Large language models powered by artificial intelligence (AI) are trained with a huge amount of text, allowing them to respond to human requests for natural language.
Researchers from Stanford University and other institutes in the US looked at 1,121,912 previous edition documents in the archives “Arxiv” and “Biorxiv” and published newspapers in nature magazines from January 2020 to September 2024.
Focusing on the frequency with which words commonly used in AI systems appear in papers, the team estimated the involvement of a large-scale language model of this study (ChatGPT in this study) when modifying the content of a research paper.
The results published in Nature Human Behavior magazine suggest that LLM (large language model) usage is steadily increasing, with the largest and fastest growth estimated on computer science papers (up to 22%).”
The researchers also estimated a greater dependence on Preprint Papers' AI systems in the archive “biorxiv” of the archive “biorxiv” written by authors from regions known to have fewer English speakers, such as China and Continental Europe.
However, papers relating to articles published via mathematics and Nature Journal show low evidence of AI use in content changes, according to analysis.
The research team said authors who submit shorter papers and preprints often show higher AI usage in more papers.
“These results could be indicators of the competitive nature of a particular field of research and the pressure to quickly publish,” the team said.
Researchers also looked at a small number of papers to understand how scholars disclose the use of AI in their writing.
An examination of 200 randomly selected computer science papers uploaded to the print archive “Arxiv” in February 2024 revealed that “only two out of 200 cases that “explicitly revealed the use of LLMS during paper writing.”
Future research looking at the disclosure statement may help you understand researchers' motivations for using AI in writing.
For example, the policy regarding disclosure of LLM use in academic writing is still unknown, or it may have other motivations to deliberately avoid disclosing AI use, the authors said.
A recent study published in Journal Science estimated that at least 13% of the 2024 research summaries could have gained help from large-scale language models, as they included more of the term “style” that is considered favored by these AI systems.
Researchers at the University of Tübingen, Germany, who analyzed more than 15 million biomedical papers published between 2010 and 2024, said that the AI model caused a dramatic change in the vocabulary used in academic writing.
