It seems Linus’s position is very clear now
My issue isn’t with using AI to work on the Linux Kernel. My issue is working on the kernel with proprietary AI which enables Sam Altman to claim Millennium Prizes.
I completely agree, using an open model would be far more inline with the spirit of open source.
“With great power comes great responsibility.”
I actually do think that someone who already has multiple decades of experience in programming without using AI is probably not producing shoddier code after incorporating it into their workflow. Important modern software projects, whose backbones are programmers fitting that description, aren’t going to be worse in the present because of Claude usage. I remain strongly against it though, because normalizing its usage guarantees the future won’t have programmers that fit that description running the shows.
I think this can be framed as a statement about generative AI in general: If you think AI has been making everything shittier today, just wait until you see what it’s been doing to everything tomorrow.
I actually do think that someone who already has multiple decades of experience in programming without using AI is probably not producing shoddier code after incorporating it into their workflow
While it’s not impossible that someone like that exists, the number of times I’ve seen someone push LLM generated code that LOOKS okay on a first glance to experienced eyes but turns out to result in things like the code as a whole suddenly ignoring all permission groups and just letting everyone through in one particular area, or allowing directory traversal, or injection, or any number of otherwise rookie mistakes is high enough that everyone, especially programmers with decades of experience should be extremely concerned.
The reactions to this have been pretty hysterical considering this is an experimental branch on his own repository. I’m not surprised he might experiment with using AI agents to solve a particular pain point to see how it does. This is using agents to experiment with things that you might not have time to address normally.
Another recent example is the build system where the author said:
An LLM was used to first determine where the bottlenecks were then to figure out how to improve them. It generated a lot of code, much of it hideous.
I extensively audited and rewrote a lot of it, and heavily edited commit messages, the cover letter and comments.
off-loading the boring grunt work of identifying bottle necks and trying new approaches seems like an ideal task for a junior agent to deal with as long is you don’t YOLO the result to the lists for other humans to de-slopify for you.
The reactions to this have been pretty hysterical
i FULLY expected more hate. lol
It’s a slippery slope. So far Linus has gone from using it to find bugs, to using it as a programming assistant, and now ostensibly to vibecode entire features.
Where will it end? Will Linus shut off external contributions and let the agents loose like Bun did? That would be a tragedy in my book.
We can just fork the latest kernel version we feel comfortable with.
The phrase vibecode has become totally meaningless. The original use was too “give in too the vibes” and let the agent go wild. There is no indication that is what Linus is doing.
The fact is LLM’s have proven to be good at finding bugs. Suitably constrained they can be diligent and helpful assistants.
The Bun experience is instructive and if that was Linus’ end goal it would be concerning but I don’t see any evidence that any of the Linux maintainers are interested in that sort of wholesale conversion. The good projects with taste will adjust to the new tools and the slop projects will eventually collapse under their own contradictory architectures.
People cannot seem to grasp that there is a difference in using LLMs to experiment and just blindly using code written by an LLM. You’d think they would realize Linus is never going to do the latter. Binary thinking leads to this…
is there a difference in the environmental destruction when used this way? no? these things have so many negative effects on the world in so many areas that making life more convenient for coddled computer touches doesn’t excuse literally any of it
is there a difference in the environmental destruction when used this way?
Yes. There is no doubt that if you use an LLM for everything that will use several fold more times electricity than someone who uses it a few times a year.
Like I said, binary thinking. You’re literally saying there is no difference in 1 and 1000000 because both are more than zero.
so the less you use it, the better? glad we can agree
Right, so we know Linus has used Claude for one project. You are saying there is no excuse for that and I don’t agree. I don’t think it makes sense to take absolute stances against things that will undoubtedly be used by most everyone eventually. Just like I think cars are generally bad, I’m not about to get pissed off at everyone who drives them.
the earth doesn’t care if you think this is a good use or linus is somehow worthy when the regular plums aren’t.
dario and his epstein wife sure do though. so keep going to bat for them
as long is you don’t YOLO the result to the lists for other humans to de-slopify for you.
(Or not de-slopify at all.)
There’s literally no one I trust more to handle llm code more than Linus.
