A community for everything relating to the GNU/Linux operating system (except the memes!)
Also, check out:
Original icon base courtesy of lewing@isc.tamu.edu and The GIMP
I'd like to call it my tireless helper, but the AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.
So the AI was wrong multiple times but it is being given credit because the human powered through the process.
If a human had done it I assume they would be called lazy for not trying hard enough. Meh.
Yeah but it's wrong FAST. That might sound like the punchline of a joke but it's really useful to see an idea written out in code to see why it was a bad idea and what to try next.
I believe it takes a deep knowledge of programming and LLMs in general, to know when they could be useful and how to use them in those cases. You first need to know they are not a magic genie, they are a machine learning thing, trained on tokens, that will output the more likely next token. And understand how the training process, the weights and the whole thing works inside to know how its going to react under certain inputs.
Then you must deeply know the problem you are dealing with. You need to know what's good, what's bad, when a solution actually works and when it does not.
If those factors are meet I don't see much issue in using them for programming.
Yeah. Linus knew what he was doing here, all the AI was doing was writing simple debug code and reading the output. It was just saving him from manually writing that code. And given the prodding he had to do, I'm not sure it actually saved him any time or energy.
And this was a debug session from hell, enormously helped by an AI doing much of the grunt-work.
I'd like to call it my tireless helper, but the AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.
I suspect those things have been trained by people who may not be quite as stubborn as I am.
But while the AI was ready to give up several times, it did keep adding debug code and analyzing it faithfully when I pushed. So credit where credit is due and I let the AI write the commit message above.
Anyone disappointed by Linus' flawed moral conpass hasn't been listening to him
Good. Glad to hear he's using the tools available to good effect
And that's where we start to toe the line on what could be deemed as an "ethical" use of LLM in free software production. Local models used to debug, but not write code? Maaaaaaybe acceptable?
Sorry but no. Contrary to what some activist types want you to believe it's not about actual ethical use when we are talking about software.
It's about LLMs being a tool that should be used where it's the proper and efficient tool for the job (see: here) and that should definitely not be used where it isn't the proper tool (vibe coded bullshit) just because some PR guy told you that this tool is your shiny new hammer and everything is a nail now.
When LLMs can produce proper reviewable code use them ffs... alas, they can't in any case complex enough to actually justify their use.
And I expected the comment and the discussion to be controversial. I am on the team that avoids AI as much as humanly possible, and I still write code the old-fashioned way (maybe with some LLM-free macros via Bash, but otherwise AI-less unless Google has inadvertently slipped something of the sort in my search results). This is because I'm well aware of the widespread dismissal of copyright law for training data, the obscene costs of training itself, the hallucinations and the sycophancy intrinsic in the way a neural network is trained. But unfortunately, the cat is out of the bag, and plenty of developers now use it routinely, and those same developers are effectively trying to toe the line to see how much are they allowed to use the cheat engine before being demanded to stop.
it’s not about actual ethical use when we are talking about software
This deeply misunderstands ethics. All actions are moral choices. Do I understand you correctly that you propose the polar opposite, that no actions are moral choices, with regards to software? I don't expect you'd say it's ethically fine to write software to orchestrate a genocide.
No, I just reject the notion that the ongoing discussion about AI use in software production is a moral/ethical one. That's a bullshit narrative to actually discredit all the good reasons against AI by making up a mostly fictional "but it's just ideology"-argument.
Yes actions include moral choices. But actions that are stupid and make no sense at all don't need to be discussed in terms of is it ethically okay to do it, because that's not the interesting point there.
I won't discuss about the moral choice of hitting my thumb with a hammer, even less to when I tried hammering in a screw like a moron. So I will also not discuss imaginary topics of ethical use of AI in coding. Because it's bullshit. It produces low quality slob. It's a scam trying to tell people with no clue about programmning that they can totally do it, too, while trying to gas-light programmers into believing that they really, really need to use that tool for coding or be left behind. Using AI for coding because you already used it for debugging tasks is not an ethical discussion. Just like using a hammer for a screw because you already hammered in nails is not one. It's a discussion about what AI CAN do and what it CAN'T. Discussions if you SHOULD use AI for a specific task have to happen AFTER you determined if it makes sense first place. And to return the hammer analogy a third time (it's a magic number after all...): If someone would try to argue that we need to discuss the ethics of hammering in a screw because it already worked with a nail, would you agree that all actions have ethical implications that need to be considered or would you dismiss him because the ethical discussion is not relevant. I did the latter in my comment.
