Well said. I remember getting caught up some in the ACAB type stuff myself in my initial transition to lefty politics. And although I’d probably still agree with it in the context of the US and the way its system operates / has operated, it’s not universally applicable and a system that was going to try to replace the US in the region would need something like cops still (but actually serving the workers, like you say).

There definitely appears to be a kind of tunnel-visioning that can go into it. Like there are valid criticisms of the way the west approaches AI and how that approach embodies, or even accelerates, capitalist tendencies. But ecological destruction didn’t start with AI, that has been going on for decades at least. Replacing workers via automation didn’t start with AI, that has been going on for… what, centuries?

If AI can be a wakeup call for some people, that helps them reach class consciousness, then great. But if they stop at “AI bad, boycott it, shame, shame, shame” and don’t go beyond that, it’s not really helping the working class.

In the storied history of the exploiting classes and their abuses and barbarism, AI’s abuses are tame, honestly. If people had the same energy for anti-imperialism as they do for anti-AI, maybe liberation would be moving faster in the imperial core. But AI comes more directly for them in a way that imperialism doesn’t. For the people living under the shadow of empire, I don’t see the same kind of kneejerk response. Like Explosive Media that I mentioned elsewhere in this thread, the Iranian group doing those anti-imperialist AI Lego videos. They are in an existential war for survival and they are taking advantage of the tools available to them. They know there’s no self-actualization for them within the imperial framework, no chance of “getting theirs” and living a quiet, safe white picket fence life. AI coming for jobs is nothing next to the terror of imperial bombs and the prospect of one’s homeland being turned into a graveyard just cause the empire is mad that a people won’t kneel.

Thanks. I’m sure I’m taking some inspiration there from Gabriel Rockhill, Jones Manoel, and others.

Yeah, most people there are alright and genuinely nice, but there is a really toxic lib minority there that’s overly represented unfortunately. Grad being a Marxist space really helps filter that sort of thing out.

That of itself says a lot about moderation on there. It’s also amusing how they couldn’t find a reason in the thread to ban me, so put it down as me having multiple accounts. Never mind the fact that my accounts predate federation and hexbear. It’s like yeah, I made two identical accounts seven years ago just for this moment. 🤣

you can really tell Americans are well represented on that instance 🤣

haha I never even noticed, I guess markdown parser thinks it’s an email

Yeah that kind of stuff is pretty cringe.

I work in tech and I develop software. LLMs generate code a lot faster than a human, but it is simply worse code; it is buggier, it is less secure, and it is less technically correct to the specifications. It takes more time to get that code up to snuff after the fact, especially given the verbose bloated code bases created by LLMs, than to write it by hand such that it was correct from the start. Also considering the real costs of running these things in such a way that you get any sort of “usable” result ends up costing more than having an engineer do it right, I stand by my point.

If my codebase needs to pass a manual code review which finds an endless slew of LLM hallucinations that need correcting, it adds an immense amount of time to the process to rectify them. And these tools create insanely long winded and bloated code bases so the review itself is much longer from the outset. We’re not talking about rigorous mathematical proofs, it’s much more akin to a smoke test, and even that is a huge burden to bear when using those tools such that I view them as useless from this caveat alone. And it’s basic best practice in the industry to perform these reviews, they are vital to quality results.

Even hobbyist software dev work would feel this impact. If you aren’t testing the functions an LLM writes you will be putting buggy work with major obvious edge cases and vulnerabilities out there. The tests they write for themselves are often incomplete, happy path, or plainly avoiding actual testing by just asserting true.

From seeing these shortcomings in a field where technical proof and automated testing is a possibility, I can only imagine the level of human review necessary on less quantitative work to ensure any sort of reliability or accuracy. They cannot be trusted as accurate or authoritative at any point in the chain. ML can be useful as statistics for fuzzy and approximate data, but that’s really where the usefulness ends

Another way of looking at it is: Are Explosive Media’s Lego videos not conveying a message? AI not having the conscious will and worldview to actively choose to generate a specific message does not prevent humans from using AI generation as a means of helping them construct a message. The idea that AI is bad for crafting a specific message is a point that has become harder to defend over time, the better AI gen has gotten at listening to highly specific prompting.

“AI” for one is not a thing. Secondly there is no way some magic technowizardry is going to solve all our problems. This is pure liberal idealism. Also it’s not even a novel concept. We have had cybersyn for like 50+ years. “AI” is just a buzzword and you all (by proxy) shilling for these grift companies is pathetic.

It’s not just marketing. China has expensive AIs as well and has been experimenting with scaling them up to large scales also. There’s a lot of corruption in the US and that leads to a lot of marketing overhype, but it’s not just that.

ANN technology in general has existed for decades, yes. The main difference recently is that most computer scientists were convinced scaling them up was pointless, because the main issue was models overfitting. Overfitting is when the model scores highly on your tests by just memorizing all the data rather than actually learning to abstract the patterns you’re trying to teach it from the data. If you give it a larger neural network, it will have more long-term memory to memorize more data, and so it would seem at first glance that scaling up a model would just make them more prone to overfitting and thus less intelligent.

However, as GPUs got better, people just started to experiment with scaling them up, and it turned out that this wasn’t true. Scaling models up does actually tend to improve their ability to learn the abstract patterns. Nobody actually knows why. The thing about ANNs is that they are kind of a black box, it’s not even fully understood how they work. But the discovery of these “scaling laws,” just as an empirical fact, is what kicked off the international race for AI.

It is kind of like when Einstein and co wrote that letter to FDR about the potential for the atomic bomb. They could see it in the data that such a thing was possible, even though no one had built it, and so they predicted it, and this made the government race towards building one before anyone else could. The discovery of the “scaling laws” implied that if you kept scaling ANNs up, they would get smarter, and so who could scale them up the largest would have the most advanced AI.

Although, in practice, it’s more complicated than this, because two ANNs can be of the same scale yet vastly different in intelligence. The work to combat overfitting is still useful, because better training methods allow you to squeeze out more intelligence out of the same number of parameters without scaling it up. Chinese researchers and companies like at Alibaba and DeepSeek have been much more focused on trying to improve the quality of the model by improving its training methodology to minimize overfitting rather than simply scaling them up infinitely, and this has allowed them to drastically improve the capabilities of even fairly small models you can run locally.

But they still do try to scale as well. Alibaba’s Qwen3.7-Max has over a trillion parameters, and it costs similarly to Anthropic’s models to use. Scaling is, in practice, empirically proven to at least be one tool in improving the capability of ANNs. Modern ANNs tend to have a development loop where they will take the latest and most up-to-date training methodologies and use that to train a model at as large of a scale as possible, but then someone will discover new training methodologies that let you improve the quality of the model without scaling it up, and so then they will go back to the beginning of the loop, training as large of a model as possible using this new training methodology.

Every time there is an improvement in the training methodology, this trickles down to models of all scales which all improve, but the largest scale models still always remain the most intelligent, and those models require an enormous amount of compute to train, and thus are intrinsically costly. While Chinese models tend to be cheaper, they are still not that cheap for the largest models. You could easily wrack up several hundred dollar bill over a month using Qwen-3.7-Max if you are building projects with it continuously.