I mean, communities work best when debates stay constructive.
In a nutshell he says that Linux is a technology project, where technology comes first. If LLMs help to improve the technology then it’s fair game. If not, then of course not.
He also says something about Linux not being a place for ‘social warriors’, which is kind of ironic, given that Linux is basically a big ‘fuck you’ against corporations and closed source, i.e., a scoial statement. Given how much beef he famously had with Nvidia, this is somewhat unexpected.
But I interpret it as coming from a place of seeing LLMs really just as a new tool and nothing more. In an interview a couple of years back, Linus also said pretty much that. He grew up with basic tools, had to create new software tools to make him more productive and improve the quality of his work. He saw LLMs just as another tool that can aid in development. I suppose this hasn’t changed.
Given, that most people usually don’t program in assembler anymore, but use high level programming languages like C++, which simplifies the process of coding enormously, or even scripting languages like Python, which present another level of abstraction and simplification, LLMs do appear somewhat like a next step in this evolution.
Issue is, you have deterministic and formal guarantees in programming languages, whereas LLMs do not. This is by far not the only issue of course, as others in this thread have also noted.I wonder whether he uses LLMs in his daily work. Might change his perspective a bit.
The thing is that they rigorously review their code internally with In the project. It needs to be reviewed and cleaned up before it gets submitted. Someone who keeps posting junk is gonna get chewed out and ignored. So, the biggest issue with LLM code is kind of irrelevant. If someone useses it responsibly and finds a way to make it an effective tool, they’re not worried about it. Banning its use is kind of pointless since the worst outcomes are already banned.
As I understand, they don’t use it to generate code, but for scanning code for bugs. This is relevant as there are now many CVEs found by AI scanning tools.
Fun fact: Based on estimates of the value of the Linux kernel (around 2200 Million Euros in 2011), the size of the kernel (in 2011, version 2.6 was current, with a bit less than 20 million lines of code), abd average salary of kernel developers (for Germany, one could estimate 60,000 - 80,000 EUR/year), one can estimate the lines of new code that a full-time kernel developer produces - less than 1000 lines per year:
value=2200000000 cost_man_year = 60000 lines_of_code= 20000000 person_years = value / cost_man_year lines_per_person_per_year = lines_of_code / person_yearsI get to 545 lines per person-year that way.
Even if AI could reduce time to type in that code to zero, it would not speed up kernel development, since developers are 99% of the time doing something different than typing in code - determining what new code should do, organizing and structuring concepts, or fixing bugs (which is determining what existing code should do, determining what it actually does, and determining how to fix differences between the former and the latter).
The only difference between using C and an LLM is C doesn’t consume thousands of gallons of water and gigawatts of electricity per second.
You forgot the part where the llm doesn’t necessarily do what you tell it to do either
Linux is basically a big ‘fuck you’ against corporations and closed source, i.e., a scoial statement.
I mean, not according to the project leader, or the contributors, who are mostly companies.
If the 3 main groups contributing to a project are Intel, Meta and IBM, it’s not really a community project, is it?
deleted by creator
He also says something about Linux not being a place for ‘social warriors’, which is kind of ironic, given that Linux is basically a big ‘fuck you’ against corporations and closed source, i.e., a scoial statement.
That’s what you want it to be. But Linus wants it to be a technology project not what you think it should be.
Given how much beef he famously had with Nvidia, this is somewhat unexpected.
Maybe because Nvidia did things that made it difficult to implement technology.
I wonder whether he uses LLMs in his daily work.
He does. He’s used it to make an interface for a side project.
It seems you’re approaching everything from a paradigm of pro or anti capitalism and think of everyone as either an ally or an adversary. You’re the “social warrior” he’s talking about. You’re unable to focus on just the technology, everything has to be put into various political boxes. It’s really exhausting for people focused on technology to deal with this way of thinking. And it’s not productive in terms of actually implementing technology.
Linus can say “go ahead and fork it” because he knows the social warriors wouldn’t be able to actually do this. They would sit around endlessly debating which hardware and network configuration passes their political purity tests and wouldn’t be able to host a fork let alone do any work on anything.
To be able to implement any kind of technology you have to be able focus on the tech only and shut out everything else.
That’s quite a lot of assumptions you’re making about me there.
And quite a lot of oversimplifications.
What assumptions? In your post you’re putting everything into either a capitalist or socialist box as if that’s all there is. You think of Linux as only a “fuck you” to corporations.
Do you think Linus doesn’t accept PRs from corporations or something? Do you think Linux isn’t used by corporations?
I think you’re the one making a lot of assumptions and oversimplifications.
Code doesn’t exist in a pure, abstract vacuum, the environment around it shapes it and makes it useful. Technology all by itself is sterile.
When Torvalds says he doesn’t want politics in his projects, he means he doesn’t want the politics he doesn’t like to come up when discussing Linux. He’s made a lot of political statements throughout the years, but it’s acceptable for him because it’s his.
The planet is burning, and government-sponsored hackers are threatening all users of one of the most widely used kernel, but sure, we need the thirsty CVE-generation machine today because he’s had his fun playing with them.
Well, good luck enforcing the GPL with the low-quality generated slop. The lack of foresight from one of the world’s top software leaders here is frankly appalling.
