appears in contexts related to rag, llm, retrieval, ai. In practice, it helps to know not only the definition, but also what this term is trying to name quickly in a conversation, design note, or document.

Nearby words often overlap and make the explanation fuzzy. It is easier to use the term well when the target, role, and typical situation are kept one step more concrete.

Reading Note

The easiest way to read this term is to look at three things first: what it is about, what nearby concept it should be separated from, and what kind of decision it usually supports. For RAG (Retrieval-Augmented Generation), the rag, llm, retrieval, ai context is already a good starting point.

It also helps not to stop at the definition alone. The more useful view is to see what the term is trying to name quickly inside a working conversation.