Two basic prompting methods:




Zero-shot (zero sample):


Describe the task directly without providing examples


Example:


"Translate the following from English into Chinese:


The weather is nice today."




Few-shot (few samples):


Provide a few examples to let the model understand the format


Example:


"Translate the following from English into Chinese:




English: Hello, how are you?


Chinese: Hello, how are you?




English: The weather is nice today.


Chinese: "




Advantages of Few-shot:


- More consistent formatting


- Specific style of output


- Reduce ambiguity

The CoT prompt lets the model demonstrate the inference process:




General tips (possible errors):


"A farm has 15 chickens and 10 rabbits,


How many legs are there in total? "




CoT Tips (more accurate):


"A farm has 15 chickens and 10 rabbits,


How many legs are there in total?


Please think step by step. "




Zero-shot CoT (universal trick):


Add after the question:


"Let's think about it step by step."


"Let's think step by step."




Applicable scenarios:


- Mathematical calculations


- Logical reasoning


- Complex analysis

Improve output quality through role setting:




System prompt example:


"You are a senior software engineer with 20 years of experience.


Focus on code reviews and best practices.


Your answer should:


1. Point out specific problems in the code


2. Provide improvement suggestions and sample code


3. Explain why this change is better


4. Use professional but understandable language"




Effect:


- The output is more professional


- More consistent formatting


- Comply with domain-specific specifications