Escribir un comentario

This paper presents a large-scale empirical study examining the influence of large language models, represented by ChatGPT, on academic writing styles in arXiv abstracts. Using a corpus of approximately one million papers submitted between 2018 and 2024, the authors analyze shifts in word-frequency patterns to estimate the prevalence of LLM-influenced writing. The scope of the dataset and the longitudinal nature of the analysis represent major strengths of the study.

A notable contribution of the work is its adaptive approach to word selection, which accounts for both increasing and decreasing term frequencies, and its careful calibration using simulated ChatGPT-modified abstracts. The finding that an estimated 35% of computer science abstracts exhibit LLM-like stylistic characteristics highlights the rapid penetration of generative models into academic communication and raises important questions about authorship, originality, and scholarly norms.

However, the interpretation of “LLM-style” writing depends heavily on the choice of baseline prompt and model (GPT-3.5), which may not fully reflect the diversity of real-world LLM usage. Additionally, stylistic similarity does not necessarily imply substantive reliance on LLM-generated content, and this distinction could be more clearly articulated.

Overall, the paper provides a rigorous and thought-provoking analysis of how generative AI is reshaping academic writing practices and offers a valuable empirical foundation for ongoing discussions around responsible and transparent use of LLMs in research.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.