Concept guide · RAG foundations
When should you use RAG instead of fine-tuning?
A slower explanation for when the short interview answer is not enough. Start with the intuition, then connect it to the technical idea and the way an interviewer expects you to reason about it.
ML area
Generative AI → RAG versus fine-tuning
Interview format
Foundations · Concept
Question
GENAI-048
Start with the idea
In simple terms
Use RAG when the model needs facts that are private, large, or changing. Use fine-tuning when you need the model to behave differently, such as following a format or style, rather than memorising a frequently changing knowledge base.
Make it concrete
A product catalogue that changes every day belongs behind retrieval. A stable support-response format that the model repeatedly misses may be a fine-tuning candidate after prompting and examples have been tested.
Build the answer
How to reason about it
- 1Ask whether the problem is missing knowledge or unreliable behaviour.
- 2Prefer retrieval for changing, private, or citation-sensitive facts.
- 3Choose fine-tuning only after measuring a stable behaviour gap and its data cost.
Go one level deeper
The technical explanation
RAG changes the evidence available at inference time, which supports freshness, source references, deletion, and access filtering. Fine-tuning changes model behaviour through additional training and can improve consistency for a narrow task, but it does not create a reliable update mechanism for current facts. The two can be combined, but the choice should follow the failure being fixed.
Say this in the interview
Use RAG when knowledge is private, changing, large, or needs source citations. Use fine-tuning mainly for behaviour, format, style, or a narrow repeated task. They can be combined, but fine-tuning is not a reliable way to keep frequently changing facts current.
Keep preparing