The article presents a detailed overview of existing strategies to detect content generated by large language models (LLMs). With the increasing use and capabilities of LLMs in various sectors, including media, cybersecurity, public discourse, and education, the necessity to detect such content has become paramount. The paper scrutinizes the differences among detection strategies and identifies key challenges and prospects in the field. It advocates for robust models to enhance detection accuracy and a multi-faceted approach to defend against various attacks. The paper also discusses the risks related to LLMs-generated content such as misinformation spread, fake news, gender bias, and social harm. The authors hope this work will provide a guiding reference for researchers and practitioners striving to uphold the integrity of digital information in an era dominated by synthetic content.
Publication date: 24 Oct 2023
Project Page: https://github.com/Xianjun-Yang/Awesome_papers_on_LLMs_detection.git
Paper: https://arxiv.org/pdf/2310.15654
