A Distribution-Free Framework for Rewrite-Based Human-text Detection via Knockoff Filtering
Researchers have developed a novel statistical framework that enhances the reliability of detecting AI-generated text. This method allows existing detection models to achieve finite-sample False Discovery Rate (FDR) guarantees without needing to be retrained. By viewing text detection as a multiple hypothesis testing problem with knockoff structure, the framework separates the detection statistic design from false discovery control, offering robust performance across various models and domains. AI
IMPACT Provides a method to improve the reliability of AI text detection without retraining existing models.