ArXiv has almost 600 submissions today and most are AI slop

30 pointsposted 19 hours ago
by marysminefnuf
(arxiv.org)

16 Comments

okaleniuk

16 hours ago

Academic publishing is in deep crisis. It didn't start with AI, and it probably wouldn't end with it. The crisis came from trying to turn academia into a factory of some sort.

I coauthored 4 papers this Summer. All bona fide: no AI slop, no fake data, all genuine research papers. But the substance of each is barely enough for a paragraph in a paper I would have gladly written instead.

But I have KPIs. And KPIs don't care about quality, quality is unaccountable. So it is in my best interest to publish a lot of "barely science" instead of spending a year or two writing a good paper that will take another year to publish, and by then my contract will end, and I'll have no numbers to show.

xdertz

16 hours ago

Then you have schools demanding a published paper as part of their masters degree. This gave rise to many conferences that are just publishing mills.

Ekaros

15 hours ago

My professor thought that there could be paper in my thesis. Thankfully they did not insist. Work was pretty crap and useless(mostly due to happening in between versions of project/idea they were working on). But well I did graduate so that is fine for me.

I can't even imagine the horrors what papers on some Master thesis would look like. Often as there is so many graduates they are not that novel topics anyway.

okaleniuk

15 hours ago

Exactly.

And if you play by the spirit and not by the rules, you're fucked. Because, with masters, it takes about a year to get to the first publishable result, and about a year to publish it in a respected journal. And that's the happy pass. You might get an extra rejection, or fall into the back of the publishing queue. This could cost you three or six month easily.

The masters program in our uni is 1.5 years long.

floro

17 hours ago

The academic writing standard is the most obtuse and painful way of writing I have ever done. I would use AI for that as well.

xarope

16 hours ago

Academic writing uses words with meaning in very domain specific ways; those in the domain understand it, and those outside find it confusing.

If an AI just randomly throws out domain specific words without that domain's specific meaning, it becomes very painful for everybody to read, and worse still adds a level of ambiguity that defeats the whole purpose of that academic writing in the first place.

floro

8 hours ago

Yes it uses domain specific words but that's not the reason why academic writing is painful to read and write.

ahartmetz

15 hours ago

In addition to all that, I find the quality of writing in average current scientific papers atrocious, too.

Look at some from earlier than mid 80s or so, they are so much better.

0xEnsp1re

13 hours ago

basically most of internet now is ai slop

insane_dreamer

17 hours ago

did anyone read through them to determine whether they're slop? or is that just someone's opinion?

regardless of whether they are or not, the speed at which they can be cranked out with the help of LLMs, means that the signal to noise ratio is getting lower by the day, and soon it might be hard to find any signal. The outcome I worry about is it destroying confidence in academic publishing (which is already suffering from a lack of confidence).

paulpauper

19 hours ago

just because it was produced with the help of Ai does not mean it's slop. Slop means it's also wrong. The math may very well be interesting and correct.

SwellJoe

17 hours ago

C'mon, be real. Do you want to be the one to wade through a mountain of garbage by someone experiencing AI psychosis and thinking they've come up with a brilliant new idea to find something interesting and correct?

The signal to noise on ArXiv has always been a problem, due to open access and no peer review, and now AI has completely buried the signal. As with everything else, I now have to assume everything is shit without some other signal of quality; when I stumble on a project on github, if the developer has no software made before 2025, I'm leaving. It's not worth my time to try to find out whether they're a slop merchant or they know how computers work. Same with writing; I have begun to assume all new writers are shoveling slop at me.

I don't like that I have to be so suspicious. I'd rather take in the world with an open mind, but I can't afford it. I have limited time on this earth (I mean it, I'm feeling my age, I'm not wasting my one brief life on reading something somebody shat out on a whim).

goolz

15 hours ago

Recently ran into someone who thought they had solved a major quantum computing problem. Needless to say, they had not, and I wasted a good hour wading through their ‘work’. Time I will never get back. Nothing more to add than to echo your thoughts. Wish I had a better solution.

bicepjai

17 hours ago

Using an LLM as the language to express human thought isn’t slop — provided the human reviewed it. Skip review/edits and you’re just publishing raw model output, which reads like slop regardless of whatever insight might be hiding in it.

This is an example expressed using llm :)

marysminefnuf

18 hours ago

Its a good point and I guess its a matter of opinion tho. I consider anything in math done with use of ai to be less quality.

rerdavies

17 hours ago

Having established that they aren't neccesarily slop, may we ask how you determined which ones were written by AI? Or is that a matter of opinion too?