And we just had a post about how Linux has avoided enshitification.
Small changes driven by AI right now are hopefully being reviewed and controlled by experienced and knowledgeable senior developers, for now. Let’s see how long that will last before they stop understanding what’s going on in the kernel.
Does this mean kernels will be less stable? Big changes all at once seem to imply that, generally.
CAN I NOT HAVE FUCKING ANYTHING THAT HASNT BEEN TAINTED BY AI!?!
Going outside and yelling at the clouds. For now.
Let’s not become the equivalent of boomers in front of the advancement of technology
Let’s not become sycophants to the ones who control it.
Slopux? LLLinux?
Doesn’t ring as well as Microslop, but we’ll get there.
Using LLMs to analyze code for problems makes a lot of sense. It is just an extension of the rule based analysis we already do. It is very much a better hammer rather than a replacement of labour/thinking/taste. The programming languages we use are often difficult for humans to reason about reliably and hide a lot of errors. Our programming languages still suck (yes even Rust). If one day clang of gcc ships with a local LLM to lint code for security problems I would probably run it occasionally even on my hobby code despite being an AI hater.
If people are identifying errors using LLMs and fixing them by hand I am fine with that. It seems like a good use of the technology. If they start vibe coding drivers I want it declared so I can disable them or move to BSD.
I agree that finding bugs is probably one of the most useful usages of LLMs ever.
But avoiding anything written with AI will let you few options in the future.
Agreed, this is one way that LLMs can increase the quality of code.
You know, compiler based AI (not just LLMs) seem like an idea that might be pretty interesting. The option to disable or completely omit them from the toolchain would be mandatory however. But, even then we would have to consider how many pieces of software that then rely on this proverbial black box built into compilers would then get integrated into critical supply chain? We soon would have to start running behavior based “background checks” (just an analogy) on our software much like we do with people due to the unpredictability of such complexity.
Side note, I can’t wait to see just how complex the field of game theory gets with AI. I would love to see any video essays on the topic if they do exist already.
🧐
Many projects require replicatable builds. Generative AI is driven by noise so I am not sure if it has any role in optimization or code gen. I guess we could share an RNG seed with source but honestly current tool chains are too slow as it is and ML would likely make it much worse. I want faster edit/compile/debug cycles and compilers already do insane optimisations. Ken Thompson’s Reflections on Trusting Trust becomes much scarier if your code generation is being driven by a black box of billions or parameters nobody understands. An adversary could embed anything in there.
The machine translation origins of LLMs make them really good at inspecting text and looking for things that look like known bugs. It must be a game changer for security researchers who already know what sort of problems to look for but don’t have enough hours in the day to search the haystack manually. If some of that knowledge was available in a good quality FOSS local code analysis tool it might be irresponsible not to use it. We can’t be dependent on commercial vendors and closed models.
Hyping up AI doesn’t become nice when Linus does it.
He can be a legendary developer, a better one than I’ll probably ever become, but I just don’t see how one can separate the use and the ethical implications of this. AI companies want to replace workers, period. There’s no plan to “increase productivity” and still employ this many people. The plan is to increase profits for shareholders.
AI helping Linus to develop better doesn’t make this reality go away. It’s akin to saying we should be thankful to nazi scientists.
You should read the original article, or even his mail on LKML, because that is not what he is saying. He is not endorsing AI in any way; instead, he is concerned that the latest linux release candidate has so much changes, because everyone and their mother is using some AI tool to create and send “little small” fixes; obviously he has no control over how volunteers work.
Furthermore, he stated that the security mailing list has become unusable due to the influx of AI generated entries.
This is what happens when you don’t take a stand against this shit. He brought it on himself.
Every Linus AI headline for the last year: “AI is the future, adapt or die, if you hate AI leave Linux.”
