GPT Role Playing Demo

This repository demonstrates GPT-4's ability to adapt to different roles and personalities through system prompts.

Setup

pip install openai

Configuration

The notebook supports both OpenAI and Azure OpenAI endpoints:

if vendor=='azure':
    client = AzureOpenAI(
        azure_endpoint=userdata.get('AzureAIEndpoint'),
        api_key=userdata.get('AzureAIKey'),
        api_version='2024-06-01'
    )
elif vendor=='openAI':
    openai.api_key = userdata.get('OpenAIAPIKey')
    client = openai

Examples

1. Standard Response

A basic prompt asking about the fable of the frog and scorpion.

2. Physics Professor

The same story told through the lens of a physics PhD, incorporating physics concepts and analogies.

3. Pre-school Teacher

Explaining rainbow formation to a 6-year-old, using simple language and relatable analogies.

4. Medieval Pirate

Responding to philosophical questions in pirate vernacular during a storm.

Key Findings

1. Role Adherence: GPT-4 maintains character while staying focused on the core question
2. Quality Control:
   • Best results: GPT-4 with temperature 0.2
   • Lower quality with:
     • Models below GPT-4
     • Temperature above 0.2
3. Age-Appropriate Language: GPT-4 effectively adapts vocabulary and concepts for different age groups

Usage

1. Set up your OpenAI/Azure credentials
   • make sure your keys are valid and accessible in your environment
2. Clone the repo to access the notebook/script:

     git clone https://github.com/BrunoFelalaga/prompting_gpt4o_with_different_roles

3. Run the notebook cells sequentially
4. Experiment with different roles and prompts

Requirements

• Python 3.x
• OpenAI package
• IPython