My background is that I am a PhD student in STEM. I’m not gonna doxx myself, but my research area is in mathematical systems theory. LLM systems are not my primary research area, but I have taken advanced coursework on classical machine learning, and LLMs will certainly affect my research area from a technical perspective. So I am obligated to understand LLMs, reinforcement learning systems, deep learning algorithms, classical machine learning systems, and basically the rest of “artificial intelligence” from a theoretical, mathematical point of view. In addition, I use LLMs extensively as a LaTeX assistant, to generate counterexamples, and to “fill in” proofs when I read math books.

I am also a lifelong anarchist.

---

Anarchist communities [especially in real life] need humans otherwise there is no community.

Absolutely. Anarchist users of LLMs do not intend to replace the human element. Literally the opposite.

GenAl is a stolen facsimile of a human who can never understand morals or praxis.

I don’t oppose proletarians “stealing” “copyrighted” work. Copyright is classically anti-anarchist rentier behavior. What I do oppose, and I believe you’re pointing to, are a few related but distinct problems with AI systems as they exist today:

1. AI systems pretend to be anything other than tools. It’s in the name, unfortunately. And it’s capitalist horseshit, but its name is actually capitalist horseshit fixed by the people who actually invented classical AI systems in the fifties. I.e., academia has had a long time to correct this terminological mistake and failed spectacularly. To some extent, the tendency of AI systems to pretend to be human is a byproduct of its training data, because it’s trained on human outputs, and except for a few weirdos, humans typically assert our autonomy in our writing. However, AI companies really should be trying to counteract that tendency; they go in the opposite direction. For this reason, I reject the notion that AI systems are “intelligent” in any way that resembles intelligent biological creatures, i.e. I reject the notion that AI systems are converging towards AGI or that they are even capable of approximating general intelligence. Furthermore, I posit that the characterization of AI systems as intelligent beings is capitalist marketing. The logic is simple: if these systems are intelligent, we would need to give them rights. Since capitalists control these machines, such a move would ultimately amplify capitalist willpower.
2. LLM companies have obtained basically all human knowledge without paying. Piracy is illegal for proletarians yet magically legal for the bourgeois when do it for some reason 🤔🤔🤔🤔. LLM companies then they paywall the sum of the world’s knowledge and refuse to even be open about the training data and resulting model weights.
3. AI agents (bots driven by LLMs) driven by thoughtless or malicious users overload conventional websites and spin up unedited AI-generated content. The proliferation of AI slop feeds back into point (3).
4. People use AI systems, like all other tools, to do heinous things. In particular, AI systems are used in the ongoing Gaza genocide, and are a critical part of ICE operations.

All of these problems are serious critiques of AI systems as tools. What you’ll notice about all four of these, however, is that they are primarily socially driven phenomena. These problems are contingent on the existence of hierarchical social relations. In fact, these problems have pre-AI counterparts:

1. ((1) is the most novel IMO.) Capitalists hire and/or bribe actual proletarians to simply repeat capitalist talking points.
2. The academic journal complex.
3. Classical bot farms and human trolling operations.
4. Basically the entire history of technology.

It can never understand freedom or control.

“AI” systems can’t understand anything. They are mathematical algorithms implemented on silicon-based computers.

It will however poison our communities (look at what SpaceX/Anthropic is doing in Memphis, TN)

This is not an inherent feature of the technology. We could have absolutely trained these models, albeit at a vastly slower sustainable pace, through the use of distributed and decentralized computing strategies. The issue is that the perverse incentivization structure of capitalism pushed AI companies to grow unsustainably, at great expense to the ecosystem and the people of the world. I.e.: capitalist interests of quick money and infinite expansion triumphed over proletarian interests of survival, dignity, freedom, and sustainability, as usual.

And it is my view, as an anarcho-communist, that by seizing the means of production from the capitalists, in particular their AI hardware, and distributing these systems more equitably around the world, we can remake these AI systems to some extent into sustainable systems at an ecologically appropriate scale. At least we would be able to try — capitalists, as monstrous as they are, never even try to do things correctly.

and isolate us from other humans within the community which is so important to us.

I cannot stress enough that I do not want to see AI replace humans in human-centered tasks. In particular, I am completely uninterested in AI-generated “drawings” or “music”. Art is an activity by sentient beings. Unedited AI “art” can be technically competent (or not), but it is ultimately just a product of a machine and therefore not tremendously compelling to me.

And I also agree that we are seeing a crisis of isolation, but my view is that:

1. This crisis predates AI.
2. It is caused by capitalism, specifically the gradual commercialization of previously free or low-cost activities and spaces.
3. It is exacerbated by ecological destruction and encroachment by the State, which simply makes the world outside your home a worse place to gather.

Obviously, this is the wrong place but I have to ask, [prior to moving the discussion to this thread] what makes you think the extreme negatives of genAl (specifically to the anarchist community) can be excused?

The extreme negatives of AI must not be excused.

They must be confronted and ruthlessly reversed. And in order to do this, we must conduct a holistic analysis of why AI technologies enable these heinous things. This is impossible to conduct, i.e. useless, without relating these technologies to the social relations that motivate these technologies to destroy for their masters. Anarchists who use AI systems do not, in general, excuse these systems or their users. To the best of our abilities, we use them with the care required not to replicate capitalist harms. And we must be held accountable for our failures on that front.

However there is a big difference between my pirating a book like the Lord of the Rings (which is credited and I can see the title of) and how Meta scraped Anna’s archive and funneled all that data into an approximation of all combined work.

Not really, no. I would have zero problems if you personally did that in a way that didn’t destroy the ecology, and then released your weights, model architecture, data, and training proces. In fact, in my view, you would be doing the world a great service. It is my personal hope that we live long enough (we as proletarians living in a world doomed by capitalists) so that I can someday train an AI on all of Anna’s Archive. I’m sorry if that’s offensive to you, but that would be so unbelievably fucking cool and useful.

What Meta did and then released as open source

* weight available, not open source. Huge difference: you cannot reasonably recreate Meta’s model weights on your own. In my view, besides being unethical, it’s bad data science because it’s not reproducible!

The works were consumed and are used in silence to camouflage as human thought. No credit will ever be contributed. And so no one can even hope to appreciate the original works that was used in training.

Yeah I can live that. E.g., I don’t cite my calculus textbook every time I use calculus in a paper. I cite the things that must be cited. And when I wrote my metal album ten years ago, I didn’t credit any artists, even though my work obviously draws from the work of dozens of metal bands. And I don’t expect to receive any credit if my work influences someone else’s music, or even if it gets sampled, and that thought doesn’t bother me at all. It would be neat if someone did shout out my work, I’d appreciate it, but really I’m just happy if someone listens to it. (Not gonna post my album here because that would insta-doxx me lol. You’re not missing out on much 😆.)

---

These are my views on AI systems:

• “AI” and “artificial intelligence” are capitalist buzzwords and should not be used. I prefer the term “statistical learning*” systems, where the asterisk is used to remind the reader that “learning” is a series of technical mathematical termd that are distinct from what people in social sciences mean by “learning”, in particular when applied to biological creatures. (See Vidyasagar’s book and Vapkis’s book for technical definitions.) I am open to terminological suggestions, as I’m not completely comfortable with the “learning*” terminology.
• AI systems can be useful tools. But that being said, not everyone is going to have a use for them, and their adoption must, like any tool, be voluntary with respect for the user’s autonomy. And that autonomy has been systematically ignored and violated by the capitalists who foist their AI products on us.
• Large language models have utility beyond auto complete for reasons we do not yet fully understand. Basically, the logic is this: they train LLMs to do tasks A, B, and C. You train a language model to do tasks A, B, and C, all well and good…but then you shove some completely different tasks X, Y, and Z into the machine. And the weird thing is, for some reason, the machine can do (say) task X. We cannot mathematically prove yet why this is the case, but this was practically seen to be the case. I personally suspect that natural language has some deep underlying structure that we cannot see yet, but informs both the AI’s ability to learn*, and our ability as humans to learn. (Anecdotally, as an autistic person, verbal language has been the key to my ability to survive for this long. Frankly, I simply lack a bunch of social skills that I can make up for by using language to read books and watch videos and just…not have to spend as much energy on being a social animal.)
• More on mathematics: classical machine learning has extremely good theory. In particular, you can use probabilistic inequalities as design tools to pick algorithms, compute asymptotic bounds as the number of samples increases, and (less classically) compute finite-sample bounds on error probabilities. The mathematical community is currently porting these tools over to deep learning algorithms and LLM systems…but it’s tough work! Even classical machine learning theory books require some measure theory and functional analysis background. It’s a lot of fun! But it’s also really really hard.
• Your understanding of AI systems and their capabilities is probably out of date. The capitalists are obviously overselling it, but there is a rapid increase in the capabilities of these systems.
• Pursuant to the previous point, commercial AI products are rarely “just” a LLM anymore. In particular, a lot of these products have access to Python environments, in particular with SymPy, compilers (e.g. they typically have access to a LaTeX compiler), and even a Linux container. Really, if you hear the phrase “agentic AI”, it basically means a bot or family of bots “driven” by LLMs. (Pursuant to my first point, agentic AI bots do not have “agency” like sentient biological creatures do! The responsibility for what these bots do lies with their human operators.) I highlighted SymPy, a Python library that does symbolic mathematics. LLMs used to be completely incapable of symbolic mathematics. While they still often make mistakes that are hard to spot, AI systems are capable of driving a Linux container. In particular, they are capable of crapping out SymPy code and running it, where SymPy does the mathematics in the background. However, the really interesting development is that AI systems can crap out Lean code. Lean is a formal proof verification programming language. It is basically the best you can get in terms of proving that a mathematical theorem is correct. (Disclaimer: I am absolutely not an expert on formal verification! My formal understanding of Lean is quite limited right now.) This will absolutely revolutionize mathematics.
• As anarchists, we ought to have sufficient technical training to be able to back up our sociopolitical ideas. You obviously don’t have to go the full PhD route like me (and I don’t recommend it lol, I’m just fucked in the head in the exact ways that make me a good PhD student 😆), but I think it would help if you tried out the HuggingFace tutorials, or if you pirated Sebastian Raschka’s book and did the exercises, to calibrate your expectations about AI. (Both of these should be completely free and open source, and doable on reasonable hardware…or Google 🤢 CoLab 🤮 if your hardware can’t handle it.) I suspect that a lot of people who are (rightly!) disgusted with the excesses of capitalist AI proliferation are not proficient in using AI systems because they haven’t practiced with it as an assertion of their rightful disgust at capitalists’ excesses. As I have said above, a lot of people will simply never have a use for AI systems, in particular if you don’t do technical work…but if you do technical work, it makes sense to at least try self-hosting these systems, and maybe putting in a little bit of time every couple months to see what the state of the art…for the times, they are a-changin’.

---

Lastly, this comment was written by a real autistic human on a cell phone with zero AI assistance, as are all my comments. Yes I actually write like this 😆

LLMs are fundamentally fascist tech that serve to poison the information ecosystem and erode people’s capacity for critical thought.

dbzer0’s general tolerance for genAI is the main reason I will never create an account on that instance. This is ridiculous. These people rant about the environment, ethics, genocide, and Big Tech, and then happily devour AI slop.

EDIT: For those of you who are ready to comment about “wonderful” local LLMs, see this.