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

Eric Mitchell, Yoonho Lee, Alexander Khazatsky et al., 2023. arXiv.

Presents DetectGPT, which flags text as machine-generated by checking whether it sits at a point of negative curvature in a language model's log-probability, without training a separate classifier.

Voynich ManuscriptPreprintChecked against the original0

What it's used for here

Cited in the methods survey on procedure fingerprinting.

Cited by on this site 1 record

1 methods surveys

Encoding procedures

SRC

Cite it

Mitchell, E., Lee, Y., Khazatsky, A., Manning, C. D., & Finn, C. (2023). DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature. arXiv. https://arxiv.org/abs/2301.11305

@article{mitchell2023detectgpt,
  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},
  journal = {arXiv},
  year = {2023},
  url = {https://arxiv.org/abs/2301.11305}
}

TY  - GEN
TI  - DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature
AU  - Mitchell, Eric
AU  - Lee, Yoonho
AU  - Khazatsky, Alexander
AU  - Manning, Christopher D.
AU  - Finn, Chelsea
PY  - 2023
JO  - arXiv
UR  - https://arxiv.org/abs/2301.11305
ER  -

Mitchell, Eric et al.. “DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature.” arXiv (2023).