Lesson 6 of 7
RAG vs fine-tuning
6 min read
You want the model to know your product and sound like your brand. One of those is a RAG problem and one is a fine-tuning problem. Which is which?
Knowledge vs behavior
RAG adds knowledge at question time — it hands the model facts to read, without changing the model. Fine-tuning changes the model itself — you keep training a foundation model on examples so it learns a new behavior: a format, a tone, a task it does reliably. RAG is about what it knows; fine-tuning is about how it acts.
RAG supplies knowledge at question time; fine-tuning reshapes behavior through training.
How to choose
Facts that change, or must be current and citable? That's RAG — update the document, not the weights. A consistent style, a strict output format, or a niche task the base model fumbles? That's fine-tuning. They aren't rivals: plenty of real systems fine-tune for behavior and use RAG for knowledge. Reach for RAG first — it's cheaper, faster to change, and easier to trace.
Changing facts → RAG. Consistent behavior or format → fine-tuning. Often you want both.
Fine-tuning bakes knowledge into the weights, where it's hard to update and impossible to cite. That's why it's the wrong tool for facts that change — and the right one for behavior that shouldn't.
The shape of it
- —RAG adds knowledge at question time without changing the model.
- —Fine-tuning changes behavior — tone, format, task — through training.
- —Changing or citable facts → RAG; consistent behavior → fine-tuning; often both.
Your assistant knows your products (they change weekly) but replies in a stiff, off-brand voice. What's the right split?
Continue in the app
Take the whole RAG & Search course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.