A Lightweight Detector: Zero-Shot Detection of Machine-Generated Text with Once Call
Jinfang Yan, Zhao We, Hong Guo · 2025
With the increasing prevalence of large language model (LLM)-based text generation technologies, an increasing number of students are leveraging AI tools to complete assignments and compose essays. This trend poses a significant challenge to the development of students' independent learning abilities and critical thinking skills. However, existing text authenticity detection methods are typically computationally expensive and resource-intensive, limiting their practicality for daily academic use. To address this issue, we propose a lightweight and efficient detection method designed to assess the originality of student-authored texts. The proposed approach enables rapid and low-cost identification of AI-generated content, making it wellsuited for routine academic evaluation scenarios. Experimental results demonstrate that the proposed detector achieves reliable performance in verifying text authenticity while substantially reducing both computational and time costs.