Try it here: https://huggingface.co/spaces/ankush29/CheckGPT
CheckGPT is an AI-powered text analysis tool that predicts if the text is generated by AI like ChatGPT. It evaluates perplexity and burstiness scores of the GPT model, providing valuable insights for investigating the origin of text content. 💡
This project is a fun hobby and WIP project, aimed at creating an open-source version of webapps like GPTZero Pro, a popular closed-source website. I think open-sourced AI text detection tools can be useful to ensure academic integrity in schools, where it's essential to verify that students are independently crafting their essays and answers for their personal growth, the transparency offered by open-source software enhances the trust and credibility of the evaluation process.
It uses GPT-2 model with an additional linear layer at the end to get the probability distribution of generating the current token sequence, which are used to estimate the perplexity and burstiness scores to classify the document.
- ✅ Evaluate the perplexity and burstiness scores of text generated by AI.
- ✅ Predict the label for text content: 'AI', 'Human + AI', 'Human'.
- ✅ Predicts the percentage of text content written by AI
- ✅ Supports PDF and Word documents for text analysis.
- ✅ Download the analysis results as a CSV file for further exploration.
To start using CheckGPT, follow these simple steps:
- Visit the hosted app on Hugging Face Spaces:
- Upload your PDF or Word documents.
- Click the "Start Checking" button to analyze the text content.
- Explore the results, including the predicted AI percentage, perplexity score, burstiness, and predicted output.
- Download the analysis results as a CSV file for further analysis.
Hugging Face hosted app is slow! If you want to experience much faster performance, deploy CheckGPT on your local device using the following steps:
- Clone this CheckGPT repository from GitHub:
git clone https://github.com/Ank-Cha/CheckGPT.gitand navigate to the project directory using the command:cd CheckGPT. - Install the required dependencies by running
pip3 install -r requirements.txt. - Run the CheckGPT app using the command
streamlit run app.py.
For MacBook Apple Silicon devices, optimize the performance even further by changing the device from "cpu" to "mps" in app.py.
For other cuda supported devices, you can try changing the device to "cuda" for enhanced performance.
in app.py
# device = "mps" # For MacBook Apple Silicon
# device = "cuda" # For other cuda supported devicesCheckGPT is built using the incredible Hugging Face Resources and Transformers library, which provides state-of-the-art natural language processing capabilities. Also checkout [this] repo which helped me to make this project.
Contributions are always welcome! If you have any ideas, improvements, or bug fixes, please open an issue or submit a pull request. Let's make CheckGPT even more amazing together!
If you find CheckGPT intriguing and valuable, I kindly request you to show your support by giving the repository a ⭐️ on GitHub. Your appreciation means a lot and encourages me to keep improving and enhancing this project. Thank you in advance for your support! 🌟
CheckGPT is developed by Ankush Chaudhari. Feel free to connect with me on LinkedIn to stay updated with the latest developments.
Please note that the results generated by CheckGPT are based on an AI model and may not be 100% accurate. It is still a work in progress project. It is advisable to use these results for investigation purposes and exercise caution when making decisions based on the analysis. The code for this project is released under the MIT License, which means you are free to use it as you see fit. However, please note that there is no warranty or guarantee provided with the code, and the developers/contributors are not liable for any issues or consequences arising from its use.