Controlling ChatGPT with Custom Instructions or API System Messages

This is the repository for the LinkedIn Learning course Controlling ChatGPT with Custom Instructions or API System Messages. The full course is available from LinkedIn Learning.

Custom Instructions in ChatGPT allow you to provide additional context and even format the type of response you want from the system every time you interact with it. In the OpenAI Chat Completions API you can access the same functionality using the System and Assistant messages. In this short course, learn how to take advantage of these features to get more consistent and customizable responses from both systems.

Instructions

This repository provides examples of how to use message roles and the instructions parameter to control the output of OpenAI's language models.

Note

The easiest way to run and interact with the examples is by opening this repository in GitHub Codespaces.

The examples are found in two folders:

• ./CompletionsAPI has examples using GitHub Models and the Completions API.
• ./ResponsesAPI has examples using OpenAI's API and the Responses API.

Setting up the Completions API examples

If you're running the repository in GitHub Codespaces, the examples in ./CompletionsAPI will work out of the box without any further setup. GitHub Models are automatically authenticated in the Codespace and the completions will generate when called.

Setting up the Responses API examples

To run the examples in ./ResponsesAPI you first need to add an OpenAI API key to your environment:

1. Go to https://platform.openai.com and sign in or sign up
2. Generate a new key at https://platform.openai.com/api-keys
3. Copy the key (you only get to see it once).
4. In Codespaces, open Terminal
5. Set up a new environment variable like this:

export OPENAI_API_KEY="your_api_key_here"

Running the examples

The examples are self-contained and run in Terminal with the Python command:

1. Open Terminal
2. Navigate to one of the folders:

cd CompletionsAPI

3. Run a script using Python:

python basic-response.py

You are free to experiment by modifying and expanding these examples in any way you like!