Lesson 1 of 6
Prompt vs RAG vs fine-tune
6 min read
Your model's answers aren't good enough. The instinct is to fine-tune. Nine times out of ten, that's the expensive wrong answer — so what's the right one?
Match the gap to the tool
Three tools fix three different gaps. Prompt engineering changes what you ask for. RAG feeds the model facts it doesn't have. Fine-tuning changes how the model behaves. The mistake is reaching for the heaviest tool first — fine-tuning is slow, costly, and freezes in knowledge that goes stale.
Ask two questions in order: is this a knowledge gap or a behavior gap? Knowledge → RAG. Behavior → maybe fine-tune. Neither → just prompt better.
The cheapest tool that works
Start at the bottom of the ladder and climb only when you must. Prompting is free and iterates in minutes. RAG grounds answers in your own documents without touching the weights. Fine-tuning is the last rung — reach for it only when you need a consistent new behavior and you have the labeled data to teach it.
Fine-tuning teaches form, not facts. If you need the model to know something, retrieve it; if you need it to consistently act a certain way, tune it.
You can combine them. A common production stack is a fine-tuned model for tone and format, with RAG supplying the up-to-date facts at query time.
The shape of it
- —Knowledge gap (new or private facts) → RAG, not fine-tuning.
- —Behavior gap (style, format, consistency) → fine-tune, but only with data.
- —Always try prompting first — it's the cheapest thing that might work.
You need the model to answer using your company's internal wiki, which changes weekly. Best first tool?
Continue in the app
Take the whole Fine-tuning & Customization course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.