Except the “social warriors” you and the incel hate, did do this. The Hurd Kernel works just fine for any business task, if Linus continues to ratfuck the Linux Kernel with LLMs, I imagine the more competent developers who don’t want their work ruined permanently will start porting over to Hurd.
But that’s beside the point. Linus wrote that the Linux project will always accept good code to improve the technology, regardless of source. If Microsoft would wake up tomorrow and switch the kernel to Linux and start to send PR for good features, Linus would add them in a heartbeat. Same as the rust decision. If it’s bad, it gets rejected, LLM or not. That is the entire fucking point.
Then you, and Linus, miss the entire fucking point. LLM code submissions look fine at first, but come in huge numbers and eventually lead to massive reverts because they’re pretty fucking useless. Code autocomplete? Maybe that’s a use for LLMs, but they are barely better than non-llm autocompletes. Actual coding LLMs? Universally shit, but are getting better and better at hiding their shit. Like how most LLM agents when asked health questions get 99% of shit right in an answer, and then fuck up so royally on the 1% you will die if you just follow their response.
And it can’t get better. LLMs don’t know know anything, they can’t. They only hallucinate. That is their sole capability.
So to sum up the danger is two fold:
-
LLM hallucinations get better at hiding behind plausible code.
-
LLM code submissions drastically increase the number of submissions of code, incentivizing using LLMs to review the code.
Fortune 500 companies are currently in this trap. Microsoft famously is having to roll back so many Windows 11 changes that they admit were primarily coded with Copilot.
FOSS is supposed to be smarter than this, not give into stupid trends that have hard fucking evidence they are temporarily, terrible trends and not a useful tool for the future.
You are using your emotions as statements of fact your terminology is buzzwords you are useless and annoying. Yea we get it you don’t like generated code that is not hard evidence of anything.
You have some surreal sentiment towards code but you seem to not understand that natural language programming, self generated code and artificial intelligence are holy grails that have been dreamed of since the start that have existed in many forms before chatbots got good enough to make waves.
They aren’t ‘making waves’ though. That’s the problem. Companies are quickly finding the permanent inherent limitations of LLMs in regards to coding and several have started desperately looking for Jr Devs again after having a policy of not hiring them for more than a year.
LLMs are objectively a dead end, all current evidence points to this. It would be great if it really were this miracle technology the people selling you the technology claims that it is, but it’s not. It’s not even competent. Bottom barrel slop devs from exploited countries like India are infinitely more capable than claude or copilot. And unfortunately they always will be.
You are just as dumb as the people that over invested you have no idea what you are talking about… Like seriously the extent of what you say is “I don’t like chatbots because they cannot solve every problem ever all at once immediately and I don’t understand code” those are the facts that can be extracted. You are a whiny bullshit artist. If you want people to take you seriously how about you take the subject seriously.
Disjkstra wrote a nice paper explaining why natural language is terrible to write code… It’s a dream for idiots, actual engineers dread the ambiguity.
Lol you are not knowledgable you think you are using words that have impact but you are just an idiot with nothing to say.
-
Ok, go ahead and run Hurd then.
I’ll be over here playing video games on Linux which actually has drivers for my video card.
I imagine the more competent developers who don’t want their work ruined permanently will start porting over to Hurd.
This will only happen in your imaginary world where being in the correct political box magically makes you better at writing code. If all of the people obsessed over the political alignment of technology (LOL!) left Linux development, nothing of value would be lost. It’s technology, it’s actually politically agnostic. But if you believe otherwise, then you’re not someone really interested in tech, you’re just interested in making tech yet another political football.
It literally isn’t politically agnostic, it’s Free and Open Source software, which is anti-capitalist.
Look, as long as the baby-shredder produces passable code, I’m all in.
🤦
I honestly expected better from him…
That headline is a hell of a cherry pick of his full email, which itself is in a very specific context.
For one thing the emphasis is very much not on ‘AI coding’ uses, instead focusing issue triage and security testing IIRC.
Is it? After reading the article and the original post in the kernel mailing list, IMO the article depicts Linus’ opinions accurately.
Was also astonished to find out that Linus considers people who advocate for sustainable tech are doing so solely out of fear and that AI can’t be criticised because humans also make errors.
Yeah, I read the mailing list as well as the article didn’t skew it much at all. Linus having hard opinions that piss is nothing new, he’s a bit of a dick even when I agree with him (to be clear, this is not one of those times). I am disappointed in his opinion, but not surprised. Linux is mostly funded by big tech, and Linus has never really been against them like the Open Source zealots like RMS. He’s just highly opinionated and historically has have little filter when sharing said opinions.
This is absolutely not true. The email is very direct, and the specific context doesn’t change anything. Here’s the full text:
I think it makes the point of sashiko - helping maintainers - unachievable. If the point to not use LLMs in general, let’s discuss this, not how to make each use case more complex.
It seems like [1] expresses a very anti-LLM position in general
Yes.
And no, that’s not the position of the Linux kernel.
I realize that some people really dislike AI, but this is an area where I’m willing to absolutely put my foot down as the top-level maintainer.
Linux is not one of those anti-AI projects, and if somebody has issues with that, they can do the open-source thing and fork it.
Or just walk away.
AI is a tool, just like other tools we use. And it’s clearly a useful one.
It may not have been that “clearly” even just a year ago, but it’s no longer in question today.
There are other questions around AI (like what the economy of it will actually look like in the end), but “is it useful” is no longer one of those questions. Anybody who doubts that clearly hasn’t actually used it.
