detect-gpt  by eric-mitchell

Research paper implementation for zero-shot machine-generated text detection

Created 2 years ago
430 stars

Top 69.0% on SourcePulse

GitHubView on GitHub
Project Summary

DetectGPT addresses the challenge of identifying machine-generated text in a zero-shot setting. It is designed for researchers and developers working on natural language processing and AI safety, offering a method to distinguish human-written content from AI-generated text without prior training on specific models.

How It Works

The core approach leverages "probability curvature" to detect AI-generated text. Instead of relying on specific model fingerprints, DetectGPT analyzes how the probability assigned to a given text changes when small perturbations are introduced. The hypothesis is that machine-generated text, often produced with higher confidence and less variance, will exhibit different curvature properties compared to human text. This method aims for a more generalizable detection capability.

Quick Start & Requirements

  • Install dependencies: python3 -m venv env && source env/bin/activate && pip install -r requirements.txt
  • Download WritingPrompts data into data/writingPrompts/ to run related experiments.
  • Prerequisites: Python 3, requirements.txt includes necessary libraries.
  • Official demo: https://detectgpt.github.io/

Highlighted Details

  • Zero-shot detection capability.
  • Utilizes probability curvature as a detection signal.
  • Official implementation of the DetectGPT paper.

Maintenance & Community

The project is the official implementation of research by Eric Mitchell, Yoonho Lee, Alexander Khazatsky, Christopher D. Manning, and Chelsea Finn. Further community engagement details are not provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. This may pose a restriction for commercial use or closed-source linking until clarified.

Limitations & Caveats

The README does not specify any limitations or known issues. The project appears to be research-focused, and its performance on diverse, real-world datasets beyond the WritingPrompts benchmark is not detailed.

Health Check
Last Commit

2 years ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
7 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.