You're reading the public-facing archive of the Category Theory Zulip server.
To join the server you need an invite. Anybody can get an invite by contacting Matteo Capucci at name dot surname at gmail dot com.
For all things related to this archive refer to the same person.
David Michael Roberts (Mar 25 2025 at 00:33):I hate to do this, but the four arXiv papers by this author, uploaded in the last week, all look wholly AI-generated to me: https://arxiv.org/search/?query=Reizi&searchtype=author
David Michael Roberts (Mar 25 2025 at 00:35):In particular the appendix of https://arxiv.org/abs/2503.16555 there is a promise of proofs, examples and so on, but the entire text of the appendix is this:
In this appendix, we present additional proofs, detailed calculations, and further examples
that complement the results in the main text. In particular, the appendix includes:
* A complete proof of the back-and-forth construction used in Lemma 5.8.
* Detailed verifications of the functoriality of the Henkin and compactness-based model constructions.
* Concrete examples illustrating the construction of models for specific theories.These supplementary materials are provided to offer deeper insight into the technical details and to demonstrate how our unified framework can be applied to various logical systems.
The next text is the bibliography and that's it. The content is also extremely banal.
Chad Nester (Mar 25 2025 at 06:48):After a cursory inspection of https://arxiv.org/abs/2503.16570, I agree.
Kevin Carlson (Mar 25 2025 at 07:33):I can't find any information about this supposed person online except an affiliation via their email, but I've made a report to the Arxiv.
fosco (Mar 25 2025 at 08:33):
yep, no way a human wrote this
David Michael Roberts (Mar 25 2025 at 09:17):Stupid LLM forgetting the syntax for bold in TeX and falling back on Markdown...
Matteo Capucci (he/him) (Mar 25 2025 at 15:39):I'm proud to say I called bullshit from the titles alone in my feed lol glad I wasn't wrong
Ryan Wisnesky (Mar 25 2025 at 16:10):Heh, we did an experiment on LLMs that produce SQL code, and for many of them, no matter how much you tell them not to format the output, they still do it. Stripping extra comments and markdown/html out of responses turned out to be the hardest part of interacting with the LLM in an automated flow.
Joe Moeller (Mar 25 2025 at 22:06):Matteo Capucci (he/him) said:
I'm proud to say I called bullshit from the titles alone in my feed lol glad I wasn't wrong
Right, natural transformations between theorems.
Joe Moeller (Mar 25 2025 at 22:07):I noticed there are two orders of the names used. Two of the papers are JRB, and two are BJR. What could be the point of that?
David Michael Roberts (Mar 26 2025 at 00:27):The email address seems to be attached to Open University Japan, so name-order may have been auto-generated differently for the different papers?
Noah Chrein (Mar 28 2025 at 15:44):fosco said:
yep, no way a human wrote this
To be fair, I have seen researchers who just learned about category theory writing this way.
Anyway, the AI-generated slop CT papers are coming. I've noticed that Qwen 2.5 is trained on a lot of higher/formal category theory. It's fun to play with and it can produce approximately accurate references to results, which can sometimes cut down on search time. It's not yet good enough to generate any meaningfully creative results, and is not enough to fool a half-keen eye, but I can imagine an undergrad using qwen to write a undergrad thesis that nobody reads.
Ivan Di Liberti (Mar 28 2025 at 15:51):Noah Chrein said:
fosco said:
yep, no way a human wrote this
To be fair, I have seen researchers who just learned about category theory writing this way.
Anyway, the AI-generated slop CT papers are coming. I've noticed that Qwen 2.5 is trained on a lot of higher/formal category theory. It's fun to play with and it can produce approximately accurate references to results, which can sometimes cut down on search time. It's not yet good enough to generate any meaningfully creative results, and is not enough to fool a half-keen eye, but I can imagine an undergrad using qwen to write a undergrad thesis that nobody reads.
What is Qwen and how happen it was trained on so much category theory?
Kevin Carlson (Mar 28 2025 at 16:07):Qwen appears to be Alibaba's language model. I hadn't heard of it till now.
Noah Chrein (Mar 28 2025 at 16:08):Perhaps the Chinese understand the importance of category theory to mathematics and hence to generalized cognition
Kevin Carlson (May 29 2025 at 22:18):There’s an interesting fake paper on the ArXiv today. I can’t really tell if it’s AI crankery or just the old fashioned kind. Did anybody glance at it? https://arxiv.org/abs/2505.22558
Cole Comfort (May 29 2025 at 22:22):Kevin Carlson said:
There’s an interesting fake paper on the ArXiv today. I can’t really tell if it’s AI crankery or just the old fashioned kind. Did anybody glance at it? https://arxiv.org/abs/2505.22558
The excessive use of lists suggests AI
Kevin Carlson (May 29 2025 at 22:47):Right, that makes sense. It was harder to find obvious local absurdities than in papers further up this thread, which is disappointing.
David Michael Roberts (May 29 2025 at 23:32):There's a whole bunch recently that I have been complaining about pointing out elsewhere. The author is uploading a new paper every couple of days, and the title names something after himself. I'm happy to see today that they've been moved to math.GM ! (as I suggested)
https://export.arxiv.org/find/math/1/au:+Alpay_F/0/1/0/all/0/1
David Michael Roberts (May 29 2025 at 23:35):And in the case at the top of the thread, namely https://arxiv.org/search/?query=Reizi&searchtype=author all these are also math.GM classified now, not math.CT.
Ryan Wisnesky (May 29 2025 at 23:55):Seems like in theory the arXiv "endorsement system" should deal with AI generated papers just like any other spam, but I guess it doesn't work in practice? https://info.arxiv.org/help/endorsement.html
Kevin Carlson (May 30 2025 at 00:35):Yes, I'm a bit confused how all these people are getting endorsements.
Mike Shulman (May 30 2025 at 00:36):At the very least it should be possible to "un-endorse" them after they've demonstrated their crankiness.
David Michael Roberts (May 30 2025 at 04:09):Another one! https://arxiv.org/abs/2505.22931
David Michael Roberts (May 30 2025 at 04:18):Maybe the arXiv needs to appoint a category theorist to the team of moderators...
Nathanael Arkor (May 30 2025 at 07:02):I thought arXiv had a strong stance against crackpottery, so why are these papers allowed to remain under math.GM, rather than being removed entirely?
John Baez (May 30 2025 at 07:10):Kevin Carlson said:
Right, that makes sense. It was harder to find obvious local absurdities than in papers further up this thread, which is disappointing.
The phrase "discrete conformal field theory" in the abstract made me raise my eyebrows. As if that were a known thing. Given how much people try everything, there probably is some work on something called discrete conformal field theory, but....
Yeah, there's a paper Conformal Field Theory at the Lattice Level: Discrete Complex Analysis and Virasoro Structure trying to understand how conformal field theory is related to field theory on a lattice. But most conformal transformations don't map a lattice to itself, so this is bound to be rough, and the idea that "Recursive Difference Categories and Topos-Theoretic Universality" would have something to say about it is, umm, questionable.
John Baez (May 30 2025 at 07:21):Nathanael Arkor said:
I thought arXiv had a strong stance against crackpottery, so why are these papers allowed to remain under math.GM, rather than being removed entirely?
It can be hard to tell whether a math paper is crazy, and people whose papers are rejected entirely complain a lot, so it seems the arXiv folks find it convenient to put borderline papers into math.GM, expecting people 'in the know' to beware of such papers. That's my impression anyway.
John Baez (May 30 2025 at 07:24):It's more diplomatic than having math.CP for crackpot math.
fosco (May 30 2025 at 08:16):This is a truly beautiful era to witness first-hand.
Areeb SM (May 30 2025 at 08:31):ViXra appears to be embracing the future...
David Michael Roberts (May 30 2025 at 08:42):But not unreservedly:
viXra.org only accept scholarly articles written without AI assistance. Please go to ai.viXra.org to submit new scholarly article written with AI assistance.
Joe Moeller (May 30 2025 at 23:38):arxiv could use the exact same disclaimer, only changing the first instance of "vixra".
John Baez (May 31 2025 at 08:49):ai.viXra.org sounds like a fascinating crackpot sociology experiment. They have 343 papers so far. Within the subject of physics, most of the papers are on "relativity and cosmology", so we can guess that part of physics attracts crackpots the most. Within mathematics, 75% of the papers are on number theory.
John Baez (May 31 2025 at 08:52):Yesterday's first submitted paper on general relativity and cosmology:
The Pi-Periodic 22/7ths Dimension: A Quantum Gravity Framework for Dark Energy
We propose a novel 4+1-dimensional quantum gravity framework incorporating a compactified extra dimension, τ , with a periodicity of π (to 22 decimal places), symbolically tied to the rational approximation 22/7.
