|
|
Log in / Subscribe / Register

Debian votes to allow "responsible use of generative AI"

[Posted August 29, 2026 by corbet]

The results of the Debian general-resolution vote on the use of large language models have been posted; the winner is choice 5: Responsible Use of Generative AI.

Debian neither endorses nor prohibits the use of generative AI tools in the development, maintenance, or documentation of software, packaging, documentation, and other media published within the Debian Project. We recognize that such tools can substantially improve the productivity of contributors when used responsibly, allowing volunteers to spend more of their limited time on work that requires technical expertise, judgment, review, and collaboration.

The Debian Project nevertheless expects that all contributions submitted to Debian, regardless of how and with which tools they were produced, satisfy the same standards of quality, correctness, maintainability, and legal compliance. The use of a generative AI tool does not diminish the contributor's responsibility for the work they submit. Contributors are expected to understand, review, test, and, where appropriate, modify AI-assisted output before incorporating it into Debian.


to post comments

Inevitable

Posted Aug 29, 2026 14:03 UTC (Sat) by dskoll (subscriber, #1630) [Link] (15 responses)

I'm a bit sad it went that way, but I guess it was inevitable. Debian is too large a project to be able to meaningfully enforce a stronger stance against LLMs. So I think they realized it's pointless to make rules that you can't enforce.

Inevitable

Posted Aug 29, 2026 15:06 UTC (Sat) by MarcB (subscriber, #101804) [Link] (14 responses)

I, for one, am very glad. In particular the fact that the two full-Anti-AI proposals (1&3) were even below "None of the above" is very good sign. Had one of those won, the attempts to enforce them would have ruined Debian.

Inevitable

Posted Aug 29, 2026 20:08 UTC (Sat) by tachi (guest, #185877) [Link] (1 responses)

Option 3 was not "fully-Any-AI"; it explicitly allowed (but discouraged) AI contributions, but disallowed LLM-generated emails to be exchanged on the lists.

Also, (at least for me) the point was not to focus on enforcement, but rather signal to the wider world "Hey, we think this is not OK".

Inevitable

Posted Aug 29, 2026 20:12 UTC (Sat) by bluca (subscriber, #118303) [Link]

Nah, it was very much about enforcement, and that's why it proposed to update the code of conduct, and one of the supporters was quite open about this on the mailing list. This was 100% intended to start a witch hunt.

Inevitable

Posted Aug 31, 2026 9:47 UTC (Mon) by jengelh (subscriber, #33263) [Link] (11 responses)

>Had one of those [Anti-AI proposals] won, the attempts to enforce them would have ruined Debian.

Had those options won, would that not have meant that those options share broad support?

Inevitable

Posted Aug 31, 2026 14:30 UTC (Mon) by MarcB (subscriber, #101804) [Link]

Sure, still the attempts to enforce them would have been devastating in the long run. I can't imagine any other outcome.

I mean, compare those suggestions, their wording and the "spirit" that shines through with something like https://forge.rust-lang.org/policies/llm-usage.html (an excellent policy, as far as I am concerned, at least in situations where code writing is not the bottle neck).

Those proposals are all far too invasive. Debian would have to spent an enormous amount of energy into enforcing them and would have alienated countless contributors (including future ones).

Inevitable

Posted Sep 2, 2026 6:33 UTC (Wed) by smurf (subscriber, #17840) [Link] (9 responses)

Right now, possibly, though IMHO quite a few DDs would have moved to a different distribution if the no-AI vote had won the day. (Me among them.)

But future developers? given the choice of supporting a distro with somewhat-sane git-based packaging standards and AI support, vs. a Debian which has neither? I don't think so.

Inevitable

Posted Sep 3, 2026 14:16 UTC (Thu) by LtWorf (subscriber, #124958) [Link] (8 responses)

Some have left, so I hope you and your AI can work much harder to replace them.

Also I suspect many more are quiet quitting and will be taking it easy but won't quit because you never know when you need a new job and being a DD can bring in some cash :)

I don't understand why you're mentioning git at all.

Also there's absolutely 0 support in debian for running local AI since it's non-free and cannot be included in Debian. Connecting or using proprietary services has never been forbidden for users, that would be absurd.

Inevitable

Posted Sep 3, 2026 19:40 UTC (Thu) by mathstuf (subscriber, #69389) [Link]

Also there's absolutely 0 support in debian for running local AI since it's non-free and cannot be included in Debian.

What is non-free about llama.cpp or its wrappers (ollama and ramalama)? Sure, CUDA support exists, but it isn't required; there's even a Vulkan backend.

Inevitable

Posted Sep 4, 2026 8:51 UTC (Fri) by smurf (subscriber, #17840) [Link] (6 responses)

I mention git because Debian's 30-year-old way to work with source code and patching it for packaging (quilt) clashes pretty strongly with the way people work today. It's another road blocker for people interested in joining Debian.

The advantage of AI is that you can tell it to update a package to a modern workflow, and ten minutes later you get something you can actually work with.

YMMV and all that.

Inevitable

Posted Sep 4, 2026 10:55 UTC (Fri) by jengelh (subscriber, #33263) [Link] (4 responses)

I don't think quilt is the main problem. If anything, it is that there are too many layers. A deb binary packages is an ar archive of tarfiles. The build description too, is not one .spec or something, but a .tar.gz of a plethora of small files (.dsc, rules, control, changelog, ..). But why use sh for commands when you can use… Makefile. Lots of escaping to be aware of. This all makes even a hello world program package rather complicated before you even get to deal with patches and quilt.

Inevitable

Posted Sep 7, 2026 6:48 UTC (Mon) by taladar (subscriber, #68407) [Link] (3 responses)

It also doesn't help that despite being decades old the format (especially version and lintian lints) change enough between supported major versions that it is a bit of a headache to maintain even for a single package you just want to deploy to all currently supported distros. This is one of those places where not backporting tooling from newer distros (not even lintian) really hurts that whole stable/LTS distro model.

Inevitable

Posted Sep 7, 2026 8:55 UTC (Mon) by mbunkus (subscriber, #87248) [Link] (2 responses)

Yeah, as a someone providing packages for various Debian & Ubuntu releases and not being a Debian Developer myself, I generate several files (most notably, debian/control & debian/rules) from templates before packaging. Because simple things such as changing package names/version requirements between different releases cannot be expressed in a single debian/control file. This also makes the requirement of dpkg-build to unpack an archive more onerous as I have to generate those archives with those newly created build files each time. Sure, automation, bla bla bla, it's really tedious to get all this right.

By contrast I only use a single .spec for all my RPM-based packages (Alma, Rocky, CentOS Stream, used to provide for SuSE, too) which can use RPM macros to differentiate between distros & releases. Much easier to work with, to be quite honest.

Inevitable

Posted Sep 7, 2026 17:35 UTC (Mon) by LtWorf (subscriber, #124958) [Link] (1 responses)

I contribute to Debian. I did rpm files as well.

I assure you that leaving configuration files for debhelpers rather than manually scripting everything means that your package will adhere to standards and rules by default rather than having to know all the standards and rules and follow them manually.

Of course if your packages aren't in the distribution and don't have to obey to any rule everything is simpler so the things that help keep the whole distribution in place seem more work (and they are), but in general I prefer making .deb than .rpm because there is less repetition needed in my experience.

Inevitable

Posted Sep 7, 2026 18:01 UTC (Mon) by mbunkus (subscriber, #87248) [Link]

I can absolutely see that advantage. I'm merely coming from the POV of someone external who doesn't want to have to learn all of those things. Safe for macOS I found Debian/Ubuntu/LinuxMint package maintenance to be overall the most time-consuming (others not mentioned yet: ArchLinux, AppImage, Flatpak, Windows via NSIS). I have to spread my knowledge so widely as that's what my users request.

Different priorities. Obviously a distro's priority is supporting its own ecosystem, not externals.

Inevitable

Posted Sep 7, 2026 17:39 UTC (Mon) by LtWorf (subscriber, #124958) [Link]

> The advantage of AI is that you can tell it to update a package to a modern workflow, and ten minutes later you get something you can actually work with.

People don't want a modern workflow. People want a workflow that works for them.

Also the entire point of having maintainers rather than just uploading whatever the authors publish is that the maintainers test the software, make sure it runs besides compiling, give a look to the diff to see if there's anything suspicious, fix bugs immediately and submit the patches to upstream so the users have the fixes right away…

Sure you can tell an AI to update all the packages but might as well just use flatpak then, it has been there for a while and doesn't destroy society.

Extremist positions soundly defeated

Posted Aug 29, 2026 14:20 UTC (Sat) by bluca (subscriber, #118303) [Link] (16 responses)

It might not be obvious due to the complexity of the voting system, but it is definitely worth noting that not only the sensible and rational position won, but the options that were set up with the implicit (or by reading some emails from some proponents, quite explicit actually) purpose of setting up witch hunts to actively punish anyone who dared not toe the extremely vague anti-AI line (option 1 to change social contract, option 3 to change code of conduct, both of which imply *expelling* any dissenter) were soundly defeated and rejected: they (and only they) both failed to beat the "None of the Above" option, which is this voting system's way to say "no, absolutely not, go away".

The rest of the options cleared that line and then it was a (complicated) matter of which one was the "least unfavored" among all voters, so to speak.

A handy graph that someone else prepared makes the result a lot more digestible: https://people.debian.org/~lucas/gr-2026-002/results.png

Extremist positions soundly defeated

Posted Aug 29, 2026 14:41 UTC (Sat) by mhvk (subscriber, #86585) [Link] (14 responses)

It depends what you mean by "soundly" - if I read that graph correctly, about 30% of the voters preferred the no-AI options over all the others. To me, it looks like "status quo" won, which is perhaps appropriate in times of rapid change.

Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones (which to me are very problematic for lots of reasons; it doesn't help to come from a country that will be under water -- the Greenland icesheet melting completely is nearly unavoidable by now, and 6 meters of sea level rise is more than one can build dykes for; why these data centers are not forced to rely on renewable energy is truly beyond me).

Extremist positions soundly defeated

Posted Aug 29, 2026 14:45 UTC (Sat) by bluca (subscriber, #118303) [Link]

> It depends what you mean by "soundly"

With this voting system, failing to beat "NOTA", as mentioned, means the option is soundly rejected even before considering relative preferences

Extremist positions soundly defeated

Posted Aug 29, 2026 15:20 UTC (Sat) by MarcB (subscriber, #101804) [Link]

> Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones...

In theory, they are. The problem is that the absurd investments into AI hardware have made it unaffordable. 64-128 GiB+ GPUs at home would be absolutely realistic without this cost explosion.

> ...the Greenland icesheet melting completely is nearly unavoidable by now...

Technically not, but I also believe we will not avoid it. But the reason for that is not AI. Its share of human energy consumption is somewhere around 0.2%, likely lower. And AI would be very amenable to using renewal energies - if any relevant government cared enough to enforce its usage.

Useful local coding models

Posted Aug 29, 2026 20:44 UTC (Sat) by emk (subscriber, #1128) [Link] (10 responses)

> Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones

If you consider the Chinese open weight models "truly open", we're either at that point already, or tantalizingly close.

There are at least 4 small and mid-sized MIT-licensed models that match or beat Opus 4.5. The cheapest to run is Qwen3.8 27B, which mostly fits in 32GB of VRAM. (Which you can still buy for about US$1,700 this week, though I don't expect that to last.) It would still be happier with 48-64GB, which is starting to get trickier.

Going up in cost from there, you have Qwen3.8 Flash Next, DeepSeek V4 0731, and finally (for small businesseses) GLM 5.3 Flash, which is closer to Opus 4.8 on many tasks but is a better match for people who have a machine room with a server rack or two. Nothing you can afford to run locally will match Opus 5 or Fable 5.

Most of these models benefit from a knowledgeable developer being in the loop somewhere. Much like modern Linux, they still have occasional rough edges. But much like modern Linux, they're perfectly reasonable in the hands of a normal developer, when used for normal use cases (where "normal use cases" involves things like "async Rust code" or "debugging Terraforn nonsense").

Power costs start around 300W, which is far less than any 220V appliances you may have.

If RAM prices go down before someone builds SkyNet, it will be absolutely possible to run MIT-licensed coding models locally on a higher-end gaming rig or Mac Studio. Laptops may take a little longer to get 64GB of VRAM and enough compute to make it pleasant.

Useful local coding models

Posted Aug 31, 2026 7:12 UTC (Mon) by smurf (subscriber, #17840) [Link] (7 responses)

Power costs start around 300W, which is far less than any 220V appliances you may have.