He’s not going to pass off shit in his name. He’ll use it in the best possible way.
Lots of reasons to be anti-AI, but it’s not all black and white. It’s here to stay and it’s useful. (But we ain’t gonna spend 400 billion a year on it.)
I have never met an engineer i trust with an llm. I doubt i’d trust linus
AI ethics aside (but not discounted overall), if there’s one person I trust above all others to use an LLM to assist in coding properly, it’s Linus Torvalds.
In my experience LLMs have been great at implementing solved problems that it has been trained on.
I’ve built a low-level media system that AFAIK exists in one 20yo unmaintained open sourced version, and ~6 proprietary versions. Even with clear specifications and baby steps Opus had a very hard time getting functional code.
use an LLM to assist in coding properly
This is categorically impossible for any ways in which the LLM could reasonably “assist”.
Putting aside any potentially elitist definitions of the word “properly” and simply taking it to mean “code that functions and isn’t malware”, you’re stuck with the same modus operandi you get copy-pasting snippets from stack overflow and hand editing them to fit your codebase.
To wit: you still need to go over every single bit of code to determine whether or not it’s safe to include in your existing codebase, and unlike stack overflow, output from an LLM isn’t being simultaneously examined by other experts for flaws at the same time you’re looking at it.
We need to stop treating LLM’s like they’re in any way a “robot programmer”.
They are database reorganization engines
(data lake of indexed, ingested material)
with a static output style
(resemble human speech patterns)
that is specifically, deliberately written to deceive about its own capability and purpose.
(speak with confidence, which is a thing that a script or a program cannot have, as though it is performing a specific requested task and not just vomiting up the contents of its proverbial shelves because someone handed it an hand crafted index card that menrions a few things from the card catalog it uses)
There should be no place for that in programming.
Linux Foundation has financial backers. He’s gotta give them lip service to keep them excited and happy. Anthropic is now a Gold Member. Oracle, Microsoft, Google, Meta, Amazon are all Platinum Members.
FOSS has always been subject to corporate meddling in their governance. There’s a reason we call it Linux and not GNU-Linux. GNU gets no money so GNU goes nowhere.
Linus gives a headline using agents, Linux gets millions to continue doing real work. That’s the game they are playing, for better or worse.
Linus regularly posted on Google+, it’s more like he himself isn’t that into foss and sees everything as a utility to use
Linus has been critical of Gplv3. I dont think he would use Gplv2 again if he could start over.
I wouldn’t even disagree with considering it. Look at how corporately independent the BSDs are… and they are just as capable in most fields as Linux.
I’m unsure which license is the best path anymore. That said the outcome would likely be the same eitherway.
the decay of humanity
Of this I agree. It may very well be the end of us.
Having said that, LLMs produce far better code and architecture than most developers.
Flame on!
How much has he dumbed down his capabilities already? When will he replace himself by an LLM?
He was already pretty dumbed down back when he used to shit on C++. From there to the neutering he went on (we don’t get middle fingers for Nvidia anymore) and now this… it’s sad to watch.
Die a hero or else, as the saying goes.
Nobody freaks out when skilled programmers/professionals use AI, it’s actually a tool for them and not full-on skill substitution.
Nobody freaks out when skilled programmers/professionals use AI
Wrong. Because there are inherent problems with ethics, such as data it has been trained on and licensing questions. Also using LLM might decrease your competence over time. This also gives too much power to those few companies who control what data and logic the Ai is fed with. Even skilled programmers, professionals or not, get lazy.
BTW I am not 100% anti Ai, but I’m not a pro vibe. I think its fine the way some developers use Ai tools. But regardless how it is used, the ethical and licensing questions remain.
What you’re saying is that if you aren’t a famous programmer you deserve to be ridiculed for using the same tool?
And how do you square this with all the crying about water?!
Nobody cares how famous a person is (in relation to this issue). If anything, the position and power a person has is much more important.
Not if you’re famous, if you’re bad at coding on your own
I can’t wait for a base linux install to be 60 GB of slop, like windows.
Year of the BSD desktop!
As long as Linus Torvalds lives, which I hope will be forever, the Linux kernel will never be AI slop.
Whether any particular distro is slop, that’s another story.
