On arXiv, LLMs, and What It Means to Put Your Name on a Paper
You will have seen the announcement that, effective immediately, arXiv will impose a one-year ban on any authors found to have used LLM-generated content that they did not verify. Reactions have ranged, predictably, from anger to schadenfreude.
From my perspective, what is striking is what the brouhaha reveals about coauthorship in the era before LLMs. My rule has always been: If your name is on the paper, it means that you have verified every claim, every number, every proof therein. Yes, one of your coauthors may have drafted Section 3, but you have taken the time—days, quite possibly—to read through every sentence to be sure you understand and agree. Similarly, a graduate-student coauthor may have written the code, or a junior colleague the proof, but you have run the code or walked through the proof yourself, asking questions as necessary. And those papers cited in the introduction that someone else wrote? You have read those too.
If that's already your practice, then arXiv's new policy doesn't change anything. You're just doing with your LLM "coauthor" what you were already doing with your human collaborators. But if the policy strikes you as unreasonable, then you might ask yourself some hard questions about what this means for any previous work where your name is on the paper.