Since OpenAI released ChatGPT in November 2022, the academic writing ecosystem has undergone a profound transformation. Recent studies reveal that scientists are increasingly relying on AI tools like ChatGPT to write research papers, a trend particularly pronounced in 2024.
Researchers at the University of Tübingen in Germany conducted an in-depth analysis of 14 million PubMed literature abstracts from 2010 to 2024. They were surprised to find that at least 10% of the research papers published in 2024 used large language models (LLMs) such as ChatGPT for assistance during the writing process. This proportion is even more astonishingly high in certain specific fields and countries.
The findings indicate that the impact of LLMs on scientific literature is profound, even surpassing the influence of the COVID-19 pandemic on academic writing. By analyzing changes in word usage frequency, researchers noticed a significant increase in the use of specific stylistic words such as "delves," "showcasing," and "underscores" since the release of ChatGPT, reflecting scientists' growing reliance on ChatGPT to refine and revise texts.
The study also revealed differences across disciplines, countries, and journals. In fields such as computer science and bioinformatics, the usage rate of LLMs reached 20%. In non-English-speaking countries like China, South Korea, and Taiwan, the usage rate of LLMs is also relatively high, while in English-speaking countries like the UK and New Zealand, it is lower. Additionally, prestigious journals like Nature, Science, and Cell have a lower usage rate of LLMs, whereas some open-access journals like Sensors and Cureus have a higher usage rate.
LLMs indeed excel in enhancing the language quality and fluency of papers, but they also raise concerns. For instance, LLMs may introduce inaccurate information or reinforce existing biases in scientific discussions. Therefore, how to use these tools appropriately to ensure the authenticity and fairness of academic writing has become an important issue to be addressed.
The academic community needs to develop clear policies and guidelines to regulate the use of LLMs in academic writing. Some publishers and funding agencies have already banned the use of LLMs in peer reviews or required public declarations of their use. As these tools continue to become more prevalent, the academic community may need to reconsider and adjust relevant policies to ensure the high quality and reliability of scientific research.
This study not only provides a comprehensive perspective on the impact of LLMs in academic writing but also serves as a wake-up call. While enjoying the convenience brought by AI, we should also reflect on how to use this tool to promote rather than hinder scientific progress.