GrapheneOS is currently defending its use of AI coding tools on Mastodon against complaints by various accounts claiming to be users.
We do not understand where you’re coming from or why you’re so incredibly angry with us. It’s not justified and does not make sense.
OSS projects have already started to recognize that LLM is just a tool and how you use that tool matters a lot more than whether you are using said tool at all. Lets bring StackOverflow for analogy: we didn’t ban browsers for code development because one can navigate to StackOverflow and copy-paste some really crappy code. LLMs are SO on steroids (actually trained on SO topics). So grabbing LLM to get a sense of direction or shape is an OK use in general, but copy-pasting it’s results blindly is silly and dangerous.
There’s a second argument that AI is ruining OSS communities which I believe is true, but refusing to accept AI code is a self-defeating action. Why? Because AI training needs unadulterated source of knowledge, one not tainted with AI and OSS communities rejecting AI would be that source. Also: if OSS community did not use AI it would hardly stop it from continuing it’s growth in Enterprise. So there’s not much of a winning strategy there either.
Which brings us to GrapheneOS decision and given above reasons I think it’s a well-balanced decision.
God. Motherfucking. Damnit. All. To. Hell.
Well capitalism is based on unethical practices so what’s the point of striving to be ethical!!!
Is what I’m reading
The thing about tools is they do what you expect. A fork picks up food. A hammer’s head doesn’t suddenly change shape or fly off. I don’t have to worry about my shoes suddenly falling apart.
AI cannot do these things, but encourages you to use them in such a way that you don’t have to oversee the output.
It is so much more work to rewrite something that has already been written poorly. It takes more time and effort than to just rewrite it from scratch. It’s not a tool as it cannot do its job. On top of that it’s just a glorified collage machine pulling code or whatever it wants from its database and smashing it together. The code is not novel or genuine.
All of that rework and it still destroys the environment and communities wherever their data centers are housed. Loud noise and now Americans don’t have the right to clean water. They’re pissing in your yard and telling you to drink it up.
Please have a spine. Or just some pride.
a hammer’s head doesn’t suddenly change shape or fly off
Actually… They do. Or did. That’s quite literally where we get the term “fly off the handle;” from hammers heads flying off.
You never used any software?
They are tools like any other. But bugs exist, and will cause it to perform in ways you don’t expect.
Your analogy is wank, is what I am saying.
They’re using AI models to find vulnerabilities. I assume the fixes are human written.
I’m anti-WhateverTheBillonairesArePushing. And by golly are they pushing AI.
You shouldn’t let billionaires control your life. They may not have your best interests at heart.
I think that’s his point
Yea, but by refusing to use anything pushed by billionaires you technically give them control. They can just start pushing anything and you would stop using it
Using tech that billionaires push makes them richer and gives them control.
Tell me how does not using their crap give them control, exactly?
It’s more of stupid technically. If a billionaire randomly decides that they want to push sliced bread really hard would you stop buying sliced bread? If so then the billionaire technically has some control over you, because they caused you to stop buying sliced bread
I mean, see where you’re going there. Bread may be a bad example because I can get sliced bread from a local or regional producer that is more conscientious. If any billionaire is pushing me to buy sliced bread its going to be some mass produced crap like Dempsters that I would never buy anyway because I do not support corporate food where I can avoid it as I’ve lived through enough mini crises in the last 30 years to have learned the hard way that when things get tough, its the local and regional creators that keep things on the shelf.
AI however, is only good in the older forms where models were made to do one thing and do it well with reasonable levels of compute. Those were sustanable. LLMs, no matter where you run them, cost disgusting amounts of compute to train and no matter how well you train them will only evey have a statistical chance of being right, and they will never have 100% accuracy without throwing even more power at them.
And, LLMs will never reach AGI. If we want something that really mimics humans, then what we need are many tiny, efficiently trained models for specific tasks with high accuracy, and then interface between all those skills with a very accurate task comprehension model, and then maybe we will get to AGI, because this is basically how humans work.
I still maintain that LLMs are the wrong tech, at the wrong time, and the only thing their proliferation is being pushed for is their questionable capability to enable ths surveillance state we were warned against for the last 40-50 years in scifi. And the CEOs want this because people are getting more and more fwd up with their shit.
All LLMs will bring the world is misery. There is no acceptable use for them and I am sticking to that line in every product decision these days. They were built on the backs of free content that the world created and a ton of copyrighted work that was just blatantly stolen.
Even a broken clock is right 2 times a day though.
