Non-English-Speaking Countries Like Korea and China Show Higher Usage
Steady Increase in Use of LLMs for Academic Paper Writing
An analysis revealed that one-fifth of biomedical research papers published last year by domestic researchers received assistance from large language models (LLMs) such as ChatGPT. On July 8, Yonhap News reported, citing a recent study published in the international journal Science Advances by Dr. Dmitry Kobak's research team at the University of Tuebingen in Germany, that 20% of biomedical research papers published by domestic researchers had used LLMs.
The research team analyzed approximately 15 million biomedical paper abstracts registered in PubMed, the U.S. National Library of Medicine's (NLM) database, from 2010 to 2024. They examined the proportion of words preferred by LLMs to determine whether LLMs had been used. As a result, they found that among the 26,657 words used in the abstracts, the frequency of 454 words preferred by LLMs?including rarely used words like "delves" and "underscores," as well as more common words like "potential" and "findings"?surged in 2024. Based on these words, the research team estimated that, as of 2024, 13.5% of abstracts were likely written with the help of LLMs.
By country, English-speaking nations such as the United Kingdom and Australia showed a usage rate of around 5%, while Korea, China, and Taiwan recorded a much higher rate of about 20%. The research team explained that in non-English-speaking countries like Korea, LLMs are likely being used practically as English proofreading tools. In contrast, in English-speaking countries, researchers may refine LLM-recommended phrases, leaving fewer visible traces of LLM use.
Academic Discussions on AI-Related Ethics and Regulations Still in Early Stages
By journal, top-tier publications such as Nature, Science, and Cell had a lower rate of 7%, and Nature's sister journals showed a rate of about 10%. However, papers published by MDPI, a Swiss publishing company suspected of running predatory journals, reached as high as 21%. This indicates that the higher the "quality" of the journal, the lower the frequency of LLM use. Among the intersections of countries and journals, 34% of Korean papers published in MDPI's journal Sensors were found to have used LLMs, marking one of the highest rates recorded.
An analysis revealed that one-fifth of biomedical research papers published last year by domestic researchers received assistance from large language models (LLMs) such as ChatGPT. Photo is unrelated to the specific content of the article. Pixabay
The research team added that the impact of LLMs on biomedical research papers is even greater than that of the emergence of COVID-19. Given this trend, they emphasized the need to reassess policies and regulations regarding the use of LLMs in scientific papers. Although the use of LLMs in academic writing is increasing with the rise of AI, discussions on related ethical regulations remain in the early stages.
According to a recent survey conducted by the international journal Nature of 5,229 researchers, 90% said they could accept the use of generative AI for editing or translating papers. However, 55% responded that it is necessary to specify AI use or to verify translations. Only 33% of researchers said that AI should not be used in paper abstracts, while 45% said it is acceptable if AI use is disclosed, and 23% said it is permissible. The proportion of researchers who considered AI use acceptable in abstracts was higher than for main sections of papers such as introductions or conclusions. Nevertheless, only 28% of researchers reported having actually used AI in writing papers, indicating that there is still considerable reluctance to use AI in practice.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
