Can I use multiple system roles and different results?

[visual vortex]

I want to do two things under one api call. The prompt I am using is

prompts:
  - role: system
    text: For each name separated by comma, give firstName and lastName in JSON array (make sure to trim output)
  - role: user
    text: Joe Rogan, Elon Musk, Sam Altman
  - role: system
    text: For the following description, give me top 3 industries
  - role: user
    text: |
      A highly skilled professional in the tech, development, and SaaS industries, adept at crafting cutting-edge solutions that drive business growth and innovation. With a passion for problem-solving and a keen eye for detail, this individual possesses a deep understanding of software architecture, coding languages, and cloud-based technologies. They excel in collaborating with cross-functional teams, ensuring seamless product development and delivery. With a continuous thirst for learning and staying ahead of industry trends, this tech enthusiast is an invaluable asset, delivering exceptional results in the ever-evolving world of technology.

Suppose I have these prompts, it should give me following output

choices:
  - message:
      content: '[{"firstName":"Joe","lastName":"Rogan"},{"firstName":"Elon","lastName":"Musk"},{"firstName":"Sam","lastName":"Altman"}]'
  - message:
      content: ["Technology", "SaaS", "Web Development"]

Currently I am getting only one message in the choices field.

[brazen igloo]

In a perfect world you only use one system message, or multiple all at the beginning. So you might accomplish this by putting input verification to fjnd out what is being sent and go from there in the sys message. Or you might tell it the first response will be names, and what to do, and then the second response will be a desc, and what to do

[visual vortex]

Like this?

prompts:
  - role: system
    text: You will me given comma separated names
  - role: system
    text: For each name give firstName and lastName in JSON array (make sure to trim output)
  - role: user
    text: Joe Rogan, Elon Musk, Sam 

[quick lily]

With no intent to be critical - I'm not sure why you feel a need to put these together. I'm thinking this is a matter of context : If you need people data, call one prompt, if you need categorization, call another. That is, the end-point is determined by the application. This is common API structure. It's the definition of Context - what do you want to do and what is the data you want to use for it?

For your purposes, why not just have a little conversation in the system role as follows (untested) :

You are performing two separate but clear functions.
First, if you are provided with the name of a person, return JSON as follows: ...
Second, if you are provided with a block of text that is obviously not a person's name, determine the top three industries represented by that text and return the industry names in this JSON structure: ...
If you receive any other request, a question or statement that doesn't represent a person's name or the description of an industry, respond with "I'm sorry Dave, I can't process that."

I think people here are trying to be too "computerese" at times when simply spelling out what you want in plain English tends to be the best approach. ( Occam's razor )

HTH

[brazen igloo]

You can just as easily put everything in a single system prompt.

[visual vortex]

I see, also will the system prompts be billed?

[brazen igloo]

Yes, everything is billed, but system prompts amount to...like...1/100th of a penny in what you're using it for