This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
This paper highlights the importance of building reliable AI-generated text detection systems, given growing concerns about the misuse of large-scale language models. To address the performance degradation of existing methods due to the lack of style modeling and the use of static thresholds, we propose the Mixture of Stylistic Experts (MoSEs) framework, which enables conditional threshold estimation for quantifying style-aware uncertainty. MoSEs consists of three core components: a Style Reference Repository (SRR), a Style-Aware Router (SAR), and a Conditional Threshold Estimator (CTE). For input text, the SRR activates appropriate reference data and provides it to the CTE, which then dynamically determines the optimal threshold by jointly modeling linguistic statistical and semantic features. MoSEs generates predicted labels with discriminant scores and corresponding confidence levels. Compared to baseline models, it achieves an average detection performance improvement of 11.34%, with a particularly significant improvement of 39.15% in low-resource environments. The source code is available at https://github.com/creator-xi/MoSEs .
