In the last few months, the problems associated with the explanded use of GPTs and LLMs via services such as ChatGPT, Copilot, Gemini, and others have come to preprint servers in mass. At Engineering Archive, we’ve seen a dramatic increase in preprints submitted, growing from an average of 60 submissions per month to closer to 200. This increase is submissions has massively increase the moderation workload as it can be difficult to filter the real human-authored work, from the slop. Often these submissions are made by so-called “independent researchers” with no institutional affiliation. They may or may not be real people. In the past, submission spam was often used as a form of citation gaming, attempts to artificially exploit Google Scholar indexing to make ones academic profile look more prestigious than is deserved. However, more recently, this is no longer the case.
As reported in Nature, this wave of AI slop is not limited to engrXiv and is hitting other preprint servers in the same way, overwhelming our volunteer moderators. Unfortunately, this wave of AI slop has real costs associated with it in terms of volunteer burnout and the costs associated with hosting this content and issuing DOIs when it slips through moderation.
At Engineering Archive, we are going to attempt to further crack down on these types of submissions. Submissions that appear as though they may be largely AI generated are going to face further scrutiny. This will slow down the timeline from submission to public posting and unfortunately, some legitimate work will be caught up as well. “Independent researchers” may also be asked to further verify their identity. We are relunctant to restrict the publication of work from authors who currently lack institutional affiliation because we don’t believe that good engineering research can only be performed within academic facilities, but something must be done mitigate the negative impacts of LLM generated content.
