Affiliation:
1. The Institute of Ethnology and Anthropology, Chinese Academy of Social Sciences, Beijing, P. R. China
Abstract
With the rapid development of large language models (LLMs), the quality of AI-generated context (AIGC) is rapidly improving, and the correctness and detection of generated content have become a global challenge. In this paper, we review the current methods of AIGC detector and introduce the definition, dataset and methods of AIGC detection, including the manual-based methods, rule-based methods, statistical learning-based methods, deep learning-based methods, knowledge enhancement-based methods and watermarking-based methods. However, these methods have very low recognition accuracy when facing the latest LLMs, such as ChatGPT and GPT-4. This paper also suggests that more work should be put into identifying AIGC quality in the future, such as whether there are logical errors, knowledge errors or data falsification, which may have more severe consequences and be more likely to be detected.
Funder
National Natural Science Foundation of China
Publisher
World Scientific Pub Co Pte Ltd
Subject
General Earth and Planetary Sciences,General Engineering,General Environmental Science