Someone is taking this 22/7 stuff very seriously! I believe Archimedes came up with this approximation to pi, and it was good enough that by the Middle Ages a bunch of mathematicians believed 22/7.
fosco (May 31 2025 at 09:13):by the Middle Ages a bunch of mathematicians believed 22/7.
:surprise: Wait, is it not? /s
James Deikun (May 31 2025 at 09:21):Archimedes squared the circle with this ONE WEIRD TRICK! Geometers hate him!
John Baez (May 31 2025 at 09:26):Actually I learned this when reading about the mathematician Franco of Liège. In 1020 he got interested in the ancient Greek problem of squaring the circle. But since he believed that pi is 22/7, he started studying the square root of 22/7. I don't know if he figured out how to construct the square root of 22/7 with straightedge and compass. But he did manage to prove that the square root of 22/7 is irrational!
Now, this is better than it sounds, because I believe the old Greek proof that is irrational had been lost in western Europe at this time. So it took some serious ingenuity.
Still, it's a sad reflection on the sorry state of mathematical knowledge in western Europe from around 500 AD to 1000 AD. It was better elsewhere at that time. I find this local collapse of civilization, and how people recovered, quite fascinating.
John Baez (May 31 2025 at 09:30):Could AI slop prompt some loss of collective intelligence now?
Fabrizio Romano Genovese (May 31 2025 at 10:15):John Baez said:
Could AI slop prompt some loss of collective intelligence now?
In general any tool that helps you thinking makes you sloppier in some respect. So yes. For instance, ancient languages are often way more complicated grammatically than new languages. One reason for this is that being able to say "Go around the mammoth, without being heard, by exactly half of a circle" in fewer words may have been a big advantage when we were hunter-gatherers, so languages tended to be more expressive. With civilization, inception of written support etc we lost the need to formulate such complicated statements in a compact way, languages became less expressive, and we probably lost some of our cognitive ability in the process as well. It's always a tradeoff.
John Baez (May 31 2025 at 11:30):I'm thinking more about how successive generations of Roman summaries of Greek scientific texts watered them down to a homeopathic dilution of their original strength. Then many of the originals were lost, at least in western Europe.
Nathanael Arkor (May 31 2025 at 11:32):@Fabrizio Romano Genovese: could you share a reference for the claim that older languages have higher entropy than modern languages?
Notification Bot (Jun 01 2025 at 08:05):13 messages were moved from this topic to #meta: off-topic > language: the rise and fall of complex grammars by John Baez.
Matteo Capucci (he/him) (Jun 02 2025 at 17:32):Fabrizio Romano Genovese said:
John Baez said:
Could AI slop prompt some loss of collective intelligence now?
In general any tool that helps you thinking makes you sloppier in some respect. So yes. For instance, ancient languages are often way more complicated grammatically than new languages. One reason for this is that being able to say "Go around the mammoth, without being heard, by exactly half of a circle" in fewer words may have been a big advantage when we were hunter-gatherers, so languages tended to be more expressive. With civilization, inception of written support etc we lost the need to formulate such complicated statements in a compact way, languages became less expressive, and we probably lost some of our cognitive ability in the process as well. It's always a tradeoff.
uuuhmm what's a reference for this? smells really funny to me...
David Michael Roberts (Jun 02 2025 at 19:44): Matteo Capucci (he/him) (Jun 12 2025 at 08:53):New AI paper up:
Matteo Capucci (he/him) (Jun 12 2025 at 08:54):This one is funny because it outs itself
image.png
Matteo Capucci (he/him) (Jun 12 2025 at 08:57):I actually approve of this way of approaching AI tools: personally, I don't think they automatically disqualify a paper. The principle should be the the author is ultimately responsible to check their results, and remains fully accountable.
Patrick Nicodemus (Jun 12 2025 at 08:57):Our results offer a formal justification for this procedure, suggesting that the analytic
continuation is not arbitrary but is in fact forced by the underlying principles of symmetry
and normalization.
Kind of a funny quote because the analytic continuation of a function is one of the most rigidly determined and least arbitrary constructions in mathematics
Matteo Capucci (he/him) (Jun 12 2025 at 08:58):Indeed, the paper is (from a quick skim) likely formally correct but basically insubtantials, it's a big cargo-cult regurgitation. The whole thing seems circular.
