Generative-AI vs. Traditional Methods for Fake News Detection

Abstract

The elevated issue of fake news dissemination led to the development of various strategies for its timely detection and containment. Ranging from traditional machine learning models to deep learning architectures, and more recently, Large Language Models (LLMs) are adopted for the detection process. Fake news detection is primarily treated as a supervised classification task, and the major part of the research is focused on this aspect. However, the emergence of LLMs has introduced a critical shift in this field. Their powerful generative capabilities enable adversarial entities to produce and spread highly realistic misinformation. Therefore, this study investigates two key issues. Firstly, the robustness of conventional detection models is assessed when exposed to cross-source data. Secondly, the effectiveness of LLMs is evaluated as a reliable defender for identifying synthetic content.