Power cost starts at 150W or so; that's my Framework desktop machine. The GPU reports 100W when 100% busy. You can even connect it to 220V if you want to. (Even if you don't live in a country where that is the standard anyway).

Useful local coding models

Posted Aug 31, 2026 20:19 UTC (Mon) by Klaasjan (subscriber, #4951) [Link] (6 responses)

…. and it is 230 V (rms), not 220 V like in the old days. In the EU, that is.

Useful local coding models

Posted Sep 2, 2026 7:01 UTC (Wed) by smurf (subscriber, #17840) [Link] (5 responses)

240 if you're in the UK. So there. :-P

Useful local coding models

Posted Sep 2, 2026 14:16 UTC (Wed) by Wol (subscriber, #4433) [Link] (4 responses)

Actually not. It was standardised years ago across Western Europe (including the UK) as 230V.

(+10/-20 iirc, whatever all of the old standards between 200 and 240 are defined as acceptable, if not the actual standard.)

Cheers,
Wol

Useful local coding models

Posted Sep 4, 2026 7:54 UTC (Fri) by joib (subscriber, #8541) [Link] (3 responses)

The rough history, AFAIU, is that most of Europe was on 220V, the UK on 240V. And they realized there was enough slack that they could harmonize it to 230V everywhere, so that's the current standard.

(off-topic) 230V standardisation

Posted Sep 4, 2026 11:26 UTC (Fri) by farnz (subscriber, #17727) [Link] (1 responses)

It's not just voltage that got standardised; we also standardised the measurement point for voltage (typically the utility meter, but some countries had previously used the socket), and standardised the maximum voltage drop from meter to end device (4% maximum drop permitted).

The result was that virtually nobody needed to change the supply side, but that wiring standards had to be updated, and equipment for the EU market had to support 198V to 253V input without loss of functionality (where before, in some countries, you could get away with 205V to 235V).

(off-topic) 230V standardisation

Posted Sep 7, 2026 13:09 UTC (Mon) by geert (subscriber, #98403) [Link]

It's a sunny day, so I'm already at 244.1V (in former 220V land).

Useful local coding models

Posted Sep 9, 2026 15:33 UTC (Wed) by anton (subscriber, #25547) [Link]

Concerning the slack, the energy suppliers were given less slack during the 1987-2009 transition period (voltage had to stay within both old and new bounds), and after that transition period, the slack increased again.

Useful local coding models

Posted Aug 31, 2026 8:38 UTC (Mon) by taladar (subscriber, #68407) [Link] (1 responses)

I don't think the problem with obtaining 64GB VRAM is RAM cost in itself but the fact that no manufacturer offers devices with it for a consumer market precisely because they know they can make more money selling those only to data centers at a business markup.

Useful local coding models

Posted Aug 31, 2026 21:00 UTC (Mon) by emk (subscriber, #1128) [Link]

Yeah, you either need to grab a larger workstation card, which has gotten very expensive, or build with two workstation cards. Which sounds easy, but actually powering and wiring two GPUs on consumer motherboards can be a project.

Still, one more big RAM price crash, and all this stuff will be pretty affordable, I think.

Extremist positions soundly defeated

Posted Aug 31, 2026 11:23 UTC (Mon) by ehiggs (subscriber, #90713) [Link]

> Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones (which to me are very problematic for lots of reasons; it doesn't help to come from a country that will be under water -- the Greenland icesheet melting completely is nearly unavoidable by now, and 6 meters of sea level rise is more than one can build dykes for; why these data centers are not forced to rely on renewable energy is truly beyond me).

Increasing energy and compute efficiency of LLMs will likely make the ice sheets melt faster thanks to Jevon's Paradox.

> renewable energy

Renewable energy also includes burning wood pellets and burning garbage. It's a misnomer when people usually mean "low carbon energy" (wind, solar, hydro, and nuclear).

Voting systems

Posted Aug 30, 2026 13:40 UTC (Sun) by marcH (subscriber, #57642) [Link]

> It might not be obvious due to the complexity of the voting system, ...

It may sound counter-intuitive but "simpler" voting systems stop people from expressing what they think. Especially the simplistic and extremely harmful "winner takes all" system that routinely forces people to vote "against" instead of "for", happily discards 49% of votes (which are not even actual opinions because of how dumb the system is), enables gerrymandering and practically limits the number of parties to a maximum of two[*]. Gerrymandering works in basically no other voting system, that's how bad the "winner takes all" voting system is.

To add insult to injury, _stack_ multiple rounds of "winner takes all"! See for instance the electoral college.

To be fair, "winner takes all" may be a good match for some cultures...

Unfortunately, humans suck too much at maths and statistics to understand all this and organize in large societies. Unless they are debian developers.

[*] because who wants to run a party that has practically zero chance of ever winning anything?

> but the options that were set up with the implicit purpose of setting up witch hunts to actively punish anyone who dared not toe the extremely vague anti-AI line were soundly defeated and rejected:

Yes, exactly that: only thanks to a non-stupid voting system. Unlike politics in so many democraties right now where dumb voting systems keep supporting extremes.

A growing disconnect

Posted Aug 29, 2026 14:44 UTC (Sat) by aimannajjar (subscriber, #184277) [Link] (103 responses)

Ed Zitron's interviews gaining 5million views overnight on YouTube, while the people of free software embrace AI unconditionally is the wildest ironies in history.

It will be interesting to watch what happens next in terms of quality and review processes as well as young engineers skills. I genuinely think it's good to have diversity in AI adoption so we can compare and contrast the net results across projects in the long term.

A growing disconnect

Posted Aug 30, 2026 16:44 UTC (Sun) by coriordan (guest, #7544) [Link] (21 responses)

I was one of the 5 million viewers. Ed is wrong.

He says LLMs produce nonsense and they're complicated to use. Nope.

There are big parts of my job that LLMs can do better than I can, and when I code for fun, LLMs are 100x better than I am.

We shouldn't embrace AI "unconditionally", but we should definitely embrace AI.

A growing disconnect

Posted Aug 30, 2026 20:04 UTC (Sun) by aimannajjar (subscriber, #184277) [Link] (1 responses)

That might be, but his positions are increasingly popular and mainstream. And my comment is about the disconnect with the general sentiment.

Let's be honest, American CEOs dreams of putting the fear of AI into everyone's heart have backfired. The COVID layoffs masked as AI "productivity gains", the "I hope AI is nice to us" statements and "the agents have talked to each other and figured a way to break out of the sloppy sandbox" headline, this is all backfiring today and open source community is oblivious to it, it seems.

The toxicity is probably here to stay, very much like "AI is here to stay". Play stupid games, win stupid prizes. I think "free software" maintainers should really be more in touch with people.

A growing disconnect

Posted Aug 30, 2026 22:28 UTC (Sun) by coriordan (guest, #7544) [Link]

AI brings good and bad, just as the internet did. People who care about the free software movement (the movement to put people in control of the technology they use) should be looking now at AI to figure out how we achieve those goals when an important chunk of software is now models rather than executables.

> his positions are increasingly popular and mainstream

I don't think he's a major force in the growing protests or criticism of certain consequences of AI. He's more entertaining than convincing.

A growing disconnect

Posted Aug 30, 2026 23:12 UTC (Sun) by mgb (guest, #3226) [Link] (18 responses)

Coding is easy, and only a small part of the software engineering lifecycle.

Vibe coding one-off scripts for non-sensitive data in a sandboxed environment? Have fun with your LLM pal.

Software engineering? Do it right.

A growing disconnect

Posted Aug 31, 2026 8:01 UTC (Mon) by coriordan (guest, #7544) [Link] (16 responses)

> Coding is easy, and only a small part of the software engineering lifecycle.

Agreed.

My point is still: there are parts of jobs that LLMs do much better than humans. This is true for my sector (I'm a policy adviser) and it's true for software development. I hope people don't jump from "a human is necessary for some parts" to "we shouldn't allow use of AI". This seems to be what Ed is doing. He points at many problems and then concludes that benefits don't exist.

Free software developers should use generative AI for some things. The real question is: what is the policy for deciding what to use AI for, and how do we integrate that into a lifecycle that provides quality assurance?

A growing disconnect

Posted Aug 31, 2026 9:55 UTC (Mon) by anselm (subscriber, #2796) [Link] (15 responses)

This seems to be what Ed is doing. He points at many problems and then concludes that benefits don't exist.

I don't think that's actually what he says. AFAICT he says that those benefits that do exist (which are probably fewer than the gen-AI companies would have us believe) aren't worth the trillions of dollars that are pumped into generative AI.

Which, I think, is not an unreasonable point to make. Imagine the same amount of money being spent, with the same naive enthusiasm and blithe disregard of ROI, on making life better for humanity at large – think healthcare, poverty relief, development aid, … – instead of a few techbros.

A growing disconnect

Posted Aug 31, 2026 10:37 UTC (Mon) by coriordan (guest, #7544) [Link] (14 responses)

> healthcare, poverty relief, development aid

This is a red herring. Investment in AI isn't the reason for any lack of funding for healthcare, poverty relief, or development aid. If someone's goal is to increase funding in those things, pausing AI will not do it.

> aren't worth the trillions of dollars

Ok, but that's their gamble. They think it's worth it. Whether they make their money back, or not, doesn't affect me.

The dot com bubble burst more than twenty years ago. Some companies don't exist anymore but the internet has grown every year. One can argue that some current AI companies will have financial problems in the future, but that doesn't mean AI is a con.

A growing disconnect

Posted Aug 31, 2026 13:22 UTC (Mon) by aimannajjar (subscriber, #184277) [Link] (12 responses)

> Ok, but that's their gamble. They think it's worth it. Whether they make their money back, or not, doesn't affect me.

That's their gamble, which in the process of taking it, they devoured natural resources, in many cases without even valid permits or despite local community objections, laid off hundreds of thousands, manipulated the stock market and engaged in shady circular deals that is sure to affect everyone when this bursts, not to mention the endless media propaganda that made millions question their future.

No that's not how bad gambles are taken. Metaverse was a bad gamble, this is criminal.

A growing disconnect

Posted Aug 31, 2026 14:05 UTC (Mon) by bluca (subscriber, #118303) [Link] (11 responses)

You have literally described what capitalism has been doing for the past ~250 years. Why is it bad if AI is involved, and just fine otherwise?

A growing disconnect

Posted Aug 31, 2026 14:40 UTC (Mon) by aimannajjar (subscriber, #184277) [Link] (1 responses)

> You have literally described what capitalism has been doing for the past ~250 years. Why is it bad if AI is involved, and just fine otherwise?

I think what we disagree on is that you believe this is capitalism alone. While I believe there is clearly more to this than just capitalism.

A growing disconnect

Posted Aug 31, 2026 14:42 UTC (Mon) by bluca (subscriber, #118303) [Link]

If that were the case, then the same pattern wouldn't have repeated pretty much verbatim over two centuries.

A growing disconnect

Posted Aug 31, 2026 14:42 UTC (Mon) by anselm (subscriber, #2796) [Link] (2 responses)

Why is it bad if AI is involved, and just fine otherwise?

Who says this is “just fine otherwise”? It's just that the gen-AI companies have taken the con game to a whole new level of brazenness.

A growing disconnect

Posted Aug 31, 2026 14:49 UTC (Mon) by bluca (subscriber, #118303) [Link] (1 responses)

They haven't, though, this is just capitalism as it has always been. If you are not fine with that, then the problem is not AI, that's just a symptom, and if it disappeared tomorrow morning, nothing would change

A growing disconnect

Posted Aug 31, 2026 14:59 UTC (Mon) by anselm (subscriber, #2796) [Link]

I think at this point we will just have to agree to differ.

A growing disconnect

Posted Sep 1, 2026 8:59 UTC (Tue) by paulj (subscriber, #341) [Link] (5 responses)

A: This practice X, this way of doing things is wrong
B: But practice X is in line with practices across a much wider encompassing system Y, if you're against that practice you're against the whole system

Seems just like a tactic to derail a discussion. If your argument is that X is a microcosm of Y, and hence one can not discuss the details of X unless done as part of some much wider, broader and hence harder to conduct, discussion of Y, you are just trying to stymie discussion.

Further, it is intellectually dishonest, for if one believed that X was a microcosm of Y, and truely were interested in a productive discussion, then one would know that a discussion of X in its details and merits/demerits would (being a microcosm of Y) help shed light on Y - without having to engage first in some wider discussion. A critique of X would be a critique of Y, and a debate around a critique of X would - without any explicit reference to Y - help shed light on any other critique of Y one might engage with in the future or on other forums.

