Kind of arbitrary since it doesn't impact their upstreams.
The very first thing I used GenAI for was fixing build and CI pipeline issues, exactly the drudgery no one cares about. Today, LLMs are excellent at the yeoman's work of maintaining patches across multiple versions and upstream, which is what I imagine a lot of Debian devs do.
Why you wouldn't want to hand that off to a computer program is beyond me.
> Why you wouldn't want to hand that off to a computer program is beyond me.
Trust, stability, security. The reasons many people run Debian.
Although I think the main incompatibility in philosophy here is in "hand that off", e.g. direct agentic usage eroding opportunity for human judgement, which seems unlikely to be adopted anyway. The realistic proposals are essentially AI use with various guard rails and rules to preserve that human judgement (my interpretation).
> If Proposal H has a weakness, it's that it does not distinguish between local and cloud-based LLMs
Is the energy usage so different between local and cloud inference? Both require electricity, the local option even more than the better optimized cloud variant even perhaps.
How either is powered makes the crucial difference I suppose. Both can potentially run on solar as well as gas or nuclear.
It's the training that takes the most energy, and that needs to happen for locally running or cloud models regardless.
What am I missing here?
EDIT:
Proposal H mentions "LLM usage accelerates the destruction of our ecosystem"
I was thinking solely about energy usage, but there is of course also water usage.
A local setup is not water-evaporator cooled most likely.
This doesn't have anything to do with water usage? The rationale of banning it is that it's copyright status is ambiguous and at the least unethical, it produces lower quality code, and it stifles new developers from getting involved.
The last option considers the environmental impact alone to be worth a total ban. The vote is essentially a list of different takes on the issue and it's obvious different people have different concerns about it.
> Is the energy usage so different between local and cloud inference?
In throughput mode for agentic loads, the energy usage (tok/s/MW) of the new NVidia Vera Rubin hardware is 30x lower than that of the B300 and perhaps 450x lower then the H200 was, which in turn is hundreds of times lower than the inference for single users at home in any non-data-center hardware. It feels like comparing the momentum of an ant to the momentum of an elephant.
People want to ban LLM contributors not because of energy usage, but because AI generated code is unmaintainable and the people who generate it tend to engage with others in domineering and brazenly manipulative ways.
Fair enough. My comment was however asking about the difference in climate damage from local vs cloud inference, is either better in that specific regard. Or even the energy usage between the two.
As for ethics, i don't see a meaningful difference between the two. Local strips out a big-tech middleman perhaps.
> Debian is a large and complicated project: the release announcement for version 13 says it has 69,830 packages, which take a total of 403 GB of disk, and contain 1,463,291,186 lines of code. So, suitably, it is a large and complex poll.
Surely these are not just Debian-owned packages. Whatever Debian decides won’t apply to packages not owned by Debian
Debian is huge - that's why so many other distros use it as a base!
So, yes, all those packages are maintained either by an individual Debian Developer or a team of them.
There are third party repos, but they're (currently) neither as widespread nor as commonly-used as those for other major distros. This vote won't affect them, and you might potentially expect to see their use grow if one of the harsher options in this GR were to pass.
Sure, this is (mostly) about direct contributions to Debian.
Proposal C does discourage use more broadly but acknowledges that many upstreams take a different view and that it would be impractical to ban them from using it.
No, otherwise one would have to remove the Linux kernel, the Hurd kernel, LLVM, Python, Chromium, Firefox, Rust and everything using these. And probably some more.
No analogy, comparison (add your own language construct) is ever 100% true. Perhaps you should consider that my intent behind this was to say, that Debian will be blind, without ANY eyes.
I do use and run my own LLMs locally. But I understand why Debian might ban it, under just 1 reason:
LLM outputs cannot be copyrighted.
That alone starts to invalidate the GPL and other FLOSS licenses. It overall weakens the project the more you accept. And companies (MS, Amazon, etm) will gleefully loot anything marked with "LLM" as a free-for-all.
Until the whole copyright situation can be finalized, I'd understand for any FLOSS org to refuse all LLM code.
(Its also the same reason why Oracle also forbids all LLM code in Java and their DB. Same reason.)
There's not yet any reason to believe that mixing LLM output with your work makes it uncopyrightable any more than hitting enter in intellisense did before that.
