Techniques & Methods
Few-Shot Learning
Few-shot learning enables models to quickly adapt to new tasks using just 2-10 examples, either through meta-learning (training specifically to learn fast) or in-context learning (providing examples in the prompt at inference time). GPT-3's surprising few-shot capabilities brought this technique mainstream.
For LLMs, few-shot prompting is the most practical form: include 2-5 examples of the desired input-output format in the prompt and the model generalizes the pattern. Performance improvements over zero-shot are often substantial, especially for structured output tasks.
Authority Links
Related Terms
Techniques & Methods
Zero-Shot Learning
Model's ability to correctly perform tasks it was not explicitly trained for.
Techniques & Methods
One-Shot Learning
Model's ability to learn and make accurate predictions from only a single example.
Techniques & Methods
Prompt Engineering
The discipline of designing input text — instructions, examples, constraints, and context — to reliably steer a language model toward accurate, well-formatted, and intent-aligned outputs without modifying model weights.