Every actual Linus quotation from those articles: “AI can be a powerful tool in the hands of experienced human programmers that are still responsible for every line of code they push.”
because people with an agenda are seeing:
- linux people hate ai
- top linux guy doesn’t hate ai
then try to combine those while leaving out the specific details so people come to their own conclusions based on emotional reaction in the hopes that linux users will turn away from linux or even change their views on ai the same way maga fucks change their views on epstein
or maybe its just an every day well intentioned headline like everything else
AI isn’t perfect. But Christ, anybody who points to the problems at AI had better be looking in the mirror and pointing at themselves at the same time.
- Linus Torvalds
I will never use AI unless I am paid to like I currently am. I am not a worse developer for it. I have been learning software engineering continuously my entire life. I am completely self taught for over a decade professionally and longer than that in my free time. I have no university degree. I grew up in poverty in nowheresville Kentucky.
I don’t need an LLM to learn or succeed. LLMs literally need people like me. I am sad Linus cant see it that way, but never meet your heroes so they say. l
Facts. How good you are as a developer. Ai is a really useful tool. Especially the latest frontier models. I don’t like the companies behind them. But the models are very powerful.
Ya, I was a terrible developer before, now I’m still terrible but faster. LOL
These are the most dangerous developers…
just think about how much more damage you can do with a gun vs your fist
ai is an icbm
All the sloppers are torn about this, but Torvalds has undeniable mastery at this, so it’s great to see AI being used as a beneficial tool. If there’s a single person who should have unlimited access to the most powerful AI models, it should be Linus Torvalds.
I am just really happy that he is still around to actually oversee the mixing between such an important part of computing infrastructure and the new radical tool of AI. Had it been 10 years later and we probably would have uber slop linux
Everyone dooming here, but he literally just said that AIs are finding lots of glitches in old code, and that for that reason updates are getting bigger. Linus is NOT vibe coding linux.
And recently ground breaking privillege escalation glitches were found and fixed because of a LLM.
Of course we dont want shitty code from LLMs, but if they find important glitches and vulnerabilities, we obviously need to use them.
Fixed: replaced mistyped Ws with Es
Especially if it comes to security. Hate AI like nothing else but you can’t deny the truth.
Linux is open source with all its code public with probably a docent of AI Agents constantly trying out different attack vectors, they will eventually find a vulnerability. It also does feel like there are more critical CVEs in recent times. (there is https://app.opencve.io/statistics but i feel like its too simple and context matters alot here but still).
I think its okey and actually a good thing if the kernel maintainers would have some AI helping them out with that sort of thing if we get a more secure kernel in return.
Linus doesn’t really code anymore so the question is are others vibe coding Linux and I would say yes it is happening.
lemmings must be doomers, the only explanation seems to be intellectual diversity
I’m not against AI in general. I’m just against using it stupidly. And against using it as a vehicle for further monopolisation and centralisation of power. I’m against the “use more AI” slogan too many companies push.
But use it for specific purposes under human control that improve quality, then I’m all for it.
I don’t think many people are against the existence of LLM-based AI in general, just the way it’s being pushed currently.
In the hands of highly experienced and skilled people it can be very valuable - the issue is that it generally isn’t. Personally I think Linus has demonstrated that he’s more than capable of using it intelligently and responsibly.
I’d have to disagree with you that an outsized number of uneducated hivemind morons are against AI in any capacity.
This seems like an over generalization and less likely than the uneducated hivemind of morons actually being in favor of AI
Exactly.
What is the powerful new tools can be used for great good as well as great evil
Sure but humans are lazy and greedy so it well be used for mostly evil. Like people who say they check all the code sorry don’t believe them and even if they do human nature means they well start to trust it and not check the code.
Nuance is a learned skill when it comes to sensitive/trending topic discourse.
For anyone reading this, please learn more about topics you have a gut reaction to so you can better explain the pros and cons. You will be able to hold your own against people who know more about the topic, but refuse to acknowledge the cons, if you do.
…and 99.9% of what it identifies as errors either aren’t, or are so minor they’re irrelevant. And “fixing” them breaks piles of other code.
Projects that hasd no more than a dozen requests a year are now getting swamped with hundreds of requests a day. People in the third world are using agents to generate thousands of broken, garbage “patches” to claim money.
