zero-shot prompting
**Zero-shot prompting** is a **technique where LLMs perform tasks without any examples in the prompt** — relying solely on instruction and pretrained knowledge, demonstrating the model's generalization capabilities.
**What Is Zero-Shot Prompting?**
- **Definition**: Give instruction without examples, model performs task.
- **Contrast**: Few-shot includes 1-5 examples, zero-shot has none.
- **Requirement**: Model must understand task from description alone.
- **Benefit**: No need to craft examples, faster prompting.
- **Trade-off**: May be less accurate than few-shot for complex tasks.
**Why Zero-Shot Matters**
- **Generalization**: Tests model's true understanding.
- **Efficiency**: No time spent crafting examples.
- **Flexibility**: Works for novel tasks without training data.
- **Scalability**: Same prompt works across variations.
- **Baseline**: Establishes minimum capability before few-shot.
**Zero-Shot Example**
```
Classify the sentiment of this review as positive, negative, or neutral:
"The product arrived quickly but the quality was disappointing."
Answer: negative
```
**vs Few-Shot**
Zero-shot: Just instruction.
Few-shot: Instruction + examples.
One-shot: Instruction + 1 example.
**When to Use**
- Simple, well-defined tasks.
- Large, capable models (GPT-4, Claude).
- When examples are hard to create.
- As baseline before trying few-shot.
Zero-shot prompting reveals **model capabilities without example engineering** — the simplest prompting approach.