DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature

Implementation of the experiments in the DetectGPT paper.

Original implementation can be found here: https://github.com/eric-mitchell/detect-gpt

Instructions

First, install the Python dependencies:

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

Second, run any of the scripts (or just individual commands) in paper_scripts/.

If you'd like to run the WritingPrompts experiments, you'll need to download the WritingPrompts data from here. Save the data into a directory data/writingPrompts.

Note: Intermediate results are saved in tmp_results/. If your experiment completes successfully, the results will be moved into the results/ directory.

Citing the paper

If our work is useful for your own, you can cite us with the following BibTex entry:

@misc{mitchell2023detectgpt,
    url = {https://arxiv.org/abs/2301.11305},
    author = {Mitchell, Eric and Lee, Yoonho and Khazatsky, Alexander and Manning, Christopher D. and Finn, Chelsea},
    title = {DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature},
    publisher = {arXiv},
    year = {2023},
}