The stylistic words that indicated LLM involvement in the study include "delves," "showcasing," and "underscores." These words showed a marked increase in frequency, suggesting the influence of Large Language Models (LLMs) like ChatGPT on the scientific texts analyzed.
The study analyzed over 14 million PubMed abstracts from 2010 to 2024 to investigate the prevalence of large language models (LLMs), such as ChatGPT, in academic writing. The researchers developed a novel methodology to track excess word usage, providing evidence that LLMs have had a notable impact on scientific literature, with at least 10% of recent biomedical abstracts showing signs of AI assistance.
The main focus of the study on LLMs in academic writing is to investigate the prevalence of AI tools like ChatGPT in scholarly literature, detect changes in writing style and vocabulary, and examine the implications for research integrity and evaluation of academic contributions. The study proposes a novel methodology to track excess word usage, providing evidence of LLMs' impact on scientific literature.