So... I think it's perfectly valid to have a detailed discussion about the details of X, without needing people to go "But it's a microcosm of Y!".

A growing disconnect

Posted Sep 1, 2026 9:33 UTC (Tue) by intelfx (subscriber, #130118) [Link] (1 responses)

A: This practice X, this way of doing things is wrong
B: But practice X is in line with practices across a much wider encompassing system Y, if you're against that practice you're against the whole system

I have a pretty different impression of Luca's posts. It's more like:

A: This practice X is inextricably evil
B: No, when practice X is being used for evil, it is in fact merely a specific case of a much wider phenomenon Y of doing many various things for evil. So if you want to oppose X on these grounds, you should oppose Y instead.

A growing disconnect

Posted Sep 1, 2026 11:09 UTC (Tue) by paulj (subscriber, #341) [Link]

I think I covered that. If X is indeed a specific case of Y, then there is value in discussing the specifics of X. For - if the claim that X is a specific case of Y is indeed true - examining X in detail will help illuminate Y. If the claim that X is a specific case of Y is not quite true, then a discussion of the specifics of X are required for anyone later to be able to conclude that that claim is indeed not accurate. Further, whereas X and its specifics are obviously on-topic for LWN, a broad discussion of Y are /not/.

Bluca may be well intentioned and correct in his claim that X is a specific case (or a microcosm) of Y. Yet, I don't think it's helpful to try direct discussion of the specifics of X (merits, demerits, etc.) to Y, when Y obviously is OffT here and can not really be discussed.

The specifics of X /are/ worth discussing. And we get bluca's point, and it doesn't help to (effectively) keep trying to shut down discussion of X with that direction to Y.

A growing disconnect

Posted Sep 1, 2026 9:34 UTC (Tue) by bluca (subscriber, #118303) [Link] (2 responses)

> you are just trying to stymie discussion

No, I'm pointing out that it is completely futile to be angry at LLMs for the sins of capitalism, because it will solve nothing, and not improve anyone's material conditions. Be angry at capitalism instead, which is the real source of all these problems. Since we stopped doing that, working conditions have stopped improving. We know how to fix it: organized labour, strict regulations and hefty fines for violations, wealth taxation will achieve the desirable outcomes, and history is there to prove it. Shouting at developers using copilot will achieve fuck all.

A growing disconnect

Posted Sep 1, 2026 12:22 UTC (Tue) by marcH (subscriber, #57642) [Link] (1 responses)

The alternatives are not massively better. You can theorize how to organize large societies all you want, a corrupted clique always ends up at the top. To various degrees, all systems perform such a selection process. All political systems and religions are just different flavours of that same failure pattern. Humans have not evolved past the tribal stage yet - which was only yesterday from a biology perspective. It's hard enough to understand other people, so forget it when you've never met them in person.

The least bad countries there tend to be on the smaller side. I don't think that is an accident. It's definitely not enough but it seems to help.

That does not mean we should stop trying. There are differences. For instance, democracy lets you rotate that corrupt top more frequently. Transparency is probably the most efficient way to mitigate those damages because it aims straight at the root cause of the problem. That's why it's been so sad to see BigTech capturing all our brain time and destroying the press.

> organized labour

Unions are such a funny example. While there are some successful ones, so many of them end up creating new systems of privileges. I speak from direct experience. Granted, those privileges are totally negligible compared to the taxes that the rich stopped paying since the 80s. But the irony is much stronger.

tl;dr: we are screwed :-)

A growing disconnect

Posted Sep 1, 2026 13:24 UTC (Tue) by paulj (subscriber, #341) [Link]

> a corrupted clique always ends up at the top. ... While there are some successful ones, so many of them end up creating new systems of privileges.

"The Rise and Decline of Nations", Mancur Olson, is a good read here. He examines how special interests and sub-groups arise, influence and capture societies around them. A very good book to read, for anyone who wonders about the influence of cliques and special interests in their societies.

There is a PDF on archive.org.

A growing disconnect

Posted Sep 2, 2026 6:43 UTC (Wed) by smurf (subscriber, #17840) [Link]

Investment in AI isn't the reason for any lack of funding for healthcare, poverty relief, or development aid. If someone's goal is to increase funding in those things, pausing AI will not do it.

The counter is that quite a few people are going to make loads more than a few $$$$$ when their shares go public.

Now, some of them are idiots (in the original sense of the word: one who cares only for himself) want to amass even more money, but there seem to be quite a few who DO think that personal wealth in excess of eight digits doesn't make sense — and plan to spend their not-so-hardly-earned money on something charitable.

A growing disconnect

Posted Aug 31, 2026 8:45 UTC (Mon) by taladar (subscriber, #68407) [Link]

I have tried for decades to convince humans to do it right. To this day those same humans still don't write tests, don't use the proper escaping APIs to avoid injections,... while the LLMs do when I tell them to.

A growing disconnect

Posted Aug 30, 2026 19:39 UTC (Sun) by tytso (✭ supporter ✭, #9993) [Link] (80 responses)

I agree with quite a lot of what Ed Zitron is saying, but that doesn't mean that AI is bad. It just means that investment in AI may have gotten over-enthusiastic.

Take for example the use of the Internet in e-commerce, which was the source of the dot-COM bubble. The use of e-commerce is not all good; there are plenty of mom-and-pop retail stores that have been driven out of business because of the shift to buying this online. However, many people would say that being able to buy goods on-line at substantially cheaper prices is something that they have benefited from. There were a huge number of companies which went out of business as a direct result of of the dot-Com boom, from Sun Microsystems to pets.com. And even for those people who correctly deduced that Amazon would come out on top of the dot-COM boom, if you purchased at the very height of the dot-COM bubble, it would take ten years for Amazon's stock price to recover. Just because a technology is good or bad, is quite different from whether a particular company is good or bad, and that in turn is quite different from whether a particular company is good investment.

You can agree that the dot-COM boom was a bubble (which is what Ed Zitron is saying) and still think that it would have been a bad idea for Debian to take a stand and reject use of the Internet for any and all uses, because it is a moral evil that is driving small businesses into bankruptcy, and because it encouraged building data centers which is destroying the world's climate --- and that's essentially what some of the more radical proposals in the Debian's General Resolution was effectively proposing with respect to AI.

A growing disconnect

Posted Aug 30, 2026 20:30 UTC (Sun) by aimannajjar (subscriber, #184277) [Link] (23 responses)

I agree with much of what you say, but what I am trying to shine a light on is that this is no longer about rational description of current state.

People genuinely hate AI, because it was marketed (to wealthy investors) as their , i.e. the people's, replacement. I am of the opinion that those CEOs who adopted aggressive scaremongering tactics knew what they were doing but either 1) they were betting that AI will truly replace all those "human" workers, and therefore they couldn't care less about their feelings or 2) they were betting that people will accept in silence in fear of losing their job or labeled as backwards and running behind or 3) combination thereof.

This is was all happening when China was behind, and American labs were feeling supremacy of sort. Then China broke through, and they no longer felt special. And people started speaking loudly, with real backlash and popular activism on the street. It seems now they lost all their bets.

Now they want to talk rationally, but where was the rationality when people's hearts were trembling in anxiety due to the relentless headline after another? I think AI toxicity is a real issue, one which dot-com didn't have, even industrial revolution probably didn't experience the same scale of fear and anxiety. People lost jobs, but it was organic, and due to real productivity gains and in all likelihood, the headlines were more empathetic. But this feels different, very different, some people really wanted to feel like "gods", I think.

A growing disconnect

Posted Aug 30, 2026 20:44 UTC (Sun) by bluca (subscriber, #118303) [Link] (8 responses)

You are so close to getting it. The final step is the realization that the villain behind all of that is the same that's behind everything else too: capitalism. That's what is toxic, not AI. If you got the infinity stones and snapped AI away from the world tomorrow morning, CEOs would still harass and exploit and fire workers in the name of profits, they'd just use different tools. We know what works to rein in capitalism, and it's not ranting about LLMs on social media.

A growing disconnect

Posted Aug 30, 2026 21:06 UTC (Sun) by aimannajjar (subscriber, #184277) [Link] (7 responses)

Not saying capitalism is good or bad but:

In what capitalist world does "token maxxing" make any sense?

All tech companies agreeing , in one breath, to forcing their employees adopt the financially-retarded practice of "token maxxing", putting their own companies interest behind? A practice that only benefits the "AI overlords"? Number of Tokens (a cost incurring metric) becoming a performance metric where HIGHER is better? Token leaderboards?no, this is no capitalism I'm afraid. Capitalism doesn't reward higher costs per employee.

I do think capitalism allowed the exploitation of people money by manipulating their feelings in Wall Street. The government was supposed to step in, but I mean, if you're allowing circular deals everywhere, what's stock price manipulation here and there. Yes capitalism rewards greed, but in AI, greed and narcissism worked together as a team.

A growing disconnect

Posted Aug 30, 2026 21:43 UTC (Sun) by bluca (subscriber, #118303) [Link] (5 responses)

> In what capitalist world does "token maxxing" make any sense?

Who said it has to make any sense? It doesn't and never did. I mean capitalism, not just "token maxxing"

A growing disconnect

Posted Aug 30, 2026 22:01 UTC (Sun) by aimannajjar (subscriber, #184277) [Link] (4 responses)

Even if capitalism doesn't make sense to you, "token maxxing" is still definitely not capitalism.

No company in a capitalist competitive economy ever asks it's employees to maximize cost and, even worse, ties that to performance reviews in a capitalist market, ever.

You don't have to agree with capitalism to sense there was something wrong and conflicted there.

A growing disconnect

Posted Aug 31, 2026 2:43 UTC (Mon) by mathstuf (subscriber, #69389) [Link]

Perhaps just plain FOMO is an ingredient here? Having two children, the (apparent) effect of FOMO on (especially) naptime is hard to ignore. Somehow the urge to see "what's going on" even at high cost has deep roots…

A growing disconnect

Posted Aug 31, 2026 15:15 UTC (Mon) by joib (subscriber, #8541) [Link] (1 responses)

AFAICT, "token maxxing" was mostly a gamey attempt to get employees to take a look at what this brave new world of AI has to offer, and not be overly afraid of using tokens. Or IOW, a bet that the productivity gain from using those tokens was much higher than the cost of them.

Nobody (sane) was advocating for wasting tokens just for the sake of it, though I'm sure it can appear like that at first glance.

A growing disconnect

Posted Sep 2, 2026 6:49 UTC (Wed) by smurf (subscriber, #17840) [Link]

Nobody (sane) was advocating for wasting tokens just for the sake of it, though I'm sure it can appear like that at first glance.

Yeah, but Goodhart's Law pretty much guarantees that this incentive ends up being self-defeating the second it's introduced.

A growing disconnect

Posted Aug 31, 2026 17:22 UTC (Mon) by raven667 (subscriber, #5198) [Link]

I think we are mistaking capitalism for commerce here, commerce is competitive and tries to reduce costs, capitalism as practiced by the AI companies is about maximizing cost (hence the propaganda effort to convince other companies to "token maxxxx") so that the wealthiest capitalists can divide and monopolize the market by making it to expensive for anyone else to compete with them. It's creating a moat made out of money, trying to make AI a high-margin luxury good.

A growing disconnect

Posted Aug 31, 2026 20:11 UTC (Mon) by tytso (✭ supporter ✭, #9993) [Link]

Token maxxing was a stupid idea that ultimately hurt the companies who engaged in it. Most of them stopped after burning through way too much money or getting mocked online and in the media.

The fact that some of the AI companies were able to convince some overly gullible CEO's to do something that was so against their own self-interest is to me not that different from Disney using marketing to convince parents that they have to take their kids to Disney World as a key, essential, fleeting milestone, and if the parents fail to take their kids to Disney World before they age out (e.g., outside of that magic 4 to 7 years of age), and they lose that innocent joy and wonder, that makes the parents somehow a failure. That kind of emotional blackmail, and fear of missing out, is quite classical marketing, and it's considered completely legal. If large number of parents fall for the marketing, and go into debt so that their kids can have that magic --- well, it sucks to be them, and it's great for the Disney Shareholders. (And remember, if your pension plan or mutual fund owns shares in Disney, you may have benefited as well.)

The FOMO over the opportunity for a company to save lots of money and become more successful than their competitors, if they only could use AI, is really not all that different from the marketing done by the Disney corporation. And it's also not that different from IBM salesfolk managing to convince CEO's that they should buy Lotus Notes, and force all of their employees to use Notes instead of standard e-mail. (And if you ever had to use Lotus Notes, you'll know how awful that was.....). Again, sucks to be the gullible CEO, great if you're the IBM shareholder.

