Research question
Can domain-specific models distinguish human-authored graduate application materials from AI-generated or AI-revised versions?
My contribution
As a coauthor, I contributed to software, data curation, validation, and analysis. The study investigates letters of recommendation and statements of intent, with a public codebase.
Evaluation context
The findings concern the domains, datasets, and models evaluated in the study. They do not establish a universal AI-text detector. The paper describes the experimental setup and limitations.