I guess it makes sense for a population raised on a diet of capitalism and imperialism to believe things like “humans are now obsolete” and “this must be how primative cultures felt when they were wiped out by superior white technology.” but damn are people easily fooled when it comes to AI hype.
I think that people who fall for the marketing around AI might be the most insufferable creeps on the planet. That includes the ones that hate AI but still believe the hype that this algorithm that is bleeding money will still be around in ten years let alone make humans obsolete, which is just all kinds of weird self-objectification. Capitalism really has fucked peoples self-worth.
Not only is the AI industry losing record amounts of money, pretty soon people are going to be too busy trying to grow food during an ecological collapse to even think about building and maintaining datacenters.
Looks like someone isn’t aware of the multiple slurp juice paradigm
When the fuck are people going to wake up and realize that YOU CAN USE MULTIPLE SLURP JUICES ON A SINGLE APE??
TIL you can use more than one slurp juice. My live reaction: :all-my-apes-gone:
Leave my Luddite ass behind and go off to your glorious techno-utopian Martian colony, that’ll sure show me.
unfortunately that’s all bullshit hype and what actually will happen is we will be stuck with them running our planet into the ground
Welp I’ll keep the concrete for their bunker air vents spinning
good luck against the drone swarms soldier o7
neither NFTs or VR promised to replace workers, that’s a pretty big difference
The only main difference is that AI can actually do tangible things. It’s just a matter of what actual value there is in those tangible things. Some things that AI can do is valuable. Most of it is not
The labor theory of value would imply that anything that AI can do doesn’t represent labor and therefore cannot be valuable. Only the models themselves, the hardware they run on, and the people thinking up the prompts, could possibly create actual value.
Automation is nothing new under a marxist framework.
LLMs are a means of production. They required a lot of labour to create (research, infrastructure, training and testing) and now embody the value created by that labour. When used to ‘create’ value, LLMs are transferring the value of the labour they embody to new commodities.
Like all automation technologies LLMs are allowing the owners of the means of production (GPU, AI companies) to extract more surplus value while driving the rate of profit closer towards zero.
I found this paper has a really clear view on it: https://www.nature.com/articles/s41599-026-07030-4
I doubt Infamousbit was referring to “value” in a strictly Marxist sense.
But to your point, yes AI models are machines (capital) that labor is applied to and they produce value together. When they’re worked as capital by labor, they produce far more value than the labor could produce on it’s own, which drives down the cost of their product. This is true of any useful machine. There are many cases where AI software produces useful output very quickly, the open question is whether they can justify their costs in the long term.
NFTs and the metaverse have already proven themselves worthless, or at least relatively worthless compared to the peak of their hype.
NFTs have worth! Very useful for money laundering
I hate what llms have done to machine learning, and what capitalism does to automation generally.
There are amazing real advanced in machine learning that are fucking mindblowing. Translation is mostly usable for communication in non technical matters and models are small enough to be practical to run we nearly have a damn bablefish (albiet one where we need to be tolerant of misunderstanding).
Computer vision has had huge advances and continues apace. So many potential benefits.
Dictating to a computer has never worked so well. Models for speech to text are tiny and fast now, for the first time you can actually use your own computer to dictate and it be ok. Massive accessibility benefits.
Language models are a neat thing. It’s kind of cool you can sort of talk to a computer and something almost like writing comes back. It’s interesting, and specialised models probably have some powerful and fun things they could be used for.
But so many of these advances are stained with grossly unethical violations of trust, and because none of them justify the trillion dollars vomited up by guliable investors everything is bundled together and thrust into every problem in an attempted at enclosure of thought itself. As a side effect building one of the largest surveillance and control systems ever.
Then in order to build a moat it’s all got to be as wastefully big and expensive as possible which heavily degrades the actual usefulness of outputs and makes things impractical to use, while also necessitating further extractive data hoarding.
Vile and sad.
I guess more broadly I like technology and automation but I like the ordinary kind. The kind where nobody is promising impossible dreams if you’ll enslave yourself. I like my dishwasher and my washing machine, I like my cake mixer, combine harvestors are awesome, diggers are an immense liberator.
I want computers to be boring again. An ordinary technology that focuses on things like economy, maintainability, reproducibility, safety, and liberation. Not this demented dream of a machine god that will crawl out of the pyre we raise with all our culture.
