A company supplies examples to make a model follow its preferred response format. How does this differ from giving the model a fresh document at request time?

1. Fine-tuning guarantees the model knows all future external facts.
2. Fine-tuning can adapt learned behavior; request context supplies information for that invocation.
3. Every document in a prompt permanently changes the model weights.
4. Prompt examples and training data are always stored and governed identically in every service.