Variation is the Key: A Variation-Based Framework for LLM-Generated Text Detection

Researchers have introduced VaryBalance, a new framework for detecting AI-generated text that reportedly outperforms the current state-of-the-art detector (Binoculars) by up to 34.3% in AUROC. The method's core insight is that human text changes more significantly when rewritten by an LLM than AI-generated text does. It quantifies this difference using mean standard deviation. The approach is robust across multiple AI models and languages and works without impractical white-box access.