Free mini-course · Lesson 5 / 6
Improve a prompt through testing and feedback
The first answer is a starting point. Improving a prompt means identifying an unmet requirement and testing a targeted change.
Editorial team: Take Two ·
The idea
If you only ask “improve this”, the model may polish the writing without fixing the error. Describe the observed defect, state the criterion and change one thing at a time. For reusable prompts, compare versions in fresh conversations with the same inputs and, where possible, the same model. One good answer does not demonstrate reliability.
A vague request
This is wrong. Do better.
Why it is incomplete
There is no testable criterion. You cannot tell which change helped and might lose correct details.
A prompt to try
Try the prompt in your usual AI tool. Select and copy the text; no coding is needed. Access, features and limits depend on the tool and plan you use.
In the previous table, you assigned restore verification to Paolo, but the notes only say that he performed the backup.
Correct that field to “to be defined”.
Keep the other information supported by the notes.
Then list, in at most three points, what information we need before assigning the tasks.What to look for
The correction must target the unsupported assignment. Preserve the completed backup and ask for the missing restore verification details.
Now try it
Your exercise
Save two versions of the lesson 3 prompt: original A and B with “Distinguish the person who performed completed work from the owner of future work”. Test on three note sets: complete, missing deadlines and ambiguous ownership. Run each case twice.
Check the result
- Does each answer meet all the criteria for those notes?
- Does B reduce the error without introducing others?
- Have you recorded the input, model, date and prompt version?
See the commented answer
Make a sheet with six results per version: 3 cases × 2 runs. Mark yes/no for fidelity to the notes, completeness and format. Compare counts per criterion and keep the failures for the next revision. These few tests guide the work; they do not certify production quality.
Examples and commented answers are teaching material written for this course. They are not recorded model outputs; answers may vary.
Sources and further reading
The exercises are original Take Two teaching examples. For technical background on instructions, examples and evaluation criteria, see: