But like... the ai was able to vibe code a shitty python application, therefore all Indians should be allowed to shit up the kernel with slop. Just look at how many bug fixes are in the kernel now!
Crazy thing is that its existed since 1996 and has been updated continuously since then yet I never hear people talk about it. L4Linux is a paravirtualized system where Linux runs as an ordinary userspace program on top of a tiny operating-system kernel called L4.
For the record BSD is awesome, NetBSD even more-so, when I tried it I had to adjust the graphics to this headache-inducing vesa configuration which isn't really meant for modern (read cheap, generic) LCD monitors. I just gotta start settling down and not be tinkering so much lol. Maybe when Intel Ice Lake gets long in the tooth enough I'll jump over specifically to NetBSD when I feel I've "outgrown" FreeBSD.
On the topic of kernels, lmfao just switch to one that works better with old hardware like LTS or use Mesa-Amber for super old GPUs.
Garuda Toolbox is one of the most useful GUI tools for Linux ever. The distro is totally underrated and lets a totally computer illiterate moron use Arch linux without knowing how to use the terminal without fucking anything up. For someone who does know how to, its Windows level ease of use.
Garuda Toolbox also has an easy "fix it" button in case a computer illiterate moron does fuck something up which will happen. Garuda also is opposed to age verification.
If you do not like the super colorful and gamer aesthetic which is understandable for productivity, you can swap to base KDE or use XFCE to not have to deal with that.
But like... the ai was able to vibe code a shitty python application, therefore all Indians should be allowed to shit up the kernel with slop. Just look at how many bug fixes are in the kernel now!
Ok so like to be fair right
Being a little devils advocate or whatever bear with me
He specifically said it's useful as a code review tool, he didn't mention vibing your PRs and I don't think that's what he meant
I mean.. I have successfully utilized AI to write a TKinter GUI for me that I felt didn't look like complete shit (which is to say, it looked basically like using a Java applet from 1995.).
A really strong password on an encrypted file system is really only good for one scenario, where someone has intermittent access to your system and are attempting passwords one by one and trying not to get caught.
In every other situation either a basic password is sufficient as the people will give up and move on, or no password is secure enough - as they have physical possession of the device and can use all resources into cracking the encryption.
It is only a matter of time before the Linux kernel becomes genuinely unmaintainable if AI contributions and code checking continues at its current velocity. Not saying it is a catch-all answer, but I believe there is a strong case to be made for more esoteric options a la *BSD. No matter how hard AI cultists try to shill this shit, the more you rely on it to generate "boilerplate" or do "menial work" o algo, the less you know your code, the less you can maintain it, and the shittier it becomes for everyone after the fact. Nevermind that mainline Linux is full of opaque proprietary trash code as is.
Linus may be keen eyed enough to review the slop into usability, but there is no guarantee of such vigilance among other kernel maintainers, especially not newer ones or glory-hungry jeets that flood every project they can get their shit smeared hands on.
No matter how hard AI cultists try to shill this shit, the more you rely on it to generate "boilerplate" or do "menial work" o algo, the less you know your code, the less you can maintain it, and the shittier it becomes for everyone after the fact.
This is nonsense. Most of the code I've worked on professionally has been code I didn't write and wasn't documented. Even on stuff I've worked on and documented, I still often don't remember much about the code if I haven't looked at it after 2 to 3 months. Most code you work with you won't know from memory anyway.
Menial work and boilerplate are the majority of dev work. AI objectively does this better than people.
There are as many more AI-refusenik cultists than AI cultists, and they often have no idea what they are talking about.
Neither is the assertion that people just don't know the code they work on true, nor is the idea that you just forget code after a while. For me, even after a time of not looking at old code, there are bits and pieces that jog my memory very quickly, something that is absent when I look over AI output. It is easy to be swayed when this demon egregore offers you easy autocomplete on simple code to then permit broad stroking the entire project into verbose vomit. Yes, humans can make garbage, but AI permits everyone to make garbage faster and with far less scrutiny because, per the Phoronix post I linked, there is an increased reliance on using AI to review AI, so you get an endlessly churning trash mill even in projects that have a (presumably) high bar for code quality like the Linux kernel.
AI makes you more productive in the short to medium term, makes you retarded in the long term, and turns your project into a stinky demonic turd. Soydevs saying they "don't code by hand anymore" is nothing short of admitting defeat and outsourcing all of your actual work to the slop factory so that you can "manage" or "delegate" or every other fancy word people use to cope with their continued retardation. Not to mention it is closed source proprietary trash, regardless of whether it is locally hosted or not, but especially frontier models like GPT or Claude that 99% of AI addicts use. I will never, ever trust anything meaningfully touched by AI the way I would something made by man.
Neither is the assertion that people just don't know the code they work on true, nor is the idea that you just forget code after a while. For me, even after a time of not looking at old code, there are bits and pieces that jog my memory very quickly, something that is absent when I look over AI output. It is easy to be swayed when this demon egregore offers you easy autocomplete on simple code to then permit broad stroking the entire project into verbose vomit.