Patrick Nicodemus (Jun 12 2025 at 09:00):Matteo Capucci (he/him) said:
I actually approve of this way of approaching AI tools: personally, I don't think they automatically disqualify a paper. The principle should be the the author is ultimately responsible to check their results, and remains fully accountable.
Yeah, I think I agree. At the very least we might have to get used to seeing that writing style everywhere, I can imagine a non-native speaker feeling a lot of pressure to use it to make their wording seem natural. It doesn't inherently disqualify the paper. But, on the other hand, it makes me suspicious and vigilant of errors, and at that point even a small error would be enough to cause me to discard it.
fosco (Jun 12 2025 at 09:18):I suspect at some point someone will try to mass-produce papers and submit them everywhere (it's almost trivial to compile a list of journal inside the math-cs area; let the machine prepare a different paper for each item of the list; let the machine submit, let the machine handle the rebuttals and modify the paper accordingly, resubmit...), relying on small probability of success after a high number of trial.
It's the academia equivalent of asking out 100 girls, one of them will say yes.
These are very interesting times to witness. Especially if you're an irredeemable nihilist.
David Michael Roberts (Jun 12 2025 at 11:56):I think there's a real risk of the image of CT being tarnished if this type of stuff becomes too common. The number theory people know how to funnel cranks away from their arXiv category, if category theorists can't do this, it's not a good look.
David Michael Roberts (Jun 12 2025 at 11:59):Also, mathematicians generally are conscious of the circle squarers and the number theory cranks and so on, and can spot this stuff pretty easily, because it's on a hot-button topic and shows the usual obvious signs. But something in category theory applied to other areas (not Applied Category Theory, but to an outsider it's not necessarily obvious) that plays to the stereotypes of CT's abstract nonsense moniker just looks like another silly CT paper that claims to revolutionise our understanding of a piece of classical mathematics when really it's empty of real content.
David Michael Roberts (Jun 12 2025 at 12:08):Perhaps not among hardcore mathematicians, who would almost surely recognise the problem and commiserate, but anyone merely adjacent, for instance someone with money who might be needed to be convinced to fund some real and good ACT may get wind of this AI nonsense.
Maybe I'm being too pessimistic here. But these are ideas that occur to me
Nathanael Arkor (Jun 12 2025 at 13:26):The number theory people know how to funnel cranks away from their arXiv category
How?
David Michael Roberts (Jun 12 2025 at 13:29):Well, I get math.NT daily announcements and I've never seen a crank number theory paper, and yet I know they do turn up in math.GM.
David Michael Roberts (Jun 12 2025 at 13:30):So somehow they manage it.
Mike Shulman (Jun 12 2025 at 15:29):Have we taken any action about these papers? Contacted anyone at arXiv about removing them and un-endorsing the submitters? That seems to me to be the obvious first step. I'd be willing to help if needed, although I don't have the time to filter the daily submissions for them myself.
Mike Shulman (Jun 12 2025 at 15:32):David Michael Roberts said:
But something in category theory applied to other areas (not Applied Category Theory, but to an outsider it's not necessarily obvious) that plays to the stereotypes of CT's abstract nonsense moniker just looks like another silly CT paper that claims to revolutionise our understanding of a piece of classical mathematics when really it's empty of real content.
I wonder if an effect like this could be what's causing the problem by making it easier for cranks to get endorsed with CT papers by lazy non-category-theorists.
Nathanael Arkor (Jun 12 2025 at 15:51):David Michael Roberts said:
Well, I get math.NT daily announcements and I've never seen a crank number theory paper, and yet I know they do turn up in math.GM.
To me, that just suggests that the arXiv editors are better at detecting crank NT papers than crank CT papers, likely because they have had more practice at it.
Kevin Carlson (Jun 12 2025 at 16:26):Mike Shulman said:
Have we taken any action about these papers? Contacted anyone at arXiv about removing them and un-endorsing the submitters? That seems to me to be the obvious first step. I'd be willing to help if needed, although I don't have the time to filter the daily submissions for them myself.
I’ve contacted the ArXiv about the first batch of these that came up. They said they’d look into it but don’t share results of investigations. I haven’t checked whether the papers are down. It feels like fingers in a dike if we can’t figure out who is endorsing these authors though!