Would you care about AI if it wasn’t pushed by billionaires?
GrapheneOS is using AI models to find vulnerabilities before others do. The Bitcoin Red Team did the same and found an absolute ton.
I hate the AI bubble, but they’re forced to use AI to defend against AI.
I’d argue that this news is over-blown. The project has more than a solid and honorable track record with insanely talented engineers behind it. AI is a new technology that is not going anywhere. I would put my faith into their decisions to improve this project by whatever means they think are best.
The truth is most commenters are not technically skilled to evaluate or critique their decisions. If GOS team, who again would be at the top % in the field, feel that this can elevate the project, should we not give them the benefit of the doubt?
AI is a new technology that is not going anywhere. I would put my faith into their decisions to improve this project by whatever means they think are best.
Aaaand just like that you disqualified yourself from having an opinion worth listening to on what “insanely talented engineers” are.
My thought it was over blown until I saw them claiming that anyone on Mastodon is “not a user” and them saying that any real users are on their Discord and not complaining.
Uh, OK - Discord was semi-excusable up until 2025, but at this point anyone using it to communicate with their users is certainly not considering how bad that channel is for user privacy. Im a Graphene User, I will not use Discord, and I am not going to continue to use an OS thats supporting the rapid expansion of datacenters through the use of AI.
I don’t use Discord myself, but I think their decision to use Discord is defensible even if I find it distasteful. Their support chat isn’t meant to be a private space. (People discussing crimes and whatnot on Discord are using poor judgement.)
I think what they are trying to say there is that there is a lot of negative commentary on the Fediverse coming from people who are not users, and that they are not really beholden to those commentators. And that’s true. They don’t really have a history of smooth PR as far as I can remember.
I also think whether their acceptance AI usage is an overblown issue or not is completely unrelated to that comment, so I’m not sure why it would change your opinion of it.
Matrix is the proper space for a privacy focused community. Discord as an application is privacy invasive and requires ID to be provided. This has nothing to do with a public community not being private. It’s the wrong technology choice for what they claim to be focused on and there is no getting around that.
AI is a new technology that is not going anywhere.
I always wonder where this sentiment comes from. Whether or not AI is going anywhere is still very much up in the air
Yeah, lots of new tech that vanished because it sucked, wasn’t profitable, destroyed the environment, etc.
It’s not going anywhere because it’s already been here for a very long time, you just didn’t know about it
What people call “AI” is just Machine Learning.
Chatbots, OCR, Text-to-Speech, Speech-to-Text, Computer Vision, Web Search, Spam filtering or Fraud detection, Personalized ADs, even your phone’s camera processing algorithm.
It’s all Machine Learning, but now under new and fancy “AI” name. And it’s been here for decades.
The reason it blew up this much only now is because people realized that the more data you feed into it, the better the output becomes. And so they started exploring it more, to the extent of causing valid ethical concerns.
Machine learning isn’t going anywhere. But it’s clear from the context here what we’re referring to when we say AI in this conversation.
Apparently, not to me. Explain.
At the very least, you’d have to convince both the US and China to stomp it out. And then there are other countries that would be willing to pick up the slack.
AI is not something self sustaining enough to necessarily require stomping out, at least that we know of at this point. Right now AI’s future relies on its operators continuing to convince people of a theoretical future scenario.
GOS is not part of a theoretical future.
You cannot put the evils back into Pandora’s Box when it was already open.
Even if these companies sink, individuals can still use their own hardware for AI. Ai is here to stay whether we like it or not.
So where are all the 3D TVs? Where’s the Nintendo Virtual Boy? Betamax? Google Wave? Google+? NFTs?
Oh not transformational enough?
Ok then how long did it take electric vehicles to come back after The fossil fuel industry killed them in like 1920?
Oh here’s a good one: thalidomide. That one went away for good.
How about the hole in the ozone layer?
Sorry, AI is only a foregone conclusion because apparently everyone’s had their consent beautifully manufactured to an extreme degree.
Sure you can. Because in this case Pandora’s box is something which exists solely through a sustained, inconceivable amount of money being continuously funneled into it without any guaranteed long term benefit to the people who do so. It’s a technology which isn’t sold as a product as it exists in its current form, but a theoretical future form.
If the bottom were to fall out from under it tomorrow, it couldn’t even continue to exist as it is now, it would disappear entirely as there would be no financial incentive to keep it running. Individuals could still run what we have now offline, but its value and usefulness would degrade rapidly as time went on, because it relies on mass-scale power consumption and data theft to remain up to date, much less improve.