Again: If using AI for coding for real world purposes (and I don't talk about non-technical people building their low quality app here... that's mostly a gimmick) was viable, sure... use it. Or let's start an ethics discussion then. That's fine with me, too.
But don't pretend that the actual discussion right know, mainly lead by clueless people evaluating a new gimmick with very little understanding of its actual capabilities (and limits) and CEOs willing to believe in any snake oil if it promises to help them get rid of those obnoxious workers demanding pay (it won't - every actual fact and number tells us that AI is in fact more expensive, even before figuring in the human employees also needed to hold the leash) is one about ethical use in coding. It isn't. It's about a discussion between those that use AI for the things it can do and avoid it where it's useless and those that somehow have found their new religion and saviour and try to make everything about a) those in tech that use AI because they obviously have seen the light and converted and b) those still left in the dark and fighting their new god for made up moral or other reasons.
That's a fairy tale completely ignoring the third group: technical people that use AI only and exactly where it makes sense. A fairy tale mostly pushed by a few people behind AI development because they already sunk such insane sums into it that they need to believe (and make you believe) that an (human level or more) artificial general intelligence is just around the corner. Just think about it: there people are actually taking about the equivalent of creating new life (new artificial conciousness) to justify their insane investments. And we should talk about the ethical implications of using AI in coding? How is that not a distraction?
Btw... now that I thought about it (and spelled some of it out) I want to walk back to my earlier statement. If AI is actually able to do soon™ what the morons behind it are proposing, that's indeed the moment to start an ethics discussion. A very complicated one about definitions of conciousness and life... good thing that this will not actually happen.
Not really. It's still not ethical to exploit the labor of others, even when our favorite computer nerd does it.
Ultimately the working people who created the data that was used to train the AI have not given consent and were not compensated for their work. Instead, Linus is paying money (probably the Linux Foundations' money?) to a large corporation that is exploiting the unpaid work, skills and culture of human beings.
In ethics the ends don't often justify the means.
Not to mention it's unclear what the LLM even did to help debug this issue, because he seems to suggest it lead him him astray multiple times claiming that it was impossible to fix.
How is that functionally different than a person who learned to code for free on Stack Overflow?
Because it's very much not a person, and it didn't learn anything.
A for-profit corporation simply scraped the net and fed a bunch of other people's work (the direct result of human labor) into a system of biased nodes and anthropomorphized it by calling it "intelligence" and "learning".
This is exploitation of labor of the working and creative class, nothing more, nothing less.
It doesn't have the agency of a human being, nor does have the creativity or individual perspective of one. It has no senses of it's own or lived experiences to draw from, and it is not sentient. Furthermore, it doesn't interact with the world or economy like a human being, it does not create things like a human being, and if has no human rights or even copyrights.
LLMs are not people, no matter how some people choose to treat them.
I am not a person inclined to believe copyright is entirely useful in the first place; but it is very illuminating that people of a certain class can essentially get away with mass theft under copyright law and be allowed to repackage that data and juice the market with the thing they've made using it and meanwhile average people are very much indeed beholden to copyright in public use cases where money is involved. Its especially crooked when you consider how the public is subsidizing so many of the underlying infrastructural costs of such a massively inefficient tool. We live in a welfare state, the average person just isn't the benefactor of it. Things haven't actually changed too much since feudalism.
What if they give the resulting LLM out for free, Apache licensed? And don't take any steps to anthropomorphize it at all?
In other contexts, that would be classified as fair-use, as long as it stays non-commercial.
There are other factors that go into the legal determination of fair use, including the amount of the work used and the potential effect of the use on the economic market for that work.
