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

TL;DR

DetectGPT is a zero-shot method that identifies machine-generated text by analyzing the negative curvature regions of an LLM's log probability function, improving detection accuracy without training classifiers.

Contribution

The paper introduces a novel curvature-based criterion for detecting LLM-generated text that does not require training data or watermarking, leveraging the structure of the model's probability function.

Findings

DetectGPT outperforms existing zero-shot detection methods.

It achieves 0.95 AUROC in identifying GPT-NeoX generated fake news.

The approach is effective without additional training or data collection.

Abstract

The increasing fluency and widespread usage of large language models (LLMs) highlight the desirability of corresponding tools aiding detection of LLM-generated text. In this paper, we identify a property of the structure of an LLM's probability function that is useful for such detection. Specifically, we demonstrate that text sampled from an LLM tends to occupy negative curvature regions of the model's log probability function. Leveraging this observation, we then define a new curvature-based criterion for judging if a passage is generated from a given LLM. This approach, which we call DetectGPT, does not require training a separate classifier, collecting a dataset of real or generated passages, or explicitly watermarking generated text. It uses only log probabilities computed by the model of interest and random perturbations of the passage from another generic pre-trained language model (e.g., T5). We find DetectGPT is more discriminative than existing zero-shot methods for model sample detection, notably improving detection of fake news articles generated by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code, data, and other project information.