I don’t need to be skilled to evaluate or critique their decisions.
AI is the stolen collective labor from all of humanity, thrown into a blender with magic mushrooms and LSD, then locked behind a paywall and operating out of massive datacenters, thats destroying communities and the planet and undoing 40 years of efficiency and environmental gains.
I dont need to be a skilled engineer, a brilliant software developer, or anything else, to know thats bad, and that every single cunt piece of shit that enables it, defends it, and worst of all, uses it are, at the very most positive outlook at it possible, thieves who are willing to sacrifice actual human life, just so they can generate a random emulsion of coding (in this case, or in other cases generate images of nude children, celebrities, your mother, or tell people on the bring of psychosis that they are actually jesus, or any of the other vile evil shit that it does) that might function, at the expense of actual human labor and well being.
And every single motherfucking one of them that make that decision deserve convicted, and to be hung from a tree.
Calling for the murder of other people in a public forum is never wise. At best, you could (should) be permanently banned. At worst, police could show up at your house and lock you away forever (the world would be better place for it!).
“Wah wah wah, someone called for criminal convictions and liability for this criminal behavior. IMMA RUN TO MOMMY MOD AND MAKE THEM DELETE IT”
So, username checks out…
¯\_(ツ)_/¯
That’s the communication problem in these conversations. Whether it’s a useful tool and whether it’s an ethical tool are two different questions. People come in and attack the developers for generating slop, and they respond that they are experienced developers who can use the tool properly to increase the security of the code or whatever. And the conversation here is about that.
And then someone like you comes in talking about ethics, which is what we should be talking about, but we are talking past each other.
The truth is most commenters are not technically skilled to evaluate or critique their decisions.
I don’t need to be skilled to evaluate or critique their decisions.
You are both correct. But the former is about the engineering, the latter is about ethics. You do need to technically skilled to evaluate whether AI is a useful tool for Graphene. You don’t need to in order to critique it on ethical grounds. These are completely different topics that have nothing to do with each other.
no, we are not talking about two different things.
We are talking about one single thing.
the theft of human potential for profit, and the sacrifice of the world for the enrichment of the few due to exploiting that theft.
You are just trying to play word games and split the discussion to defend the indefensible.
Lol you are so argumentative that this is how you respond to someone agreeing with you? Forcibly putting me into the “opposing camp” against my will. All right then, live your delusion while we fight AI without you.
Exactly this. As much as I hate the current AI scenario across the board, the GOS team has shown, time and again, that they are stellar developers and coders, that they genuinely care about security, and that they are the best at countering anti-privacy on the technology front.
If any team can use AI responsibly to improve all this, it’s the GOS team. And if it ends up failing on it’s face, I expect them to move away from it if things go south. Now, if they don’t because of ‘AI addiction’, then I will be looking at what my alternative options are. Until them, I’m 100% with them, as they have not given me any reason yet to doubt their judgement.
A bit unrelated, but I wonder if I’m the only one left that is:
- Using small, open-weight models
- Asking it to only write small snippets that I can easily go through, check and understand.
- Copying pasting from good old web frontends instead of integrating it into my editor or whatever else people do with agents nowadays.
I feel like when we’re talking about AI this sort of usage gets lumped into the same pile as agentic vibe coding and it’s not helpful for discussion. Surely most people wouldn’t find this problematic?
No, there are plenty on the left doing exactly that. Plenty don’t mention it because there’s been so much AI pushback they don’t want to be written off, others don’t mention it because they know it doesn’t really contribute to the conversation most times. It’s unlikely that your use is reflective of most AI use in these large projects.
Realistically, even your use doesn’t address some people’s issues. People dislike that the training data was used without consent, that’s not really addressed by open weight models. You checking and understanding the output is great, but a lot of people take issue with having offloaded that to the AI in the first place. Maybe your use doesn’t have the same negative environmental impact as some of the other models, but I’m not sure it really matters. If we’re using AI then inevitably some outputs will have larger environmental impact than others.
I am not saying your AI use is the problem with humanity or anything, but no, you’re not alone. Most people just aren’t talking about people like you and your specific situation doesn’t actually fully address people’s issues with AI regardless.
I don’t get the copy pasting. That’s just a convenience thing. You can easily integrate local models into an IDE.
