Zero-Shot Detection of LLM-Generated Text using Token Cohesiveness
This addresses the need for practical, training-free detection of AI-generated text, which is incremental as it builds on existing zero-shot detectors by adding a plug-and-play module.
The paper tackled the problem of detecting LLM-generated text by introducing token cohesiveness as a new feature, showing that LLM-generated text has higher token cohesiveness than human-written text, and demonstrated that TOCSIN improves existing zero-shot detectors with extensive experiments across various settings.
The increasing capability and widespread usage of large language models (LLMs) highlight the desirability of automatic detection of LLM-generated text. Zero-shot detectors, due to their training-free nature, have received considerable attention and notable success. In this paper, we identify a new feature, token cohesiveness, that is useful for zero-shot detection, and we demonstrate that LLM-generated text tends to exhibit higher token cohesiveness than human-written text. Based on this observation, we devise TOCSIN, a generic dual-channel detection paradigm that uses token cohesiveness as a plug-and-play module to improve existing zero-shot detectors. To calculate token cohesiveness, TOCSIN only requires a few rounds of random token deletion and semantic difference measurement, making it particularly suitable for a practical black-box setting where the source model used for generation is not accessible. Extensive experiments with four state-of-the-art base detectors on various datasets, source models, and evaluation settings demonstrate the effectiveness and generality of the proposed approach. Code available at: \url{https://github.com/Shixuan-Ma/TOCSIN}.