GPT Editors, Not Authors: The Stylistic Footprint of LLMs in Academic Preprints

2025-05-22

AI Summary

This paper addresses the critical question of whether large language models (LLMs) are employed for global polishing or local content generation in academic preprints. We propose an interpretable bibliometric analysis framework that integrates PELT changepoint detection with a Bayesian classifier, leveraging stylistic feature modeling and GPT-regenerated texts to construct training data and quantify the spatial distribution of LLM traces. Empirical results reveal strong global uniformity in LLM intervention—indicating predominant use for holistic revision (e.g., grammatical correction, stylistic harmonization) rather than piecemeal content generation. This finding mitigates the risk of hallucinated text infiltrating preprints and establishes the first interpretable, reproducible detection paradigm for academic integrity assessment, grounded in stylistic segmentation and changepoint analysis.

Technology Category

Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument MiningMachine Learning: Large Multimodal Models (LMMs)Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

Web Mining and Content Analysis: Large pretrained models with web dataUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationGraph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The proliferation of Large Language Models (LLMs) in late 2022 has impacted academic writing, threatening credibility, and causing institutional uncertainty. We seek to determine the degree to which LLMs are used to generate critical text as opposed to being used for editing, such as checking for grammar errors or inappropriate phrasing. In our study, we analyze arXiv papers for stylistic segmentation, which we measure by varying a PELT threshold against a Bayesian classifier trained on GPT-regenerated text. We find that LLM-attributed language is not predictive of stylistic segmentation, suggesting that when authors use LLMs, they do so uniformly, reducing the risk of hallucinations being introduced into academic preprints.

Problem

• Assess LLM usage in academic writing vs editing
• Measure stylistic impact of LLMs on arXiv papers
• Determine uniformity of LLM use to reduce hallucinations

Innovation

• Stylistic segmentation analysis of arXiv papers
• Bayesian classifier trained on GPT-regenerated text
• PELT threshold variation for measurement