I prefer it less for the writing part (unless big refactors that aren’t actually complex) and more for analysis. If you got a couple million lines of code upstream that you depend on heavily (ERP modules) then it’s nice to fire off a prompt and come back to “you got weird behavior x because upstream changed y, here’s the exact file and lines and how it flows there”
Probably.
My company is currently pushing “loop engineering” on us… They are in love with Claude and keeps referring to it as “him”. Very proudly announce that “I set ut a loop, and he closed 40 PRs while I slept!” (without review, of course). Based on vibe written tickets as well, which contains many hallucinations.
The poor QA department (2 people) is now trying to figure out how to test all the slop.
Things like this radicalize people against all AI.
We usually say “the other guy”, or sometimes we say “I & AI” (a poor twst on a Rastafarian saying)
Sooner than later most things the average person does will sort of be vibe coded like compiling to an intermediate language. The idea of apps and walled gardens will not be as important. Instead of approved apps, it will be approved models and agents. Even the idea of an operating system is going to change. Linux is a great space for integrating the computing environment with AI coding. The full source is available to the model and any change you ask it to make can just be done for you. The idea of running a “Linux” or “Windows” could be limited to certain software and domains.
I think that certain things will be fairly resistant to forking, even by AI. Foundational software like the kernel, compilers, basic tools, databases, and so on will likely benefit more from being used as the standard/vanilla version rather than a modded version (especially a vibe-modded one).
But people and companies will experiment with this for sure and it wouldn’t surprise me if some of the less software-competent hardware companies start making their Linux forks or drivers with LLMs, at least for a couple of years in the near future.
I also think that this will shift lots of companies further away from GPL toward MIT/BSD, because they’ll want a fully moddable source but they won’t want to share what they’re vibecoding.
There really is a difference between AI-assisted coding and vibe coding, and I don’t know enough about actually coding to look at their repos and tell you what’s happening.
“AI-assisted” will inevitably turn into AI-generated over time as the babysitters get complacent.
The issue is, there’s also difference between how people cram “AI-assistance” into their previously well-refined workflows. Did see the odd person ending up with the same issues as straight-out vibe coding but without the speed, including the eventual (mild) AI psychosis and later switching to vibe coding after the chatbot convincing him.
Also I think it would be pretty trivial to generate an “understanding” of the code, yet AI boosters never bring up it as a concern.
Reading the comments, wow AI is polarizing. Neither side will budge.
I guess only time will tell, to see if all the AI code will actually come crashing down.
Non AI code has been crashing down for decades, controlled chaos, continuous improvement. We have lots of tools to put metrics on that, and AI assisted code is fairly well crushing those metrics. It’s not 100% self sufficient code creation Nirvana, but it might be fairly compared to the Cotton Gin in terms of making the field workers more productive in the delivery of market ready code/cotton.
but it might be fairly compared to the Cotton Gin in terms of making the field workers more productive in the delivery of market ready code/cotton.
Last i heard was that developers FELT like it made them more productive but actually not so much.
Last i heard was that developers FELT
Several of those studies were published on early 2025 data, LLM agents have been rapidly improving since then. In early 2025 I would agree with the sentiment - it was kind of equivocal for overall productiveness. By early 2026 I’d say it was a much clearer advantage, both in doing the same kind of work faster, but more importantly: in doing better work at the same speed or even a little faster.
Non AI code has been crashing down for decades,
i.e. the code that made literally everything until a few years ago? That was crashing?
the code that made literally everything until a few years ago? That was crashing?
Quite a bit more than it should.
Yes, which means most every code crashes at some point. Whether it is made by humans or not.
Yes
We have lots of tools to put metrics on that, and AI assisted code is fairly well crushing those metrics.
You mean more lines of code?
Not all metrics are good. LOC is about as bad as they get.
Bugs per merge is a much better one.
Are you trying to make a joke or a serious contribution
If it was about real AI – not stochastic parrots – I would be all for it. What happened to those developments that started in the 70s where you have automatic formal proofs that a piece of code does what it is supposed to do? That is intelligent. Producing slop more voluminous and convoluted than humans could produce or understand is not intelligent software architecture just like fixing a hole in a roof by throwing what sticks at it is not intelligent building architecture. Both may do the job for some time. But what are the long-term costs and who is capable to clean up? In my opinion, what’s most desired is tools that are bug free, do what they should without annoying the user, and are radically simplified in their architecture. (For example, I would be happy with a perfectly working product that does what NFS or Samba aspire and fail to do for decades.) Apparently, AI can’t deliver that. To have the billionth app that is a browser wrapper and shows me exactly one website is much more important.
