DEER: Disentangled Mixture of Experts with Instance-Adaptive Routing for Generalizable Machine-Generated Text Detection

A new framework called DEER has been introduced to tackle the growing challenge of detecting machine-generated text, which is becoming increasingly realistic due to advancements in large language models. Current detection methods often struggle when faced with different domains, but DEER aims to improve accuracy by capturing both domain-specific and general patterns. This development is significant as it enhances our ability to discern between human and machine-generated content, which is crucial in various fields such as education, journalism, and online content moderation.