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.
Ben Kaminsky (Jan 21 2026 at 19:51):"Think of attention as constructing what Grothendieck called a ๐๐ผ๐ฝ๐ผ๐ ,a mathematical universe where meaningful computation can occur. In this view, the attention weights are not just numbers; they are building:"
"Iโve been working on a framework that treats transformers not as black boxes, but as computational gauge fields, where reasoning, performance, and interpretability come with mathematical guarantees."
So I recently come across a rather interesting line of thought being peddled based on applying non-abelian gauge theory to create robust AI models.
I can't help but wonder though...is this guy...for real? I personally don't feel well read up enough on modern AI to really make a judgement here, but his line of thought is something that naturally appeals to me.
Apologies, but one but one must be cautious when seeing posts on linkedn discussing sheaf cohomology. Anyone more learned than me have any thoughts? I'd rather hear the truth than get suckered in based on my own biases.
fosco (Jan 21 2026 at 20:13):
fosco (Jan 21 2026 at 20:15):on top of that, this text is using a trick predatory emails use to bypass the spam filter; see that the "p" of "๐๐ผ๐ฝ๐ผ๐" is a bit weird, lower than usual?
fosco (Jan 21 2026 at 20:16):so yeah, I put my money on "crackpot"
fosco (Jan 21 2026 at 20:19):"๐๐ฎ๐ธ๐ฒ๐ฟโ๐๐ฎ๐บ๐ฝ๐ฏ๐ฒ๐น๐นโ๐๐ฎ๐๐๐ฑ๐ผ๐ฟ๐ณ๐ณ (๐๐๐) ๐ฝ๐น๐ฎ๐พ๐๐ฒ๐๐๐ฒ๐ show us that attention heads organize into algebraic structures, sometimes collapsing into abelian subalgebras, something the industry once said was impossible."
Google isn't returning any result for "BCH plaquette", "Baker-Campbell-Hausdorff plaquette", nor I manage to find meaning in this jumble of words. So I suspect the concept doesn't exist; even if it does exist, I bet it is misused.
Kevin Carlson (Jan 21 2026 at 20:52):This guy is definitely not for real, and is in some sense not even a guy: this is very clearly AI-written, and doesnโt contain anything that even begins to look like a real idea.
Ben Kaminsky (Jan 21 2026 at 21:06):fosco said:
on top of that, this text is using a trick predatory emails use to bypass the spam filter; see that the "p" of "๐๐ผ๐ฝ๐ผ๐" is a bit weird, lower than usual?
Ah, very nice catch. I see.
John Baez (Jan 21 2026 at 21:47):Most importantly, none of this shit makes sense. And I know higher gauge theory: I helped write the paper Higher gauge theory which helped start the subject.
John Baez (Jan 21 2026 at 21:50):Don't try to learn math on LinkedIn: learn it from textbooks, then when you've got the basics down, move on to journal papers and the arXiv.
David Michael Roberts (Jan 21 2026 at 23:54):https://en.wikipedia.org/wiki/Betteridge%27s_law_of_headlines
Ben Kaminsky (Jan 22 2026 at 03:20):Well, my main aim is trying to understand the physical limits of how computation can happen. I got interested in understanding this emerging trend of geometry+AI because my gut feeling is that I could find examples of computational geometric bordisms/decorated homotopy 1-types "in the wild."
The professional version: https://arxiv.org/abs/2502.00873 (Sorry, helices again. You can't avoid them.)
If I had to sum up what's going on here in a few words..."Curvature as Computation"?
The state of this intersection, call it what you might, is similar to HTT in like the mid 90's? But much more fragmented. It's split between AI, CT, gauge theory, and diff geometry, existing in a kind of...primordial math soup. It will probably coalesce into a more manageable form by around 2030 or so.
I looked further afield than the usual places to see if I could find useful ideas "on the fringe", I guess not in this particular case.