I’ll be honest, I think we can oppose AI as much as we want, but we need to have an understanding of its capabilities and how it works.
With the amount of cash the pentagon and 3-letter agencies are pouring into it, I think assuming it’s all going to be gone in 10 years is kind of naive. I mean, the pentagon supposedly used Claude to choose that school in Minab to target. Hegseth is firing people for not using AI in their daily tasks. Add to that the announcement a couple weeks ago from the heads of the big companies that they want to slow down and get governmental regulation sorted out, and I don’t really see it going away entirely anytime soon.yeah the US MIC actually benefits from genAI there’s no reason it would ever go away.
I haven’t been able to quite articulate my thoughts yet, but the all-or-nothing attitude I’ve seen from a lot on the left just feels misguided to me. At the barest minimum, we need to develop an understanding of its capabilities, which is incompatible with total rejection.
As long as capitalism exists and a data center draws more electricity than a mid-sized city, there can be no compromise.
I just don’t see how this is a useful attitude. AI is going to be used to monitor, identify, exploit and punish you. Without any knowledge of its capabilities and operation, you won’t be able to even begin to strategize how to counteract this. I’m not saying YOU must use it every day, but ignorance is exploitable.
I think it is obvious that there are productive applications for machine learning. As a subset of machine learning algorithms, I am sure there are productive applications for LLMs and adjacent technologies as well. As long as the person fishing for a convincing argument (maybe they can help find bugs! maybe they can create programming language bindings for us!) is the same person who is insisting we also use them as a substitute for human art, literature, translations, educational materials, doctors, teachers, and therapists, I think all related infrastructure should be burned to the ground. Until the critics are the ones driving development, all development should be resisted.
In terms of these tools being used by anti-capitalists in a struggle for liberation, dream on. We can find some marginal applications for small models, surely (I run LibreTranslate as a service on matapacos.dog, for instance), but the deficit in resources here is on the order of trillions of dollars. The capitalists are in the process of building out hundreds of gigawatts worth of power stations to run hundreds of square miles of data centers. Understanding the technology doesn’t do anything for us beyond recognizing what the function of a Robocop is, or what the implications for automated precrime detection are.
The axis of resistance understands what an aircraft carrier is, but that doesn’t mean aircraft carriers are a useful tool by any means in their anti-imperialist struggle. It doesn’t mean their time should be wasted studying the minutia of preventative maintenance tasks carried out aboard an aircraft carrier. The economics of the empire and the resistance are polar opposites.
This is how we should approach the AI build-out. We should recognize it as a weapon particularly suited to (and reflective of) the economics of our enemy. We should understand its capabilities as well as its limitations, weaknesses, and specifically how to paralyze and destroy it.
These multi-billion dollar weapons are a form of commodity fetishism. We look at them at abstract and are awe-struck by their capabilities, but their ability to operate in any capacity relies on extremely vast networks of obscured social relations. We cannot expect them to continue functioning as such when those social relations are disrupted or transformed. In fact, we already see the cracks manifesting in the example of the aircraft carrier.
These feel contradictory, but the second is basically my point.
Understanding the technology doesn’t do anything for us beyond recognizing what the function of a Robocop is, or what the implications for automated precrime detection are
We should understand its capabilities as well as its limitations, weaknesses, and specifically how to paralyze and destroy it.
I think the point that we are burying our heads in the sand and ignoring the capabilities of this technology is the part I disagree with mainly. This technology has been aggressively shoved into every aspect of social life and hyped in the media for several years. It has transformed the economy, the labor market, and even the foundations of epistemology. I reject the assumption that people who resist it unconditionally are choosing ignorance.
Look, it’s just big talk. The ALF was big talk and, in my eyes, morally completely correct! But there was also a complete failure to strategize and poverty of political philosophy and understanding. Class war isn’t a fucking game, but by taking an all or nothing approach one neuters themselves and their own stances viability. I’m not saying to be a social Democrat asking for moderate reform; but you’ve got to be able to articulate it better and in a more honest way.
I’m not opposed to it being available for scientists to study and research. I just don’t think it should be privatized or used for anything outside of learning to spot when our enemies ise it for propaganda.
