iiinova — where your digital self keeps evolving

Reposted by Gianni Gastaldi

Hagen Blix @hagenblix.bsky.social · 22/07/2026

Why Bernie Sanders is wrong - and nationalizing AI is a bad idea "AI is, above all else, a weapon of class war from above. Whether it’s the state that wields it or corporations does little to change that fact."

jacobin.com

The Socialist Case Against Nationalizing AI

AI is not a neutral technology whose benefits need to be redistributed. It is a weapon of class war from above, designed to cheapen labor and concentrate power. Nationalizing it does nothing to change...

718262

Gianni Gastaldi @giannig.bsky.social · 20/07/2026

My talk at the Literary AI Symposium organised by @monashuniversity.bsky.social is now online: literaryai.org/talk-gastaldi Feedback is most welcome, as I'm currently finishing a paper related to the content of this talk. The other talks are online as well; check them out! #CriticalFormalism

literaryai.org

Wherefore art thou critique? From literary criticism to critical formalism in the age of AI — Literary AI Symposium 2026

021

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

But that’s where the work is, indeed

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

That’s exactly the kind of formal work I want to think my colleagues and I have been doing (we’re not the only ones, of course, and that doesn’t mean we’ll succeed)

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

If one gave 1% of the budget these companies spend on empirical exploration to a bunch of mathematicians collaborating with social scientists, my bet is that results like this would quickly come up for transformers. Mech-int is just not the right research orientation from a formal viewpoint

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

After that, you know that you can't get more structure in your (word2vec) embedding space than the one encoded by the first k singular vectors of a pmi matrix you can construct by hand. The neural model is there for no other reason than optimized computation. The formal object is independent from it

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

Before that, everyone was swimming in the embedding space for structural snippets, medievally praising the magic of neural nets for the obscure properties it could make emerge (well, they still are)

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

100% agreed. But my point is precisely that this is bad *for maths reasons* (only then the question of why it exists anyway can be asked). My preferred example (and model) is the result by Levy and Goldberg, showing formally (empirical confirmation came after) that word2vec is implicit SVD

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

I'm with you in that LLMs—which are truly amazing—might be the occasion to interrupt critical theory's materialist slumber. But we can do a lot better than the pseudoscience accompanying product dev in tech corps. That's why I see a huge critical power in formalist perspectives at this point in time

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

The follow-up question is why this tech corps are doing bad maths, and the answer might be: because they can afford it. And only here is where material conditions might kick in again (and why my point might have come across as a materialist argument against your formalist-leaning claim)

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

The reasons for this being bad maths are above all formal, not material, and anyone who knows the foundations of arithmetic would see it

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

...(because for some unknown reason, that's what the model happens to be doing), and justify this by calling all this computation some fancy name ("reasoning", "prepostdevelopmental arithmeticity consciousness", or what not), then we should not but see bad maths disguised into bad philosophy.

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

If based on arithmetic data (say, billions of strings like: 2+3=5) someone trains a model and through empirical exploration of the model itself defines even numbers as those numbers that when you multiply them by 35, then divide by 7, then by 5 and then by 2, result in an integer...

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

If only because, whatever the epistemological vibe of the ML community is now, (statistical) models are not empirical objects themselves, so doing empirical research on models has limited value for establishing formal properties, if any at all.

100

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

But my point—which might have come across as skeptical without this context—was that, if one takes the formalist position seriously, big tech mech-int work (including the J-Space paper) is not the right place to look at. This is just bad formal/scientific work.