We’re almost to the point where the average non-technical person uses AI daily (just like phones and the internet were once only used by the technocrats) without blinking, it will be ubiquitous and universal in adoption, yet comments here on HN from people that have the vantage point of seeing AND understanding how AI works, and who can also predict the inevitable integration of AI in society, have got their collective heads in the sand to the point they’re soon going to sound like conspiracy theorists.
The fact remains that it's what I see around me, and even if I'm in an inverse bubble, it shows that "the average non-technical person uses AI daily without blinking" is probably a huge overstatement.
Which is gonna be tricky as I am confident you don't have a time machine.
I think generative AI as a whole is a total scam. It has uses, but not many, and outside of translation, not good ones... and yet, it is running at truly epic financial losses. Multiple trillions of dollars are being pumped into this.
I think the next AI Winter, the 3rd big one, is coming very soon.
There's going to be a market crash to dwarf the dot-com crash and 20087 financial crisis (both of which I worked through as an employed adult), and a lot of the current tech industry will be wiped out.
I've watched that happen before as well.
I am quite looking forward to it. I have no stocks or shares in anything anywhere, few savings, no loans, no debt... but all my immediate family own their own homes outright, and I am close to retirement age.
I know few CEOs/Business owners who are non programmers (well one was exposed to it/learning, but didn't go down that path) and they do use AI for various work and non work related things. One told me how at manufacturing plant someone solved some unpleasant process to input specs into machine for manufacturing by having AI creating the automation.
His comment to me was: People who don't embrace AI will eventually be outcompeted by the ones who do.
And what happens when the price goes up by an extra zero on the end?
And then another one?
And then the cloud stuff goes away because all the vendors go broke, despite the price rises, and you have to switch to using local models?
That is, models that are hugely slower, unless you have a computer with hardware tensor-math acceleration, which costs 10x more because of the chip famine?
And then, suddenly, those local models cost an absolute bomb to license, because the companies that bought up the wreckage of the original model creators are trying to recoup their spend?
Meanwhile the company is being hit by nuisance litigation from people whose code the bots plagiarised, or customers who are annoyed it doesn't work?
There are so many ways all this stuff can just go away.
>And what happens when the price goes up by an extra zero on the end?
>And then another one?
Well so far the price trends towards going down by the zero at the end. Inference isn’t expensive, there’s a whole laundry list of providers to choose from. It’s training (for everybody) and salaries (for US labs) that are expensive.
>and you have to switch to using local models?
Many companies are interested in that, if for no other reason than to avoid uploading secret sauce to we-totally-dont-train-on-your-chats-pinky-promise API endpoints. A mid sized company can cough up $500k for dxg b200 which will be useful for years and years and the only ongoing cost is electricity. Put glm 5.3 on it and it’ll zip along for a number of employees at the same time.
>And then, suddenly, those local models cost an absolute bomb to license, because the companies that bought up the wreckage of the original model creators are trying to recoup their spend?
How are they recouping anything on a mit licensed model released into the wild? There are plenty of models released by "who?" companies and the cost to download those is zero.
>Meanwhile the company is being hit by nuisance litigation from people whose code the bots plagiarised
Any day now
The doom scenario makes no sense in the race to the bottom that’s happening now. China being the other big entity that’s invested in AI makes all the what-ifs about litigation and plagiarism moot; they will gladly offer everybody inference for dirt cheap, if for no other reason that they’d like it if you gifted them your IP that they will then use to outcompete you with their manufacturing prowess.
The genie’s out of the bag. AI is never going away, and we all need to adapt whether we like it or not.
But what when salaries would increase by factor 100-1000? This would make AI use even cheaper. (About as realistic...)
> And then the cloud stuff goes away because all the vendors go broke, despite the price rises, and you have to switch to using local models?
In this case there will be cheap inference hardware on the market as someone will sell all those assets. They don't disappear when a company goes broke (object permanence).
We could also speculate what would happen if all hyperscalers (AWS, GCP, Azure, Alibaba, ...) would suddenly be broke. Again, about as realistic a scenario.
There do seem to be pretty wildly varying views on how heavily it is used and what value it is providing. I see a lot of people getting a lot of value from it, and some people avoiding it like the plague. I think it's getting to be a more divisive topic over time, as well.
It also seems to be fast moving and emotionally charged enough that there is often disagreement on what should be fairly objective facts. E.g. whether there is any value to generative AI tools at all, and the magnitude of their impact on the environment.
