Framework
Zero-shot versus few-shot prompting
The question it answers
When a manager asks an AI tool to draft emails or sort customer feedback, is a plain instruction enough, or should they add a few worked examples to get the result they want?
- Origin
- Brown et al., 2020
- Diagram type
- Comparison table
- Level of analysis
- Individual
- Topics
- AI for leaders
Diagram
Zero-shot versus few-shot prompting
| Criterion | Zero-shot | Few-shot |
|---|---|---|
| Input | the instruction only | the instruction plus a few sample input-output pairs |
| Suits | tasks with an obvious meaning | a style or classification that is hard to describe |
| Risk | misreading the intent | poor examples skew the result |
| Example | classifying emails as urgent or routine | supplying an old report as a model to keep the tone |
Sources
References
Listed A–Z by author surname, Harvard style. A source with a DOI links to https://doi.org/…; one without links to its page or says plainly that it has no DOI.
- Brown, T.B., Mann, B. and Ryder, N. (2020) Language models are few-shot learners. (arXiv preprint arXiv:2005.14165).doi:10.48550/arXiv.2005.14165Source page