A broad prompt can produce a polished response while leaving the real task unresolved. The model needs useful context, but the person reviewing the answer also needs a clear account of what was requested.
State the job, the relevant input, and the expected form of the result. Include important constraints and explain what should remain unknown rather than filled in. A small example can clarify a format or tone that is otherwise difficult to describe.
Review the output against the request and revise the prompt around the first meaningful mismatch. A longer prompt is not automatically a better one. The practical goal is a clear task and a result whose quality can be inspected.
Imagine a practical setting.
Imagine asking for a short introduction that uses only the supplied notes and identifies missing information. The request gives the draft a clearer boundary.
Before the next experiment.
Give review a specific responsibility. Checking a source, testing an instruction, and editing a sentence are different jobs and may need different evidence.
- State the job and relevant input.
- Describe the expected result.
- Leave unknown facts explicit.
Follow a related question
Define the shared starting condition.
Compare groups that share a startChoose one question the trial can answer.
Try a big idea on a small scaleKeep learning
Related background to continue exploring this subject.
Google: prompt design strategies Google: an introduction to language models