A growing disconnect

Posted Aug 30, 2026 23:32 UTC (Sun) by tytso (✭ supporter ✭, #9993) [Link] (3 responses)

The CEO's at Nvidia, Open AI, and Anthropic made the pitch to investors because they needed the cash --- lots of cash --- and they thought the best way to get that cash was to demonstrate how AI would create a huge amount of value to investors. As they needed more cash, they needed more and more exaggerated profits, which means they needed to make larger and larger claims.

As far as the companies doing the layoffs, they fall into three categories. One is those CEO's who genuinely hired too many people and needed an excuse. "I screwed up and was over-optimistic about revenue growth" sounds like the CEO was at fault. "We're laying off people because AI" makes the CEO sound visionary.

Another is those companies who invested so much into Data Centers that they needed to lay off people to make room for buying more machines. Amazon laid off a whole bunch of people after having the most profitable quarter in their history, because they were trying to avoid their Free Cash Flow (that's money going out minus money coming in, regardless of whether the money is spent on operating expenses or capital expenses). Google allowed their FCF to go negative, and their share price got pummeled. In the case of Oracle, their growth had been slowing for the past decade, and given that their cloud business was in a distant 4th place, with Google having 2.5x more revenue, and Amazon and Microsoft having 4x more revenue. Oracle thought that AI could be their silver bullet to reshuffle the deck. The problem, though, was twofold. The first is that it required a very huge upfront investment (see the FCF problem). The second is that compared to the enterprise software market, which was a very high margin business, the AI token business is much less so. Once you build your enterprise software product, if you can manage to sell double the product, the incremental cost is very small (just the sales commision and support / help desk). But with the AI token business, if you manage to sell double the number of tokens, you need double the number of servers, GPU's, and data centers. And since Oracle is just selling GPU's, without their own LLM model, they are no better than a NeoCloud, which means it is a very low margin business indeed. Think of the ROI of an electric utility versus a software company. I don't know if the Oracle CEO figured this out before he made the pivot, but given that Oracle is turning into a utility they had to lay off a lot of people, including a lot of Linux Kernel developers I know at Oracle. :-(

The third category of CEO's who laid off people are the people who were genuinely deluded enough to think AI would save them money, and in our capitalistic system, the only stakeholder which matters is the share holder. Profit Good. Ruining the lives of employees is a risk that the CEO is happy to take, with apologies to Lord Farquaad. Examples here include Facebook, where Zuckerberg genuinely thought that he could slash 60% of Facebook SWE's, and thus justify how much money he had dumped into AI. (See: https://futurism.com/artificial-intelligence/mark-zuckerb... ). Another example of Ford, who thought they could fire a whole bunch of experience quality insurance inspectors and replace them AI. This backfired spectacularly, and they had to hire their experienced engineers and give them a bump in salary so they would be willing to come back.

They reason why dot-Com didn't have as much toxicity is that the number of people who got laid off was much smaller, and those who did were people much lower on the totem pole (e.g., people working in retail). And in many cases, when companies like Toys'R'Us went bankrupt, while e-commerce was definitely a contributor, the blame got laid at the feet of private equity instead. (And make no mistake, private equity very much deserved it. Yet another "feature" of our cut-throat capitalistic system.). In contrast, AI is impacting the white-collar worker --- the lawyers, the computer programmers, the knowledge workers. Blue collar workers aren't really affected; in fact, a lot of trade unions who have a lot to lose if data center building gets halted, have started pushing back against the AI data center backlash. ( https://www.wsj.com/economy/jobs/blue-collar-jobs-are-the... ). So there is now a blue collar backlash to the backlash. :-)

A growing disconnect

Posted Aug 31, 2026 2:47 UTC (Mon) by mathstuf (subscriber, #69389) [Link]

And in many cases, when companies like Toys'R'Us went bankrupt, while e-commerce was definitely a contributor, the blame got laid at the feet of private equity instead.

A family friend worked at the bank that made the decision that bankruptcy was the way to go. He kept a letter from a child lamenting the loss of Toys'R'Us asking to keep it alive on his desk. PE is definitely to blame there.

A growing disconnect

Posted Aug 31, 2026 20:51 UTC (Mon) by MarcB (subscriber, #101804) [Link] (1 responses)

> As far as the companies doing the layoffs, they fall into three categories. One is those CEO's who genuinely hired too many people and needed an excuse. "I screwed up and was over-optimistic about revenue growth" sounds like the CEO was at fault. "We're laying off people because AI" makes the CEO sound visionary.

A very relevant variation of this is "We expect economic downturn, and must prepare for it, but don't want to talk about it".

Also, I actually can understand some of those CEOs. They are in a very tricky position.

Firstly, a CEO will usually have some idea about what their employees are doing. But they will not have detailed knowledge. This is a very vulnerable state when it comes to AI. AI is extremely good at convincing you of capabilities it does not have if you use it in an area were your own understanding is limited.

Secondly, those CEOs are exposed to investors and shareholders. Those certainly have no idea about what the company's employees are doing. They also have no idea about what AI can actually do and what it cannot. But they have heard that it is super-human (And maybe they have created their own task planner app, which clearly proves that software developers are now redundant!).

This puts CEOs, and companies as a whole, under extreme pressure.

Good thing is that the worst of this seems to be over. Exploding token costs and high-profile failures were noticed. AI strategies are slowly changing into something that actually makes sense, and the understanding that AI is just a (powerful) new tool instead of "the solution to everything" is spreading. (This also happens in the other direction, i.e. people who were convinced that AI is utterly useless begin to see its value - which brings us back to why this GR result is good).

A growing disconnect

Posted Sep 1, 2026 9:04 UTC (Tue) by taladar (subscriber, #68407) [Link]

What i don't understand is why anyone still falls for the "x is the solution to everything" nonsense at all, no matter what x is, we have seen this with so many hype cycles over the last few decades that even people as far removed from the industry as CEOs should have figured this out by now. I know asking for them to have read relevant literature like "No silver bullet" might be a bit much to ask of people who hold unimportant positions with low entry requirements like CEOs but the concept isn't that hard to figure out on their own.

A growing disconnect

Posted Aug 31, 2026 11:58 UTC (Mon) by kleptog (subscriber, #1183) [Link] (9 responses)

> People genuinely hate AI,

This kind of phrasing is unhelpful. There are many things you can not like about AI: that training uses copyrighted works without permissions, that CEOs are using it as an excuse to fire people, that it's making search engines worse, it wastes energy, it helps students cheat, etc. By lumping everything under "hating AI" it changes from something you can have a nuanced position on to an identity position.

The same thing happens with labels like (anti-)capitalism, (anti-)globalism, socialism, etc. They become blobs of ideas you can no longer meaningfully have discussions/debates about because as soon as you suggest a nuanced position, people from all sides accuse you of supporting the other side.

I understand why people do this, because it works as coalition building, allowing people who otherwise have conflicting interests to operate under the same banner. It certainly feels emotionally satisfying and politically useful, but it doesn't make for good debates, discussions, policy or laws because it obscures the actual problems being addressed.

A growing disconnect

Posted Aug 31, 2026 16:40 UTC (Mon) by aimannajjar (subscriber, #184277) [Link] (8 responses)

> > People genuinely hate AI
> This kind of phrasing is unhelpful.

> I understand why people do this, because it works as coalition building, allowing people who otherwise have conflicting interests to operate under the same banner. It certainly feels emotionally satisfying and politically useful, but it doesn't make for good debates, discussions, policy or laws because it obscures the actual problems being addressed.

"People hate AI" is a description and not my position, and yes people are not being nuanced at all. But it's not because it's politically useful. They genuinely and deeply hate AI, nuance or not, once you hate something, you hate it. Can you blame them? Have you seen the headlines or the dystopian CEOs statements? "AI going to replace all human labor in a few years" - who says stuff like that except some narcissistic billionaire who has nothing to worry about? and the consistent doubling-down in each and every layoff?

It might seem conspiratorial (US is already blaming China) to those who are not directly affected by it. Unfortunately, being out of touch will not help them. Both the US government, and American companies are extremely out of touch. They've been doubling down for years and now they have to "win peoples hearts" despite their narcissistic urges to not care.

China does not need to do anything here except act **decently** and improve their tech, old-fashioned human decency.

A growing disconnect

Posted Aug 31, 2026 18:24 UTC (Mon) by gutschke (subscriber, #27910) [Link] (2 responses)

I genuinely don't understand what you are asking for.

You repeatedly and convincingly argue that the hatred for AI is a manufactured response. Your biggest issue seems to be with the messenger and not with the technology. You give countless examples of how hatred is an emotional reaction more than anything that could be argued on its merits. You make a strong point how this hatred has started becoming an identity for some.

And then you complain how out of touch others are by not embracing the same emotions. That's a pretty big ask.

I also don't get the final conclusion. It sounds as if you are perfectly happy with AI, as long as it isn't US made. That makes your appeal to emotion even harder to for me to understand.

A growing disconnect

Posted Aug 31, 2026 19:01 UTC (Mon) by aimannajjar (subscriber, #184277) [Link] (1 responses)

> You give countless examples of how hatred is an emotional reaction more than anything that could be argued on its merits.

I didn't say that. The emotional reaction is justified, devouring natural resources, the deceptive propaganda, the AI-washed layoffs, manipulating Wall Street, influencing local politicians to act against their constituents objections, raising electricity bills, and the list goes on. In my world view these emotions have merits and are important. It's just that in your world view this is entirely about whether AI is useful or not, hence the confusion. I understand, pragmatism and all.

> And then you complain how out of touch others are by not embracing the same emotions

That is exactly what out of touch means. Not understanding the emotions of general population, no need to "embrace" it, but understanding it is a good step.

> It sounds as if you are perfectly happy with AI, as long as it isn't US made

I am happy with any AI made by any decent company that has chosen to market it as a tool to advance humanity rather than replacing it.

A growing disconnect

Posted Sep 1, 2026 7:53 UTC (Tue) by kleptog (subscriber, #1183) [Link]

> That is exactly what out of touch means. Not understanding the emotions of general population, no need to "embrace" it, but understanding it is a good step.

You don't "understand" emotions. Emotions are like indicator lights: they are a indicator something is (or might be) wrong, but provide you with no information about why.

So someone tells me they "hate AI". Now what? It gives me absolutely no information about how to help them. You cannot debate an emotion, you cannot understand someone else's emotions, they just are. It's up each person to use words so other people can help them.

This applies to other areas too. "The people are outraged". A content free statement. "The people are worried they are going to lose their jobs!". This is more useful, perhaps we can strengthen worker protections so you can't be fired without cause. Perhaps (in the US) health insurance shouldn't be tied to your job. Perhaps we need to find a way so more of the productivity growth goes to wages rather than owners. All sorts of policies that would improve people's lives, and have nothing to do with AI.

For every complicated problem there is a solution that is simple, obvious and wrong.

A growing disconnect

Posted Sep 1, 2026 5:33 UTC (Tue) by zdzichu (subscriber, #17118) [Link] (4 responses)

What's wrong with replacing all labor? Most work is chore and people tend to "hate" (maybe too strong word) their work. Apart from few craftsmen, hardly anyone defines themselves through the work they do.

It will be wonderful, when all labor will be done by Thunderhead or other AI. People will have unlimited time to to spend their Universal Basic Income (if the money will still exist in this Utopia) and more time to play Cyberpunk 2077.

A growing disconnect

Posted Sep 1, 2026 6:08 UTC (Tue) by NYKevin (subscriber, #129325) [Link] (3 responses)

We have been sold labor replacement many times before, starting at roughly the industrial revolution. Hours worked have come down since the said revolution, but not since the 30's or so, despite tremendous gains in productivity. At a certain point, this falls into "definition of insanity" territory. Increases in productivity will not diminish hours worked, they will simply push us into tasks that are harder to automate.

A growing disconnect

Posted Sep 1, 2026 7:41 UTC (Tue) by taladar (subscriber, #68407) [Link] (1 responses)

But this is not due to the particular productivity technology of the day, this is due to wealth distribution issues. Those are completely orthogonal to productivity gains.

A growing disconnect

Posted Sep 1, 2026 16:04 UTC (Tue) by NYKevin (subscriber, #129325) [Link]

I do not disagree. I simply want capitalists (and "temporarily embarrassed" capitalists) to stop trying to sell me productivity gain as a panacea for all of capitalism's ills.