110

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

I've been defending such a claim for some time now (e.g., at one of @mttpsq.bsky.social workshops: www.youtube.com/watch?v=Nhrb...) including my own talk at this event.

youtube.com

Juan Luis (Gianni) Gastaldi - Toward a Critical Formalism

YouTube video by ERC project AIMODELS

120

Gianni Gastaldi @giannig.bsky.social · 08/07/2026

Hey Matthew, if I may extend this morning's discussion, I, for one, couldn't agree more with the claim that there are computational dimensions—in LLMs as much as elsewhere—that are irreducible to material conditions (in both senses of the latter expression)

120

Gianni Gastaldi @giannig.bsky.social · 01/07/2026

3 days, 7 scholars, one question: what happens to language when writing goes computational? Organised by Mujie Li at the School of Arts and Social Sciences of @monashuniversity.bsky.social #MonashUni

000

Gianni Gastaldi @giannig.bsky.social · 01/07/2026

Thrilled to be speaking at the Literary AI Symposium alongside @davidgunkel.bsky.social, John Cayley, Søren Bro Pold, @bengrosser.bsky.social, Dennis Tenen, and @mkirschenbaum.bsky.social 🗓 6-8 July 2026 · Online · Free - literaryai.org #LiteraryAI #DigitalHumanities #CriticalFormalism

literaryai.org

Literary AI — Symposium 2026

261

Reposted by Gianni Gastaldi

Thomas Seiller @seiller.bsky.social · 09/06/2026

It took longer than initially planned, but our preprint on the connection between Isbell nuclei and linear realisability models is now online: hal.science/hal-05640508! This is a collaboration with @giannig.bsky.social, S. Jarvis and J. Terilla (both from @thegraduatecenter.bsky.social).

hal.science

Making sure you're not a bot!

132

Reposted by Gianni Gastaldi

Thomas Seiller @seiller.bsky.social · 12/01/2026

New paper! We have been interested in R-enriched adjunctions, motivated to understand structures arising from statistical data over textual corpora. This first paper delves into the geometric structure. A companion paper devoted to the logical structure is being finalised. Stay tuned!

seiller.org

121

Gianni Gastaldi @giannig.bsky.social · 05/02/2025

If you’re in NYC, come tomorrow and Friday to discuss Venarcular AI at the Digital Theory Lab NYU. @leifw.bsky.social has put up a fantastique program, including @davidchalmers.bsky.social @whkchun.bsky.social @tallinzen.bsky.social and many other brilliant colleagues. You can also attend online

as.nyu.edu

Vernacular AI, 2-day Symposium

052

Reposted by Gianni Gastaldi

maybe: leif weatherby @leifw.bsky.social · 19/12/2024

on Feb 6-7, the Digital Theory Lab at NYU will host an all-star cast of talks for "Vernacular AI"

0194

Gianni Gastaldi @giannig.bsky.social · 17/01/2025

I'm excited to participate in this Symposium organized by the great @leifw.bsky.social at NYU, alongside so many outstanding colleagues. Drop by if you're in NYC on Feb 6-7! mailchi.mp/nyu/vernacul...

mailchi.mp

Vernacular AI: A Digital Theory Lab Symposium | Feb 6-7

031

Gianni Gastaldi @giannig.bsky.social · 16/08/2024

In NYC from Aug 19th. DM me if you’re around!

000

Gianni Gastaldi @giannig.bsky.social · 19/03/2024

I'll ask around, someone near me should know

010

Gianni Gastaldi @giannig.bsky.social · 19/03/2024

Not really ML/subsymbolic, but definitively worth taking a look if you don't know them, since they are surprisingly early: en.wikipedia.org/wiki/Strache... Troy Astarte has been giving a few talks on this

110

Gianni Gastaldi @giannig.bsky.social · 19/01/2024

Thrilled to see this paper finally out at the AMS Notices! Fantastic experience as a philosopher to work hand in hand with the talented mathematicians J Terilla and T-D Bradley on what a structuralist theory of language could be today. A modest proof that philo doesn’t have to be confined to ethics

https://www.ams.org/journals/notices/202402/rnoti-p174.pdf

140

Reposted by Gianni Gastaldi

Ted Underwood @tedunderwood.com · 10/09/2023

New blog post, trying to get beyond a debate that has so far treated students mostly as consumers of AI. In the twenty-first century, critical thinking will require understanding how models work.

tedunderwood.com

Liberally-educated students need to be more than consumers of AI

If I'm buying a thinking process, I really need to understand what I see when I look under the hood.

67331