Few-Shot Learning

Few-shot learning is the ability of an AI model to learn a new task from just a few examples — sometimes as few as one to five. This contrasts sharply with traditional machine learning, which typically requires thousands or millions of labelled examples.

How few-shot learning works

In the context of large language models, few-shot learning means providing a few examples of the desired input-output pattern in your prompt. The model recognises the pattern and applies it to new inputs:

Prompt: "Classify the sentiment. 'Great product, love it!' → Positive. 'Terrible experience.' → Negative. 'Works as expected, nothing special.' →"

The model has seen only two examples but can correctly classify the third as Neutral.

Types by number of examples

• Zero-shot — no examples provided. The model relies entirely on its training knowledge and your instructions.
• One-shot — a single example demonstrates the task.
• Few-shot — two to ten examples demonstrate the task.

Why few-shot learning is revolutionary

Before large language models, adapting AI to a new task meant collecting training data, labelling it, training a model, and evaluating it — a process taking weeks or months. Few-shot learning lets you prototype a task in minutes by crafting a prompt with examples.

When few-shot learning works well

• Classification tasks with clear categories
• Format transformation (converting between data formats)
• Style matching (writing in a specific tone or format)
• Simple extraction tasks (pulling specific information from text)

When you need more than few-shot

• Complex reasoning that requires deep domain knowledge
• Tasks where edge cases are common and important
• High-stakes applications where consistency is critical
• Tasks requiring knowledge the model was not trained on

Few-shot vs. fine-tuning

Few-shot learning is a prompting technique — no model weights are changed. Fine-tuning actually retrains the model on your examples. Fine-tuning produces more reliable, consistent results but is more expensive and time-consuming. Few-shot is the right starting point; fine-tune only when few-shot is insufficient.