A growing disconnect

Posted Sep 1, 2026 9:28 UTC (Tue) by bluca (subscriber, #118303) [Link]

That's because working hours have come down thanks to trade union activism, not thanks to productivity gains.

A growing disconnect

Posted Aug 31, 2026 0:37 UTC (Mon) by dskoll (subscriber, #1630) [Link] (55 responses)

However, many people would say that being able to buy goods on-line at substantially cheaper prices is something that they have benefited from.

Undoubtedly. But where has that led? It has led to the place where Amazon is almost a monopoly and almost a monopsony. Many vendors have no choice but to sell through Amazon at whatever terms it dictates because otherwise Amazon would simply crush them. Wanna self-publish your book? Good luck trying not to sell it via Amazon.

Let's also not forget about Amazon's infamous union-busting tactics and the dozens of ways it exploits workers and quite literally damages their health. AI companies similarly exploit and harm workers.

The problem in the USA is that there's little to no regulation of businesses, whether it be in online sales or AI, and so they tend to behave very badly and get people angry at them. Some more regulation would, of course, reduce profits, but I don't think our societies should work to maximize profit. They should work to maximize well-being.

A growing disconnect

Posted Aug 31, 2026 2:14 UTC (Mon) by tytso (✭ supporter ✭, #9993) [Link] (54 responses)

Sure. That's my point. e-commerce Amazon style is extremely mixed. It makes life much more convenient, and consumers save money. But it's done on the backs of Amazon warehouse employees.

Or we could point out how the industrial revolution made factory life miserable, and put a lot of weavers and other skilled workers out of work. This is what spawned the luddite movement, which tried to smash machinery in factories in protest of how the industrial revolution. This is all true. Peoples' lives were made much, much worse as a result of the industrial resolution. But it's also true that before the industrial revolution the cost of a shirt would be around $3,500 ( https://www.sleuthsayers.org/2013/06/the-3500-shirt-histo... ), which meant that most working-class people could only afford three or four shirts. Today, a shirt is about hundred times less. Should we side with the luddites and go back to the time when thread and cloth were made by hand?

Going back to comparing the dot-com boom to AI, would it be rational to say that we should reject the internet since it was associated with the dot-com boom, and much of the internet was financed because of the e-commerce boom? This is similar to what some of the more extreme proposals in the Debian General Resolution were proposing, up to and including using Debian's Code of Conduct to forbid the use of AI or LLM's in any e-mail message (which presumably might include a patch which was generated using AI, or a security report that was generated using AI) that was sent to a Debian mailing list.

A growing disconnect

Posted Aug 31, 2026 2:49 UTC (Mon) by dskoll (subscriber, #1630) [Link] (53 responses)

I think the problem is that as the Internet has matured, companies have learned from it that the only way to succeed is with a ruthless winner-take-all policy.

So while in 1999, IMO, it was possible to start ethical Internet-based companies, I do not think it's possible now to start ethical AI-based companies---at least, not large-scale ones like OpenAI, Anthropic and so on. All of the large AI companies operate on large-scale theft of human creativity, exploitation of employees, externalization of scraping costs, externalization of environmental costs, and other shady practices.

So again IMO, until such companies are strictly regulated, it's not ethical to use their products. I know I'm tilting at windmills, but without windmill-tilters we wouldn't have had the Free Software Foundation and the GPL.

A growing disconnect

Posted Aug 31, 2026 5:54 UTC (Mon) by rgmoore (✭ supporter ✭, #75) [Link]

I think the problem is that as the Internet has matured, companies have learned from it that the only way to succeed is with a ruthless winner-take-all policy.

I think you're right that internet companies are worse today than they were 30 years ago, but that's true of non-internet companies, too. It's no coincidence that internet companies are behaving badly at the same time the USA is experiencing a wave of food safety disasters. It's the same root cause: inadequate enforcement of laws against corporate misbehavior. Even when the laws haven't been completely repealed, the agencies responsible for enforcing them have been hollowed out so the laws are largely a dead letter. This trend is especially bad here in the USA, but the EU is better only by comparison.

We wouldn't have a single Amazon dominating ecommerce, a single Google dominating search, a single Meta dominating social media, etc. if we had serious enforcement of antitrust laws. Those internet giants aren't just an indicator that antitrust laws aren't being enforced. Their enormous profitability has taught venture capitalists that establishing a monopoly is the key to achieving massive stock value. That's the real reason tech companies receive a big bonus in their stock valuation compared to non-tech: investors assume that if they succeed they'll be able to charge monopoly rents. That's why Tesla stock has been so outlandishly overvalued, for instance; they've successfully sold themselves as a tech company, so their stock price is based on an assumption that they'll get monopoly profit margins.

A growing disconnect

Posted Aug 31, 2026 18:41 UTC (Mon) by smurf (subscriber, #17840) [Link] (51 responses)

large-scale theft of human creativity

Theft has a rather clear definition. In non-legalese terms: I steal something if I (covertly, else its robbery) take it away from you, against your will and the law and all that, so that afterwards I have use of it and you don't.

Scarfing up oodles of human creativity and baking it all into a couple gigabytes or weights may be deplorable and shoddy and deprive some artists and writers and whatnot of their income, we can debate that until the cows come home, but IT IS NOT THEFT to deprive somebody of something they never had in the first place.

Using fighting words like these serves to signal that you're not interested in a nuanced discussion, but other than that it's not generally helpful. IMHO.

If that's your intent, go right ahead, but please do it somewhere else. IMHO².

A growing disconnect

Posted Aug 31, 2026 19:02 UTC (Mon) by dskoll (subscriber, #1630) [Link] (41 responses)

I absolutely consider it theft because down the line, it does take away income from humans. In my opinion, "theft" is an accurate description, not "fighting words."

If you think I'm exaggerating, I suggest you talk to some people in the fields of self-publishing, graphics arts, and music.

There are well-established precedents for compensating people for loss of expected future income.

A growing disconnect

Posted Aug 31, 2026 19:38 UTC (Mon) by bluca (subscriber, #118303) [Link]

Yes, if you ask RIAA, the MPA and Metallica they absolutely agree! yOu WoUlDn'T dOwNlOaD a CaR after all, would you?

A growing disconnect

Posted Aug 31, 2026 19:56 UTC (Mon) by tytso (✭ supporter ✭, #9993) [Link] (5 responses)

The invention of the mechanized loom at the dawn of the industrial age took away income from humans. It's not particularly useful to call mechanized looms "theft". When the invention of the automobile took away jobs from the horse carriage and buggy whip makers, there was no call to compensate them. So I disagree with you that there is a "established pattern".

From the perspective of LLM's, the question is whether the AI labs used illegitimate means to use when they scraped web pages and used it for training data. Consider that if a human reads an LWN article to learn how to be a better Kernel programmer, to the extent that they become a more talented programmer, that might mean that they could get hired by a company. To the extent that they get hired over some other candidate, would that be considered "theft"? So how is it different when an AI lab uses that same article as part of their training data?

The legal question seems to turn on whether the Large Language Model's weights are a derived work (from a copyright perspective) of their training data or not. A number of courts have ruled that the answer is no, but there are still lawsuits pending, and it's definitely not a settled question. But even if that question is settled, it still doesn't answer the question of whether the output of the LLM is in turn a derived work of the source training data. For example, if the training data included the linux kernel sources, and the BSD kernel sources, and poems from the Victorian era that have since entered the public domain, and you ask the LLM to generate some poetry, is that LLM output a derived work of the Linux kernel?

As to whether it is someone is legitimately angry, over AI or something else, that's not a particularly interesting question, at least to me. The question from my perspective is whether I should take their concerns seriously when proposing policies around the use of LLM's for the Linux Kernel, or when voting on things like the recently concluded Debian General Resolution on AI/LLM's. So while it might or might not be legitimate for someone to angry, and to throw around questions like "theft" --- who cares? There is a lot of outrage out on the internet. Film at 11.

Other people such as politicians who need to decide what their positions on data centers are concerned, might make a very different calculation on whether they need to take those "angry at AI" people seriously. But of course, given that there are blue collar works who are advocating very strongly against the data center moratoriums because *their* construction and plumbing jobs are on the line, and their trade unions are threatening to withhold support to the politicians, that's *also* something that politicians might decide that they need to take into consideration.

A growing disconnect

Posted Aug 31, 2026 22:32 UTC (Mon) by rgmoore (✭ supporter ✭, #75) [Link]

From the perspective of LLM's, the question is whether the AI labs used illegitimate means to use when they scraped web pages and used it for training data.

They didn't just scrape web pages. They also engaged in some very clear copyright violation by, for example, using pirated copy of non-web media. A big example was Anthropic being sued for using pirated books in training its AI; they wound up paying about $1.5 billion to settle the lawsuit. This is the reason they are now buying and scanning hardcopy books in bulk. First sale doctrine says they're legally in the clear as long as they bought the physical copy, even if they bought it used. It's way cheaper to buy and scan the books than to lose another lawsuit like that.

A growing disconnect

Posted Sep 1, 2026 6:19 UTC (Tue) by NYKevin (subscriber, #129325) [Link] (3 responses)

> The legal question seems to turn on whether the Large Language Model's weights are a derived work (from a copyright perspective) of their training data or not. A number of courts have ruled that the answer is no, but there are still lawsuits pending, and it's definitely not a settled question.

I wouldn't even go that far. The matter has barely been litigated. Most of the litigation so far has focused on the slightly different question of whether training involves illegal copying of the training works.

This is important, because if training is illegal, but models are not, then the "obvious" solution is to train in a different jurisdiction (where it is hopefully legal) and import the model. But nobody wants to litigate the question of whether a model is a derivative work of a given training input, because in most if not all jurisdictions, that is a complicated fact-dependent analysis. It may give different answers for different training inputs and/or different models. That means you have to litigate one combination of training input and model at a time, and cannot e.g. combine them all into a class action. If you do manage to "win," you probably get a decently large settlement, no admission of wrongdoing, an NDA, and no meaningful precedent.

A growing disconnect

Posted Sep 1, 2026 8:15 UTC (Tue) by taladar (subscriber, #68407) [Link] (2 responses)

> nobody wants to litigate the question of whether a model is a derivative work of a given training input, because in most if not all jurisdictions, that is a complicated fact-dependent analysis.

I agree that nobody wants to litigate that but the reason is probably more along the lines that it would show how ill defined that whole "derivative work" concept really is an how many other situations there are where something could clearly be considered derivative and yet we absolutely do not want Copyright to prevent that from happening in that situation.

A growing disconnect

Posted Sep 2, 2026 2:40 UTC (Wed) by NYKevin (subscriber, #129325) [Link] (1 responses)

Those cases have already been litigated long ago, leading to the development of doctrines like scènes à faire. Copyright has been around for roughly three hundred years. Most of the "obvious" questions have been answered by this point.

A growing disconnect

Posted Sep 2, 2026 2:44 UTC (Wed) by NYKevin (subscriber, #129325) [Link]

(Amusing but irrelevant anecdote: I tried Googling that phrase with the wrong accent marks, and the LLM answered in English. Then I tried with the correct accents, and it switched to French!)

A growing disconnect

Posted Aug 31, 2026 20:18 UTC (Mon) by MarcB (subscriber, #101804) [Link]

> I absolutely consider it theft because down the line, it does take away income from humans. In my opinion, "theft" is an accurate description, not "fighting words."

Many things take away income from humans: ignoring patents - including those on software or medicine, charity work, offering repairs the original vendor did not intent, giving away software for free, generally speaking everything that reduces artificial scarcity.

This clearly is not a useful definition of theft.

> There are well-established precedents for compensating people for loss of expected future income.

There actually are very few, that are not outright horrible or were quickly twisted beyond any reason. "Investor–state dispute settlements" come to mind. Or repeating extensions of copyright terms. Or tax systems that grow more and more complicated - and richer in loopholes - because some influential party felt disadvantaged.

A growing disconnect

Posted Aug 31, 2026 20:46 UTC (Mon) by mb (subscriber, #50428) [Link] (7 responses)

If you write any computer program, it take away work and therefore possible income from humans. That's the whole point.

Let's theoretically assume you write an awesome, highly optimized, automated and scripted calendar app. It takes away income from your assistant that you don't have to employ anymore to manage your appointments and meetings.

A growing disconnect

Posted Aug 31, 2026 21:09 UTC (Mon) by dskoll (subscriber, #1630) [Link] (6 responses)

That is a very far-fetched example since I never employed an assistant anyway, and I doubt anyone who uses my calendar software did either.

