The Adaptation Toolbox

1. What does the system prompt primarily change in a model's performance?

The model's weights The behavior of the model The knowledge available to the model The task demonstration format

2. When are few-shot examples particularly useful?

When training a model from scratch When demonstrating a specific task For changing underlying knowledge For permanent model updates

3. What is the primary benefit of using RAG (Retrieval-Augmented Generation)?

Modifying the model's weights Injecting up-to-date knowledge at query time Changing the output format consistently Demonstrating examples of desired outputs

4. Why might fine-tuning be the wrong choice for integrating policy documents into a model?

It can change behavior but not facts It creates outdated weights It makes knowledge unverifiable All of the above

Continue: Prompt Injection Drills → Review this lesson Retake quiz