What an interesting cultural artifact! I note no proposals in favour of "let it rip" which is presumably because that would not be a reflection of Debian philosophy. I do think the aggressively anti AI proposals are themselves un-cautious in the extreme.
It's one thing for a maintainer to use AI, and another to review AI code sent your way. The latter has been proven to be unsustainable, low value, and ultimately a challenge for open source. Everyone acknowledges this problem exists and some corrective action is required to mitigate it.
Kind of arbitrary since it doesn't impact their upstreams.
The very first thing I used GenAI for was fixing build and CI pipeline issues, exactly the drudgery no one cares about. Today, LLMs are excellent at the yeoman's work of maintaining patches across multiple versions and upstream, which is what I imagine a lot of Debian devs do.
Why you wouldn't want to hand that off to a computer program is beyond me.
> Why you wouldn't want to hand that off to a computer program is beyond me.
Trust, stability, security. The reasons many people run Debian.
Although I think the main incompatibility in philosophy here is in "hand that off", e.g. direct agentic usage eroding opportunity for human judgement, which seems unlikely to be adopted anyway. The realistic proposals are essentially AI use with various guard rails and rules to preserve that human judgement (my interpretation).
Actual Proposals (scroll down): https://www.debian.org/vote/2026/vote_002
There are ethical and environmental concerns arising simply from use and contribution to the industry which many people take seriously
> If Proposal H has a weakness, it's that it does not distinguish between local and cloud-based LLMs
Is the energy usage so different between local and cloud inference? Both require electricity, the local option even more than the better optimized cloud variant even perhaps. How either is powered makes the crucial difference I suppose. Both can potentially run on solar as well as gas or nuclear.
It's the training that takes the most energy, and that needs to happen for locally running or cloud models regardless.
What am I missing here? EDIT:
Proposal H mentions "LLM usage accelerates the destruction of our ecosystem"
I was thinking solely about energy usage, but there is of course also water usage. A local setup is not water-evaporator cooled most likely.
This doesn't have anything to do with water usage? The rationale of banning it is that it's copyright status is ambiguous and at the least unethical, it produces lower quality code, and it stifles new developers from getting involved.
The last option considers the environmental impact alone to be worth a total ban. The vote is essentially a list of different takes on the issue and it's obvious different people have different concerns about it.
> Is the energy usage so different between local and cloud inference?
In throughput mode for agentic loads, the energy usage (tok/s/MW) of the new NVidia Vera Rubin hardware is 30x lower than that of the B300 and perhaps 450x lower then the H200 was, which in turn is hundreds of times lower than the inference for single users at home in any non-data-center hardware. It feels like comparing the momentum of an ant to the momentum of an elephant.
People want to ban LLM contributors not because of energy usage, but because AI generated code is unmaintainable and the people who generate it tend to engage with others in domineering and brazenly manipulative ways.
Fair enough. My comment was however asking about the difference in climate damage from local vs cloud inference, is either better in that specific regard. Or even the energy usage between the two.
As for ethics, i don't see a meaningful difference between the two. Local strips out a big-tech middleman perhaps.
"¿Porque no los dos?"
"Why not both?"
> Debian is a large and complicated project: the release announcement for version 13 says it has 69,830 packages, which take a total of 403 GB of disk, and contain 1,463,291,186 lines of code. So, suitably, it is a large and complex poll.
Surely these are not just Debian-owned packages. Whatever Debian decides won’t apply to packages not owned by Debian
Debian is huge - that's why so many other distros use it as a base!
So, yes, all those packages are maintained either by an individual Debian Developer or a team of them.
There are third party repos, but they're (currently) neither as widespread nor as commonly-used as those for other major distros. This vote won't affect them, and you might potentially expect to see their use grow if one of the harsher options in this GR were to pass.
The point though, is that they package 3rd party applications and libraries.
If they ban AI, they won't stop packaging apps and libraries from parties that do use AI. They wouldn't even know who does or doesn't.
Sure, this is (mostly) about direct contributions to Debian.
Proposal C does discourage use more broadly but acknowledges that many upstreams take a different view and that it would be impractical to ban them from using it.
Posting this because I want people to vote -- and ideally vote for a total ban.
I'm curious about enforcement. Are they going to ban packages with a whiff of llm? If so, how will they reliably audit?
No, otherwise one would have to remove the Linux kernel, the Hurd kernel, LLVM, Python, Chromium, Firefox, Rust and everything using these. And probably some more.
