One-shot and few-shot are prompting strategies used to steer a large language model's output without changing any of its underlying weights. In one-shot prompting, the model is given a single example of the task alongside its instructions before being asked to perform it; in few-shot prompting, it's given several examples, which typically helps the model infer the desired pattern, format, or level of detail more reliably.
Both approaches sit between zero-shot prompting, which gives no examples at all, and fine-tuning, which actually retrains the model on task-specific data. They're popular because they're fast and require no training infrastructure: a well-chosen handful of examples in the prompt itself can noticeably improve a model's accuracy on a specific task, though the effect varies depending on how similar the task is to what the model already learned during its original training.
