CH 06 · Prompt Architecture · 10/10
Read as one pageTesting prompts
Prompts deserve evaluation like any other component (Chapter 15). Maintain a set of representative inputs and check that the model's behaviour matches the rules: does it cite correctly, does it apply the template when triggered, does it avoid the forbidden outputs? When you change a rule, re-run the set to catch regressions. Prompt changes are deploys; treat them with the same care.
Finished this idea?
Your learning progress stays in this browser.