[dead]
What are your arguments for a ban? Why do you expect someone to listen if you didn't provide a justification?
It does not appear that normal people can. There is no vote page, no information on how to, and we can only view results.
> and ideally vote for a total ban.
If a total ban were to happen, wouldn't that make Debian a target for hackers who can use LLMs to find vulnerabilities?
How would Debian keep up with the influx of AI assisted 0 days?
Not using AI to write production code doesn't mean you can't use AI proactively to find vulnerabilities.
Poster wants a "total ban", you can only deduce that means vulnerability management also.
Or you could read the proposal and deduce nothing.
It explicitly excludes security fixes.
Oh the hypocrisy...
Exactly, agreed 100% https://en.wiktionary.org/wiki/in_the_land_of_the_blind,_the...
H. G. Wells wrote about how this proverb has limitations [1].
[1] https://en.wikipedia.org/wiki/The_Country_of_the_Blind
No analogy, comparison (add your own language construct) is ever 100% true. Perhaps you should consider that my intent behind this was to say, that Debian will be blind, without ANY eyes.
Given almost all people here cannot vote, does this mean you posted non-AI-generated slop?
Wouldn't it be better to reduce slop, no matter how it was produced, instead of only AI-generated contributions, including non-slop contributions?
Are you going to fork kernel? If not then you are just a bunch of stupid hypocrites.
> Posting this because I want people to vote
Irrelevant to post it here then with your reasons because it only concerns Debian developers
I wrote it. I am fully aware of that, and I said so in the piece.
I want to spread awareness that it's happening -- it only has a couple of days to run.
So far there are only ~350 votes:
https://vote.debian.org/~secretary/gr_llm/
Circa 270 people have voted:
https://vote.debian.org/~secretary/gr_llm/tally.txt
(My reckoning; could well be wrong!)
With a project the size of Debian, I feel sure that is a tiny fraction of eligible developers.
I do use and run my own LLMs locally. But I understand why Debian might ban it, under just 1 reason:
LLM outputs cannot be copyrighted.
That alone starts to invalidate the GPL and other FLOSS licenses. It overall weakens the project the more you accept. And companies (MS, Amazon, etm) will gleefully loot anything marked with "LLM" as a free-for-all.
Until the whole copyright situation can be finalized, I'd understand for any FLOSS org to refuse all LLM code.
(Its also the same reason why Oracle also forbids all LLM code in Java and their DB. Same reason.)
There's not yet any reason to believe that mixing LLM output with your work makes it uncopyrightable any more than hitting enter in intellisense did before that.
Ban.
The options range from is A to H. It's not yes or no.
Crazy how divided AI is here on HN…
We’re almost to the point where the average non-technical person uses AI daily (just like phones and the internet were once only used by the technocrats) without blinking, it will be ubiquitous and universal in adoption, yet comments here on HN from people that have the vantage point of seeing AND understanding how AI works, and who can also predict the inevitable integration of AI in society, have got their collective heads in the sand to the point they’re soon going to sound like conspiracy theorists.
Every non-technical person I know hates genAI and goes far out of their way to avoid it.
I suspect your sample is a bit skewed, then. I see a fair bit more diversity of genAI use amongst technical and non-technical people I know.
It may be.
The fact remains that it's what I see around me, and even if I'm in an inverse bubble, it shows that "the average non-technical person uses AI daily without blinking" is probably a huge overstatement.
> it will be ubiquitous and universal in adoption
[[citation needed]]
Which is gonna be tricky as I am confident you don't have a time machine.
I think generative AI as a whole is a total scam. It has uses, but not many, and outside of translation, not good ones... and yet, it is running at truly epic financial losses. Multiple trillions of dollars are being pumped into this.
I think the next AI Winter, the 3rd big one, is coming very soon.
There's going to be a market crash to dwarf the dot-com crash and 20087 financial crisis (both of which I worked through as an employed adult), and a lot of the current tech industry will be wiped out.
I've watched that happen before as well.
I am quite looking forward to it. I have no stocks or shares in anything anywhere, few savings, no loans, no debt... but all my immediate family own their own homes outright, and I am close to retirement age.
Let it all burn.
I know few CEOs/Business owners who are non programmers (well one was exposed to it/learning, but didn't go down that path) and they do use AI for various work and non work related things. One told me how at manufacturing plant someone solved some unpleasant process to input specs into machine for manufacturing by having AI creating the automation.
