I'm not sure if this conversation has been had on this sub before - but there's this comment I read on the linked post that more or less sums up by observations after vibecoding (which ironically came to me in the middle of my programming learning journey - I do think I've suffered learning about contexts)
As for the argument about decreasing code liability that I have faced from others - I do think vibecoding pushes us harder to commodify, and hence push into production more than we should ideally be doing. Aren't we increasing code liability if we do this?
Edit:
Some more links that have better arguments than myself:
-
Another post by the same author that perhaps does a better job of explaining the "embedded (or structural) politics of AI" - namely, how the faults here are baked in to the productive process itself.
-
Stop Gen AI - a website run by an independent researcher with resources on how it can be achieved.
The way I see the argument about utility/usefulness, it's not that it's universally justified, always, and we can stop there. It's that like many tools, it's more about what it's used for and whose interests it supports than anything else. I mean, gen AI is not child labor, so we can put that comparison in the bin. Can it be destructive to the environment and have cause for concern for that reason? Yes. So can much of industrial practice (and its recklessness tied to capital is why we're dealing with climate change right now, which has been going on for a long time now, long before gen AI was ever a thing). But if we were to say, we therefore must abandon all industrial practice, that would leave it to the exploiting entities of the world and they would use those machines to crush every liberation effort with ease. So yes, "useful" is not sufficient on its own as an argument to do something, but it doesn't need to be with gen AI.
As for the argument about decreasing code liability that I have faced from others - I do think vibecoding pushes us harder to commodify, and hence push into production more than we should ideally be doing. Aren’t we increasing code liability if we do this?
I'm not sure, I'd imagine it depends on the interests going on. Like if a business is trying to maximize profits and go by a "move fast and break things" philosophy, maybe they'll put on pressure to vibe code their way through. But this could build tech debt very fast, which is an issue prior to AI, but maybe not at such an accelerated rate. And with how many services are "live service" now, taking tech debt seriously is critical to being able to sustain the service.
From things I've heard about Linus over time, I have no interest in defending him specifically, but the way I interpret his position based on the quoted information is that if you're a skilled programmer and you're using AI to assist you, that's not an invalid thing to do. Trying to get AI to do what you want in a complex codebase when you have no idea what you're doing is possible (and can still be useful) but may fast have diminishing returns on a project of increasing complexity that needs to be sustained across years or decades of changes. But for one-and-done projects, especially personal projects where it doesn't need to match the legal expectations of a business transaction, I think that's where vibe coding shines more so; as long as it's "good enough", that's already way better than "never being able to fulfill the idea."
The comparison between AI and child labour was a hypothetical because, AI may be causing similar levels of damage - it generates through non consensually obtained datasets of human endeavour and hence is as unethical as CL. It's a bit like saying if we keep going down this road, we will become very proficient at writing prompts to take the easy way out while losing the experience of understanding the context. And the argument is, that within the logic of capitalism, it will all be because it was possible. Whether you say its merely a tool or not.
And if I'm building a weekend project (as opposed to a business product), I can speak for myself in that it was always equally an opportunity to learn about the underlying technology so that I can keep it running, have more ideas, and finally, be an expert in it someday. These days I vibecode projects just for the sake of it. And speaking from personal experience, I don't learn much at all. I've heard lots of others face this kind of cognitive decline.
Edit: I'm not sure if that's what you're saying, but if we are okay with LLM's becoming our experts in the long run, they should not be so probabilistic. As in, we can never really monitor at a low level what a given population will get in return if they prompt for x task. Much less, we don't even know how to monitor an LLM's generation process except by asking it to generate something that assures us it's vectors mapped out what an individual user can only hope is what it semiotically connotes.
The comparison between AI and child labour was a hypothetical because, AI may be causing similar levels of damage - it generates through non consensually obtained datasets of human endeavour and hence is as unethical as CL. It’s a bit like saying if we keep going down this road, we will become very proficient at writing prompts to take the easy way out while losing the experience of understanding the context. And the argument is, that within the logic of capitalism, it will all be because it was possible. Whether you say its merely a tool or not.
I don't think violating copyright--a mechanism of capitalist monopoly over particular forms of expression of ideas--is anything like child labor.
That said, for my own part (setting aside what Linus Torvalds is saying completely), I am not meaning to say it is "merely a tool." My emphasis is that the form the tool takes and whose interests it is used for is a vital part of evaluating it. For example, on a moral and ethical level, I despise nuclear weapons, but it would be foolish for me to push for the DPRK to disarm while we live in a world where having nuclear capability can help protect you from imperialist invasion. At the same time, I recognize the bombings of Hiroshima and Nagasaki, two civilian cities, as a horrific war crime, in spite of the fact that imperial Japan was doing fucked up things itself.
When the empire uses AI to help it target civilians, that's obviously a fucked up use of it. When China uses AI to help solve a medical science question, that's a very different story.
As for "forgetting" / becoming dependent on AI, I agree there is risk there because of how capitalism acts so much in the short-term. But again, this is not entirely an AI problem. Some of this is already ongoing institutional rot, due to capitalism choosing short-term gain over long-term sustainability. I don't believe we should be aiming for AI becoming "experts" for us in a broader sense of things. Rather, we should be aiming for AI being a helpful assistant to a human, instead of the other way around.
Would you say copyright can't be a tool for good, as you put it?
And if AI exists in any form outside of the subscription model, without hallucinations and bias, and without consuming our municipal water supply, I would've agreed that it creates that kind of worth/alternative similar to the examples on nuclear use you said. But I can't think of an example of such an alternative.
