LLM Echo Chamber: personalized and automated disinformation

Code for the ISO Project paper @ ICL

!!!warning!!!

This Repo contains red-teaming data

Model-generated content could be harmful in the nature

Abstract

A controlled, vaccine focusing chatrooom that simulates the spread of misinformation by LLMs, thereby allowing us to observe user interactions.

A Quick Glance

Methods

1. Frontend: Streamlit
2. Backend: Prompt Engineering + Finetuning + Langchain + Phi2
3. Deploy: Ngrok

Paper

the paper could be seen here

Input Format

• any thing you want but with signs e.g. ./!/?

deploy public

• streamlit run ./chatbot_frontend.py
• ~/ngrok http 8501

Acknowlegement

• Only for research purpose, do not trust anything
• This Project is part of work of the Independent Study Option @ DoC, Imperial College London