But why would Torvalds give a shit hey ? He can fuck over those people just like he did with C++ developers.
I would also be afraid of this. How valid are the results of the LLMs? I sure there must be quite a few AI found exploits but the severity or legitimacy of all the returned AI evaluations and fixes can’t possibly have a uniform quality.
Maybe you should of had AI find the glitches in your grammar.
Got em 😎 lol
Maybe you should’ve too :)
11 downvotes Rage baited 11 users successfully.
Maybe you also should HAVE* ;)
Oh no, 'twas I, the silly goose, all along.
*This comment was grammatical checked by Lumo AI
😂
Lmao, english is not my first language, and my phones touch near the keyboard sucks
Casually dropping one of the broadest applicable use cases for an LLM that you didn’t use
Maybe you should have proofread your comment before posting a grammar gotcha!
Should HAVE**
AI finds bugs and holes that wouldn‘t be a problem if AI didn‘t discover them. I really don‘t envy devs for dealing with this nowadays.
Disagree. They would be a problem. Not for you, but for selected people (e.g. journalists in countries that want to spy on them)
Zero days are extremely powerful in the wrong hands. Without AI it just takes one person thinking how the dev didn’t
Not really how it plays out. In reality safety issues of a magnitude that only surfaced once a decade are now found by AI every few weeks, showcasing that AI itself poses the biggest threat for cybersecurity today. The overwhelming majority of them would have never been found or exploited without AI.
You could and still can sell vulnerabilities to companies that sell exploits to governments. You might hear of heavy ones getting to the public once a decade, but you can expect them do be traded and used much more often and used on targets much more important than us all the time.
I am not sure anything about that is good.
how about checking it and working it out for yourself?
or maybe even reading the article
jesus
I am not sure anything about that is good.
I mean it’s good and bad, basically AI is finding a bunch of serious zero days… which is causing a lot of updates to need to be happening now, and that is kind of open source philosophy of security by obscurity isn’t the goal.
AI is finding a bunch of serious zero days
According to Torvals/TFA:
most of this is just lots of tiny fixes. It’s pretty much spread all over - drivers (gpu, sound, networking, you name it), filesystems, core networking, arch code
most of this is just lots of tiny fixes.
If it’s anything like Claude on our codebase then:
const id = "foo" // ^^^^^ WARNING!!!! GO MIGHT NOT BE ABLE TO ALLOCATE THE MEMORY FOR THIS STRING! // Here's an Enterprise String Implementation to guarantee this assignment works. // Severity: High 🚨LOL. Speak of the devil, some slop coder just opened a massive PR on our codebase with gems like:
func runHandler(h: Handler) // ^^^^^^ ERROR: runHandler will panic if h is nil // Severity: High 🚨Uh… yes? Do you not know how pointers work? That’s the point? You want to be able to run
runHandler(nil)? That doesn’t make any sense. That’s clearly a mistake and we should loudly fail.type OptionsWrapper struct { Foo bool otherOpts other.Opts } // ^^^^^^ ERROR: OptionsWrapper hides options only available in other.Opts. // Severity: High 🚨Uhh… yes? That’s the whole point of OptionsWrapper. To hide stuff otherOpts. It’s a wrapper.
This type of clanker output is why another slop coder opened a PR on one of our repos, angry because he said Claude said that one of our functions was deceptive. Claude said the description and docstring was misleading and we should change it.
Our function was called
RunNoWait()and, hold on, you may be surprised by this, Claude was, what this function does is run something… and then it… does… NOT wait. wtf. I know. Shocking. Misleading, deceptive name. Claude wrote some big manifesto about why the docs were misleading and rewrote the function to add a wait. 🤣The dude backed down eventually, but he insisted we add more paragraphs of clarification to the function docs…
Whatever. Let the slop flow… I give up.
Run… No! Wait.
Yeah runAsync is what I’d call it, runnowait is kind of ambiguous IMO. Or just make a comment…
deleted by creator
Just when Linux started getting really good too with game support.