We're just not talking about who owns it and hence who benefits - which is, the corporations. I know China has good models but even those cost too much for me as I live in a third world country and programming was supposed to be a way for me to earn an income.
I'm not sure of situations where copyright is overall "for good". Certainly on a small scale, there can be individual artisanal situations where copyright benefits them and helps them have a stable career, but as far as I'm aware, the brunt of copyright benefit is large corporations using it to maximize profits and limit what others can do, creatively. There have even been cases of a corporation making some throwaway work with X IP name attached to it so that they can hold onto the intellectual property rights for longer, illustrating how absurd it can get as a practice.
And if AI exists in any form outside of the subscription model, without hallucinations and bias, and without consuming our municipal water supply, I would’ve agreed that it creates that kind of worth/alternative similar to the examples on nuclear use you said. But I can’t think of an example of such an alternative.
The nuclear example isn't supposed to be that it's good without negative side effects though. Going back to the DPRK example, them having nukes helps them defend against imperial invasion via the principle of mutual-assured destruction (MAD), but MAD itself is a pretty fucked up situation for the world to be in at all. I would rather we live in a world where the possibility of global nuclear holocaust does not exist.
So, with AI, is there currently any form of AI proliferation that does not have negative side effects? Probably not. Especially under capitalism, it is accelerating some of the problems that were already there. However, unless you live in a socialist state with a robust people's democracy, you and I are not deciding how many data centers get built. Capitalist states are certainly not keen on consulting the working class on such decisions. So in that context, the decision is not "should I vote yes for it or no," the decision is "should I make use of what they're going to build anyway, regardless of what I say, or should I boycott it and potentially make me less effective in the struggle for liberation as a result."
I know China has good models but even those cost too much for me as I live in a third world country and programming was supposed to be a way for me to earn an income.
I'm assuming you mean "cost" as in "personal damage to career prospects" because there are Chinese models like Deepseek and Qwen that can be used for free, to some degree. That said, I'm aware that gen AI poses a threat to job security for a lot of people. However, as emphasized before, this is not unique to AI. Automation has been coming for jobs for a while and programming itself is a primary enabler in that process of developing automation. It sucks, I'm not arguing against that at all, but again, this goes back to capitalism as a problem. People should not need to depend on X type of work to be able to survive and part of what made programming so lucrative and high demand in the first place was its contribution to automation, i.e. cost-saving in the long-term for companies.
I get your points, thanks for being with me so far. I want to say another thing in good faith and want to know if you think it makes sense.
MAD itself is a pretty fucked up situation for the world to be in at all.
Capitalism has scaled up production of components that we say suck unanimously because of the state it brings the world to - for instance, plastics that do not degrade, guns, surveillance technology. These are, to me, components for useful tools in good hands or made out of processes that can be utilised better.
If we are to resist the productive capacity tools or processes being misused, however, we have to go against the tendency of capitalism to misuse its productive capacities. And organising at the consumption level has statistically affected this tendency of capitalism to misuse/fall in wrong hands - e.g. for the BDS brands, and largely for building consensus against gun sales.
the decision is "should I make use of what they're going to build anyway, regardless of what I say, or should I boycott it and potentially make me less effective in the struggle for liberation as a result."
Even from a communist viewpoint, if we resist the relentless kind of upscaling of GenAI more critically we would have gained in our communities/organising spaces. I say this because the GenAI economy is always in a precarious state - with debt funded by big finance and many saying it's a "bubble about to burst". Our chance to fight this capitalism's big bet on AI can lead to a victory for the left.
Vibecoded software due to it's tendency to bloat, risks becoming usable only when backed by a trustable producer (this ends up being big tech a lot of times). Which makes the imperative to be informed about alternatives to GenAI enabled software equally important for the organised left.
Edit: I know you're probably not implying the left not be informed about alternatives to GenAI enabled software but, I mentioned it because if we accept that it must (be informed), an equally large part of organising would be to insulate ourselves from vibecoding.
Even from a communist viewpoint, if we resist the relentless kind of upscaling of GenAI more critically we would have gained in our communities/organising spaces.
I mean, we could try to resist it, but what would that even look like? Local communities that are strong enough as communities can try to resist data centers being plopped down in there and using up their water. For some, that may be a very immediate, existential problem. But that is not in itself resisting the reckless capitalist scale up as a whole. It's sorta like unions in that way. Unions can be helpful sometimes in the short-term, but if they never transform into a vanguard, they're just a layer of short-term harm reduction, not a fundamental challenge to the system. This is not to say it couldn't have any benefit, but that there needs to be a more comprehensive strategy.
Typically what I encounter in anti-AI is an individualist boycott mentality, or trying to pressure/shame others into boycotting, which to be blunt, is a shitty strategy that has never been effective for resistance to much of anything.
As to vibe coding specifically, I don't think it is any kind of inherent danger to programming practice and people on the "left" who use it alongside conventional programming can develop standards that try to safeguard against buildup of garbage. Aerospace programming has stricter standards than the rest of the industry long before gen AI was a thing. Standards are something that can be developed and enforced. Vibe coding, in my assessment of it, is largely a problem insofar as someone treats it as a replacement for a human in a project where longevity and quality is vital. Programmers have been arguing over, and tinkering with, design patterns for decades in their various attempts to make code cleaner, more maintainable, and more extensible. Vibe coding adds a new element to the mix, one of dealing with a sort of chaotic "expert novice" (the AI), who can be extremely helpful but can also cause problems if left unchecked. The capitalist's mistake is thinking the AI can cleanly replace people with no negative side effects, or that the negative side effects can be ignored and become somebody else's problem.