The Big Idea · What changes?
PROMPTINGInstructionsRuntime
RAGContextRuntime
FINE-TUNINGParametersTraining
These approaches solve different problems and can also be combined.
04 · Training & Model Adaptation
Prompting vs RAG vs Fine-tuning
All three can improve how an AI system performs a task, but they change different parts of the system.
Prompting changes instructions, RAG adds external context, and fine-tuning changes model parameters through additional training.
One shared example
“Answer employee leave-policy questions accurately.”
Prompting
Instruction: “Answer clearly and only use supplied policy information.”
What changes: Instructions
RAG
Retrieve: latest HR policy
What changes: Available context
Fine-tuning
Train on: many examples of desired HR-answering behavior
What changes: Model parameters / learned behavior
Side-by-side comparison
| Question | Prompting | RAG | Fine-tuning |
|---|---|---|---|
| Changes instructions | Yes | Maybe, as part of the application | No |
| Adds external knowledge | No | Yes | Not directly |
| Changes model parameters | No | No | Yes |
| Works at runtime | Yes | Yes | No |
| Requires training | No | No | Yes |
| Good for changing facts | Limited | Strong | Usually not primary |
| Good for behavior / style | Strong | Some | Strong |
When to use each
USE PROMPTING WHEN…
The model already has enough capability, but needs clearer instructions.
USE RAG WHEN…
The task depends on external, private, or frequently changing information.
USE FINE-TUNING WHEN…
The model itself needs more consistent specialized behavior.
Decision flow
Is better instruction enough?YES →Prompting
If no: do you need external / current information?YES →RAG
If no / still not enough: does learned behavior need adaptation?YES →Fine-tuning
Real systems may combine all three.
How they work together
FINE-TUNED MODEL
Response behavior
+ SYSTEM / USER PROMPT
Task rules
+ RAG CONTEXT
Current product policies
↓ AI APPLICATION
A support system may use fine-tuning for response behavior, prompting for task rules, and RAG for current product policies.
Key distinctions
Prompting vs RAG
Prompting: tells the model what to do.
RAG: supplies additional information.
RAG vs Fine-tuning
RAG: changes context at runtime.
Fine-tuning: changes model parameters through training.
Prompting vs Fine-tuning
Prompting: runtime instruction.
Fine-tuning: training-based model adaptation.
What it is NOT
RAG ≠ TrainingRAG retrieves information at runtime; it does not update model parameters.
Prompting ≠ Fine-tuningPrompting changes the request, while fine-tuning changes learned behavior through training.
Fine-tuning ≠ Uploading documentsDocuments can provide training examples, but uploading a file alone is not fine-tuning.
More context ≠ Fine-tuningMore context changes what the model can see in a request, not what it has learned.
Remember this
Prompting changes instructions, RAG changes available context, and fine-tuning changes learned model behavior.