Prompt engineering is the process of optimizing text input to a language model to guide its behavior and improve the quality of its output.
Core Techniques
- Zero-shot Prompting: Asking the model to perform a task without providing any examples.
- Few-shot Prompting: Providing a few examples of input-output pairs to demonstrate the desired behavior.
- Chain-of-Thought (CoT): Encouraging the model to 'think step-by-step' to solve complex reasoning problems.
- System Prompts: Setting the high-level behavior or 'persona' of the model (e.g., 'You are a helpful coding assistant').
cot_example.txttext
User: If I have 3 apples and I buy 2 more, then eat 1, how many do I have?
Let's think step-by-step.
Model: 1. You started with 3 apples.
2. You bought 2 more, so 3 + 2 = 5 apples.
3. You ate 1, so 5 - 1 = 4 apples.
Final answer: 4Best Practices
- Be specific and clear about the desired output format.
- Provide context and constraints (e.g., 'Summarize in 50 words').
- Use delimiters like triple quotes or XML tags to separate instructions from content.
- Iterate and refine prompts based on model responses.