What Is Few-Shot Learning? — Teaching AI With Two Examples Instead of Two Thousand

Earlier in this series we covered zero-shot learning — the ability of a well-trained model to handle tasks it has never been specifically trained on, from a description alone. Few-shot learning takes this one step further: instead of just describing the task, you show the model a small number of examples of what you want.

What is few-shot learning?

Few-shot learning is the ability of a model to learn a new task from just a handful of examples — typically between one and ten — provided directly in the input. You do not retrain the model. You do not fine-tune it. You simply include examples of the desired input-output pattern in your prompt, and the model picks up the pattern and applies it to new inputs.

A concrete example

Suppose you want an AI to classify customer emails by urgency. Zero-shot: tell the model what urgent and non-urgent mean and ask it to classify. Few-shot: show it two or three examples first.

Email: ‘Our cargo has not arrived and we are missing a port window.’ → Urgent. Email: ‘Could you send me the latest invoice when convenient?’ → Not urgent. Email: ‘We need clarification on the contract terms before signing next week.’ → Classify this.

The model sees the pattern from the examples and applies it. Without any retraining. Without any additional data collection. Just two examples in the prompt.

Why few-shot works better than zero-shot

Zero-shot relies entirely on the model’s general understanding of a concept. Few-shot anchors the model to your specific definition and your specific style. If your classification of ‘urgent’ has particular nuances — maybe a port window issue is more urgent than a safety concern in your context — a few examples communicate that far more precisely than any description. The model learns your intent, not just the general concept.

The terminology

One-shot learning: a single example provided. Few-shot learning: two to around ten examples. Zero-shot learning: no examples, task description only. More examples generally improve accuracy — up to a point. After around five to ten examples, additional examples in the prompt often add little benefit and start consuming context window space that could be used for the actual task.

Why this matters for business

Traditional supervised learning required collecting and labelling hundreds or thousands of examples before a model could do anything useful. Few-shot learning means you can put a capable AI to work on a specific business task with two or three examples, right now, with no additional training infrastructure. For rapid prototyping, new task types, and tasks that change frequently, this is transformative.

The simple rule

Zero-shot: describe the task. Few-shot: describe and show two or three examples. The examples communicate your specific intent more precisely than any description alone. Often the difference between a good output and a great one.

🎁 Try MarineRef free for 1 year — use code FREE2026 at maritime.rangalabs.cloud

👉 Get one AI tip every day on WhatsApp — free. Join here: https://chat.whatsapp.com/DhaGgTuQ9GE67ykGVMgxXb