Yes, it can also be a somewhat painful tool, both for maintainer workloads and just from a “it keeps finding embarrassing bugs” standpoint.
But the solution is not to put your head in the sand and sing “La La La, I can’t hear you” at the top of your voice like some people seem to do.
The solution is to make sure those LLM tools help maintainers instead of just causing them pain. There’s no question on that side.
We’re not forcing anybody to use it, but I will very loudly ignore people who try to argue against other people from using it.
And no, AI isn’t perfect. But Christ, anybody who points to the problems at AI had better be looking in the mirror and pointing at themselves at the same time.
Because it’s not like natural intelligence is always all that great either.
The kernel project has been and will continue to be about the technology.
Sure, the social angle of working on open source is important and often a very motivating part of the project, but in the end that’s a side benefit, not the point of the project.
This is NOT some kind of “social warrior” project, never has been, and never will be.
In the kernel community we do open source because it results in better technology, not because of religious reasons.
And so we make decisions primarily based on technical merit. Not fear of new tools.
He is such a back stabbing condescending out of touch deplorable libturd. He is the only tool I see. In the movie “their will be blood” when Eli slaps his dad for pulling down his pants… and then he slaps the shit out of him… Thats what I want to do to this fool. Computers are for humans NOT humans for computers. Linux is awsome because of humans… No more with the gipity… I’d rather be 2nd place then juiced with a limp dick. HOLD THE LINE AGAINST THE SWINE
My brother im christ
Are you ok?
You do realize who you are talking about. Love him or hate him but Linus Torvalds is THE Linux guy
I hate that these discussion often seem to happen in a vaccuum regarding the person.
This man has spent over 30 years writing/maintaining the Kernel. He most certainly knows more than yoi about the kernel the maintainablility and maybe most importantly: The community
Take it down a notch will ya?
deleted by creator
People are disappointed that Linus is pro ai
Being “pro AI” in my book goes a lot further than what they are doing. Using generative “AI” for coding with human oversight and the high standard of code review they have is a far cry from wanting to cram LLM slop into everything as much as possible. I think it’s good that they are trying to find security issues like this, for example.
deleted by creator
Yeah I think you’re right, when it comes to AI as a coding tool. I heard someone ask something like “where on the scale from flashlight to violin would you put LLM tools for coding”. By that, they meant, if you use a flashlight, and it doesn’t work, you’re pretty sure it’s not your lack of experience with the tool - the tool is just broken. But if you play a violin and it sounds like shit, it’s probably because you’re not versed enough to play it (or at the very least you don’t know if it’s you or the violin who’s at fault), assuming you don’t play the violin.
If one finds LLMs unhelpful for coding, is it them not knowing how to play it, or is the tool inherently broken? It’s very arrogant to believe one is so knowledgeable about a tool, one which is in constant development too, that it is not them who is using it wrong, but rather the tool that is wrong.
There’s a plethora of other areas of just criticism of the technology, but I think how useful it is for coding is not one of them.
Aside: your “manmade” line breaks make your comment harder to read on a phone. I guess it’s a hard habit to break, but I would suggest you try. Most places people type text to be read by other these days, one should account for vastly different resolutions and aspect ratios (e.g. vertical & horisontal monitors). Adding in line breaks where it looks good on your screen is like writing text for yourself to read. That’s my personal opinion at least
Reading comprehension is at such a massive cratering low, you really gave criticism on writing style for prose copy pasted from an entirely separate medium written by a completely different author
Jesus Christ people
Take it easy. And don’t just take my one comment as anecdotal evidence of reading comprehension. I will say, first off, I wrote the above shortly after waking up - perhaps not my most analytical piece ever. Furthermore, English is not my first language (kinda not even second). And finally, I have dyslexia, so I literally have worse than average reading comprehension skills.
Also, from my device, it seems only the first 2 paragraphs are in quotation, and everything after “yes” has regular formatting, which gave me the impression of that being the commenters own words, and with the weird line breaks. (I’ll add I know people who have typed like that but have switched, hence I made the remark.)
All I want to say is you shouldn’t use my misunderstanding of the comment I replied to as any evidence of general reading comprehension. It shouldn’t be. And you seem overly affected by this.
The fact that you have to “prompt-engineer” means that the tool has a terrible interface.
In your flaslight analogy, there’s an on/off button that doesn’t work. You have to flip the flashlight, tap it twice, check the batteries and flip it twice again. Then it turns on.
The fact that LLM coding tools are basically micro-managing with a special twist “knowing how to prompt”, makes it a shit tool for a lot of people. And they are justified in thinking so.
You can’t hone a skill you have little control over. It’s like a hammer that strikes a random area within a certain radius of a nail.
I both agree and disagree.
As for having little control, I feel like that happens in other disciplines too. The best archers need to account for the stochastic nature of wind patterns, and thus they do not always hit their intended target. But still, their accuracy is far greater than a novice. There is both skill and “luck” - if you will - involved in the process. You definitely can hone a skill you have little control over in the beginning, I believe. Maybe this is the analogy that should’ve been made.
I will say, “knowing how to prompt” feels like a weird skill, and I think the way we interface with these tools is kinda wacky. Regular text feels to fuzzy.
