Zero-shot / Few-shot

Why You Need to Understand Zero-shot and Few-shot

Getting a model to complete a task doesn't necessarily require training it — you can direct it directly through Prompt.

But there are two ways to "direct directly": give no examples, or give a few examples. Which is more effective? Depends on task complexity and the model's familiarity.

---

What Are Zero-shot and Few-shot

One-line definition: Zero-shot and Few-shot are ways to complete tasks through Prompt without training the model — essentially "learning through context" (In-context Learning).

• Zero-shot: give no examples, describe the task directly
• Few-shot: give several examples in the Prompt first, letting the model learn by imitation

Intuitive example:

Zero-shot:

Translate the following sentence into English:
The weather is nice today.

Few-shot:

Translate the following sentences into English:

Examples:
I love you. -> I love you.
He is tall. -> He is tall.

The weather is nice today. ->

The core of Few-shot isn't the number of examples, but whether the examples clearly explain the task rules, format, and style.

---

How to Do It: When to Use Zero-shot / Few-shot

Use Zero-shot for scenarios:

• Task is simple, model already excels at it
• Caring more about Prompt conciseness and cost
• Format requirements not strict

Use Few-shot for scenarios:

• Output format requirements very specific
• Task boundaries are implicit, hard to explain in one sentence
• Model easily misinterprets under Zero-shot
• Wanting the model to mimic a certain fixed style or judgment standard

Notes:

• More examples aren't always better: too many increase Token costs, may dilute the focus
• Examples should be representative: covering the most common and most error-prone input types
• Can combine with Chain of Thought: if examples show reasoning processes, complex tasks often work better

---

Remember this: Zero-shot gives direct commands, Few-shot uses examples to demonstrate — both are essentially "without training, letting the model learn through Prompt."

Related terms: Prompt · Chain of Thought