Few-shot learning guides an AI model by placing a small number of examples directly in the prompt. Multiple examples reveal both the expected pattern and acceptable variations, making the approach useful for subtle or judgment-heavy tasks. section 4.2.3
Use few-shot prompting when you need to:
- Show several examples instead of relying on a single demonstration. section 4.2
- Teach a temporary style or preference, even when that style is not part of the model’s original training. section 4.2.3
- Define how results should be presented as models become better at handling basic instructions. section 4.2.5
Two poem styles
Provide examples for two invented forms: a “Halo” poem about nature using a 5-7-5 syllable structure, and a humorous “Lemon” poem using an AABBA rhyme scheme. Then ask the model to write both forms about a cat named Mittens that loves pizza and explain their differences. section 4.2.3
