Аңдатпа
With the advent of large language models such as GPT-3.5, GPT-4, and the latest variants, the task of creating text that can pass off as human-written has never been simpler. Although the potential for text creation has never been more exciting, the implications of such technology in terms of authoring, academic honesty, and the spread of misinformation are very serious. The purpose of this paper is to analyze the existing 27 research publications related to the task of identifying AI-written text. Various approaches, datasets, evaluation schemes, and issues associated with the task will also be covered in the paper. The approaches used for text identification can be broadly classified into several main categories, namely the use of transformer networks, stylometry, watermarking, and robust approaches. Although the approaches used for text identification are very effective in the usual environments, there are several issues associated with cross-domain text identification, multilingual text identification, and adversarial attacks.