Published On Jan 26, 2024
Learn how to make AI chatbots talk back and forth to each other about any topic. The agents will pull research from the web sources you specify, and there's no need to preprogram responses.
This video provides a demo walkthrough of some code available on Google Colab where I've programmed an AI accelerationist chatbot and an AI alarmist to debate the future of AI, using information from the web as the source for their responses.
I'll show you how to customize it to provide your own debate topic, agent personas, and rules for debate, as well as how to customize the model to be more creative or restrictive in its responses.
The code uses Python, Langchain, and OpenAI, and leverages searches from ArXiv, Wikipedia, and DuckDuckGo.
Run / modify the code: https://colab.research.google.com/dri...
Inspect the output in langview: https://smith.langchain.com/public/f6...
View the output: https://docs.google.com/document/d/1x...
Read the Original article: / two-llm-based-autonomous-agents-debate-eac...
About Me: I am a Staff Design Technologist on Mozilla's Innovation Team. Learn more about Mozilla Innovation at future.mozilla.org
0:00 - Project Overview
0:24 - See the AI debate
1:37 - Ideas for Agent debate usage
2:03 - Implications of non-human mediated AI interaction
2:38 - Agent setup and prompting
5:00 - Agent generated personas and output
7:53 - Colab API Keys
8:56 - Understanding the code
11:24 - Using python libraries in Colab
13:27 - Modifying the code
16:58 - Ideas for extended the functionality
18:00 - Conclusion