Few-shot prompting

Definition

Providing a model with a small number of examples in the prompt to guide its behavior, without any fine-tuning. Few-shot is a fast, low-cost way to adapt a general model to a specific task.

Why it matters

Few-shot prompting is the fastest path from idea to working prototype. Instead of spending weeks on fine-tuning, you drop 3-5 examples into your prompt and the model adapts instantly. This makes AI accessible to teams without ML engineering expertise. The strategic advantage is speed: you can test ten different AI features in a day using few-shot prompting, while fine-tuning each would take weeks. The limitation is context window cost, as examples consume tokens, and reliability: few-shot is less consistent than fine-tuning for production workloads. Smart teams use few-shot for rapid prototyping and graduate to fine-tuning only when they have validated the use case.

In practice

GPT-3's original paper (Brown et al., 2020) demonstrated that few-shot prompting could match fine-tuned models on many benchmarks. In practice, companies use few-shot for everything from sentiment classification to data extraction. A common pattern: include 5 examples of the desired input-output format in the system prompt, then process new inputs. Anthropic's Claude documentation recommends 3-5 examples for most tasks. The technique is often combined with chain-of-thought: showing the model examples of step-by-step reasoning produces better results than showing only input-output pairs.