Few-Shot Learning
Few-shot learning is learning a task from a small number of examples. In language-model prompting, few-shot examples go into the prompt without changing the model's weights; the broader machine-learning term also covers methods that train or adapt models from limited examples.
Few-shot learning describes the small number of examples available for a task, not one particular way of updating a model. Google’s machine-learning glossary distinguishes the broader learning problem from few-shot prompting.
For an LLM, few-shot prompting supplies several demonstrations in the input. For example, a support-ticket classifier might see “charged twice: billing” and “cannot sign in: account access” before classifying a new ticket. The model uses those demonstrations during inference; this does not update its weights. Brown and colleagues used this setup in Language Models are Few-Shot Learners.
Zero-shot prompting supplies the task instruction without demonstrations. It is a comparison to few-shot prompting, not an alias for few-shot learning. In-context learning describes learning from the supplied context; it is not a synonym for every method covered by few-shot learning.
For a builder choosing between prompts, compare the same held-out tasks with and without examples. Examples can clarify labels or output formats, but they also consume context window space. Measure whether they help your task. Fine-tuning changes weights and is a separate option, rather than a mandatory next step after adding examples.