Yes, humans can make garbage, but AI permits everyone to make garbage faster and with far less scrutiny because, per the Phoronix post I linked, there is an increased reliance on using AI to review AI, so you get an endlessly churning trash mill even in projects that have a (presumably) high bar for code quality like the Linux kernel.
AI makes you more productive in the short to medium term, makes you retarded in the long term, and turns your project into a stinky demonic turd. Soydevs saying they "don't code by hand anymore" is nothing short of admitting defeat and outsourcing all of your actual work to the slop factory so that you can "manage" or "delegate" or every other fancy word people use to cope with their continued retardation.
AI makes you more productive in the short, medium and long term. Using pejoratives such as "soydevs" and "demonic" constantly doesn't change that fact, surprised you haven't used Microsoft, iToys, Winblows and other cringe terms.
I can code fine, btw, after months of not doing any manual code. This idea that you suddenly become helpless and cannot review the output is nonsense, and it is repeated by people like yourself that don't know what they are talking about.
Not to mention it is closed source proprietary trash, regardless of whether it is locally hosted or not, but especially frontier models like GPT or Claude that 99% of AI addicts use. I will never, ever trust anything meaningfully touched by AI the way I would something made by man.
What matters is if your data / code / whatever is tied into the proprietary platform, not that you happen to be using it. BTW, even Stallman would agree with me on this assessment (because there is a direct quote from him where he says as much).
I use a local open-source AI, and I use proprietary tools and AI when it makes sense. Being a GNU zealot is retarded and has always been retarded.
Every single one of your arguments against AI usage is emotional and made by people that have no idea what they are talking about.
So if I'm understanding correctly, the reason people should use LLMs is... as a somewhat more intelligent copy+paste, keyboard macro, and refactoring tool?
Programming languages were a mistake. Syntax never should have become more irregular than Lisp, and frankly probably never should have even been primarily textual. Now you've got people saying that they need a bajillion teraflops of compute power to perform basic program transformations.
If you're not using it to do any sort of understanding, then it's a massive waste, and if you are using it to do any sort of understanding, it's redundant, because you need your own understanding anyway. This redundancy can be useful to achieve a higher confidence in quality - more understandings get tested - but not higher throughput. Higher throughput is exactly what Linus and everyone else who advocates LLM usage describe as being the goal, which means either they have absolutely zero good tools for menial work or they are compromising on their own understanding. There is no substitute for understanding, and reviewing is almost exclusively understanding.
A certain Minnesotan (IIRC) university has the opportunity to do some very funny things in the coming years.
What matters is if your data / code / whatever is tied into the proprietary platform, not that you happen to be using it. BTW, even Stallman would agree with me on this assessment (because there is a direct quote from him where he says as much).
"Stallman says X does not automatically make it nonfree" is hardly an endorsement. The GPL considers "obfuscated source code" to be an oxymoron, but what exactly counts as "obfuscated" is not clear. It is entirely possible for something to be free software and still be completely impenetrable, especially if it is massive and tightly-coupled. If, in addition, it can only feasibly be modified and understood via a proprietary tool, then it is of course "tied into" that proprietary tool. The extent to which this actually occurs is no doubt something you and your interlocutor disagree on, but let's not pretend that it's impossible for a proprietary tool to beget dependencies on itself. There are already many free software projects whose complexity poses a sufficiently high barrier to entry that exercising the four freedoms in practice is difficult. The usage of generative models to enable more complexity can most certainly make this worse.
Higher throughput is exactly what Linus and everyone else who advocates LLM usage describe as being the goal, which means either they have absolutely zero good tools for menial work or they are compromising on their own understanding. There is no substitute for understanding, and reviewing is almost exclusively understanding.
There's you suggesting this, and then you've got Con Kolivas who's started maintaining his old scheduler again because LLMs take the burdensome grind of maintenance off of him. And so, if I take you at face value, you're saying Con Kolivas is compromising on his own understanding of his own project. This is just one example of a professional, high-capability programmer dissenting from your view. Not hard to find more. So I think you're painting with overly broad strokes here.
As someone who will celebrate a decade of being free from institutional code monkey work soon, LLMs help me achieve more with less effort. It doesn't stop me grappling with the difficult engineering decisions. If anything, it makes them harder, because now I don't even have my own ignorance to insulate me from the choices I'm trying to make.
So if I'm understanding correctly, the reason people should use LLMs is... as a somewhat more intelligent copy+paste, keyboard macro, and refactoring tool?
Only if you have the worst possible interpretation of what I said. If you've ever had to upgrade dependencies in some languages, you will know that often you get into dependency hell and/or do refactors when updating dependencies. This can be done in minutes with an AI, and there is no copy, paste, keyboard macro or refactoring tool that can do these effectively.
I suggest you actually use it and see what it is good at and what it isn't. I really get bored with having conversations with people that either haven't used a thing or are haters that just wish to dismiss any advantage because you can conceivably do the same task using some complex combination of scripting and existing tools that would only be used once.
