I recently posted a pre-print about reasoning and large-language models....
trying to get that piece on a pre-print server itself was a bit of a journey which illustrates concerns I've had for the last 18 months or so: namely that AI will kill publishing reform.
here a blog post about it:
#academic #openScience
https://write.as/ulrikehahn/is-ai-killing-scientific-reform
@UlrikeHahn I completely agree that we need to reflect on what role we want arxiv and other preprint servers to play in research. At the same time, I don't consider this to be a novel problem (see this from 1999; also, the DOI on this record is extremely messed up for some reason...: https://asistdl.onlinelibrary.wiley.com/doi/abs/10.1002/(SICI)1097-4571(1999)50:10%3C890::AID-ASI6%3E3.0.CO;2-8). AI, Covid, and other crises may have made it more visible again, but I haven't really seen Arxiv take action in relation to critical research about their role in scholarly communications
@zackbatist I agree that the pressures on academic publishing are not new, but I think the bit that’s new is the tools for the unlimited generation of scientific sounding text without the existence of corresponding tools that could keep up with the checking
so whatever problems we structurally had flowing from the limited supply of expertise before, are now being made infinitely worse
@UlrikeHahn I'm not convinced that the issue is proportion of resources to make : validate trustworthiness of truth claims. I think we have fooled ourselves into thinking that (a) one form of review is a sufficient marker of quality, and (b) creating other kinds of formal review processes will be any different. Journal clubs and internal/institutional review are great alternatives, but sadly not taken seriously. And what makes them great is they are community-minded, not generic.
@UlrikeHahn In other words, if we're chasing scale, we have to deal with the problems of scale. But that's not the only way.
@zackbatist so I do think scale is the real issue here (and one we were already chasing without much success before LLMs….), and think that, unfortunately, every single “solution” I can envision for dealing with this problem is going to end up doubling down on the very things we wanted to eliminate with publishing reform…
I hope I’m wrong…
@UlrikeHahn I agree, and this is why I think that chasing scale is not the best way forward
@zackbatist I think I disagree with you a little. I feel like the narrative that we had fooled ourselves into thinking that peer review is a sufficient quality indicator is at odds with my experience as an academic. I think everyone has always been able to think of published papers that they thought were not worth the paper they were printed on…
but I think also that peer review had in implicit line it was trying to maintain: reject obvious nonsense while leaving for the subsequent literature to sort out problem papers that themselves required intelligent/detailed argument to reject.
even that line was porous, but scale now makes it totally untenable
@UlrikeHahn sounds a lot like how you envision arxiv, from your original post
So totally agree!
The fact that arxiv actually feels like they need to review preprints, in a very ironic way means we need a pre-preprint server? 🤣
And as if to add insult to injury, our publication system was already FUBAR before LLMs were a thing, but people kept pretending the problem will go away by itself by ignoring it. That ostrich attitude is so going to blow up in everybody's faces now.