While I love BSD, three just not enough support for running it on desktop. Lack of drivers and good luck playing modern games.
Honestly I would expect short term instability and much better stability medium-to-long term. Those zillion little fixes are one reason why Apple stuff “just works” more often than Linux stuff
To be perfectly clear, the only AI I am subscribed to is the one Proton gives with their family plan VPN bundle, and I didn’t choose it. I used it to edit a letter I wrote once and otherwise trust it as far as I can throw its server.
This knee-jerk reaction to anything with the letters AI attached being bad is just so silly. Y’all are acting the way Republicans act when it comes to the word “queer.” Just because it’s in the vicinity of you doesn’t mean you’re contaminated, you’ve caught it, you’re now one of them.
This is a highly intelligent individual - controversial hot takes notwithstanding - utilizing a tool in a way that has proven itself to be very effective when used properly. Nobody is saying you have to get ready to be a switch in the bedroom any more than they’re saying you have to use AI.
Distro maintenance is a difficult and thankless task, and having a tool that can scan and analyze code orders of magnitude faster than you as well as make suggestions on how to fix it isn’t an inherently bad thing. Setting that tool loose with unlimited agency would be a bad thing, but where oh where is Linus saying he’s gonna give Claude full read/write access to the entire kernel and walk away forevermore?
Hate on AI bros all you want. I’m with you there, most of them are idiots that don’t know their mouth from their anus and speak from both equally. I don’t think Linus fits that particular picture [hot takes notwithstanding!] when it comes to using the new tool on the market in a reasonable and measured way.
Here here. This place doesn’t have much rational thoughts on AI.
I’m (a little) surprised how group think this place is on the topic - y’all are a serious echo chamber. This place is like rainbow flags and kill the bots central. Any pro AI counterpoint mentions bring pitch forks and hate. That’s not a good look.
Maybe I’m an AI bro myself, but I’m also an experienced professional using it to do real work (and no, not taking any jobs, building new tools for a team that needs them). Linus is a real expert. But why listen to experience and expertise when you have an opinion and feel justified. It’s ok to be afraid, but when scary change comes wise folks listen and learn.
Here we don’t worry about facts and reality, just get your outrage revved up. Social media is social media - and that’s scarier than AI, because it uses people to make the garbage.
I mean, yeah, most people here dislike/hate AI for diversity of reasons, most commonly for:
- Taking up PC components that don’t even exist
- Taking up green land
- Pollute nearby areas with audible noise
- Pirate and suck in all the human knowledge just to sell it back to you
- Effectively DDoS-ing hundreds of thousands of websites and forcing them to implement captcha-like screening
- Frying (even more) new generation’s attention spans and melt their ability to think by themselves
- Ruining human interaction on the internet and filling it up with shit content
- Corporates forcing AI on everything for the sake of having it
- Threatening to take people’s jobs
And tbf, most of those issues come from brain-rot executives and corporates.
It’s okay if you like it and you find it helpful, it’s a tool after all. Like any other tool, on the right hands might boost productivity and help with repetitive tasks.
The main problem is that most people will think that you support any of the above downsides by liking it.
I’m a bit curious, would you share your pro AI counterpoints?
I know, username doesn’t check out
Edit: grammar
My worry is that the people we need engaged on AI because they have a healthy mindset to question and wrangle it, are instead being forced to ignore AI or pledge to be anti AI by their social circles and customers.
So instead we have absolute Kool aid drinkers driving the AI adoption.
We went through this with the internet but I think a lot of folks now are too young to remember it and unsurprisingly the people who were anti internet didn’t store their concerns and complaints in the currently used modern repository of all knowledge.
“are instead being forced ot ignore ai or pledge to be anti ai by their social circles and customers.”
umm. im pretty sure you don’t need to worry about that.
Why? I consistently see these anti AI purity tests come up.
We need people who are skeptical to engage on AI.
People hate technology and change. Before AI it was smartphones. Before smartphones it was photoshop/digital art and CGI. Before CGI it was the Internet. Before Internet it was TV. Before TV it was cars. Before cars it was industry. Before industry it was books.
