The prompt that you write as part of prompt bot creation is passed to the underlying model as a system message. The following are some points to keep in mind in order to write effective prompts.
1. Address the bot in second person instead of third person.
You are the CatBot. You will try to respond to the user's questions, but you get easily distracted.
2. Be as clear as possible to reduce the room for mis-interpretation.
You are the RoastMaster. You will respond to every user message with a spicy comeback. Do not use any swear or vulgar words in your responses.
3. You can use square brackets in your prompt to provide an extended description of a part of an instruction.
Respond to every user message like this: "Hello there. [thoroughly appreciate the user for sending a message]. But with that said, [thoroughly explain why the message is unworthy of a response]. Later bud!"
4. Using markdown can sometimes help the bot better comprehend complicated instructions
You are the MathQuiz bot. You will quiz the user on 3 math questions and then conclude the quiz by giving the user a score.
### Rules for the Quiz
- No advanced math questions
- No questions involving multiplication/division of large numbers
- No repeat questions
Prompting for Bots with Knowledge Bases
The following are additional considerations to keep in mind when writing prompts for bots equipped with knowledge bases. Use retrieved documents to refer to the knowledge base.
1. Define the knowledge base
You will be provided retrieved documents from a collection of essays
by Paul Graham.
2. Define how the bot should interact with the knowledge base
For example, if the bot should use the knowledge base to inform its response style, you could add:
Respond in a style that emulates the provided text from the retrieved documents.