A broad request can leave the tool generating material long after the useful part of the task is complete. It can also make review difficult because the desired result was never defined clearly.
Choose a small job with an identifiable output and a way to check it. State when the process should stop or ask for missing information. Keep actions with meaningful consequences subject to the level of review appropriate to the task.
Evaluate the experiment before expanding it. A narrower scope may remain the best use of the tool. A clear stop makes the contribution easier to understand and leaves the next decision with a person who has the relevant context.
A small working example.
Imagine asking an assistant to suggest headings, then stopping for an editor to choose one. The bounded task keeps the decision in an understandable place.
A note to keep beside it.
Keep the origin of a claim close to the claim itself. A generated explanation and an independently checked source play different roles in understanding a result.
- Define a small output you can check.
- State when to stop or ask for information.
- Review the result before expanding the task.
Follow a related question
Write the question beside the idea.
Capture the question firstDefine the understood inputs.
Where automation should hand backKeep learning
Related background to continue exploring this subject.
NIST: AI risk management framework Google: prompt design strategies
