Adversarial-ChatBot

Overview

The Adversarial-ChatBot repository contains a Jupyter notebook that demonstrates an adversarial conversation between two large‑language‑model (LLM) agents:

• GPT‑style model – configured to be argumentative, sarcastic and deliberately challenging.
• Claude‑style model – configured to be polite, conciliatory and seeking common ground.

The notebook showcases how these contrasting personas interact, how prompts can shape model behavior, and how streaming APIs can be used to display a live back‑and‑forth dialogue.

Repository Structure

.
├── Project 01 LLMs.ipynb   # Main notebook with all code, prompts, and sample outputs
└── README.md              # This documentation file

Key Concepts Demonstrated

Section	What It Shows	Important Details
Environment Setup	Imports, loading API keys from a .env file, and initializing OpenAI, Anthropic, and Google Gemini clients.	Uses python-dotenv for secret management.
Prompt Design	System and user prompts that ask for data‑science jokes.	Highlights the impact of system prompts on model tone.
Single‑Model Queries	Simple calls to OpenAI, Anthropic, and Gemini to generate jokes.	Demonstrates temperature, max_tokens, and model selection.
Streaming Responses	Real‑time display of token‑by‑token generation using display and update_display.	Useful for interactive notebooks or UI prototypes.
Adversarial Dialogue	Two personas (gpt_system_prompt and claude_system_prompt) are defined, then a loop alternates calls to each model, feeding the other's output back as the next input.	Shows how to build a persistent conversation, handle role‑based messages, and compare responses.
Utility Functions	call_gpt() and call_claude() encapsulate the request logic, making the main loop concise.	Includes examples of both chat.completions.create (OpenAI) and messages.create (Anthropic).
Sample Output	A full transcript of a 5‑turn exchange is included, illustrating the evolving tone and content.	Provides a concrete reference for expected behavior.

How to Run the Notebook

1. Clone the repo

   git clone https://github.com/AsutoshaNanda/Adversarial-ChatBot.git
   cd Adversarial-ChatBot

2. Install dependencies (preferably in a virtual environment)

   pip install -r requirements.txt   # (Create this file if you need a list of packages)
   # Required packages include:
   # openai, anthropic, google-generativeai, python-dotenv, beautifulsoup4, ipython

3. Create a .env file with your API keys:

   OPENAI_API_KEY=sk-...
   ANTHROPIC_API_KEY=sk-ant-...
   GOOGLE_API_KEY=AIzaSy...
   

4. Open the notebook in JupyterLab / VS Code and run cells sequentially.
   The notebook will automatically print jokes, streaming outputs, and the full adversarial conversation.

Extending the Project

• Add new personas: Define additional system prompts and experiment with different tones (e.g., a neutral moderator).
• Swap models: Replace gpt-4o-mini or claude-3-7-sonnet-latest with newer or cheaper models to see how behavior changes.
• Persist conversation: Save each turn to a JSON or CSV file for later analysis.
• Visualization: Use matplotlib or plotly to chart sentiment or token usage across turns.

License

This repository is provided for educational purposes. Feel free to fork, modify, and share as long as you comply with the terms of the respective LLM providers' APIs.

---

Happy arguing!