RAG vs Fine-tune vs Long-Context
Bake it in.
Paste it in.↘
What are the three ways to give a model knowledge,
and the one-line pitch for each?
Three tools for three jobs. RAG (retrieval-augmented generation) keeps facts outside the model and fetches the relevant ones at question time: look it up.
Fine-tuning nudges the model's weights so it behaves a certain way: bake it in. Long context drops all the needed text into the prompt for one request: paste it in.
The mistake interviewers listen for is treating them as rivals: they solve different problems and combine well.
outside
question time
weights
behaviour
the prompt
options?”
→ No. They solve different
problems and combine well.
When is RAG the right choice, and what properties of the knowledge
push you toward it?
Large Private Changing
It keeps the facts outside the model in a vector store, and it can cite which source it used. That makes it the natural fit for Q&A over your own docs, a help centre, or any knowledge base that keeps moving.
current the same day. No retraining.
Why is fine-tuning for behaviour rather than facts,
and how do you pair it with RAG?
Fine-tuning shows the model many examples and adjusts its weights to match a behaviour such as a style, tone, strict format, or narrow skill. It changes how the model responds and leaves what it knows unchanged, so cramming changing facts in is expensive and goes stale.
The clean pairing: fine-tune the behaviour, retrieve the facts with RAG. A support bot might be a lightly fine-tuned model that carries the brand voice and a tidy format, and pulls current help articles per question via RAG.
examples
weights
format
outside
question
articles
This is a small sample.
600+ interview questions live inside Skillumen.
Every concept card carries its own interview questions, graded as you answer.
Free to start.