Few-shot learning is when you include a small number of input/output examples directly in your prompt to demonstrate the task. You're not changing any weights — the model learns the pattern from examples in context, then applies it to your actual input.
It works surprisingly well and is almost always worth trying before fine-tuning. Three to five good examples typically outperform three paragraphs of instruction prose for tasks with a specific output format. The examples need to be representative, correctly labeled, and formatted exactly the way you want the output.
Common pitfall: using unbalanced examples (all positive examples for a classification task, no negatives). The model will learn to always predict the majority class. Randomize and balance your few-shot set.
Bring this to your business
Knowing the term is one thing. Shipping it is another.
We do two-week AI Sprints — one term, one workflow, into production by Day 10.