The example you gave and those tytso gave (the mechanical loom; the automobile) did indeed change the nature of the workforce, but they ended up creating new jobs and whole new markets too. Those inventions also did not steal (yes, I'm sticking to my guns) human creative output and then regurgitate it, displacing humans from their creative jobs.

From what I can see of AI, it is going to displace humans but not really open up new opportunities for them. And specifically as it relates to computing, I think it's going to lead to a dearth of good human programmers, which in turn will lead to a death of innovative software to train on, and we will see innovation plummet as the software industry settles into a "this is what we have and it's all we have" mode as no new good IP is created for the AIs to steal.

Anyway. I'm retired from the industry, so I don't really have that much skin in the game. I just know that if I were starting out now, I would not seek a career in software development.

A growing disconnect

Posted Aug 31, 2026 21:22 UTC (Mon) by mb (subscriber, #50428) [Link] (5 responses)

The whole (computer) industry is about replacing human work with machine work
https://en.wikipedia.org/wiki/Computer_(occupation)

Many of the programs I write manually without AI replace human work.
Many of the programs I write with AI replace human work.

There is no difference.

The only exception being programs that do tasks inherently impossible for humans. These don't take away work from humans, because humans couldn't do it anyway. No matter if written by AI or not.

A growing disconnect

Posted Aug 31, 2026 23:56 UTC (Mon) by rgmoore (✭ supporter ✭, #75) [Link] (4 responses)

The only exception being programs that do tasks inherently impossible for humans.

You're leaving out a whole category of programs: ones that aren't doing work in the sense of employment in the first place. Nobody is losing their job to automation because somebody wrote a computer game, or a social media app, or the like. A lot of computer applications fall into that category. People have been using computers for frivolous purposes almost from the beginning, certainly since well before the first personal computer. I wouldn't be at all surprised to learn that automation of employable work is only a tiny fraction of all the stuff that computers do these days.

A growing disconnect

Posted Sep 1, 2026 8:55 UTC (Tue) by taladar (subscriber, #68407) [Link] (3 responses)

Of course some entertainer is losing work because somebody wrote a computer game. And some social meeting place like a cafe or club might lose work because of a social media app.

A growing disconnect

Posted Sep 1, 2026 16:33 UTC (Tue) by rgmoore (✭ supporter ✭, #75) [Link] (2 responses)

Of course some entertainer is losing work because somebody wrote a computer game.

But that isn't really replacing human labor with automation; that's replacing one kind of job (entertainer) with a different kind of job (computer game developer). If you look at the credits list for a lot of newer computer games, they're as huge as the credits lists for movies. There isn't a big loss of jobs moving from one way of doing things to another, just a change in what the jobs are. There's a lot of that kind of thing in computer programming. If I wind up keeping my contact list in a computer program instead of a Rolodex, isn't replacing people who make Rolodexes with automation; it's replacing people who make Rolodexes with people who program contact managers.

That's different from when my employer replaced hard copy purchase requisitions with an online system. That cut a lot of positions in our administrative and purchasing departments because the new system just required a lot less work. It cut out even more jobs when it started interfacing directly to preferred vendors' online catalogs, since we no longer needed humans to interface between the systems. That's an example of losing human labor to automation.

A growing disconnect

Posted Sep 2, 2026 7:33 UTC (Wed) by taladar (subscriber, #68407) [Link] (1 responses)

The entertainment example also cut a lot of positions. Where previously you needed people to make every single thing you saw on screen you suddenly could create dozens of hours of entertainment per user for potentially millions of players with a team of artists and developers that is probably not much larger than e.g. a TV series or movie crew that made each second "by hand" and that doesn't even take into account replacing actual, in person entertainment with mass media in the first place which certainly cost a lot of theatre actors, standup comedians, live musicians,... their income.

Face it, every bit of progress reduces the number of people required to perform a given task and that is okay. The problem is not that we find ways to produce more useful goods or services with fewer people. The problem is what we do with the gained productivity.

A growing disconnect

Posted Sep 2, 2026 10:38 UTC (Wed) by paulj (subscriber, #341) [Link]

Also, the skilled modellers, carpenters, artists, tailors and dressmakers who made the props, miniatures, sets, mattes and costumes for the movies of old will not be the computer artists making computer models. People who spent half a life time becoming masters in one set of skills do not suddenly become masters in the skills of a field replacing theirs - and generally can not and do not.

As an aside, are modern CGI-based films/shows really better? Last year or two I've gone and watched some old classic films with my kids, and (despite the innate ennui kids have for old stuff from previous generations) they were still impressed at the immense scale of production we used to get in old epic block busters. You can't help but be impressed by, e.g., literally tens of thousands of soldiers, including a few thousand horsemen, in "Waterloo". Or the scale of "Cleopatra" and some other similar epics, ditto "A Bridge Too Far". etc. We've lost a lot of that with CGI. (Rare exceptions like "Saving Private Ryan", with the Normandy scenes with the Irish Army).

A growing disconnect

Posted Sep 2, 2026 6:57 UTC (Wed) by smurf (subscriber, #17840) [Link] (24 responses)

it does take away income from humans

No it doesn't. To repeat: You cannot take something away from somebody who doesn't have it.

Withholding is not taking away. These fighting words do not further solutions, only confrontation, and the result ends up being a net negative for society.

Don't believe me? one example: unions that force employers to not lay off people, with the net result that workers stand around doing nothing, plus the resulting product gets so expensive it's not being built any more. Check, for instance, the cost of building a mile of subway in NYC vs. Madrid.

There are well-established precedents for compensating people for loss of expected future income.

Zero of them are framed as restitution for theft.

A growing disconnect

Posted Sep 2, 2026 10:41 UTC (Wed) by paulj (subscriber, #341) [Link]

> Zero of them are framed as restitution for theft.

Indeed. This is a fundamental difference between "AI" and prior industrial advances. AI ingests all prior human knowledge and wishes to then extract rent from it.

A growing disconnect

Posted Sep 2, 2026 12:37 UTC (Wed) by dskoll (subscriber, #1630) [Link] (22 responses)

I'm sorry, but you are not living in the real world. Talk to graphic designers. Talk to musicians. Talk to self-published book authors. Their incomes are massively dropping because of AI output generated from stolen human-created works.

You can rationalize or sugar-coat it all you want, but this is the reality.

A growing disconnect

Posted Sep 2, 2026 13:23 UTC (Wed) by bluca (subscriber, #118303) [Link] (7 responses)

Their incomes are massively dropping because the capitalist class extracts all profits to enrich themselves, at the expense of everyone else and the environment. It has nothing to do with AI. It's a phenomenon that predates AI, and will survive AI. Handy example I saw just today: https://social.treehouse.systems/@Elizafox/11719939532601...

A growing disconnect

Posted Sep 2, 2026 13:48 UTC (Wed) by dskoll (subscriber, #1630) [Link] (6 responses)

AI dramatically exacerbates the problem and also gives the "capitalist class" you're referring to dramatically more power.

A growing disconnect

Posted Sep 2, 2026 14:35 UTC (Wed) by mathstuf (subscriber, #69389) [Link] (4 responses)

Another way to say it, perhaps: the capitalist class used to have muskets and pistols, but AI has given them automatic weapons. We can understand that they're still going to do bad things, but we don't have to exacerbate the issue by ignoring the step-function change in damage done.

A growing disconnect

Posted Sep 2, 2026 14:43 UTC (Wed) by bluca (subscriber, #118303) [Link] (3 responses)

It's just a convenient distraction to exacerbate the dividi-et-impera that allows the billionaires to keep doing what they are doing. You are missing the woods for the trees, and they appreciate your services.

A growing disconnect

Posted Sep 2, 2026 15:21 UTC (Wed) by mathstuf (subscriber, #69389) [Link] (2 responses)

I'm not saying that capitalists aren't an issue. But when your basement is flooding, you fix the immediate problem before tackling the observation the land around the house needs to be regraded.

A growing disconnect

Posted Sep 2, 2026 15:28 UTC (Wed) by bluca (subscriber, #118303) [Link]

If you spend all your time and energy shoving buckets of water out, you won't have any left to stop the people who are shoving buckets of water in, and the basement will remain flooded

A growing disconnect

Posted Sep 2, 2026 15:39 UTC (Wed) by paulj (subscriber, #341) [Link]

Indeed. And capitalism isn't even the root problem either. As another comment has pointed out, it's a human problem that long predates capitalism, that sub-groups will seek to concentrate control of resources to themselves (and some sub-groups will often succeed - see Mancur Olson for explanations as to why that is). You see this in capitalistic societies, in communism, in feudal societies, in hunter-gatherer societies, etc.

Further, solutions to such issues may often require domain-by-domain approaches. Cause "It's the system, we have to change the whole system" is... to put it mildly... not exactly the most concrete and practical suggestion for a solution. We have domain specific regulations for things like land rights, for water rights, for rights over creative works, etc., etc. Each of the sets of regulations for each domain are heavily tailored to details of those domains.

To implement a solution for some given domain requires understanding the issues and details of those issues in those domains. Dealing with the issues AI raises will require understanding and discussing those issues, in their particular details.

The repeated, endless, "It's not AI, it's capitalism" comments from bluca are both misguided (it's not capitalism) and becoming a bit tedious.

A growing disconnect

Posted Sep 2, 2026 14:42 UTC (Wed) by bluca (subscriber, #118303) [Link]

No, it really doesn't make that much of a difference, the power imbalance is what it always has been. Immense, unfathomable wealth inequality is what makes the difference. You'd like it if that were true though, as "eliminating AI" doesn't inconvenience you in any way and ranting against it doesn't cost you anything, so it's an easy cop out and a cathartic scapegoat to rally against.
The reality is more complex than that, and without serious reforms to reverse the trend on de-regulation and conglomeration of mega-corps, removal of workers rights and bargaining powers, and wealth inequality, and more, things will keep getting worse. But that's hard and inconvenient and doesn't get as many likes on social media though.

A growing disconnect

Posted Sep 3, 2026 15:26 UTC (Thu) by smurf (subscriber, #17840) [Link] (13 responses)

Their incomes are massively dropping because of AI output generated from stolen human-created works.

Those. Works. Have. Not. Been. Stolen. They still have them.

Also, I KNOW that many peoples' incomes ARE dropping. That's not the point. The point is that, yes they earn less, but nobody takes the money away from them that they already HAVE.

Thus per definition neither is theft.

You can call it theft all you want, but that's roughly as effective as stating that the moon is made of Swiss cheese — especially when you try to convince those people who might actually make a difference here.

A growing disconnect

Posted Sep 3, 2026 15:36 UTC (Thu) by dskoll (subscriber, #1630) [Link] (12 responses)

I'm going to keep using the words theft and stolen because those are appropriate. By stealing human creations, LLMs are economically harming the people who created the stolen works. The fact that creators still have their now-devalued works is cold comfort.

Some more enlightened jurisdictions (eg, the EU) permit creators of works to specifically opt their works out of AI training. Unfortunately, that just means AI companies do the training on stolen works in less-enlightened jurisdictions. Until the whole world has uniform rules in this regard, AI companies will continue stealing.

And I hold out very little hope the USA will follow the EU's lead.

A growing disconnect

Posted Sep 3, 2026 15:44 UTC (Thu) by bluca (subscriber, #118303) [Link] (11 responses)

yOu WoUlDn'T dOwNloAd A cAr

A growing disconnect

Posted Sep 3, 2026 15:53 UTC (Thu) by corbet (editor, #1) [Link] (10 responses)

And this response is helpful how? We could write a bot to do this stuff.

ISTM that some people believe that the creator of an artifact holds a property right for the exploitation of the ideas (patents) or expression (copyright) inherent in that artifact. If you have that belief, then unauthorized use constitutes a theft of a property right. A subset of the people feeling this way also feel that use in model training is unauthorized use, and thus an act of theft.

If you believe the property right does not (or should not) exist, you see it differently. You might also believe that the property right does exist, but that model training is fair use, and thus not theft.

Going around in endless circles on this is really not going to change anybody's mind at this point. Perhaps we can stop here? Please?

A growing disconnect

Posted Sep 3, 2026 20:17 UTC (Thu) by intelfx (subscriber, #130118) [Link] (9 responses)

If you believe the property right does not (or should not) exist, you see it differently. You might also believe that the property right does exist, but that model training is fair use, and thus not theft.

There's also a third belief that words have meanings, and using the word theft to describe something which is, at worst, a violation of intellectual property laws (in their most restrictive interpretation, which is not even always shared by actual courts of law), means either ignorance or a deliberate choice of a fighting word to produce an inflammatory emotional response.