John Baez (Jun 12 2025 at 16:32):Did your first batch include Recursive Difference Categories and Topos-Theoretic Universality by Andreu Ballus Santacana? That was a crank paper discussed here earlier. It's still up! Santacana is also responsible for the new one you folks are talking about today, Analytic Uniqueness of Ball Volume Interpolation: Categorical Invariance and Universal Characterization.
John Baez (Jun 12 2025 at 16:33):I checked earlier, and Santacana appears to be in the department of philosophy of UAB Barcelona.
(He's definitely got the Grothendieck bald-head thing going on.)
Kevin Carlson (Jun 12 2025 at 17:26):I reported the papers of Barreto that David Roberts opened this thread with. Unfortunately they're still up and there have been two more since then. They're all in GM now, though, which I guess is the best it seems we can generally hope for.
David Michael Roberts (Jun 12 2025 at 23:27):The moving to math.GM has been patchy. Some of the ones by one author whose primary listing is CS.lo haven't moved, while those that were listed under math.CT have. Presumably because computer scientists are even less well-equipped than a generic mathematician to judge what CT is actually AI-generated crank material.
Mike Shulman (Jun 12 2025 at 23:44):If the people submitting these things are actually employed by reputable institutions, perhaps we should contact their employers.
Kevin Carlson (Jun 13 2025 at 02:25):The first person I reported is unlocatable online, IIRC. But that’s apparently not the case for everyone.
Zoltan A. Kocsis (Z.A.K.) (Jun 13 2025 at 10:29):David Michael Roberts said:
I think there's a real risk of the image of CT being tarnished if this type of stuff becomes too common. The number theory people know how to funnel cranks away from their arXiv category, if category theorists can't do this, it's not a good look.
Every week, one or two of these papers make it into the math.LO/cs.LO announcements, which is frankly ridiculous. We had a person who just had a couple of their articles GM-holed last week get through to cs.LO again this week. Especially disappointing since at the same time, I know multiple people with solid academic affiliations, long records in logic, and academic email addresses who've seen their announcements blocked/delayed while they appealed (e.g. conference extended abstracts misclassified and rejected as "abstract-only submissions", or a PhD thesis randomly rejected) :/
I don't think this tarnishes the image of logic itself, but it's certainly a big source of noise and not a good look for arXiv moderation.
fosco (Jun 13 2025 at 11:33):A possible solution is to set up a small website that collects these papers and flag them as "probably bollocks". A small number of us, committed to express a judgment evaluate these submissions pointing out "this passage is AI generated" "the second sentence at page 2 doesn't make any sense" etc
It takes a lot of work, but we all know what's the rule here:
image.png
I agree that this state of affairs tarnishes the reputation of category theory/ists and I think there is only one way to nip the problem in the bud, that is taking responsibility and vehemently assert that "yeah, no, we do not recognize this shit as category theory or even as decent math"
Chad Nester (Jun 13 2025 at 11:53):I'm not sure this kind of "negative curation", in which we maintain lists of things that are bad, is the way to go.
In an ideal world the function of journals is to be lists of things that are good, or at least probably not bad.
David Michael Roberts (Jun 13 2025 at 12:13):One option is to make sure all these dodgy AI-generated papers have comments on PubPeer. See eg https://www.pubpeer.com/publications/D52D1CC22593701472A83CFB9C2FD8 If the obvious red flags are documented here, then a list of links can be curated in a place category theorists have control over, or sent to arXiv admins, or employers of people making this nonsense.
fosco (Jun 13 2025 at 13:52):Chad Nester said:
I'm not sure this kind of "negative curation", in which we maintain lists of things that are bad, is the way to go.
History disagrees https://en.wikipedia.org/wiki/Index_Librorum_Prohibitorum lists of things that are bad can be used to repress heresy.
Matteo Capucci (he/him) (Jun 13 2025 at 13:56):fosco said:
Chad Nester said:
I'm not sure this kind of "negative curation", in which we maintain lists of things that are bad, is the way to go.
History disagrees https://en.wikipedia.org/wiki/Index_Librorum_Prohibitorum lists of things that are bad can be used to repress heresy.
Index Paperorum Crackpoti
Chad Nester (Jun 15 2025 at 07:18):fosco said:
Chad Nester said:
I'm not sure this kind of "negative curation", in which we maintain lists of things that are bad, is the way to go.
History disagrees https://en.wikipedia.org/wiki/Index_Librorum_Prohibitorum lists of things that are bad can be used to repress heresy.
An inquisition would, at least, be entertaining :)