It's not a correct interpretation of fair use to say that anything non-commercial is legit. (And, even then, in the case of corporate AI it is very commercial...)
https://fairuse.stanford.edu/overview/fair-use/four-factors/
But beyond that legal definition, let's talk about what seems "fair" in a basic human sense:
If you were a painter who who makes art for a living, does it really seem "fair" for someone else to take your work, without any sort of consent or compensation, and use it to produce other works that are in direct competition (for time, money and attention) with you? How is that "fair" in any sense of the word?
That sounds like exploitation to me.
Call me crazy, but I think the working people whose labor and creativity have made "AI" possible should be reaping the rewards of its existence, not the capitalists and oligarchs in control of the big tech companies.
But what does it matter if:
The model weights are distributed, for free.
Run by the user, not profiting the trainer at all.
Its output is strictly used for noncommercial use; nothing the user ever makes a dime from, nor that competes with the original artist?
Thats the same thing as a fanfic to me. Or open software. Or per your analogy, a personal (noncommercial) painting inspired by another's work.
I'm not simping for big tech or LLMs or anything, I'm just saying if one rips out the commercial aspect completely, that takes a lot of the bad ethics with it.
Personally I think everyone should be benefitting, and I don't think that's gonna happen if the big corpos lobby successfully to ban foss alternatives or if hardware stays perpetually expensive, but for coders as a whole it wouldn't be the worst if we can simply run a foss model locally for a use case that makes sense, especially stuff like debugging. I do agree with most of your points, for what is worth. I just don't agree with copyright law or IP law a whole lot even if it sometimes benefits regular workers, these laws were reshaped long ago to protect big owners of capital and we need to move away from them at some point.
If you were a painter who who makes art for a living, does it really seem “fair” for someone else to take your work, without any sort of consent or compensation, and use it to produce other works that are in direct competition (for time, money and attention) with you? How is that “fair” in any sense of the word?
This is exactly how it works in the real world. I can go teach a class* on how to imitate the art style and techniques of any living famous artist, and as long as I’m not knocking off a direct copy of their pieces, it’s kosher per copyright law and per social norms.
I could even show the artist’s works unaltered in those classes as long as I’m doing so to comment educationally.
That’s been the standard for literal centuries and most creatives are not offended or upset by it. I personally would not at all be offended or considered it unfair if someone taught a class on how I made things I showed publicly.
No, that's not actually how it works in the real world. You're equating the human process of learning and practicing art to scraping and feeding a bunch of stuff into a machine learning algorithm, but they couldn't be more different.
Of course a human artist learns from studying the works of other artists, but not at all in the same way that AI does. A human artist has a lifetime of lived experiences which contain a perspective that is unique to their place and time, as well as direct sensory observations and a worldview than informs how they learn and create. Human artists also have individual tastes and personal agency that inform the kinds of work that they like and want to study from.
In other words, when a person draws a cat, for example, they aren't just recycling images of cats that they've seen in other people's artwork, they are also imbuing their artwork with their own understanding of what a cat is, how it looks and acts, etc.
People don't function like statistical models. They don't experience the universe in the same way, they don't learn in the same way, and they certainly don't create or interact with artistic media in the same way. Even a digital painting made by a human is not analogous to generative AI output despite the fact that they are both pixels in the end. It's an entirely different process from start to finish. Would anyone seriously try to argue that what Suno is doing with music in any way equivalent to a person learning how to play the piano or the guitar?
On top of that, a human is obviously not going to have the same potential impact on the market that generative AI would have. Even the most experienced writer can only thoughtfully write a book so quickly, while an LLM can be used to generate countless books in a single day. The market impact of a use is one of the key pillars to making a determination of "fairness" in a legal/copyright sense, and it's abundantly clear that human learning and "machine learning" are not comparable in terms of their market impact.
In other words, when people equate AI to human learning, it is nothing more than anthropomorphization and personification. What is fair for a human to do one way is not necessarily fair for a corporate computer system to do in an entirely different way. It's comparing apples to oranges.
If we're going to talk about the fairness of using copyrighted works for AI training, we have to talk about that specific process, as it really is, and how it affects society. People who equate a person trying to learn how to manga with a corporate image generation machine strip mining every image on the internet are doing nothing but providing cover for some of the biggest, richest corporations on Earth while they exploit the work of unpaid creative labor.
