Fine-tuning vs RAG for daily-updated docs

WHAT THIS TESTS This checks whether you can separate two often-confused techniques: fine-tuning changes a model's weights to shape behavior and style, while RAG supplies fresh external knowledge at inference time. The daily-update detail is the decisive clue.

A GOOD ANSWER COVERS RAG is the right primary choice. Because the documentation changes every day, you want a system where updating knowledge means re-indexing documents, not retraining a model. The architecture: split documents into chunks, generate an embedding vector for each chunk with an embedding model, and store those vectors in a vector database. At query time, embed the user's question, retrieve the top-k most similar chunks, and inject them into the prompt as context. The LLM then answers grounded in the retrieved text, ideally with citations back to source documents. This keeps answers current, reduces hallucination, and lets you update knowledge in minutes. Fine-tuning would bake yesterday's facts into the weights, going stale immediately and requiring expensive retraining each day.

COMMON WRONG ANSWERS Choosing fine-tuning for fresh facts; claiming fine-tuning prevents hallucination; ignoring chunking and embedding; suggesting you stuff the entire corpus into every prompt; not addressing how updates propagate.

LIKELY FOLLOW-UPS How do you chunk documents? What embedding model and vector store? How do you handle retrieval failures or no-match cases? When would you combine RAG with light fine-tuning for tone?

ONE CONCRETE EXAMPLE An engineer updates a policy document at noon. A nightly or near-real-time pipeline re-embeds the changed chunks and upserts them into the vector index. When a user asks about the policy that afternoon, the retriever pulls the new chunk and the model answers correctly, with a citation, without any model retraining.