“AI is not democratizing art and knowledge; it is privatizing and automating it under the control of billionaires who, like the personality cults enforced by the führers of Benjamin’s era, demand that we view them as geniuses to whom we owe deference – and even, in the age of ChatGPT and social media, our very words and identities.”
Typically, LLM coders don’t use local LLMs. Or even if they do, they didn’t train them. And even if they did, the lack of social cohesion remains.
That’s not really true though. There’s hundreds of open source models.
Most people don’t self host things in general. We like to pay others so we don’t have to worry about things. That doesn’t change the fact that you can run these models yourself if you wish and they will continue to become easier to run yourself.
I think there’s a right way and a wrong way to use it. Linux for example is allowing it, but in a sensible way that maintains cohesion. But I hear you on the abuses. I’m not defending those. If I was an artist that got my work consumed into the Borg I would be furious. There needs to be more lawsuits.
This is where all the exciting stuff is happening lately (for me, at least). When Apple announces a local model on the iPhone, I think even more money is going to pour into that research area (smaller, more efficiency, etc).
So if you’d be furious as an artist, why is Linux using it against people’s will fine? I don’t see how they avoid the cohesion problems, or what some lawyers brought up.
Disclaimer: I’m not a lawyer, this isn’t legal advice.
I think we need to be very careful with the rhetoric of it being okay for code but not for art. This is all the same technology and we need to maintain solidarity so we can have a unified front of all professions against it.
Your core error here is thinking code and art are in different mediums that play by different rules & letting borg scrapers loose on one of them is fine but on the other it’s not. That when a musician complains about being put in the training dataset without consent the complaint is valid, and when a programmer does it’s not. Or that ethics somehow apply differently depending on the type of output. Whichever it is.
They are both acts of communication and social acts. They create objects that express their authors. They are collaborative games of interpretation. You don’t just write the code and cause material effect, you write it for someone else to read and interpret, you write it to express a bridge between a problem and a solution which priorly existed only in your mind, you form attachment to the thing you create, sometimes uniquely intimate. The subjects do indeed not realize that is the case a lot of the time, and do not target themselves as such. That doesn’t change the nature of the process.
I don’t know any open source developer that thinks of themselves as artists. They’re all just making tools to make life easier.
Are you sure you’ve asked more than two? A majority that i know (including myself) do think of themselves that way. I’m a writer, both for the english language and computer languages, though sometimes I think of myself as more of a toolsmith instead (also a form of art).
Also I would say this says a lot more about you than it does about coding. It neither has to he tedious nor drudgery; the only times it was purely that for me were contributing to an oracle codebase many many years ago and a codebase in a language i did not understand. If coding is a job to you that you want automated then do yourself and everyone else a favour and only code for compensation.
Even if I were to steel man your argument that code is not art, I feel it is still important that we acknowledge all the other harms of the technology and stand in solidarity with everyone else that is affected.
Writers, artists, journalists and many other professions are affected by this deeply unethical technology, and we should make a stand for them as well. GenAI is not okay.
Even if we ignore the solidarity argument, we can also talk about environmental harms, who is pushing this technology, its effect on labor, resource usage, political effects, etc.
There are so, so, so many reasons why this technology is unethical and we should be resisting it, that there is no excuse to ever allow it.
This metaphor being used in inevitability/defeatism is always very funny to me because a tube of colgate is, like, the most defeatable object ever. Wash it down the sink and toss the tube in the trash. It doesn’t have agency. It’s only as much of a problem to your bathroom as you make it.
No, it isn’t. It does not liberate the means of production, it obfuscates them. A large ML model is a black box, it is not auditable, it is not understandable by its user, it can not be predicted or reasoned about, its own production is inherently opaque and centralized due to needing a shit ton of training data and compute to be born. Just because it gives you code does not mean the production of that code is under your control, it just moves the control point from “what writes the code” to “what writes the thing that writes the code”.
This is a much deeper disconnect than the usual dependency chain depth. If you’re trying to build something with npm, yes, sure, you have seven thousand abstraction layers under your feet and some of them are deeply demented and counterintuitive, but fundamentally they’re still something that can be inspected, something that (probably) has docs, something you can ask people about and get actual answers. LLM-assisted coding is worrying if the tone of english in your instruction files is fucking up the output some subtle way. “Make no mistakes”. Sending a plea and hoping you’re graced. Faith-driven development.
Again, training these things is still entirely out of grasp for majority of people due to inherent computational demands. Control of their creation is limited to those who can afford like, actual computers, with actual components inside them, and not a banged up school netbook with 2G RAM and 128G emmc drive. This is an even harder barrier than bloated build dependencies of some modern open source projects, because it’s inherent to the medium, you can’t just switch to a model that has lighter build process, because there isn’t one, there can’t be one.
You cannot inspect. You cannot understand. You cannot modify. Is this libre software? Is this your Robin Hood?
This was true even before the chip shortage. It’s getting much worse. I don’t have a computer capable of building these things and I won’t for years to come, possibly ever. If I were to use them, the people in control of my toolset would be hard-filtered to “has a european upper middle class salary or better, is into tech, is into AI, is into AI as a provider”. That’s not a lot of people. If I were unhappy about how the toolset works on some non-trivial level, my only option would be to come to these people’s metaphorical doorstep and plead politely. That’s not tool liberation, that’s tool colonialism.
simple question - why do you need AI in the delta?
are you not literate? this can be solved without AI…
More reasons for AI in the delta? delta is about mail and accessibility where even the internet is a luxury!