Fake News Detection Using (Large) Language Models

• Abstract: Fake news detection has become a challenge with the rise of misinformation on social media, politics, and biased news reporting. Various approaches involving machine learning and deep learning have been proposed to detect fake news, but these methods have mostly been limited to specific domains and the patterns in the text on which the models are trained. We find that Large Language Models, with their ability to reason about the context and facts involved in news statements, achieve the best performance in fake news detection, surpassing Small Language Models, as well as traditional machine learning and deep learning methods. We also compare various fine-tuning methods and discuss their trade offs in terms of accuracy and efficiency.