Glossary term
Glossary term
Agentic Systems
Providing a model with a small number of labelled examples in the prompt to guide output format or behavior.
A machine learning approach, often used for object classification, designed to train effective classification models from only a small number of training examples.
See also one-shot learning and zero-shot learning.
GitHub Copilot uses few-shot examples embedded in the prompt for code-completion tasks - showing 2-3 examples of the project's coding style so the model matches naming conventions and comment patterns.
GPT-4 is deployed by a data-entry automation company with 3 labelled examples of invoice-to-JSON parsing in the prompt, achieving 97% field-extraction accuracy without any fine-tuning.
A customer-support classification system uses 5-shot prompting to classify ticket intents - one example per category (billing, technical, account, complaint, general) - matching fine-tuned model accuracy at a fraction of the cost.
Created for this library
A legal-tech vendor uses few-shot learning to spin up new clause classifiers from a handful of labeled examples per client.
A medical AI team uses few-shot learning to extend its triage model to a new rare condition with only a few dozen labeled cases.
A customer support team uses few-shot learning to launch intent classifiers for new product features with a small set of curated labels.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License