Show, Don't Tell: Prompting With Examples
Use a few well chosen examples to get the format, tone and detail you want without long explanations.
A taste of a lesson
I gave it three example product blurbs and now every new one starts with 'Discover'. How do I stop that?
That is the model copying a surface pattern, which is very common. If all your examples open with similar words, it treats that opening as part of the rule. Rewrite your examples so each starts differently, for instance one with the main benefit, one with a question, one with a concrete detail. You can also add a short instruction: 'Vary the opening line.' Then run five new blurbs and check. Try it now: rewrite just the first line of each of your three examples so no two begin the same way.
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Write a few-shot prompt with clearly labelled examples
- Choose varied examples that teach the pattern rather than the topic
- Spot when the model copies quirks or reuses content from examples
- Decide when examples help and when instructions alone are better
Lesson plan
- 1 Describing versus showing See how examples communicate format and tone faster than instructions. Start
- 2 Building a clean example set Format examples so the model clearly sees the pattern and the real request. Start
- 3 Variety and edge cases Choose examples that teach the general pattern instead of one topic. Start
- 4 Side effects to watch for Catch copying, leakage and bias introduced by your examples. Start
- 5 Examples for sorting and labelling Use few-shot prompts to classify messages consistently. Start
- 6 Your example library Save example sets you can reuse for recurring tasks. Start
Try asking
About this tutor
Some things are hard to describe but easy to show: the tone of your product descriptions, the way your team labels support tickets, the exact layout of a summary. This tutor teaches few-shot prompting, which means giving the model a handful of examples of input and output before your real request. You will learn how many examples to use, how to choose varied ones, how to stop the model copying their surface quirks, and when examples hurt more than they help. Each lesson uses tasks you bring, so by the end you have a small library of example sets for your own recurring work.
Reviews
4.5
2 ratingsSample
- Sofia R.Sample
We label support emails into six buckets. Balancing the examples across categories fixed the bias towards 'billing' that I had not even noticed.
- Tomasz W.Sample
Clear and practical. The content leakage lesson was a surprise, it had been copying a fake customer name from my example into real replies.
About the teacher
Teaches prompting one technique at a time, with plenty of before and after examples
9 tutors 339 lessons taught Sample
I teach people to write prompts the way I once taught people to write briefs: say what you want, who it is for, and what good looks like. I came to AI from editing and corporate training, so I care more about clear thinking than clever tricks. In my lessons we take one technique at a time, try it on...
See Chiara's profile and tutorsMore like this
Other tutors on the same or nearby topics.