If you're not using it to do any sort of understanding, then it's a massive waste, and if you are using it to do any sort of understanding, it's redundant, because you need your own understanding anyway. This redundancy can be useful to achieve a higher confidence in quality - more understandings get tested - but not higher throughput.
I've worked throughout my career on large codebases with little or no documentation. At best you may understand how a few modules work in the system, but often knowing how these monstrosities work is impossible. Nonetheless, work got done, and releases went out the door.
When I am using an AI, I can interact with it in a way where I can gain a greater understanding of the project as a whole. It does get stuff wrong, so if things seem odd or they don't line up with other parts of the system, you have to manually review it. Part of being a good engineer is understanding when and where you should expend your efforts, not pretending to be some savant.
Higher throughput is exactly what Linus and everyone else who advocates LLM usage describe as being the goal, which means either they have absolutely zero good tools for menial work or they are compromising on their own understanding. There is no substitute for understanding, and reviewing is almost exclusively understanding.
You also don't need to know every intricate detail of how everything works to understand how the system and/or subsystem works as a whole.
When people make arguments like this, I wonder who they picture in their mind, the systems they are working on. I don't memorise code. I look at the functions, classes, modules, etc. and look at the docs / behaviours, and I debug and understand the stuff that is going wrong and understand only what needs to be understood.
With an AI I can quiz it and have it list out all the points of note before I make an engineering decision. In the past I would have to traverse the same codebase myself to come to the same conclusions; however, the process would take hours rather than minutes. It is an interactive process. This is different from pure vibe coding, where you just get it to do stuff and you see if the output looks right.
Nobody said it did. The point being made is that the man who created the movement says that this is actually more important than whether the progra is GPL'd. His voice, whether you like him or not, is influential and authoritative. BTW, many other well-known engineers have said similar things to him. It isn't a controversial claim.
It is pretty clear what obfuscated code is. It's only not clear if people engage in stupid games where they pretend to not understand what the word 'obfuscated' means.
What I was alluding to in my post is something like if I use JetBrains IDEs (which are proprietary) as a code editor, that doesn't mean that you can't use Vim to edit the same codebase. The proprietary tool (the IDE) doesn't tie me into the IDE at all. Because the important thing is the code and not the editor/IDE.
Similarly, if I use Claude in a project, that doesn't stop me from using another AI, including those that are locally hosted. In fact, I do this all the time. I use Claude for some problems. Claude can do better than Qwen 3.5b.
Contrast this to if I were video editing in something like Adobe Premiere Pro; I probably couldn't take the project file (or however it stores it) and continue editing in another program, e.g., KdenLive.
In your scenario. Sure, that would be effectively a proprietary format. Ironically, AI could understand a proprietary file format, reverse engineer it, document it and create an implementation to deal with it much faster than you could. Which then effectively removes the tie-in on the proprietary tool.
There are already many free software projects whose complexity poses a sufficiently high barrier to entry that exercising the four freedoms in practice is difficult. The usage of generative models to enable more complexity can most certainly make this worse.
No. I think the opposite will happen. I think it can make it better.
Firstly, you can have the AI scan the entire project and explain major pieces to you; you can interactively quiz it; you can have it find dead code (that often other tools won't see); you can have it find duplicate code or near-duplicate code and a whole bunch of other stuff that could simplify the code. There is so much you can have it do to help you understand a complex project.
Whenever I have any of these conversations with anyone that makes any of these arguments, they are pretending that everyone is just pure vibe-coding and the AI is just spitting out code that cannot be read. You are pretending that everything goes in one direction, when it doesn't. Often the code that is produced is often the same and/or similar to what I would have written and is perfectly fine to read.
It is only a matter of time before the Linux kernel becomes genuinely unmaintainable if AI contributions and code checking continues at its current velocity. Not saying it is a catch-all answer, but I believe there is a strong case to be made for more esoteric options a la *BSD. No matter how hard AI cultists try to shill this shit, the more you rely on it to generate "boilerplate" or do "menial work" o algo, the less you know your code, the less you can maintain it, and the shittier it becomes for everyone after the fact. Nevermind that mainline Linux is full of opaque proprietary trash code as is.
Linus may be keen eyed enough to review the slop into usability, but there is no guarantee of such vigilance among other kernel maintainers, especially not newer ones or glory-hungry jeets that flood every project they can get their shit smeared hands on.
Even with ai enshittification on Linux it doesn't magically make *bsd usable, absolutely 0 of my computers can even run a bsd. Imagine trying to run Linux in the days where Ubuntu wasn't really a thing yet, that's what it's like trying to run a bsd on modern hardware. Just a complete fucking tossup on if it'll decide it wants to even boot or install or anything.
Ah yes the age old "Ubuntu is based on Debian and I don't understand under the hood changes"
If you ever wonder why lmde is still not a full drop in replacement for mint Ubuntu yet after years of effort that's why that comment is incorrect
