K-Culture GlossaryㅍWords you meet while using AI
Few-shot · Zero-shot
The ability to do a new task after seeing just a few examples (few-shot), or with no examples at all (zero-shot).
In plain words
When you ask an AI to do something new, showing it a few examples first is called few-shot. Giving it just an instruction with no examples is zero-shot. "Sort these into 3 categories like this — example: ..." is few-shot. "Classify the sentiment as positive or negative" is zero-shot.
This was one of the amazing things about large language models: older AI systems needed retraining for every new task, but LLMs can pick up a new task on the spot from just a handful of examples. It's also a practical tip — if you don't like the output you're getting, try attaching one or two examples of the result you actually want.
Try it yourself
- Try it with no examples: "Classify this review: 'Shipping was fast, but the box arrived crushed.'" — the AI will probably use a generic standard like positive/negative.
- Now add examples: "Classify it like these examples. 'Tastes good but the portion is small' → Disappointing / 'Would definitely buy again' → Satisfied / Now this one: 'Shipping was fast, but the box arrived crushed.'"
- With just two examples, the classification criteria itself shifts to the categories you defined (Satisfied/Disappointing) — that's the power of few-shot, and a real-world trick: if you have a specific format in mind, show examples of it.