ChatGPT Prompt Engineering

ChatGPT Prompt Engineering for Developers

从Andrew Ng和Isa Fulford的联合课程中学习所获。

更加适合开发者使用，对于目前ChatGPT、Claude这样trained模型，可以不使用特别复杂的prompt校正。

1 Guidelines for Prompting

1.1 Principles of Prompting

Principle 1. Write clear and specific instructions.

clear$\ne$short

1. Use delimiters

   • Triple quotes: “”"
   • Triple backticks: ```
   • Triple dashes: ---
   • Angle brackets: <>
   • XML tages: <tag></tag>

作用：

• Make it clear t the model that this is a seperate section.
• Avoid prompt injections.

2. Ask for structured output

HTML, JSON

3. Check whether conditions are satisfied

Check assumptions required to do the task.

4. Few-shot prompting

Give successful examples of completing tasks.

Then ask model to perform the task.

Principle 2. Give the model time to think.

1. Specify the steps to complete a task

Step 1 - …
Step 2 - …
…
Step N - …

2. Instruct the model to work out its own solution before rushing to a conclusion.

1.2 Model Limitations

Hallucination

Makes statements that sound plausible but are not true

Reducing hallucinations:

First find relevant information,

then answer the question based on the relevant information.

2 Iterative Prompt Development

Iterative Process

• Try something
• Analyze where the result does not give what you want
• Clarify instructions, give more time to think
• Refine prompts with a batch of examples

3 Summarizing

4 Inferring

5 Transforming

6 Expanding

Temperature

• Temperature = 0: For tasks that require reliability, predictability
• Temperature = 0.3 (0.7, or even higher): For tasks that require variety

7 Conclusion

Role

messages = 
[
	{"role": "system"
	 "content": "You are an assistant… "},
	{"role": "user"
	 "content": "tell me a joke "},
	{"role": "assistant"
	 "content": "Why did the chicken… "},
	…
]

• system: Sets behavior of assitant
• assistant: Chat model
• user: You