I also don’t think there is much point because it isn’t something that can last given the money and resources it consumes and thr current trajectory of our collapse
Oh they won’t want it to go away, but they will no longer have the means to run it
we need to have an understanding of its capabilities and how it works.
Let me take a shot. Here’s some broad strokes.
Often the task of a computer is to take in some data and transform it into another kind of data. A program like Photoshop can take a large RAW photo from a nice camera transform it into a smaller PNG suitable for dank :logo: memes. However you need the work of a specialist to build such a program. Just as often it is beyond those specialists’ capabilities to build such a program; many such cases! :all-my-apes-gone:
Trying to get a computer to create their own programs is as old as computer science itself. Neural Networks for example are very well studied since the 50s, and the recent ML techniques today are a part of that active area of research. Neural networks (including Attention networks[1]) are really neat because, with good data[2], you can “train” them to transform one set of data into another. In computer science terms we can train models that mimic any function. This includes creating models of programs that are beyond the capabilities of specialists, such as working sound recognition. With good data and powerful hardware, you too can
whip your computer repeatedly“train” a model that recognizes the sound of bird farts. :bird-wat:It’s one thing to understand neural networks, but quite another to have hardware to train and run them. It was only recently that computer hardware became powerful and ubiquitous to train and run these models. Yes,
gamers are the reason why rocks think. But it’s not just GPUs; storage, networking, RAM, and more became more powerful and less expensive[3].Just like the Schwartz, there’s upsides and downsides. Just like teaching regular people, how you train a model is just as important. You can train a model too hard and Overfit a model, where it learns the test data too well but it fails when out in the wild. Its analogous to Teaching To The Test in education. The other problem is getting useful training data, and the real world is anything but. The dirty secret of ML is 80% of the work is scrubbing data. Much of this work is outsourced at poverty wages. And that is not talking about the spiraling prices from the token sellers putting pressure on business budgets.
I think assuming it’s all going to be gone in 10 years is kind of naive.
Yes. Nothing in Computer Science ever goes away. It’s got that Charlie Sheen
briefcases of cocainetiger blood; It just waits. This is AI’s third hype wave. The last two blew up so badly the resulting pullback in funding is called AI Winter. What will go away is the unlimited money. When all this investment doesn’t have the promised returns, investors will stop investing and with that all the major companies and SOTA models will be DOA and there will be yet another crater in the industry. At this point the industry is mimicking the Open Marriage scene from Arrested Development :dead-dove-3:While I see the demise of commercial AI companies built around huge closed investor-subsidized SOTA models, I do not see the end of these technologies. Personal computers are still getting more powerful, and local models[4] are increasingly capable. SOTA models are required to be experts at everything which bloats their size and training exponentially, whereas more narrow-scoped models[5] are far smaller to train and run. There is also continuing research for taking existing models and fine-tuning them without requiring a full retraining[6] lowering the hardware requirements dramatically.
For us on the left “AI” and all its runty children are just another Means of Production waiting to be turned towards the interests of the working class.
Attention Is All You Need is the major paper that led to our current Neural Network renaissance, but hardly the only one ↩︎
and I cannot emphasize enough it must be good data. ↩︎
Recent hardware price spikes because of industry circlejerking notwithstanding ↩︎
For my purposes, Local Models are any model individuals or organizations can run and manage. Closed Models are those that are managed my others and lack meaningful control for the end user, such as OpenAI and Anthropic models. ↩︎
Such as the aforementioned model tuned to the music of Bird Farts :bird-screm1: ↩︎
we could fine-tune our Bird Fart model to look for the beautiful butt music of @Snort_Owl@hexbear.net :bird-screm-2: ↩︎
You’re assuming the pentagon isn’t delusional. It’s going to be gone in 10 years because the current collapse will reach a point where the resources we have can no longer physically support the infrastructure and costs required to run it
Capital is dead labour, that, vampire-like, only lives by sucking living labour, and lives the more, the more labour it sucks.
Proof that just saying it doesn’t make it correct. That being said, AI is a bit different than the first two examples you used. They have the corporate overlords pushing us to use AI on a massive scale.
Take my work for instance. I am
toldencouraged to use AI at work as a tool. This of course makes it so dumbfucks use it heavily to the point that it’s generating garbage essentially.Fun times, we are fucked as a species.
Skipped cryptocurrency