But it does help immensely, I find, to take every wrong agentic coding output as a learning experience. When I know the answer, and the agent failed, I ask myself why it didn’t fint it. Taking these opportunities has led me to be more proficient in the use of skill files, agent files, subagents, context window management, etc., all which have improved the agents output massively. I feel like a year ago, it was just prompting and copy-pasting (at least for me), but now, with tools like OpenCode, I can get a better looping effect with fewer errors.
I am curious about your findings
Don’t forget we are talking about a tool.
A good tool is easily documentable, and then usable with said documentation. A good tool has perfect repeatability It is easy to predict the outcome of using a good tool.
Current LLMs have neither. For all the user knows, it’s a black box that takes input and poops out output. You can only steer the output after some output has already been given.
I think we disagree on the nature of a tool. Or maybe I am thinking about more than a “tool”, but not using the appropriate word for it.
Image riding a very fast horse. There is no perfect repeatability, it might veer off, or react to something you didn’t see, but a good rider would probably manage to arrive at the intended destination. I think I view it the same with agentic coding.
Now, I agree a horse is not a tool, and that’s why I mentioned that my language is perhaps a bit imprecise. (I guess you could even call it an agent.) But maybe an “instrument” or an “implement” fits, although that’s a bit cold wording for a horse in general, as they have other qualities, as being cute etc… But I think you get my point.
I very much think LLMs can fall under the description of “useful implement for coding” today. But I do think it is a very poor knowledge retrieval implement. In fact I sucks for all these one-off questions, I my opinion.
I think the more interesting discussion is about how these AI companies consolidate power even more, how they steal data, how they pollute like crazy. But I think people who dismiss their use in coding perhaps view LLMs more like a flashlight and less like a very fast horse moving you from A to B.
Context engineering is a skill analogous to communicating a problem to fellow maintainers. This typically involves articulating the issue within an issue tracker, pull request, or comment using written text. Therefore, proficiency in “prompt engineering” is essentially a measure of one’s articulacy.
So it’s not a “useful tool for coding” then, if the quality of the output depends of one’s ability to articulate in a way that tickles the LLM “the right way”.
Also, no matter how well you articulate, the output is still somewhat random. You can produce two different sections of code with (essentially) the same prompt, which also just means that theres an amount of luck and percieved randomness in a tool.
Imagine you want a piece of software written but you’re hiring it out, like a lot of corporations do. You write up a loose specification and hand it to two different dev shops and you’ll get two wildly different results. The tighter the spec, the closer the results will be.
It’s not much different from that.
Indeed. I agree that it’s not a tool, but a type of outsourcing.
Following my analogy with the archer, promting the right way is like accounting for wind. And again, no matter how much the best archers in the world account for wind, they do sometimes miss. Still very impressive how high their skill is, even with randomness involved. And with agentic coding, one can always shoot twice.
TLDR: In short, like a lot of AI use the biggest problem is more how other people are using it that I have little control over.
The thing is that CodeGen utility varies greatly on what is being advocated and what situation it is being applied to. CodeGen getting things going when the operator isn’t strongly opinionated about the details for fairly common patterns is fairly strongly in the wheelhouse. It is somewhat worse at amending a project in the same broad ways it was able to do when starting from scratch, even a project the model itself generated. So as you progress, it is more and more likely that a human will need to understand what is going on to be effective in modification. This presents a problem as people work themselves into a mess and can’t get out of it and this causes a fair number of projects to just get abandoned because they can’t go anywhere.
Even as it has demonstrable utility, the real world implications can be a mess. For example, the other week someone used agentic AI to open up 70 ‘security findings’ on a project I work on. To it’s credit, it found one actual issue, and while another issue was incorrect, looking into it I did find a separate issue it didn’t notice, so I got two fixes out of it. However I had to deal with 68 completely stupid things that weren’t anything. The operator at least up front sent an email that they didn’t understand any of it and how much was real, but wanted to share in case any of it is useful. But I guess that’s the price I pay for the two valid issues that might have otherwise not been caught. Then a few days after settling those, another user opened up over 60 and they were all dupes. Then the next week another user opened up a bunch that were all dupes too. Now it’s ridiculous. One could argue that I could fight fire with fire and put an LLM on triaging the issues and closing out the dupes, but that means random github users can now make me spend my money on LLM services just by opening issues. Further, after being on the receiving end of AI chatbot triage in customer support, I hate inflicting that on the humans opening issues.
Then of course there’s the code submissions and expectations around how I should handle them. There was a longstanding understanding that folks try their best, but implementing feature requests takes effort and folks are broadly understanding at delays or being a bit down the priority list. Now someone will Claude up a merge request instead of an issue and be impatient because “Claude already did the hard work, all you have to do is accept it”, and it’s a mess of code to review. It’s not like the code at the hands of the operator is exactly good, for example someone sent a merge request because some feature raised an exception for them and blocked it from working and Claude “fixed” it. Problem is Claude changed the code to catch the exception, do nothing, and just say that it worked. In their test scenario, the feature was trying to make things the way they already were, so a no-op was no problem, but when I made their test case actually start from another state, it failed but still said it worked.
