Over 30% of new ArXiv submissions now read as AI-written
A new study analyzing 12,750 ArXiv papers suggests that approximately 32% of new academic submissions exhibit characteristics of AI-generated text. The researchers used a detector calibrated against pre-ChatGPT data to ensure a low false-positive rate.
Why it matters
The findings raise significant concerns regarding the integrity of scientific publishing and the potential for AI to flood academic repositories with low-effort content.
We scored the full text of 12,750 arXiv papers and found that about a third of new ones read as machine-written. Here is the method, the results, and an honest account of the limitations.
There is a genre of headline that says "N% of X is now AI," and most are not worth reading, because the detector behind the number also flags some share of genuine human writing. If a tool marks 40% of new papers as machine-written but also marks 20% of papers written before ChatGPT existed, the real story is the 20% nobody mentioned.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in