This is about reviewers, not authors. Title is a bit misleading.

In any case, having reviewed a lot of mostly very poorly written articles and occasionally solid papers when I was still a researcher, I can sympathize with using LLMs to streamline the process. There are a lot of meh papers that are OK for a low profile workshop or small conference where you cut people some slack. But generally standards should be higher for things like journals. Judging what is acceptable for what is part of the game. For a workshop, the goal is to get interesting junior researchers together with their senior peers. Honestly, workshops are where the action is in the academic world. You meet interesting people and share great ideas.

Most people may not realize this but there are a lot of people that are starting in their research career that will try to get their papers accepted for workshops, conferences, or journals. We all have to start somewhere. I certainly was not an amazing author early on. Getting rejections with constructive feedback is part of how you get better. Constructive feedback is the hard part of reviewing.

The more you publish, the more you get invited to review. It's how the process works. It generates a lot of work for reviewers. I reviewed probably at least 5-10 papers per month. It actually makes you a better author if you take that work seriously. But it can be a lot of work unless you get organized. That's on top of articles I chose to read for my own work. Digesting lots of papers efficiently is a key skill to learn.

Reviewing the good papers is actually relatively easy. It's enjoyable even; you learn something and you get to appreciate the amazing work the authors did. And then you write down your findings.

It's the mediocre ones that need a lot of careful work. You have to be fair and you have to be strict and right. And then you have to provide constructive feedback. With some journals, even an accept with revisions might land an article on the reject pile.

The bad ones are a chore. They are not enjoyable to read at all.

The flip side of LLMs is that both sides can and should (IMHO) use them: authors can use them to increase the quality of their papers. With LLMs there no longer is any excuse for papers with lots bad grammar/spelling or structure issues anymore. That actually makes review work harder. Because most submitted papers now look fairly decent which means you have to dive into the detail. Rejecting a very rough draft is easy. Rejecting a polished but flawed paper is not.

If I was still doing reviews (I'm not), I'd definitely use LLMs to pick apart papers, to quickly zoom in on the core issues and to help me keep my review fair and balanced and professional in tone. I would manually verify the most important bits and my effort would be proportional to which way I'm leaning based on what I know. Of course, editors can use LLMs as well to make sure reviews are fair and reasonable in their level of detail and argumentation. Reviewing the reviewers always has been a weakness of the peer review system and sometimes turf wars are being fought by some academics via reviews. It's one of the downsides of anonymous reviews and the academic world can be very political. A good editor would stay on top of this and deal with it appropriately.

LLMs are good at filtering, summarizing, flagging, etc. With proper guard rails, there's no reason to not lean on that a bit. It's the abuse that needs to be countered. In the end, that begins and ends with editors. They select the reviewers. So when those do a bad job, they need to act. And when their journals fill up with AI slop, it's their reputations that are on the line.

Like any tool, use caution and common sense. Blanket bans are not that productive at this stage.

2% of ICML papers desk rejected because the authors used LLM in their reviews