Glossary · Term

Few shot

Also known as: few-shot, few-shot prompting

The method is to show them a few examples of the desired answer and then have them do the work. This is especially effective when formatting your answers.

Fewshot is a method of asking AI to do a task by first showing it a few examples of the desired answer and then having it do the actual task. It's like handing a new employee two or three good report samples and asking them to write something like this instead of giving a long verbal explanation.

LLM has the ability to identify patterns by just looking at examples in the prompt without any additional re-learning, and FewShot takes advantage of this. It is especially effective when the format, tone of voice, and classification criteria of the answer must be matched, so it is a basic prompting technique widely used in practical automation such as data organization or customer inquiry classification.

Ask without examples is called a zero shot, and the quality of the examples directly leads to the quality of the results. Giving ambiguous or contradictory examples can actually worsen the results, so care must be taken when choosing examples.

✅ Why it matters

You can match the desired format and standard with just a prompt without relearning
Easily improves the accuracy of practical tasks such as classification and format conversion
This is the most basic prompting technique that anyone can use right away.

⚠️ Limits and debates

As the example becomes longer, it takes up more context and costs more. If the quality of the example is poor or biased, the results also worsen. In complex inference tasks, examples alone can have limitations.