At arXiv, we are seeing submissions that consist of many short sections and bullet points rather than a fluent narrative. These submissions are more like outlines than actual papers. For those of you who play with LLMs, is this writing style produced by some of the current LLMs?
Comments
So clearly "peer review" is broadly broken wrt validating papers for scientific merit, but maybe there's an opportunity for some type of "community review" system that presents newly submitted papers double blind to volunteers who score them wrt "this looks like AIGC slop"?
arXiv doesn't review submissions; we just check that they have the syntactic form of scientific papers ready for review.
So it will stay hard to detect LLM generated text with just a quick look.