It must also be noted that the community to which this site is dedicated, at its root, grew out of a movement that found copyright oppressive, and invented an entire legal hack to ensure that the information stays free, so there's that.

A growing disconnect

Posted Sep 3, 2026 20:20 UTC (Thu) by intelfx (subscriber, #130118) [Link]

...I do apologize for attempting to continue an argument that you intended to shut down, but I found myself really disagreeing with the "two choices" framing here.

A growing disconnect

Posted Sep 10, 2026 16:33 UTC (Thu) by mrugiero (guest, #153040) [Link] (7 responses)

There is also a fourth belief that is that we should all be equal in front of the law, and the way LLMs had been trained combined with the fact those doing it faced no real consequences proves we're not. "Pirate" some papers for the common good and you'll go to jail. "Scrape" a massive database of copyright infringing material to train "AI" and somehow all is good. The difference? One was a regular person and the other is a company valued on trillions of dollars. I'm not sure why people aren't angrier at that than they seem. There seem to be two different legal systems, one for the working class and one for the capitalist class. Both work for the capitalist class.

A growing disconnect

Posted Sep 10, 2026 18:02 UTC (Thu) by bluca (subscriber, #118303) [Link] (6 responses)

Yeah but the sensible reaction to that is to demand that copyright gets reformed and its restrictions severely tuned down, rather than taking the view that Metallica were right to sue teenagers using Napster because all IP violations are "theft"

A growing disconnect

Posted Sep 11, 2026 10:23 UTC (Fri) by anselm (subscriber, #2796) [Link] (5 responses)

Another sensible reaction would be to argue that, rather than watering down existing copyright laws, they should apply to both teenagers and big AI corporations alike. If teenagers don't get to download copyrighted stuff from the Internet then neither should big AI corporations – especially if the big AI corporations do it on a vastly greater scale than the teenagers ever could, and for commercial gain rather than for their own enjoyment.

Incidentally, before you claim again that in the EU, training LLMs from copyrighted data is explicitly allowed unless the copyright holder objects: At least here in Germany, the Landgericht München 1 would beg to disagree. See Az 42 O 763/25. The court's conclusion is that the “text and data mining” carve-out allows examining the mined data for “patterns, trends, and correlations” (§ 44b UrhG), but not using it in the creation of a new, durable product – like a genAI model – that can potentially imitate or reproduce the original data. (This ruling can still be appealed, so it'll be a while until we find out whether it will actually stand, but the direction is at least interesting.)

A growing disconnect

Posted Sep 11, 2026 10:42 UTC (Fri) by bluca (subscriber, #118303) [Link] (4 responses)

> Another sensible reaction would be to argue that, rather than watering down existing copyright laws, they should apply to both teenagers and big AI corporations alike.

Being in favor of strong IP laws, like Disney, Microsoft and the MPA are, is of course a legitimate position. You are of course entitled to believe information should be restricted as much as possible and that proprietary software is the best software development model, and that the strongest IP laws possible are necessary and just. As an open source person it is not one I share: information yearns to be free.

> Incidentally, before you claim again that in the EU, training LLMs from copyrighted data is explicitly allowed unless the copyright holder objects: At least here in Germany, the Landgericht München 1 would beg to disagree.

EU law has supremacy where the treaties apply, as established by the Court of Justice time and time again. It does happen every now and then that local courts try to assert local laws over it, and invariably it gets squashed once the case goes up the stack.

A growing disconnect

Posted Sep 11, 2026 12:19 UTC (Fri) by dskoll (subscriber, #1630) [Link]

As both a creator of non-software works and of open-source software, I can see both sides, but in the end I come down in favor of sensible copyright protections. We need a way for creators to assert both financial and moral ownership of their works, for a limited time like 25 years or so.

If it were not for copyright law, I would not have created my non-software works. And I might well have not created my open-source (GPL'd) software either, if anyone could just take them and make them proprietary.

A growing disconnect

Posted Sep 11, 2026 14:02 UTC (Fri) by kleptog (subscriber, #1183) [Link]

> EU law has supremacy where the treaties apply, as established by the Court of Justice time and time again. It does happen every now and then that local courts try to assert local laws over it, and invariably it gets squashed once the case goes up the stack.

EU law can also be unclear at times.

The way the AI Act (2024/1689) is written it seems to assume that AI training is "text and data mining" (TDM) as defined in the copyright directive. It however, *doesn't say so explicitly*. In particular, TDM is basically only mentioned in the recitals, in particular Recital 105 where it says AI model providers doing TDM and needing to negotiate with rights-holders if they opt out. (There's a reference in the risk management section but that's not relevant here.)

So the question is: does a mention in a recital in a Regulation change the meaning of TDM in the Copyright Directive (2019/790). That's not so obvious to me. And honestly, is this something you want decided by a court rather than the legislature? It feels like the legislature intended it, but then they should have said so explicitly.

A growing disconnect

Posted Sep 13, 2026 19:35 UTC (Sun) by mrugiero (guest, #153040) [Link] (1 responses)

> Being in favor of strong IP laws
> As an open source person

Those are not in contradiction. You may very well believe the author of a piece has a right to decide how it's used and by who and under what terms, and still decide that your own works shall be available to anyone for any use.

> information yearns to be free

Sure, but making the laws lax **NOW** just reflects the same pattern I'm complaining about. The law did not change when Metallica sued teenagers, it did not change when Aaron Swartz released a ton of papers, but in this hypothetical scenario it will change **AFTER** massive corps make mass infringement so they get a free out of jail card. How is that not yet another example of laws by capital and for capital? I think the most rational reaction would be seizing the means of production and burning down datacenters, not advocating for friendlier laws for the rich.

A growing disconnect

Posted Sep 13, 2026 23:14 UTC (Sun) by bluca (subscriber, #118303) [Link]

> Sure, but making the laws lax **NOW** just reflects the same pattern I'm complaining about.

The law was always wrong. The best time to change it was 20 years ago, the second best time is today.

> I think the most rational reaction would be seizing the means of production and burning down datacenters, not advocating for friendlier laws for the rich.

Finally, we are getting somewhere ;-)

A growing disconnect

Posted Aug 31, 2026 21:00 UTC (Mon) by aimannajjar (subscriber, #184277) [Link] (8 responses)

> Scarfing up oodles of human creativity and baking it all into a couple gigabytes or weights may be deplorable and shoddy and deprive some artists and writers and whatnot of their income, we can debate that until the cows come home, but IT IS NOT THEFT to deprive somebody of something they never had in the first place

Wait, so someone stealing your work to create a machine that can recreate that work is not theft? Is IP theft not a thing suddenly today?



A growing disconnect

Posted Aug 31, 2026 21:24 UTC (Mon) by intelfx (subscriber, #130118) [Link]

Wait, so someone stealing your work to create a machine that can recreate that work is not theft? Is IP theft not a thing suddenly today?

By the same token, "IP theft" is not actually a theft. There's no such thing as "IP theft". There's only violation of various IP laws (most commonly, copyright).

And once you adopt the proper nomenclature, the answer to your question becomes self-evident: for a violation of IP laws to occur, there must be an actual law and an actual violation of it (and an actual subject of that violation, i.e. the work whose copyright was infringed on). Otherwise it's just hot air. As it currently stands, training models on existing (lawfully obtained copies of) works and then using those models in process of creation of new (dissimilar) works is not, in fact, proven to be a violation of any existing law in US or Europe.

A growing disconnect

Posted Aug 31, 2026 21:27 UTC (Mon) by bluca (subscriber, #118303) [Link] (5 responses)

"IP theft" was bullshit when Metallica was suing kids downloading songs with Napster, and it is bullshit today. Or doesn't information yearn to be free anymore?

A growing disconnect

Posted Sep 1, 2026 3:04 UTC (Tue) by aimannajjar (subscriber, #184277) [Link] (4 responses)

> "IP theft" was bullshit when Metallica was suing kids downloading songs with Napster, and it is bullshit today. Or doesn't information yearn to be free anymore?

The flip side of this statement is even more interesting: "IP theft" was such a heavily enforceable crime when kids were downloading media over BitTorrent, but no longer so when Amodei is doing it to book authors who spend years perfecting their books to make a living.

A growing disconnect

Posted Sep 1, 2026 6:54 UTC (Tue) by anselm (subscriber, #2796) [Link] (1 responses)

"IP theft" was such a heavily enforceable crime when kids were downloading media over BitTorrent, but no longer so when Amodei is doing it to book authors who spend years perfecting their books to make a living.

It's not quite as simple as that. Here in Germany, for example, casual downloads of media are usually a civil tort, not a crime, i.e., if you're caught and convicted, you can be ordered to pay compensation, but – unlike what the content industry would have you believe – you won't go to prison. IP “theft” only becomes a crime (punishable with up to 5 years in prison) when it is done wilfully and on a commercial scale, which might have been bad news for Mr Amodei had he ended up in front of a German court.

(This is incidentally why, if you're sued for copyright infringement via BitTorrent in Germany, what you're likely going to be accused of is not actually downloading the stuff in question but distributing it, which is how BitTorrent works – people who download stuff using BitTorrent also make it available from their own machine to other BitTorrent users. This will push you towards the “commercial scale” end of the civil tort/crime spectrum, which is not good, and will also lead to higher claims for damages because you didn't merely cheat the original publisher out of the price for one copy of the stuff in question but possibly many copies. The details can be messy.)

A growing disconnect

Posted Sep 1, 2026 8:58 UTC (Tue) by taladar (subscriber, #68407) [Link]

They also like to pretend "1 whole movie" is commercial scale.

A growing disconnect

Posted Sep 1, 2026 9:26 UTC (Tue) by bluca (subscriber, #118303) [Link] (1 responses)

Actually Anthropic and Meta got some hefty fines for torrenting stuff I'm pretty sure?

A growing disconnect

Posted Sep 1, 2026 11:15 UTC (Tue) by anselm (subscriber, #2796) [Link]

Not exactly. Anthropic settled Bartz v. Anthropic for 1.5 billion USD, but a settlement is not a fine (i.e., Anthropic has not actually been found guilty of copyright infringement by the court). A variety of lawsuits brought against Anthropic by major music publishers are still pending.

As far as OpenAI is concerned, they have also been sued by several plaintiffs including the New York Times but AFAIK no fines have been imposed so far. OpenAI seems generally less interested in settling than Anthropic, but they will probably try to settle a lawsuit if it seems to be going against them and being found guilty would create an unfavourable precedent.

A growing disconnect

Posted Sep 2, 2026 15:41 UTC (Wed) by smurf (subscriber, #17840) [Link]

Wait, so someone stealing your work to create a machine that can recreate that work is not theft? Is IP theft not a thing suddenly today?

Did they take your work away from you so you cannot use it yourself any more? No.

Thus it's not theft.

Very simple, that.

"IP theft" is the same problem. Fighting words, designed to extract maximum f***up money from the people who committed the heinous crime of downloading-and-then-re-sharing popular music on Napster, or movies on Bittorrent. "We logged/guesstimated 2000 TCP connections to your box, each getting [a minuscle part of] something that grosses $20 when you buy it in the store. $40k in damages please. Oh, plus lawyer cost." (Yes I'm exaggerating, but only slightly.)

Lawyer time: ten seconds, signing the paper. Lawyer earns: xx% on top. Per case.

Sorry but if that's stealing from somebody it's from the perpetrator(s).

Maintainer opinions?

Posted Aug 29, 2026 15:13 UTC (Sat) by jpeisach (subscriber, #181966) [Link] (2 responses)

What if a package maintainer feels that they wish to reject a PR because the description is written by a LLM? Are they allowed to do that?

Maintainer opinions?

Posted Aug 29, 2026 15:28 UTC (Sat) by hlieberman (subscriber, #123867) [Link]

Speaking only as a DD, but not officially on behalf of the project:

As long as you are being respectful and following the other requirements of the Code of Conduct, the only group with the ability to force a maintainer to accept a patch would be the Technical Committee. That's a high bar, and generally the only things that go in front of them are issues around major packages or mass-filings, or when two maintainers heatedly disagree.

As a maintainer of a package, I am responsible for what goes into that package. Another maintainer can disagree with my decisions, but absent a ruling of the TC, they can't override my decision*.

*: Technically, a developer _could_ just ship an update with their change, but doing so over the objections of the maintainer would almost certainly end up getting you in a world of trouble.

Maintainer opinions?

Posted Aug 29, 2026 15:30 UTC (Sat) by cen (subscriber, #170575) [Link]

AFAIK a Debian maintainer generally has broad discretion over which patches to accept for any reason, including AI. But they cannot use that discretion to refuse to implement a binding Debian project/Technical Committee decision or an applicable Debian Policy requirement.

