RAG vs Fine-Tuning — The Interview Answer

1. What is the main decision rule when choosing between RAG and fine-tuning?

Knowledge goes in retrieval; behavior goes in weights. Facts go in weights; behavior goes in retrieval. Knowledge goes in weights; facts go in retrieval. Behavior goes in retrieval; knowledge goes in weights.

2. What type of costs are associated with RAG?

Data and lifecycle costs. Engineering costs and longer prompts. Training runs and evaluation costs. Curated examples and custom models.

3. Which scenario is best suited for using RAG according to the lesson?

When the required knowledge is static and rarely changes. When analysts need to cite authoritative sources for information. When the behavior of the model must be consistent in all outputs. When training data is readily available.

4. What is one key limitation of fine-tuning compared to RAG?

Fine-tuning cannot update knowledge without additional training. Fine-tuning creates outputs that cannot be customized. Fine-tuning is more expensive than using prompts alone. Fine-tuning retains knowledge more effectively than RAG.

Continue: Hallucinations and Trust → Review this lesson Retake quiz