This project provides an open-source PyTorch implementation of GPTZero, an AI model designed to detect whether text was generated by AI or a human. It aims to offer an alternative for educational institutions and researchers concerned about AI-generated content, replicating the functionality of the popular but closed-source GPTZero service.

How It Works

The implementation leverages the perplexity scoring mechanism, specifically using a RoBERTa model from Hugging Face Transformers. By calculating the perplexity of a given text, the model can infer the likelihood of it being AI-generated, aiming to match the results of the original GPTZero service.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: Python 3.x, PyTorch, Hugging Face Transformers.
• Usage:
  • Python function: from model import GPT2PPL; model = GPT2PPL(); model("your text here")
  • Local inference: python3 local_infer.py

Highlighted Details

• Claims 100% identical results to gptzero.me through extensive corpus comparison.
• Based on Hugging Face Transformers and RoBERTa.

Maintenance & Community

No specific community channels or maintenance details are provided in the README.

Licensing & Compatibility

The README does not explicitly state a license. The project relies on Hugging Face Transformers, which is typically Apache 2.0 licensed. Compatibility for commercial use would require explicit licensing confirmation.

Limitations & Caveats

The project's claims of 100% identical results should be independently verified. The README does not detail the specific model versions or datasets used for comparison, nor does it mention potential limitations in detection accuracy across different AI models or text types.