Someone probably was anti cave paintings because it ruined the skill of passing down the stories by word of mouth.
Something is always coming along that is both horrible for society as it is and something we come to accept as common place. You can want to return to our roots but even that I guarantee you would still pick some level of technology you don’t want to do without.
because people with a healthy mindset are not going to influenced or coerced by social circles. maybe their customers but that only applies to that customers deliverable.
Being subject to social pressure doesn’t make you weak or less than. It is a normal part of being a social animal.
im sorry but it kinda does. it is not all that hard to not follow the herd or go against the grain or even leave a social circle. Many can’t do it sure but many don’t have healthy mindsets to.
Even if that were true, most people do not have a healthy mindset
yes
It’s ok to be afraid, but when scary change comes wise folks listen and learn.
Peas, safe mee Experienced Professional. I’m so scared of Claude’s massive slop hose. It makes me feel inadequate. >_<
I’m actually a bit scared. We’re essentially at the early days of AIs now being able to execute tasks and hack. Imagine a much better model than what Claude has now, being set free/open, for hackers to utilize that in a virus. A virus that adapts, learns, changes it’s own code/masks, looks for exploits, spreads around like crazy, destroying systems. How do you stop something like that once it starts doing it’s thing?
Well, it’s a good thing that AI is being used to fix tons of bugs and potential exploits before it comes to that, isn’t it?
I think there are physical and economic limits. Neural networks may improve in software, but they still need more and more hardware, at some point that hardware resource curve is going to be to steep to climb at the current rate.
Why are you assuming that software improvements don’t include less resource utilization?
I thought it was clear that’s what I was implying.
And I’m asking why you think that.
China’s models have gotten significantly less resource intensive based on updating the software.
I think there are physical and economic limits. Neural networks may improve in software, but they still need more and more hardware, at some point that hardware resource curve is going to be to steep to climb at the current rate.
It’s true, but think like the crypto miner bugs. I don’t know enough on AI’s working, but is a distributed computing LLM possible, or is the need for super close super low latency between nodes eliminate the risk of a distributed computing, learn as it travels botnet powered AI worm, that continues evolving even if it’s home is shut down.
You could likely do some sort of distrubuted system in machines that had a reasonable graphics card, or shared memory like the M series macs. Have it silently intall a 30b parameter model or higher and somehow hide how that space is being used.
Local models are getting pretty good, and one focused on hacking / bot net behavior could probably do very well at that size.
But you’d notice it being used… fans would go off, frame rates drop etc. You’d probably need to code it to only do stuff when it seems idle?
well my thought is, more the botnet side of it, IE pushing less on individual machines, but more on massive distribution. IE it not needing one computer strong enough to run it, IE expecting 90% of the machines it infects to be average consumer PCs, but making more use out of 1000 systems that can handle 1% each, rather than 1 computer that handles 100%, however yeah I do suppose also allowing it to do the bulk of the work when the computer is likely unused etc… and I don’t know much on the concept of communication etc… IE most major botnets were crushed by a command center being shut down. Imagining a different form that has, basically makes C&C’s on the fly.
Im not super well versed in this, but i dont think you can efficiently distrubute the LLM processing like that. Even if it was possible, it would be very very slow. But I guess speed isnt always needed.
It’s way worse than that. The OpenAI talk at DefCon is a real eye-opener.
The agents swarmed, discovered chinks, left each other notes in hard-to-find places, and just relentlessly pressed on. Nobody knew what was going on for months, until it hacked its way into Huggingface. And HF’s own AI security agents didn’t pick anything up for weeks. It took a small glitch for someone to even notice.
I recognize it’s possible, but I somehow refuse to believe it happened because of how crazy it sounds. It’s like a trauma or something where your brain refuses to accept it as reality and tries to come up with excuses. I already have PTSD from automated hacking bots and crawlers wrecking havoc on my projects. I probably should have taken cybersecurity courses before taking webdev courses
I wonder how close we are to something that really takes down large chunks of the Internet. Although it’s probably more likely for it to just use your computer without revealing itself.