His comment to me was: People who don't embrace AI will eventually be outcompeted by the ones who do.
Uhuh.
And what happens when the price goes up by an extra zero on the end?
And then another one?
And then the cloud stuff goes away because all the vendors go broke, despite the price rises, and you have to switch to using local models?
That is, models that are hugely slower, unless you have a computer with hardware tensor-math acceleration, which costs 10x more because of the chip famine?
And then, suddenly, those local models cost an absolute bomb to license, because the companies that bought up the wreckage of the original model creators are trying to recoup their spend?
Meanwhile the company is being hit by nuisance litigation from people whose code the bots plagiarised, or customers who are annoyed it doesn't work?
There are so many ways all this stuff can just go away.
>And what happens when the price goes up by an extra zero on the end?
>And then another one?
Well so far the price trends towards going down by the zero at the end. Inference isn’t expensive, there’s a whole laundry list of providers to choose from. It’s training (for everybody) and salaries (for US labs) that are expensive.
>and you have to switch to using local models?
Many companies are interested in that, if for no other reason than to avoid uploading secret sauce to we-totally-dont-train-on-your-chats-pinky-promise API endpoints. A mid sized company can cough up $500k for dxg b200 which will be useful for years and years and the only ongoing cost is electricity. Put glm 5.3 on it and it’ll zip along for a number of employees at the same time.
>And then, suddenly, those local models cost an absolute bomb to license, because the companies that bought up the wreckage of the original model creators are trying to recoup their spend?
How are they recouping anything on a mit licensed model released into the wild? There are plenty of models released by "who?" companies and the cost to download those is zero.
>Meanwhile the company is being hit by nuisance litigation from people whose code the bots plagiarised
Any day now
The doom scenario makes no sense in the race to the bottom that’s happening now. China being the other big entity that’s invested in AI makes all the what-ifs about litigation and plagiarism moot; they will gladly offer everybody inference for dirt cheap, if for no other reason that they’d like it if you gifted them your IP that they will then use to outcompete you with their manufacturing prowess.
The genie’s out of the bag. AI is never going away, and we all need to adapt whether we like it or not.
> And then another one?
But what when salaries would increase by factor 100-1000? This would make AI use even cheaper. (About as realistic...)
> And then the cloud stuff goes away because all the vendors go broke, despite the price rises, and you have to switch to using local models?
In this case there will be cheap inference hardware on the market as someone will sell all those assets. They don't disappear when a company goes broke (object permanence).
We could also speculate what would happen if all hyperscalers (AWS, GCP, Azure, Alibaba, ...) would suddenly be broke. Again, about as realistic a scenario.
> the average non-technical person uses AI daily
Citation needed.
"Having one's head in the sand" means being unaware of something; everybody is aware of it. "Conspiracy theorists" is just a slur.
If you have anything substantive to say about it that other people haven't heard, we're listening.
There do seem to be pretty wildly varying views on how heavily it is used and what value it is providing. I see a lot of people getting a lot of value from it, and some people avoiding it like the plague. I think it's getting to be a more divisive topic over time, as well.
It also seems to be fast moving and emotionally charged enough that there is often disagreement on what should be fairly objective facts. E.g. whether there is any value to generative AI tools at all, and the magnitude of their impact on the environment.
[dead]
Total ban please. This nonsense needs to end. Linus Torvalds is weakly vibe coding the kernel, that's bad enough.
Then you cannot use Linux, Firefox, Chromium, Python, Rust, LLVM and so on. Hurd can't be used either.
Maybe you have to switch to FreeDOS?
NetBSD is right there.
NetBSD has no alternative to Chromium, Firefox, LLVM, Rust or Python. They use AI-generated software too, including in the base system.
What an interesting cultural artifact! I note no proposals in favour of "let it rip" which is presumably because that would not be a reflection of Debian philosophy. I do think the aggressively anti AI proposals are themselves un-cautious in the extreme.
It's one thing for a maintainer to use AI, and another to review AI code sent your way. The latter has been proven to be unsustainable, low value, and ultimately a challenge for open source. Everyone acknowledges this problem exists and some corrective action is required to mitigate it.
Thats certainly an argument about the problems with LLM usage. But it's not a meaningful reply to what I said unless I am missing something.