It’s called symbolic AI, or sometimes GOFAI for good old-fashioned AI. It’s no longer considered AI by the wider public, it seems. I believe those techniques are still widely used, but they have reached their limit quite some time ago. Cyc was maybe the high water mark of that approach. https://en.wikipedia.org/wiki/Cyc
What happened to those developments that started in the 70s where you have automatic formal proofs that a piece of code does what it is supposed to do?
The stochastic parrots we have today are pretty good at doing that kind of tedious, extensive, franky mind numbingly boring stuff. They’re also fast enough that you can “afford” to take the time / effort required to do those kinds of proofs where they are warranted - it makes them “a good idea” in a lot more places than they used to be when they were so much more expensive to perform.
modern AI can be used to generate formal lean proofs:
There is no making ai go away, it’s never going away. Now it’s a tool to divide us.
People still seem to think AI is a fad that will go away. I don’t see this happening (for code) at all. Not ever
Like every other recent bubble, the output is not monolithic and uniform. Like all before it, all the stupid/silly/wasteful use cases will dry up, and something will remain, but nothing like what we see at the apex of the hype cycle.
See the dot-com and crypto bubbles for the most directly applicable examples. Both yielded useful technological advancements still in use today, but those are only a fraction of a speck of a ghost of the use cases being touted at their peak.
People don’t really understand this yet. There are LLMs that can do useful work running locally on 5 year old laptops (ok, only if it’s a Macbook Pro, but still). Qwen on consumer hardware isn’t close to Claude Code using cloud models, but it can do real work and this is the floor, the absolute minimum of AI we will have going forward. No matter what happens to the industry and the bubble, it will not go away. With any other outcome other than a total collapse of the bubble, AI will be more powerful than that.
If they’re just using it to find and patch bugs/exploits? Fine by me.
Vibe coding? Get the fuck outta here.
If Vibe coding means people who don’t know how to code creating code… yeah, that’s kind of like giving people who don’t know how to paint an airless gun and 5 gallon bucket and telling 'em to go for it on this building here…
Even people who can code tend get worse over time the more and more they rely on “AI” to avoid having to think too hard.
Studies have shown doctors who rely heavily on “AI” are similarly affected.
People who rely too heavily on anything get brain atrophy with net negative consequence.
Those same tools can be used for learning and exploration and strengthening skills - all a matter of how they’re used.
This is the key point.
From my reading that is not the limit in how they are using AI. It seems like they are accepting LLM generated code as inputs.
This pretty much compromises the whole project as far as I am concerned.
Based on my understanding, the “core” team uses AI to detect vulnerabilities.
They also take codes from paid third party devs, and they say they don’t mind how they wrote the code, but it must be clear and each contributor needs to be able to explain it.
They follow the same rules on code conventions including readability. So basically, it’s less that they accept it than they don’t care as long as it matches their quality expectations. If you can’t tell, you can’t ban…
Considering Graphene’s entire reputation is staked on extremely precise definitions of security, this is a little concerning.
They’re the best. If the best degrades, I don’t foresee some Graphene alternative stepping up very quickly (see also: cross-platform alternatives to Google-made Chromium and Google-funded Gecko)
is the linux kernel “the best”? because they are using AI tooling. Linus actually threatened to use LLM to maintain sparse if nobody stepped up…
is curl the best? because the creator of curl has also used/accepted LLM code…
if the creators of two of “the best” pieces of software ever made are accepting (NOT embracing or promoting… just accepting) LLM code… maybe it’s time for us to think about the possibility that this MIGHT be a useful tool instead of everyone just yelling “SLOP SLOP SLOP” everywhere…
People trusted Linux to be better than Microsoft Windows, and saw the effect of a company with tons of money and first-class AI access exponentially worsening their product.
It has been disappointing to see people like Linus Torvalds turn aggressive against people who do not agree with his ideology, and I find it distasteful for you to use reductive attacks against strawmen as well.
Where is he aggressive? What ideology?
The only ideology linus follows is “dont break userspace”
Someone commented on here the other day that they don’t use anything that uses AI. I had to point out that they were on Lemmy, in a contemporary web browser, connecting to a Linux server running inside of a docker container. I’m still not sure if they understood my point, but I’m glad someone else on here is seeing the ripple effect.
If it’s thoroughly vetted, it might still be OK, but it’s concerning…
@bekkaboo@girlcock.club Linux itself is heavily using AI models.
For security analysis and uncovering bugs.
