Core Viewpoint - The emergence of large language models (LLMs) like ChatGPT has significantly transformed academic writing, particularly in the biomedical field, raising concerns about research integrity and the accuracy of generated content [2][5]. Group 1: Impact of LLMs on Academic Writing - A study published in July 2025 revealed that approximately 200,000 out of 1.5 million biomedical papers indexed by PubMed in 2024 showed signs of LLM-generated text, accounting for about 1/7 of the abstracts [3][4]. - The use of LLM-assisted writing in biomedical publications is accelerating, with an earlier assessment indicating that about 1/9 of abstracts in the first half of 2024 exhibited similar signs [4]. - The study found that at least 13.5% of abstracts in 2024 were processed using LLMs, with some sub-corpora reaching as high as 40%, indicating a profound impact on scientific writing [5][10]. Group 2: Methodology and Findings - Researchers analyzed over 15 million abstracts from PubMed between 2010 and 2024, identifying 454 words that appeared significantly more frequently in 2024 compared to previous years, many of which were stylistic rather than content-related [7][9]. - The study highlighted that the vocabulary changes were more pronounced following the rise of LLMs than during significant events like the COVID-19 pandemic, with a notable increase in the use of adjectives and verbs [9][10]. - The proportion of LLM-assisted writing varies across disciplines, countries, and journals, with over 20% of abstracts in certain regions and fields utilizing LLMs [10]. Group 3: Challenges and Adaptations - Previous attempts to assess the impact of LLMs on academic writing faced challenges due to the lack of disclosure from users, making it difficult to evaluate the true extent of LLM usage [6]. - As authors become aware of specific vocabulary associated with AI-generated text, such as "delves," its usage may decline, complicating the assessment of AI's influence on academic writing [12]. - While using AI for text refinement or translation is deemed reasonable, generating large volumes of text without oversight raises concerns regarding research integrity [13].