GLOSSARY
Few-Shot Prompting
Teaching by example inside the prompt: show the model a few input-output pairs, and it imitates the pattern on new inputs — no retraining.
Zero-shot means asking directly (“classify this review as positive or negative”); few-shot adds three or four worked examples first, and accuracy on the real task often jumps. The examples define format, tone, edge-case handling and the unstated rules that a description alone fails to convey — it is the cheapest and most underrated lever in applied prompting.
Costs scale with examples: every pair occupies context window and bills tokens forever. Later work (instruction tuning, fine-tuning) internalizes what few-shot demonstrates in-context, which is why mature products often start few-shot and graduate to a fine-tune when volume justifies it.