ArXiv announces a ban on AI content and the responses are hilarious.

[jamesmarshall@sfba.social]

@docpop from his profile, Miller has a PhD from U. Chicago, a JD from Stanford, and is a professor at Smith College. I think this reflects badly on every one of those-- what is research from these places actually worth, if fake citations are common there?

He and his defenders are basically saying "everyone does it, and you're naive to complain." If it's true that everyone does it, that reflects badly on modern academia as a whole.

Any academic researchers want to chime in here?

[marta@corteximplant.com]

@docpop to add, broken trust from your colleagues has been a problem before AI too, here's a particularly popular story https://laskowskilab.faculty.ucdavis.edu/2020/01/29/retractions/

[rjblaskiewicz@mstdn.social]

@docpop Um....this raises so many questions. Like where did they buy their advanced degrees?

[tpfto@mathstodon.xyz]

@docpop good grief, these dudes are definitely telling on themselves. If they can't find the time or means to vet everything they've written, why should we spend the time or means reading their crap?

(See also: https://bsky.app/profile/manigarm.bsky.social/post/3mm2g7372yc2o )

[ggmcbg@mstdn.plus]

@docpop
"But I only had SuckJobAI create the fake citations to support what you and I both know is true anyway that I wrote about, if you'd just do your fuckin' reasear..."

[unchartedworlds@scicomm.xyz]

@FediThing @docpop

He said further down in the thread what he thinks citations are for:

"The citations are there to help readers who want to learn more about a sub topic, quickly locate new papers. They also function as a business suit, signaling that you're a serious person."

(yikes)

and various people have replied to that saying no that is not, in fact, what they are for!

[damonhd@mastodon.social]

@FediThing @unchartedworlds @docpop The bar to be one is maybe not as high as it should be? (That's how I hope to sneak in as one. maybe!)

[schnedan@social.tchncs.de]

@docpop Guess this James Miller is not skilled, qualified and educated enough to file papers... even without AI content.

[schnedan@social.tchncs.de]

@FediThing @unchartedworlds @docpop guess its a similar process as granting a patent in the USA... (like the ones our specialist tells us: these patent is valid even if it would never get approval in Germany, Europe, China,...), or safety certificates for Boeing...

[schnedan@social.tchncs.de]

@adriano @docpop @FediThing well how we did it at university back then? We did all the work alone, when there was no trustworthy partner for group-work. If you wanted to ensure quality and deadlines, you needed the one person you can trust: Yourself.

[schnedan@social.tchncs.de]

@adriano @docpop @FediThing Naggers?

[schnedan@social.tchncs.de]

@jackeric @FediThing @docpop I meet PhD which barely were able to fasten one's shoes... a PhD just means: get some 20+% more money for the same work.

[temptoetiam@eldritch.cafe]

@rjblaskiewicz @docpop even worse: the guy teaches. In college.
His ratemyprofessor page is... interesting.

[rickf@indieweb.social]

@docpop @digitalsnow

I both cringe and laugh in professor.

[alicemcalicepants@ohai.social]

@docpop I read at least part of every book in the bibliography of my PhD thesis. Did I do it wrong?

[paul_ipv6@infosec.exchange]

@alicemcalicepants @docpop

no, you did the self-torture that is a key part of a PhD program.

[wuweiwolf@tech.lgbt]

@docpop

- The cited work had an error that required a later correction to be mentioned in an erratum or subsequent article (e.g. fixing a typo in a formula), and the citing researchers were unaware of the correction, so their own work reproduces the same error.
- The cited work had an error that was later corrected, and the citing researchers knew about the correction, but they didn't bother to mention it or cite the later work where the correction appears, so anyone trying to reproduce their results would have no idea that there's an inaccuracy that needs to be corrected.
- The work that should have been cited is out of print and was never digitized, so rather than work with a research library to try to track down a copy and see what it says, the authors either don't double-check what it said, or cite some other work by the same author.

This is a major problem! LLMs are both directly making the problem worse, and encouraging scientists to churn out publications without putting in the work to verify.

[wuweiwolf@tech.lgbt]

@docpop
Unfortunately, I think this is a common attitude in many fields. As someone who has often needed to reproduce other scientists' work, I can confirm that I frequently check other peoples' citations in methodology that turn out to just be wrong. E.g., "As in Smith (2004), we decide the value of parameter X by applying formula Y to input Z." Often I find one of the following:

- The citation does not spell out important details, which have never made it into the scientific record at all, but are spread informally among a particular clique of researchers who frequently collaborate. (Or they are in documentation that was not publicly archived, e.g. internal wikis, proprietary software code, etc.)
- The citation is just to the wrong paper (e.g. the actual source is a different paper by the same author the year before, or the cited paper itself doesn't contain the full details and cites some earlier paper instead).

[quietewe@urbanists.social]

@adriano @docpop @FediThing “perhaps you could speak to your co-authors, my guy”

[abazigal@mastodon.social]

@docpop I almost read it in a sarcastic tone. Would be funny if he was deliberately trying to be contrarian (in a bad way) just to make the OP’s point.