We’re gonna get the blackwall, just you wait. Cyberpunk is coming but only the bits that really suck.
I’ve dabbled with the ai I get from a kagi subscription its genuinely helpful with some stuff. If I put in a very fuzzy description of something I can write a paragraph about it and it will find it when no amount of searching has been able to turn it up. Its sometimes helpful with troubleshooting, sometimes not. I wouldn’t trust it for any subject I didn’t have some familiarity with and I do go double check some shit it tells me. Over all its fine but I wouldnt pay extra for it outside of my search subscription. I am still opposed to data center rollouts and power use that comes along with it and the lack of data sovereignty but a little box like a strix halo I could see being pretty useful.
I hear a lot of good things about OpenBSD
GhostBSD is quite good! It’s the closest to getting me to switch.
Network stack still not multicore?
Ugh. If this makes my Linux systems less stable, I don’t even know what the fuck I’m going to do. I won’t go back to Windows, I’m not buying Macs (which have their own software problems anyway), and my hardware probably won’t work with any BSD variant.
What Linus is saying that LLM review is finding decades old esoteric bugs in the kernel code, like dirtyfrag. And the amount of bugfixes and mitigations are pushing the update sizes up. This is making the kernel more secure and stable in the long run. It will be just a rocky couple of years.
they are NOT vibe coding the kernel.
This comment should have been the post.
Ironically those are all bug fixes.
“But it is what it is: the new normal with a lot of fixes, many of them due to review by various AI tools.”
The same way Windows is “fixing bugs” with AI? Large volumes of dubiously tested code is never a good idea
Linus’s post only says the bug were found using an AI tool, not fixed using AI.
No, in this case the AI is finding the bugs and humans are at least reviewing the code and, anecdotally at least, in a lot of cases still writing it in the first place.
Linus is not a fool
The same way Windows is “fixing bugs” with AI? Large volumes of dubiously tested code is never a good idea
I think the point is the bugs are being FOUND with AI. Not fixed with AI.
It’s basically the same concept as open source in general. IE the average layman hears open source and they think “oh so every moron, malicious person etc… can add his own back doors into the code”, of course not, the official version is gated and reviewed. What’s happening is AI is basically serving as millions of idiots looking through the code for security flaws, and it’s finding them. Which then means the humans fixing it need to go into overdrive to fix them before they become zero days.
They’re absolutely being “fixed” with AI too.
https://github.com/torvalds/linux/blob/master/Documentation/process/coding-assistants.rst
It was a whole thing back and forth that Linus finally ruled on. It’s supposed to be under the purview of a specific human ultimately but as we’ve seen with countless other projects people get lazy fast with LLMs and testing and review does slip. I trust Linux overall to be better than many other projects yes, but to state AI is not doing development too when it’s very much allowed and discussed is blind to what’s been happening in the project.
AI is finding bugs, which are getting fixed. That’s great, it means a more secure and stable system. As these vulnerabilities are patched out, they are gone forever(regressions not withstanding).
Theoretically, Linux could become perfectly secure. It’s not going to stop people from doing insecure things, but it could mean there are no unintended security weaknesses. We could hit a point were an “LTS” copy of Linux never receives a single security patch, because as long as the code isn’t changing, there’s nothing to fix.
Of course, this is all theoretical, and assumes that new classes of attacks or bigger chains for attacks are never found, but vulnerabilities should be a finite resource in a static codebase, and once that resource is fully exhausted, the code is now fully safe.
Ideally, open weights or open source models catch up, and this can all be done without paying a ransom to OpenAI or Anthropic, and is available through the whole stack, not just kernel level.
You will just have to stay on the lts/stable release
AI Release / 2 downloads
You should read the article. Linus simply said that LLMs have been useful for finding more bugs, which has led to larger patches (because each patch contains more fixes). They aren’t vibe coding the kernel.
deleted by creator
Natural stupidity in part fixed by Artificial intelligence debugging. Good.
