1. ChatGPT und halluzinierte Referenzen in Artikeln aus ausgewählten Bereichen der Betriebswirtschaftslehre.
- Author
-
Schlögl †, Christian
- Subjects
- *
CHATGPT , *LANGUAGE models , *ARTIFICIAL intelligence , *TECHNICAL writing , *BIBLIOGRAPHY - Abstract
This paper examines the problem of hallucinations in four scientific articles generated by ChatGPT-4 from various sub-fields of business administration. Specifically, the aim is to check whether the references generated by ChatGPT exist and, if this is the case, whether they are correct. A first surprising result is that nearly 40 Prozent of the 174 sources referenced in the articles, whose bibliographic details were listed after the individual main chapters, were not included in the specially for each article generated bibliography. A check of the bibliographic entries revealed that between 18 % and 100 % of the sources in the individual articles could not be found with Google Scholar. Furthermore, half of the sources found were incorrect to varying degrees (wrong document type, wrong journal name, wrong page reference). On average, only a quarter of the bibliographic references were completely correct. In Web of Science, only a quarter of the sources could be verified, but most of them were correct. The above results show that ChatGPT-4 – besides ethical and legal concerns – should definitely not be used for writing scientific papers, as a large proportion of the references generated by ChatGPT (valid at the time of writing this paper) does not exist or is partially incorrect. Furthermore, it can be assumed that some of the article texts generated are also "invented" or come from unreferenced sources (plagiarism). Texts generated by ChatGPT must therefore be subject to a comprehensive review, which may not always be easy. The authors therefore agree with the recommendation of Gimpel et al. (2023) that ChatGPT should primarily be used as a supplement to writing (e. g., spell-checking, creating summaries, supporting the generation of ideas). [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF