Use of large language models to identify fake information

Abstract

In recent years, the field of artificial intelligence has undergone a true revolution with the emergence of large language models (LLMs) such as GPT-4, Llama-3, Gemini, and others, which have been successfully applied across a wide range of tasks – from text generation to data analysis. This article examines how these models can be effectively used for detecting fake information. This study explores the use of the ChatGPT chatbot for identifying fake information in the context of cybersecurity. Using a large language model, a swarm of virtual experts was created, which generated informational messages on the topic of cybersecurity (both fake and truthful) and assessed them as either “fake” or “true.” For analysis, a semantic network was constructed and subsequently visualized using Gephi. The research analyzed two datasets of messages: one created by human experts and the other by artificial experts. Each message was rated and converted into a numerical format for further analysis. Using the Hamming distance, the results were validated, and the accuracy of matches between assessments was determined. As a result of building the semantic network, key concepts in the field of cybersecurity were identified, along with the relationships between them. A swarm of artificial experts generated a dataset of messages with fake and truthful content, which was assessed both by the artificial experts themselves and by a human expert. Analysis of the Hamming distance between these assessments demonstrated that artificial intelligence has potential in detecting fake information; however, at this stage, its performance requires human oversight and adjustments.