arXiv Brings Down the Hammer

Send the arXiv AI-generated slop, get a yearlong vacation from submissions

When I was in academia - doing my Master's degree, and starting my Ph.D. before dropping out - it was critical, I was told, that all my research was my own, all my work my own. There didn't need to be anything else said. You heard about students who were forced out of programs for violations of academic honesty. The deal was simple: in exchange for the poverty wages that was your stipend (I made $15k between my scholarship and fellowship), you did your own research, wrote your own papers.

And in a sense, it's not hard right? You just have to do your job. I know a lot of people hated the writing side of CS, who outright struggled with it (Engish often not their first language), but as someone who almost majored in English, I loved it. My writing was solid and concise. I managed to get work accepted at a prestigious conference. I learned how to use LaTeX and love BibTeX.

But apparently all stuff around academic honesty is just a lie to tell grad students, because arXiv has a huge AI slop problem. Fake citations, placeholder text, all kinds of stuff. But apparently if the researchers themselves won't hold themselves to the highest standard, then arXiv will. They just announced that any issues arising from AI-produced content will make the authors subject to a one-year ban, and a permanent requirement that future publications undergo peer review before being submitted to arXiv.

This is harsh and wonderful. Academic journals and conferences are where research is published. Thomas Dietterich, a member of arXiv's editorial council, notes that all authors of a manuscript are responsible for its content, and so if issues related to AI use slip in - Dietterich provides the examples of "inappropriate language, plagiarized content, biased content, errors, mistakes, incorrect references, or misleading content" - then they're responsible, and will be held responsible.

I'm hoping this is a step towards the Finding Out era (we've clearly been in Fucking Around for a while). The genie's out of the bottle. I don't think you can stop people from using AI. But you can enforce consequences. If your name's on the paper, it's your paper. If your name's on the commit, it's your code. I think in future there'll be a reckoning around how we let this poison our creativity and flow. But that's a conversation for another day. And at least for now, at least someone's finally saying that if you put your name on it, it's yours, and no excuses.

gemlog