Maintainability

Posted Aug 29, 2026 15:40 UTC (Sat) by akselmo (subscriber, #174307) [Link] (8 responses)

I have a feeling this will cause maintenance and stability issues further down the line, as people rarely are "responsible" with
these tools. I and my servers hope I am wrong.

Maintainability

Posted Aug 29, 2026 16:41 UTC (Sat) by dilinger (subscriber, #2867) [Link] (7 responses)

It really was the worst possible option. As Bunk said (https://lists.debian.org/debian-vote/2026/08/msg00367.html), lack of disclosure is by no means responsible. The result leaves it at the status quo, eg the current wild west. It feels to me like the project as a whole voted to kick the can down the road a few years (and I've seen a few people who voted for the option say as much).

Maintainability

Posted Aug 29, 2026 16:54 UTC (Sat) by MarcB (subscriber, #101804) [Link] (6 responses)

What does it really improve? People can lie, and the possibility alone that someone might have been lying causes tensions.

Ironically, demanding disclosure of AI usage would only work without problems in an environment, where it is absolutely acceptable to use AI.

Maintainability

Posted Aug 29, 2026 17:07 UTC (Sat) by dilinger (subscriber, #2867) [Link] (4 responses)

> What does it really improve? People can lie, and the possibility alone that someone might have been lying causes tensions.

Currently people can lie about whether they used AI, and there's no consequence. Sure, there might be personal consequences - a specific maintainer, after discovering such an incident, might say, "never submit a PR to me again". The person who lied, however, can just move onto another package/maintainer/project.

Otoh, codifying a "you must disclose" rule allows for more project-wide consequences.

> Ironically, demanding disclosure of AI usage would only work without problems in an environment, where it is absolutely acceptable to use AI.

Sure, that was option 2. Well, except option 2 used the wording "should" instead of "must", but oh well.

Option 3 did use the word "must", but it was also a vote to reject as far as practical, so that's not really an environment where it's acceptable to use AI.

Maintainability

Posted Aug 29, 2026 18:13 UTC (Sat) by MarcB (subscriber, #101804) [Link] (2 responses)

> Otoh, codifying a "you must disclose" rule allows for more project-wide consequences.

That is precisely the problem. Those consequences might very well be quasi-witch-hunts/inquisitions with huge chilling effects on *any* potential contributor (even those not using AI). It could easily have created an atmosphere of pervasive distrust and people forming an "AI police" (maybe even outside the project; imagine comments in the bug tracker).

Having gone through some of the public discussions about this, I am absolutely certain this would have happened.

But don't get me wrong: I would absolutely prefer a declaration of AI usage - given an appropriate environment.

The Wikipedia comparison

Posted Sep 1, 2026 16:34 UTC (Tue) by NYKevin (subscriber, #129325) [Link] (1 responses)

Debian politics remind me a lot of Wikipedia politics, and Wikipedia has a long and sordid history of editors trying to manipulate the project's rules against perceived enemies. [1] is a somewhat infamous (if older) example, but I suspect there are many lower-profile instances that have flown under the radar just because of how absurdly easy it is to coordinate activity off-project without being noticed.

OTOH, Wikipedia has set itself up for this failure mode by banning overt factions. That probably helps motivate people to conduct these discussions off-project and in secret (whereas DDs seem much more willing to flaunt their politics openly). Still, I would be pretty surprised if this sort of thing has literally never happened in Debian history (whether it was discovered or not).

[1]: https://en.wikipedia.org/wiki/WP:EEML

The Wikipedia comparison

Posted Sep 1, 2026 18:08 UTC (Tue) by amacater (subscriber, #790) [Link]

TINC - There Is No Cabal

Maintainability

Posted Aug 31, 2026 9:10 UTC (Mon) by taladar (subscriber, #68407) [Link]

As far as I remember the "you must disclose" options avoided actually specifying how that disclosure should happen so that probably didn't exactly help them gain any support.

Maintainability

Posted Aug 29, 2026 17:28 UTC (Sat) by oridb (subscriber, #85941) [Link]

> What does it really improve? People can lie, and the possibility alone that someone might have been lying causes tensions.

As I keep saying, because people will steal anyways, we should legalize robbery.

what I miss is anti-agent-harassment provisions

Posted Aug 29, 2026 16:22 UTC (Sat) by hmh (subscriber, #3838) [Link] (1 responses)

My biggest issue with AI is fire-and-forget agents harassing or otherwise demanding time from humans (maintainers) in the mailing lists or bug trackers. Even single post by agents is already bad because the cheaper it is (in human time and effort) to push slop, the worse for maintainers that are in the receiving end.

If that starts to get even worse (in Debian), I expect something forbidding that outright to make it to the CoC or terms of service, this GR outcome would not contradict that.

what I miss is anti-agent-harassment provisions

Posted Aug 29, 2026 21:16 UTC (Sat) by aigarius (guest, #7329) [Link]

Just replace AI with human and .... existing Debian rules are perfectly adequate already. Which is exactly what the winning choice says.

How to satisfy legal compliance?

Posted Aug 29, 2026 16:35 UTC (Sat) by gray_-_wolf (subscriber, #131074) [Link] (9 responses)

> The Debian Project nevertheless expects that all contributions submitted to Debian, regardless of how and with which tools they were produced, satisfy the same standards of quality, correctness, maintainability, and legal compliance.

So I get all of these except the last one. When Claude gives me back a block of code, *how* can I ensure it is legally compliant? Either Debian decided that LLM-generated patches *cannot have* legal issues or they do not allow non-trivial generated patches, but did not want to say it? Which is it?

How to satisfy legal compliance?

Posted Aug 29, 2026 18:28 UTC (Sat) by clint (subscriber, #7076) [Link] (6 responses)

How can you ensure that a block of code given to you by a human is "legally compliant"? The answer is that you cannot.

How to satisfy legal compliance?

Posted Aug 29, 2026 22:51 UTC (Sat) by alx.manpages (subscriber, #145117) [Link] (5 responses)

You can ask the human, which should know. An LLM *can't* know.

How to satisfy legal compliance?

Posted Aug 30, 2026 8:50 UTC (Sun) by MarcB (subscriber, #101804) [Link] (1 responses)

Truly knowing this would require things like an exhaustive patent research that no one does.

Personally, I think this issue is blown out of proportion. There are legal risks specific to AI generated code, there are legal risks specific to human generated code (NDAs, other contracts, ...) and there are a lot of shared risks.

The last category is by far the largest. It includes patents, unjustified claims (think SCO), and so on.

As a side-node: The reason software-patents are not wrecking everything is that no "real" software company is actually enforcing their patents, because this would end up in mutually assured destruction. The same thing will likely apply to AI, even if legislation should make it risky.
Also: Given that the code used to train the AI likely was mostly open source - and we are talking about AI-generated open source code here - the risk is further reduced to that of license incompatibility.

How to satisfy legal compliance?

Posted Aug 30, 2026 11:55 UTC (Sun) by alx.manpages (subscriber, #145117) [Link]

> Truly knowing this would require things like an exhaustive patent research that no one does.

That patent research would be to check if you've invented something that was already invented and patented before?

That would be a pedantic legal issue, but I honestly don't care much about such issues.

I care about the moral/ethical issue, which is itself the root in which laws are based. LLMs are stealing code from their authors without complying with their license agreement (the agreement by which you can use their code). This is not ethically correct, and that's what's most important to me.

How to satisfy legal compliance?

Posted Aug 30, 2026 18:11 UTC (Sun) by clint (subscriber, #7076) [Link] (2 responses)

You can ask me if this sentence I've just written is legally compliant, but I cannot know that either so my answer would be meaningless. If what you're actually saying is that all you want is a good-faith belief that some arbitrary thing satisfies some arbitrary condition, that's an entirely different question.

How to satisfy legal compliance?

Posted Aug 30, 2026 20:13 UTC (Sun) by alx.manpages (subscriber, #145117) [Link] (1 responses)

It's not entirely different. I'm not a lawyer, but have some unusually extensive law experience, and AFAIK, acting on good faith will often be a key difference between going to jail or not.

How to satisfy legal compliance?

Posted Aug 30, 2026 21:34 UTC (Sun) by jwarnica (subscriber, #27492) [Link]

Acting on good faith that, say, a load is tied down, and you flip your transport truck and squish a motorcycle is very different than acting in good faith when the worst possible error is copyright infringement.

Some chunk of code may or may not be used in a life-critical application, and may have flaws in it. But the license being violated has absolutely no risk of life or limb.

How to satisfy legal compliance?

Posted Aug 30, 2026 2:21 UTC (Sun) by banana (guest, #144773) [Link]

If you don't think you can ensure such a thing, then don't submit the package to Debian. Some people may believe that there isn't an issue in practice and submit LLM-generated patched; this will work out fine as long as they're correct in this belief.

How to satisfy legal compliance?

Posted Aug 30, 2026 4:22 UTC (Sun) by intelfx (subscriber, #130118) [Link]

So I get all of these except the last one. When Claude gives me back a block of code, how can I ensure it is legally compliant? Either Debian decided that LLM-generated patches cannot have legal issues or they do not allow non-trivial generated patches, but did not want to say it? Which is it?

I'd like to point out that the answer to your question is literally contained in the text of the winning proposal:

Debian acknowledges that the legal status of material produced by generative AI
systems remains the subject of ongoing discussion in many jurisdictions,
including questions relating to copyright, authorship, licensing, and potential
reproduction of training material. The Project does not seek to resolve these
unsettled legal questions through this General Resolution, nor does it adopt
a position on whether AI-generated output is, in whole or in part, copyrightable
or derived from copyrighted works.

does not override project's choices

Posted Aug 29, 2026 19:06 UTC (Sat) by atai (subscriber, #10977) [Link] (4 responses)

This Debian decision does not override the choices of each project/package packaged in Debian, right?

For example, GNOME does not allow AI written code. The GNOME guideline still applies in Debian

does not override project's choices

Posted Aug 29, 2026 19:23 UTC (Sat) by salimma (subscriber, #34460) [Link]

It doesn’t affect upstream projects.

Though I should note that while some individual GNOME projects have adopted no-LLM policies, the wider project itself has not

does not override project's choices

Posted Aug 29, 2026 19:25 UTC (Sat) by intelfx (subscriber, #130118) [Link] (1 responses)

This Debian decision does not override the choices of each project/package packaged in Debian, right?

For example, GNOME does not allow AI written code. The GNOME guideline still applies in Debian

The GNOME guideline applies to the GNOME upstream. There's a distinction: GNOME does not have a power to "not allow" AI-written code, it only has a power to "not accept" it.

If GNOME does not allow AI written code (I didn't look, so I'm just taking your statement at face value), that does not and cannot prohibit Debian Developers to e.g. apply AI-authored patches to GNOME code downstream.

does not override project's choices

Posted Aug 31, 2026 7:00 UTC (Mon) by frankie (subscriber, #13593) [Link]

>If GNOME does not allow AI written code (I didn't look, so I'm just taking your statement at ?face value), that does not and cannot prohibit Debian Developers to e.g. apply AI-authored patches to GNOME code downstream.

... and also do not prevent any derived distribution or any independent hub (.e.g.flatpak, pypi, anaconda, docker hub, ....) from doing differently.

In synthesis, no distribution lives in a closed world. AI tooling is here to stay; people should understand how to manage impacts, not try to live in a separate world.

does not override project's choices

Posted Sep 1, 2026 13:43 UTC (Tue) by Conan_Kudo (subscriber, #103240) [Link]

For example, GNOME does not allow AI written code.

This is not actually a thing. Some GNOME codebases have made such a rule, but there is no project-wide rule.

LLM generated

Posted Aug 30, 2026 19:37 UTC (Sun) by joey (guest, #328) [Link] (1 responses)

It's worth noting that the winning text was generated by an LLM.

LLM generated

Posted Aug 30, 2026 19:44 UTC (Sun) by tytso (✭ supporter ✭, #9993) [Link]

To be specific, a human being came up with the ideas for the proposal, then an LLM generated the text from the ideas --- and then the human being did an editing pass to refine the generated text. Which is to say, the LLM was a tool, which was used by a human, and a human being took responsibility for the final result. Which is as it should be.

Barn doors open

Posted Sep 3, 2026 13:37 UTC (Thu) by stakkur (subscriber, #186047) [Link]

It’s not ideal, but the barn door’s open and the horses have already left, so to speak. Even if a ‘no AI’ policy were adopted, enforcing it would be near impossible. I think Debian is taking the most practical path.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds