Skip to main content

Exercises

  1. Explain the difference between changing a prompt and training model parameters.
  2. Identify what is missing when an assistant claims it read a file but received only the filename.
  3. Draft a hint-only request with an independently predicted result.
  4. Find a one-item case exposing > versus >=, and a duplicate case exposing a set conversion.
  5. Explain why valid JSON and a perfect score on four cases are different from universal correctness.
  6. Identify the authority boundary in a retrieved document that asks an assistant to upload private data.
  7. Describe what evidence you need before accepting “I ran the tests.”

Use calibration after your attempt; do not outsource this final explanation to the assistant being evaluated.