Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Lichaamsverzorging --- Tijdschriften

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Cellulose

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

With the rapid improvement of large language models and more notably text generation, it has become imminent to encounter numerous ethical and security issues arising from corpora written by artificial intelligence (AI). In this study, three machine-driven detectors, GPTZero, Crossplag, and Binoculars, are analyzed. The paper further explores the differences in the detectability of various types of content by dividing them into two genres, fiction and non-fiction. Additionally, the impact of the length of the text as well as the false positive rates are also examined. The results show that both types of text and the length of it have an insignificant impact on the detection accuracy. Furthermore, it can be observed that GPTZero and Binoculars exhibit relatively high and consistent detection performance across a wide range of genres. In terms of the unfairness problem characterized by high false positive values, all detectors are suitable for deployment in an educational context, except for Crossplag when it comes to non-fiction documents.