Another project I used to be a part of recently got handed over from the maintainers that long maintained it and lost issue to a super Slop-happy AI user. They got it rigged so it’s all agentic and “addresses all issues and resolves all pull requests automatically from anyone”. The project had some poor issue hygeine, so this meant the agentic code was doing things like fixing an issue from 2017 where the user complained it didn’t work with Ubuntu 16.04. No idea if the fix actually worked, but it did make code changes that realistically no one would care about if they only did what they want. Except now they are getting bunch of new issues because they say “hey, the latest update broke a bunch of stuff I relied on”, and another said “hey, it was great to see my issue get fixed last week, but this week the issue is back?” and messages saying “the update pace is ridiculous, why are there 3-4 releases a day, the former cadence was once a quarter”. It’s become a broiling mess of chaos because some AI enthusiast decided to make a nearly abandoned project that people were using a playground for AI usage. Dude even said point blank it’s fine, just open an issue and AI will fix it for you and if it breaks, then someone else just open another issue and it’ll get fixed and he is confident that eventually the users will shape the project into some equilibrium that way, so he has to do very little except pay for the tokens.
Buncha reddit rejects sounding off on a headline.
Excuss me… who the fuck do you are? Do you know were you are dorthy? This ain’t Kansas.
K
Saying “just fork it” is obviously disingenuous anyway. Linux is gigantic at this point and would be completely infeasible to maintain without a huge group of talented people and funding. He knows very well no one can just fork Linux in a meaningful way, just like you can’t just easily fork Chromium if you disagree with Google’s high level shitbaggery. Even MS threw in the towel on maintaining their own browser for the same reason.
So, is Linus just dictator-for-life? Does he have the final say on everything?
More or less, yes
I mean, pretty much yes, as far as I am aware.
I meam have you ever heard linus talk about other controversial topics? He is not one to give a fuck or mince words.
He knows very well no one can just fork Linux in a meaningful way
Fair enough, Linux-libre is technically an example of a fork. That being said, that project has a very specific goal of just removing proprietary packages and blobs, it’s a lot more simple and direct to maintain. I don’t know that you could truly do such a thing with AI/LLM tainted code without a hard fork and essentially no longer merging in from upstream. Vim Classic is trying to this right now, and it is a lot more doable since that’s a much smaller project.
Wow, they are at best cherry picking the article they cite on productivity with AI, if not outright trying to lie. They say
Last year, an METR study found that open source coders using AI tools were 19 percent less productive than those who didn’t use them, even as those AI-using coders said they felt 20 percent more productive. But in a February update on a follow-up study, those same researchers said that “we believe it is likely that developers are more sped up from AI tools now—in early 2026—compared to our estimates from early 2025,” citing early raw results and conversations with study participants.
But if you, y’know, actually read this February update, you would see that the researchers are lamenting the fact that they cannot draw conclusions effectively because of the proliferation of AI use. To wit, they say:
Altogether, these issues make it challenging to interpret our central estimate, and we believe it is likely a bad proxy for the real productivity impact of AI tools on these developers.
In fact here is the full passage Arstechnica cherry-picked from (emphasis mine):
Based on conversations with study participants, we believe it is likely that developers are more sped up from AI tools now — in early 2026 — compared to our estimates from early 2025. However, because of the selection effects in our experiment, our data is only very weak evidence for the size of this increase.
Based on conversations, which is to say, not on their data. This after the study a year prior which suggested developers are not reliable reporters of their own productivity.
Yeah, METR is an industry group funded by Google, Microsoft, Anthropic etc. I happened to summarize their work just yesterday actually: https://lemmy.blahaj.zone/comment/21358850 but one thing I think it’s important to point out is that there is no second study, not yet. The blog, and the open source dataset, are all we have, because the researchers decided to adjust their method. So the intimation arstechnica gives that they completed a second study is false.
And the selection effects they are talking about are pro-ai software developers being unwilling or unable to complete complex tasks without their precious ai anymore, whereas a year ago they were more willing to do both ways in order to prove how powerful ai is. In my opinion, as I stated, this reflects poorly on their levels of self confidence from using ai, more than saying anything about a speed boost/decline.
Ahaha. The research showed that developers don’t know how fast they are, but when we talked to them they said that they’re pretty fast, so we’d rather report on that. I see no problems or contradictions with this.
It would make sense that it’s better now though?
Claude Code came out in Feb 2025, widely available in May 2025. The tool itself has had many improvements as has the model behind it. It’s pretty much THE tool people get into AI with currently.
Shit, I can get bugfixes as oneshot prompts now with Fable 5 using the superpowers plugin (not really superpowers, just skill files and stuff to tell it how to use test-driven development and other nice practices). I describe the bug in some detail, it’ll create a failing test, then fix the issue. Some of those have not been trivial to track down and fix at all, but I can give it to an LLM and go work on another issue at the same time.
That’s irrelevant. I’m talking about what the researchers wrote, how that relates to their own prior research, and how Arstechnica reported on it. I am not making and have not made any claims about whether or not developers are actually more productive.
sees list of sponsors
If not corruption, why corruption-shaped?
Sad this article doesn’t talk about why open source projects reject AI, specifically because of the legal implications of all the copyright violations baked into their training data
Linus himself doesn’t talk about it, and pretends he doesn’t know.
I would have to look it up but I am pretty sure this has been discussed in the past
If you find anything interesting, please let me know, I’m interested!
None of that will ever be prosecuted, because it was done through the wealthy and their favored praxis.
That is only an excuse.
That is not how copyright works. But if you expand IP law that way, open source projects will be the first victim.
I don’t understand why so many down votes. It is like a rock and a hard place this IP but yet crapitalism snickers at the top. It is like " Thanks for the free work turds" like we are entering the new dark ages of Korpo Kyber Krime
Yeah. I think anything that contradicts that AI “steals” (not learns and not copies) is downvoted as heretical. I’ve thought about this quite a bit.
They never had anything copyrightable taken away from them by an AI. But they might be afraid of loosing their jobs. So they ignore facts, deny reality and logic. I think this type of “post truth world” behavior really took over in the last decade.
I also have a more crazy theory but it's a bit silly lol
I think for many people LLMs and generative AI represent this “deep insult” to who we are as humans. LLMs can be more creative than something like 50% of humans. Like if you’d prompt 1000 humans to draw something creative and then let other humans evaluate their results with that of LLMs, the LLMs would probably win in terms of “this is more creative”. Maybe it’s only 10% or maybe it’s even more. Maybe it’ll stall and maybe it’ll continue to improve. But it’s a question that can be scientifically answered.
And many hollywood movies, romcoms, sitcoms, dramadies, they all have the protagonist do something “creative” and sometimes it’s even a plot point that they should get to do their creative thing professionally. It’s idealized as the ultimate expression of humanity. And then LLMs came and took a massive shit on that idea lol. A program replacing someone writing or drawing something is seen as something far “worse” than a robot or machine replacing someone doing manual labor. One is liberation of dirty work, the other is supplanting and making irrelevant.
So it can’t be. The insult and psychological wound is so deep that they now must deny reality and attack anything that contradicts them lol.
Obviously actual “high art” or “high culture” is really something completely different than the content and entertainment slop that is produced today, even before AI. Actual high art isn’t in danger of being replaced by LLM until we create an AI that is sapient.
Actually that is how copyright law could work.
On the ingest/training side I think the courts have been mostly consistent with your view and sided with the AI companies, though some lawyers are still opening new cases so at least some of them aren’t so sure.
The problem is the defense the AI companies have given is still a potential problem for software. They state that even as the ingest of works allow the model to generate material that would violate the copyright of trained data, that it’s the fault of the operator, not the AI company. So even if a user oblivious to the original material manages to get a model to create a knock-off of a copyrighted work, it is still the user’s fault, not the AI company.
This implication to my knowledge has not been tested in court. I haven’t seen cases where someone found and sued an AI user for plausibly oblivious copyright violation. Now I cannot imagine that the consequences can be very punitive if they can show the AI models will produce such a violation without the user indicating any awareness of the source material, but they may still demand it be ripped out, which depending on age of code could be a huge mess and a great loss of functionality of other things that depend on the offending code. Even if the particular case could be won on arguments about whether it meets the criteria of an infringement in that case, fighting the case is hard enough.
The question is whether the risk of accidental copyright violation and the risk of such a violation even being noticed is significant enough. Currently in the GenAI hype world where tons of money is seen to be had from fueling the phenomenon, that risk appears low. Question is if the hype dies down and the money suckers want more money, will they at least start making copyright claims against folks focusing on how the AI defense was to throw the users under the bus. Again, open source projects themselves are probably low on the list of targets, as financial penalties seem unlikely, but it can’t be ruled out.
This seems equivalent to someone buying a hammer from a pawn shop. Upon authorities finding out that hammer had been stolen or used in a crime instead of charging or investigating the Pawn Shop owner, they charge the person who bought the hammer because they made things with the assistance of “stolen goods”.
Used car / firearm / AI-assisted-code / other-example-here
Except the pawn shop is known for stealing from the community and the person that bought the hammer knew it was stolen.
Kind of complicates things.
No, it’s equivalent to a loaf of bread that can be magically copied for no cost infinitely. But someone lays claim on that particular loaf on the basis of “intellectual property”.
Someone who once studied how this magic loaf of bread works made a magic cake, so we can all eat cake instead. Obviously the owner of the loaf doesn’t like that and sues, shouting “he’s a thief!”. Millions of morons start to pick up the chant “Thief! Thief! He stole the fire of the Gods! Lock him up!”.
Now the obvious question becomes: When do analogies spiral out of control?
Anyone thinking about walking away might be interested in Haiku OS, which does not accept AI code
NetBSD is against tainted commits, including LLM-generated code.
Is it usable as a daily driver? Do you use it? I am genuinely curious and exploring Haiku and some BSDs Would love to hear the experience of users
I’ve only played with it in a VM for half an hour or so.
My gut feel is it’d be ok for most consumer uses (maybe not gaming?) but for my type of development work it wouldn’t be ideal.
Commenter is right. It remains to be seen whether or not it’s actually possible to license code generated this way as anything other than like MIT.
A person didn’t really make it, and so a person doesn’t really own it.
All AI generated code should be GPL. It was trained on all of us, and its output belongs to all of us.
If the model was trained on GPL code then one could argue anything the model outputs is a derivative.
That would be a copyright violation.
You can’t take closed-source code, make a derivative of it, and then GPL it. That’s illegal
MIT code can be used in GPL projects though. It simply becomes GPL code.
The hiccup is that the statement is that pure AI output is not subject to copyright, but once a human does something ‘transformative’ to it, then that is. So if a human can work back to the untouched AI output, they can go to town but they can’t reproduce any modification that was done after the AI generation. Since in practice, the GenAI row output is not provided it’s impossible for the human to know what the human touches were to undo, so that scenario does not present a whole lot of realistic implications.
a person pays another person to build them a website. who owns it?
edit: [everyone disliked that]
The person who built it owns it, and can transfer that ownership to the person that paid for it.
If a machine made it, no one owns it.
Copyright applies to creative works, which are defined as a manifestation of creative effort in the world through a creative process involving one or more individuals. It can be argued that since no human creativity was involved, copyright does not apply.
If a machine made it, no one owns it.
so, everything’s fair game since the industrial revolution?
look, I understand what you’re trying to say and for the most part I agree but my point is that AI is fundamentally just another tool doing the same thing as other tools but unified and using natural language. what’s the difference if someones builds a website using a website building app like Wix, or uses AI to build one?
i still fundamentally disagree with the enormous environmental costs of AI, the infringements of creatives and contributors in its training, the economic cliff its disruption has put us on, the rapid expansion of energy-hungry infrastructure data centres, and the health reports of local residents who have been forced to endure the 24-hour operation of them.
I probably could have worded it better.
To answer your point, creating an object using a machine is not a creative process by the definition used in this context, unless you put your creativity in it too. It’s the difference between creating a dress based on your own design and making it based on someone else’s. The ownership comes from the ideation, not the machanical process used to make the dress.
The key difference between AI and other tools is that it replaces the creative part of the process, thus removing the human factor.
As a side note, most of the time the end product contains both elements generated with AI and others done by humans. How much human contribution is required for it to be considered a creative work will probably be up to courts and lawmakers I’m afraid.
On a more ideological note, one could argue that AI models were created using basically the entirety of humanity works, often without compensating the ones holding the rights to such works, and thus should be owned by all humanity.
“creativity” is not as easy to define as you seem to imply. When photography replaced a lot of portrait painters, does that mean it wasn’t creative work? I mean sure painters can still paint other things, but likewise, coders can still code without AI. Does this just mean that there is “non-creative” code (the code that AI can generate) and “creative” code?
I said in my reply that determining whether something that was made with both AI and human contributions should be deemed a work of human creativity is not trivial. That being said, I can’t consider a piece of code entirely written by AI to fit the criteria for copyright protection.
And, dude, photography is 100% creative work and recognized as such everywhere. Yes, the machine takes the picture, but there’s a human who decided the subject, how to frame it, light, composition, not to mention the artistic intent, what he means to convey with that picture.
Then is AI code creative because the human wrote the prompt or guided the agent? But as you already said you agree that creativity is hard to define, I’m ok with us disagreeing on the exact boundary.
criteria for copyright protection.
honestly copyright was already BS as is. “Fair use” is such a mess
So what you’re saying is that there’s a tipping point in “intellectual property” where after a certain amount of creative help from a machine it stops being your idea/property?
My mind goes to real play DND podcasts. There is a framework that allows them to tell their story, but the story still belongs to them.
So like if an LLM writes a function to make the program work properly, like to avoid a logic error you didn’t forsee without you asking it too. You don’t own that function? Would it be different if you asked it to fix that bug, or do you have to say how to fix the bug?
Well, probably? I’m by no means an expert, so I don’t know how that could be defined or work in practice. And I think it’ll take us a long time to figure this one out
Sounds like prompt logs need to be tracked in git with reference. But also it sounds basically unenforceable.
In the usa, it has been ruled in the past that only human can have copyright. And if a human is not the author, they don’t have the copyright.
https://en.wikipedia.org/wiki/Monkey_selfie_copyright_dispute
This has also come up for generative AI with consistent result.
Problem in those cases is that they hinged upon the human passing through the non-human product without even claiming to have modified it. If human modification has been done, then that modification is subject to copyright. In the case of a composite of human and non-human origination, good luck extracting the non-human portion of the work to take advantage of that copyright opening.
Yes. My point was that the person’s comparison is shit. That genAI’s and human’s output aren’t the same.
I agree it is a crap comparison, just warning that even as the slop straight from the LLM is not copyrightable, blend in a little human touch up and it is “blessed” into being copyrighted. So it’s of little consequence unless the operator just passes it through untouched or makes it somehow easy to identify and remove the human tweaks.
In terms of slopping up a rip off under the assumption that it would obviously be the same as the original slop operators starting point, AI companies in their defense have already stated that operators can violate copyright implicitly as they legally declared they aren’t on the hook for any resultant infringement…
I wonder if their legal “protection” will work in court.
Intentionally absurd comparison: if I would host all the batman movies on a public website but make people agree to they are legally responsible for illegally downloading the movies, would I get away with that? I don’t think so.
Consequently, how much distance does there need to be between illegal activity and the company/person facilitate that activity?
Man, that’s so disappointing.
Eeh, I don’t think he is going to stop tearing people a new one for submitting bad code, though. He lives for the flame.
And the problem with LLM code is that it’s often bad and has a lot of volume. If their process for rejecting bad human code can scale to also reject bad LLM code, I don’t see a problem with allowing for the possibility that someone submits LLM code worth accepting.
just use an LLM to review the code, duh /s
Last time I saw Linus in the news he was ripping into someone submitting a huge blob of code after the deadline, and said that would not be going into the next kernel. So he’s at least expecting humans to review things thoroughly, otherwise why allocate time for human review?