People don't function like statistical models. They don't experience the universe in the same way, they don't learn in the same way, and they certainly don't create or interact with artistic media in the same way
Why is this relevant to ethics. Why do you need to learn like a human does, in order for it to be ethical. Why is it only exploitation when a machine does it.
You’re lumping all of machine learning into the bucket of tech bros, ChatGPT, Facebook and such.
There was a thriving, small scale world of machine learning stuff that existed before they hijacked the public’s attention, and it never went away. It still exists. It will continue to exist when this bubble at least deflates some.
That's not my intent. Machine learning as a method of computation isn't the problem.
Among other things, the problem is how the business of AI is sourcing their input training data, how these outputs are being used to manipulate people and markets, etc.
ML is sometimes the best tool for certain computing tasks and succeeds where algorithmic programming fails. Sometimes traditional logic and algorithms is preferable for it's determinism and easily understandable behavior.
The AI industry however is built on exploitation of human and natural resources, and it also built on top of a massive economic bubble and is hurting the market for jobs and human made things.
Comparing AI to human learning is neglecting to acknowledge the problems that we currently face with how AI is currently being trained and deployed by the biggest players in the industry.
Okay, well, what about Nvidia Nemotron’s series as a counter example?
It’s built on an open dataset; its training regime isn’t secret, it’s reproducible. It’s Apache licensed. It’s not published to manipulate anyone; it’s mostly presented as a research tool, or a starting point for users to customize for mundane text processing tasks. It runs reasonably efficiently on a large variety of software and hardware, with a standardized architecture.
Of course AI, as it is mostly presented to the public, is an existential problem.
But you’re trying to fundamentally tie text-based machine learning to the tech bros as a root for its ethical issues. It’s not that simple. Even if Sam Altman and all his kind keel over tomorrow, the complications the existence of LLMs and other generative models present is not going away, and we will have to deal with models that can do unethical things without anything unethical in their training regime.
If you were a painter who who makes art for a living, does it really seem “fair” for someone else to take your work, without any sort of consent or compensation, and use it to produce other works that are in direct competition (for time, money and attention) with you? How is that “fair” in any sense of the word?
You’re equating the human process of learning and practicing art to scraping and feeding a bunch of stuff into a machine learning algorithm
You were not talking at all about LLMs in your rhetoric: you were talking about artists painting and I responded to what you wrote. You made the analogy linking LLM output to an artists output; I did not.
Please engage in good faith or self-ban.
It doesn't have the agency of a human being, nor does have the creativity or individual perspective of one. It has no senses of it's own or lived experiences to draw from, and it is not sentient. Furthermore, it doesn't interact with the world or economy like a human being, it does not create things like a human being, and if has no human rights or even copyrights.
We don't know enough about creativity and sentience and agency to know any of these things. And we definitely don't know whether they are relevant to the ethics of intellectual property. Most people don't even agree on whether intellectual property is ethical in the first place.
A for-profit corporation simply scraped the net and fed a bunch of other people's work (the direct result of human labor) into a system of biased nodes and anthropomorphized it by calling it "intelligence" and "learning".
I'm interested in how you know this when it didn't state what LLM was used. Ethically trained models exist.
But regardless, the function isn't very different. Both ingest works made by other people and produce something using those works as reference materials.
The mechanisms in which this is achieved might be different, but they'd have to be since we're dealing with binary switches instead of cells. But people tend to forget that neural networks are an attempt at programmatically mimicking the way the human brain works.
The human brain is just a meat computer, after all.
Stack Overflow was always about sharing your learning and knowledge with other humans. Just like open source. We didn't anticipate, and didn't give permission, for it being used to train AI.
So the problem is that silicone is doing it instead of carbon?
We dont even know what LLM he used. It's entirely possible it's an ethically trained model. I know for a fact that Amazon has ones that were trained on licensed works. Adobe has one for image gen too.
Professional athlete uses bracelet to run fast.
If you buy the same bracelet, you'll be just as fast!
I admire Linus Torvolds a lot. His works, the orginazations he's built, and his public speaking are all really well thought out imho, but I don't care when he poops and how much the log weighed.
Every time he does anything AI now a couple or articles crop up and get passed around. Its honestly annoying.