Skip to content

Large Language Models in Academia: Boosting Productivity but Reinforcing Inequality

Summary

Using data on 4,582 computer-science scholars from 194 top U.S. universities and 218,723 papers (2019–2024), the authors provide a large-scale empirical assessment of LLMs' effects on research productivity. Following LLMs' introduction, publication rates rose by roughly 8%, accelerating to 3.2% in 2023 and 12.8% in 2024, and junior scholars benefited more than seniors (productivity gain declining ~1% per year of experience). Difference-in-differences and generalized synthetic control analyses show that native English-speaking (NES) researchers published more than non-native English-speaking (NNES) peers, widening linguistic disparities. Overall, LLMs boost scholarly productivity and lower barriers for early-career researchers, while also reinforcing inequities rooted in language proficiency.

Contribution

Presents what the authors describe as the first large-scale empirical assessment of LLM impacts on research productivity, documenting overall productivity gains, an experience gradient favoring junior scholars, and a widening NES-vs-NNES gap.

Method

Empirical observational study: panel of 4,582 CS scholars at 194 top U.S. universities and 218,723 papers (2019–2024), analyzed with difference-in-differences and generalized synthetic control designs.

Relevance to RISE

ICIS 2025 paper documenting the inequality side of AI-in-academia productivity gains. Companion to 1; cited together they sketch a productivity-with-distributional-cost story for RISE deployment.

Critique / open questions

Sample restricted to computer-science scholars at 194 top U.S. universities, so generalization to other disciplines, ranks of institutions, or non-U.S. settings is unwarranted. LLM "exposure" appears to be inferred from a post-introduction time cut rather than measured individual use, which the abstract does not address.

  1. Filimonovic, D., Rutzer, C., & Wunsch, C. (2025). Can GenAI improve academic performance? Evidence from the social and behavioral sciences. https://arxiv.org/abs/2510.02408 ↩