Janne M. Korhonen10h ago
After thinking more about this & thanks to folks who could phrase my concerns better than I (@[email protected] & @[email protected] especially, also @[email protected]): IF unchecked #LLM use means a blanket ban on *all* the authors, I think this is a bad solution to a real problem. #science
Show threadJanne M. Korhonen10h ago
Note the big IF - I didn't find anything about this from ArXiv web pages. But since even I was favourably disposed towards this idea, and many others seem to be so as well, I think this discussion is worth having even if it turns out that ArXiv policy is different. bsky.app/profile/jmko...

RE: https://bsky.app/profile/did:plc:l5qgtdxd462q6cz5hbt25edk/post/3mmaz23b7n22h
Show threadJanne M. Korhonen
Unchecked #LLM use is a real problem. It *must* be policed. But almost all the posts I've seen, especially those quoting this post where I discovered the screenshot, seem to assume *every* author of a scientific paper should know *every* reference cited. THIS IS NOT TRUE. bsky.app/profile/mani...

RE: https://bsky.app/profile/did:plc:y4kznwazezwlrpvo65qnt7s7/post/3mm2g7372yc2o
8:35amWeb
Show threadJanne M. Korhonen10h ago
In fields where #research is generally produced by a single or at most a few persons, it is certainly reasonable to assume and demand that the author(s) are familiar with all the works they cite. I said so myself. But. *Not all fields work like that.* And many of those use ArXiv *a lot.*

RE: https://bsky.app/profile/did:plc:l5qgtdxd462q6cz5hbt25edk/post/3mmazkm6xns2h
Show threadJanne M. Korhonen10h ago
Here @[email protected] explains one fairly common way how an author in a co-authored paper might legitimately not know even nearly every reference in the paper. Banning the service provider for misconduct of people he trusted seems excessive punishment to me at least. bsky.app/profile/tmkd...

RE: https://bsky.app/profile/did:plc:riwhbtutln67n2goe5dnchgt/post/3mmbatrhrec26
Show threadJanne M. Korhonen10h ago
Nevertheless, maybe it is acceptable, and perhaps necessary, to demand every author at least google the references to check they exist, and are not made up by an #LLM. Maybe also demand they read the paper and flag e.g. obvious copy pastes from an LLM. But what if confabulation is more subtle?
Show threadJanne M. Korhonen10h ago
These issues and risks are likely to be more common in physical and experimental sciences - which AFAIK are heavy users of ArXiv - but they do crop up in other fields too. Basically every time when doing more holistic interdisciplinary research. Which practically *everybody* wants more of.

RE: https://bsky.app/profile/did:plc:riwhbtutln67n2goe5dnchgt/post/3mmbatrosx226
Show threadJanne M. Korhonen10h ago
If research outlets adopt draconian practices like banning every author of a paper with questionable #research practices, that inevitably adds yet another risk to all multi- and #interdisciplinary collaboration. Which is difficult and risky already. And this will increase #inequality in research.
Show threadJanne M. Korhonen10h ago
If I risk a ban from a publication venue important for me if a collaborator from another field misbehaves, I have to be *really* sure about the collaborator. Junior scholars are a particular risk, simply because of their inexperience and lack of record in the field.
Show threadJanne M. Korhonen10h ago
And scholars hailing from countries or universities with sketchy or no reputation will be even more suspect than they are today. Not to claim that Ivy League grads don't cheat - of course they do - but let's be honest: reputation, connections and brand value matter. In #science, perhaps more.
Show threadJanne M. Korhonen10h ago
But, as I said twice already, the problem *is* real and it *must* be policed. I think one fairly obvious solution would be to make the *lead author* the one ultimately responsible for the paper, at least unless someone else confesses. (Maybe even then, to some extent.) bsky.app/profile/osma...

RE: https://bsky.app/profile/did:plc:v4urwsnfxsx6bdrtouappmdy/post/3mmb65ekuiww2
Show threadJanne M. Korhonen10h ago
Also potentially worth censuring in some cases are the PIs, department heads, and the like. Of course, trying to lead a group of researchers is like herding cats, only harder, so we should not overestimate the department head's capability to ensure good conduct. bsky.app/profile/tmkd...

RE: https://bsky.app/profile/did:plc:riwhbtutln67n2goe5dnchgt/post/3mmbeq2aj6k2n
Show threadJanne M. Korhonen10h ago
So I would start with an experiment: a policy of making the lead author the one who is by default responsible for ensuring there is no scientific misconduct. If that isn't enough of a deterrent, then we need to think of other measures. Though I also agree in principle with @[email protected] :

RE: https://bsky.app/profile/did:plc:hmtynq24viq5yycureab7inc/post/3mmb2jzcquk2p