Skip to main content

Zero-Shot Prompting

Zero-Shot CoT Framework

Zero-Shot Prompting is a technique where a language model is asked to perform a task without being given any explicit examples or demonstrations. Instead, the model relies on its pre-trained knowledge and the clarity of the instruction to generate a response. This approach is especially useful for straightforward tasks or when you want to quickly test a model’s generalization ability.

Zero-shot prompting is foundational in prompt engineering and is often the first step before exploring more advanced techniques like few-shot or chain-of-thought prompting.

Use When​

  • The task is simple, well-known, or commonly encountered by language models.
  • You want to quickly test a model’s ability to generalize.
  • There is no need to guide the model with specific examples or demonstrations.

Pattern​

Directly instruct the model to perform a task using a clear, concise prompt. Avoid providing any examples or demonstrations in the prompt.

Examples​

Example 1: Sentiment Classification​

Classify the text into neutral, negative, or positive. Text: I think the vacation is okay.

Model output:

Neutral

Example 2: Translation​

Translate the following sentence to French: "Where is the library?"

Model output:

Où est la bibliothèque ?

Example 3: Code Explanation​

Explain what this Python function does:
```python
def fibonacci(n):
    if n <= 1:
        return n
    return fibonacci(n-1) + fibonacci(n-2)
*Model output:*

This function calculates the nth number in the Fibonacci sequence using recursion. It returns the input number if it's 0 or 1, otherwise it adds the two preceding Fibonacci numbers. For example, fibonacci(5) would return 5 (the sequence: 0, 1, 1, 2, 3, 5).


## Benefits
- **Simplicity:** No need to curate or maintain example sets.
- **Speed:** Quick to implement and test.
- **Generalization:** Useful for tasks that are well-represented in the model’s training data.

## Pitfalls
- May produce less accurate results for complex or ambiguous tasks.
- Lacks context for nuanced or domain-specific instructions.
- Performance can vary depending on how well the task matches the model’s pre-training.

## References
- Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (2020). [Language Models are Few-Shot Learners](https://arxiv.org/abs/2005.14165). *NeurIPS 2020*.