https://github.com/shohail-ismail/atomic-de-fake
A prototype made for the Idiap Research Institute's 'Idiap Create Challenge' that detects misinformation in text posts by using AI to generate assessment questions then aggregating human fact-checkers' responses to either assign a final trustworthiness seal or provide corrective feedback to the original poster.
https://github.com/shohail-ismail/atomic-de-fake
ai averitec human-ai-teaming mistral-api streamlit
Last synced: 5 months ago
JSON representation
A prototype made for the Idiap Research Institute's 'Idiap Create Challenge' that detects misinformation in text posts by using AI to generate assessment questions then aggregating human fact-checkers' responses to either assign a final trustworthiness seal or provide corrective feedback to the original poster.
- Host: GitHub
- URL: https://github.com/shohail-ismail/atomic-de-fake
- Owner: Shohail-Ismail
- Created: 2024-09-01T16:51:42.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-09-11T21:19:32.000Z (8 months ago)
- Last Synced: 2024-09-12T07:43:36.113Z (8 months ago)
- Topics: ai, averitec, human-ai-teaming, mistral-api, streamlit
- Language: Python
- Homepage:
- Size: 54.7 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Atomic DeFake
## Requirements
* Anaconda
* Python 3.11
## Installation
```
conda create -n atomic-defake python=3.11
conda activate atomic-defake
pip install -r requirements.txt
```
## Setup
Create a `.env` file in the project root and put in your Mistral API key as follows:
```
MISTRAL_API_KEY=
```
## Running the application
```
streamlit run ui.py
```
After running the app, a Streamlit interface will open in your browser. After logging in (credentials aren't needed), click 'User post' from the sidebar and input text to verify for misinformation. After clicking the 'AtomicDeFake' button, you will be taken to the 'Contributor' screen where you can answer questions about the post, along with a final confidence score (you will need to do this twice so as to simulate more than one fact-checker). After doing this, the page will run the aggregator, then output the result (with feedback if the post is deemed to contain misinformation).