A community to post scientific articles, news, and civil discussion.
dart board;; science bs
rule #1: be kind
The citation format needs an overhaul.
Make citations hyperlinked and publicly accessible. Past a certain date (2004?), make it mandatory. And if the research is mega paywalled, well… perhaps we should do something about that, too.
Then they’d be machine-verifiable.
The system has been dysfunctional. As it is elsewhere, the convenience of AI fraud/slop is simply exacerbating the existing issue to breaking points.
But if it’s not paywalled, how will the poor publishers make even more money off the labor of the researchers?
When I was in college in like 2005, we had to go through a citations course. It seemed outdated to me at the time, what with the way literally everything is searchable now(then.)
I was swiftly berated for suggesting that there was anything wrong with the way we do citations. Because APA has it alllll figured out. If it's a website they even added a website citation tool!
When I asked if we could cite Wikipedia then, I was berated again. Obviously you would never do anything so idiotic as cite Wikipedia! Why, anybody can say anything on Wikipedia.
So what's the difference between Wikipedia and any other website? She couldn't say.
I just copied wikipedias citations the entire time I was in college. My citations were never questioned.
Wikipedia is not a primary source. They have an explicit rule against original research: https://en.wikipedia.org/wiki/Wikipedia:No_original_research
So it's in the same class as your own papers. Citing it directly is regurgitation. The more this happens, the worse the quality gets. Same as saving and resaving a jpeg, or when AI gets trained on AI slop.
Sp you are unaware that wikipedia ALSO has sources
Make citations hyperlinked and publicly accessible.
It's already the standard in many journals that citations should include a hyperlinked DOI.
And if the research is mega paywalled, well… perhaps we should do something about that, too.
That's also increasingly the standard, for example all of my publications (and I left science years ago) can be accessed in preprint format without a paywall.
Then machine checking should be implemented
Whatever needs to be restructured to make citations automatically verifiable needs to happen, and then this will be less of an issue.
Frankly, this is something that already should be, and to a large degree is, manually checked by editors and referees.
While this is a problem that should be taken seriously, it's also something that mainly affects trash-tier journals. You won't find many hallucinated citations in Nature or Science (I doubt there has been a single one), and authors have strong incentives to prevent it from happening as they risk their reputation (and with it future grants).
Still, what about citations of articles that themselves contain hallucinated citations? It's a food chain problem.
I guess what I’m saying is the checking should be more… accessible? And less costly, hence machine automatable. This would increase the quality of journals that, for whatever reason, don’t do enough human verification, and it would allow bigger journals to do deeper checks “down the chain” with the same labor.
Maybe start with "LLM mark" ruling?
Lmao. the reviewers are using ai to try and detect ai without questioning if ther own ai is hallucinating either.
Ban AI slop maybe?
Be reasonable. Maybe implement a second AI to fact check the first one?
IDK, who will check to make sure the second AI is doing a good job?
It’s AI all the way down! We’re gonna need a lot of data centers