RAG vs Fine-tuning vs Prompting

Question 1

Which approach best handles knowledge that changes in real time?

Accepted Answer

Retrieval augmentation, which fetches live documents at query time — RAG retrieves current data at query time, so it handles fresh/changing knowledge without retraining — unlike fine-tuning, which bakes in a snapshot.

Question 2

Prompt engineering is best when you need…

Accepted Answer

Fast iteration, no training needed — Prompting is the cheapest, fastest lever — no infrastructure or data pipeline — ideal for quick adaptation and prototyping.

Question 3

RAG is the right choice when you need to…

Accepted Answer

Injecting external, changing facts — RAG supplies up-to-date/proprietary facts via retrieval, keeping knowledge current by updating an index rather than the model.

Question 4

Fine-tuning is most appropriate when you need to…

Accepted Answer

Teaching behavior/style with labels — Fine-tuning adapts HOW the model responds (tone, format, task skill); it excels at behavior, not at injecting constantly-changing facts.

Question 5

The key distinction is that RAG adds ___ while fine-tuning changes ___.

Accepted Answer

knowledge access; model behavior — RAG = give the model the right information; fine-tuning = change how the model behaves. They address different problems.

Question 6

Why is fine-tuning a POOR way to add frequently-changing facts?

Accepted Answer

Each update to facts requires a full retraining cycle — Baking facts into weights is expensive to update and unreliable for precise recall; RAG handles changing knowledge far better.

Question 7

For a chatbot needing your company's LATEST documentation, the best approach is…

Accepted Answer

Retrieval augmentation with a live-synced document index — RAG keeps answers current by retrieving from a maintained index, avoiding constant re-training as documentation evolves.

Question 8

To make a model reliably output a specific JSON format or adopt a brand voice across all answers, consider…

Accepted Answer

Fine-tuning on labeled examples — Consistent format/style is a behavior; fine-tuning (or careful prompting + structured output) shapes it, whereas RAG supplies facts, not form.

Question 9

These three approaches are…

Accepted Answer

Complementary — stack them by need — Real systems mix them: prompt well, retrieve for knowledge, and optionally fine-tune for behavior — each addresses a different need.

Question 10

A cost/latency consideration for very long prompts is that…

Accepted Answer

Long context adds token cost and latency — If a behavior is needed on every call, baking it in via fine-tuning can reduce prompt length and per-call cost versus repeating long instructions.

Key insight

Which lever fixes this?

Test your knowledge

Which approach best handles knowledge that changes in real time?

The short version

Interview Questions

References & Further Reading

Related topics

Love this visualization?