The context of the quote is not code generation, it is an AI-based code review tool which caught real issues which made it past human review. He’s not arguing for lowering the code quality standards, he’s acknowledging that in the past year, AI-based code review tools have started to provide real value.
He has vibe-coded at least one non-kernel hobby project, though.
This is one scenario where the GenAI has been pretty useful, to a point.
It can help someone review code. It has a lot of false positives, but it does occasionally catch something, and I can forgive the false positives, even if they are super annoying and time consuming, so long as it is catching something on occasion.
In open source, a problem is so many people in your community now want to ‘help’ by using their AI subscriptions to review the existing code and then spam the project with their ‘findings’ because they don’t understand well enough to double check the AI. This greatly amplifies the false positive problem with mind numbing volume of dupes.
My employer rolled out an AI-based bot which commented on code reviews. It was absolute garbage, and every single comment I ever saw it make was a false positive. Disproving the false positives was a huge waste of time. I hated it with a passion.
Recently it was upgraded with a modern model. Now I’ve only ever seen it give legitimate feedback. It’s like night and day. Sometimes the feedback is overly pedantic, and more often than not suggesting an improvement slated for later, but still legitimate feedback with an appropriate tone for the severity.
I believe recent sentiment from the curl project (who previously complained about drowning in slop PRs), is that their AI-based pull requests have also improved significantly.
I’m still not a fan of AI-generated code, since I mostly think in terms of code and AI fucks up the process for me, but I have to acknowledge that quality of output has improved greatly the past year or so.
I switched to using Linux only a couple of decades ago. The reason was solely the GPL. I liked the idea behind it enough to put up with any difficulties compared with other OSes. Linus has fence sat on many “political” ideas. I guess this is a long way of saying, I agree with you.
Seems like this is in line with the Software Freedom Conservancy’s statement, then? It’s up to the individual. If you don’t like it, you can just not use it, and if you really don’t want others using it, you can just… fork it like he says?
That was exactly Linus’ point, especially as their comment came from the context of people debating AI bad/good in the kernel mailing list (which is totally not the place to be having that discussion).
You can’t fork it if it’s now containing illegal copyright violations
“Fork off”, apparently
Haiku it is, then?
(Also, HURD can’t get here soon enough)
I tried Guix/Hurd and Gentoo/Hurd in qemu yesterday, both booted nicely and have been shown to run on pre 2013 hardware. If they get proper contributions, the Linux kernel’s days are numbered in some circumstancies
Oh God, what fresh new Hell is this HURD thing?
Oh, that’s an old hell.
“Better the devil you know”, as they say.
Redox is where I’m hoping to move but also this is an overreaction, it’s just about code review tools he’s not vibe coding the kernel.
Oooh? First time I hear of Redox.
It will be a straight upgrade to the linux kernel when it reaches some level of driver parity and maturity… gonna be a while though, if ever
So Linux is no longer open source?
Because we can’t fork something that’s not open source. And just 1 AI contribution means it’s a copyright violation, so its no longer copyleft.
That is quite the statement. Is there any legal precedence for that? Yes, I know the technical why, but a blanket “one AI contribution and its over” is a little over the top, to put it mildly, and sounds like an extreme over generalization of a complicated problem
Nope, it hasn’t been litigated yet.
But it stands to reason that AI is a derivative of copyrighted material (unless the training set was released and confirmed to only contain otherwise). And there is loads of case precedent for fucking defendants hard on that violation
Well, it has been litigated and it continues to be litigated so the legal community doesn’t consider it completely settled, but so far court behavior seems inclined to agree with the AI companies.
That said, the AI company defense has broadly been “training is fair use, and even if an operator manages to get the model to produce an infringing work, the user is liable, not us”. So that “user is on the hook part” should be worrying people, but no one has yet tried to test this sort of oblivious infringement. I would expect it should at least be plausible that the operator was oblivious to the copyright violation they induced the model to commit, but still demand the offending material be removed.
Currently the industry is betting that it is “too big to fail” at this point and no one would dare pull that trigger on an unsuspecting violator. Further, copyright doesn’t cover “trivial” lifts from a source and the likelihood favors an accidental copy being a trivial part of a whole.
Hi, I’m a software engineer.
I’ve also been trained on copyrighted material, because every codebase I’ve worked on for money has been proprietary.
Does that mean I can’t touch open source code? I could accidentally use a snippet I’ve written before in some other setting.
The real licensing issue of AI use shouldn’t even apply to Linux because it’s already GPL. GPL infects anything it touches so if you use GPL-trained AI on an MIT project and it spits out something resembling the GPL code it was trained on, that would make that MIT project a GPL project. It would also make your proprietary projects GPL projects.
You’re unlikely to spit out code from your corpus verbatim, so the answer to your first question is a resounding no. If you did copy from your previous job’s, you’d be liable.
So funny you should mention that, I worked at a company that dealt with Linux and third-party proprietary code. They kept the software developers highly segregated because they feared accidental copyright infringement. They thought at the time that even a human learning and accidentally reproducing something they remembered from working on proprietary code too risky.
The practical difference is that the third-parties were likely to sue, but the corpus of trained data is pretty much open source projects that may have a right to sue but in practice everyone knows they aren’t going to be able to chase down violations due to lack of resource.
Whole thing is a nightmare and I hate it. Luckily nothing I do for work is open source so even if my code accidentally resembles some other codebase I’ve worked on, nobody will find out lol
