“nothing that can be done about this” says guy who caused this and derogatorily called anyone who doesn’t like it an SJW
-
@zzt don't forget the Gas Town guy's pet project, a million LOC Java codebase for a 30 year old gMUD which was apparently impossible to develop in, to which he added over a dozen "AI agents," which proceeded to introduce tons of regressions to things which had worked fine for decades and straight up hallucinate patch notes
@tael @zzt at this point, as someone with ADHD who has experienced the "slot machine effect" of LLMs (not code) and realized, with horror, what exactly that neuro feedback was doing in my brain...how many of these folks going in on it in this coding way are undiagnosed, and thus unaware of the impacts, or maybe are diagnosed and lack the mindfulness practices to start becoming aware of the way their brains handle these sorts of stimuli.
Like, I only learned about that kind of practice since 2019 and I've found it invaluable in recognizing when something is causing horrible problems for me. I don't always like to see it, but man Linus etc need to actually have the A awareness Of Self to understand the way their brain handles the feedback from the slot machine.
-
and again I’ll point out how fucking quick this shit is happening
torvalds posts a defense of AI slop in the kernel with a right-wing conspiracy theory as the reason
a bunch of other open source projects just go “well the kernel says it’s ok so we’re doing slop now”
barely a couple of weeks later the kernel is changing its release cadence and quality standards entirely and the man in charge is shrugging and literally going “what can you do”
you can cut the crap, asshole. it’s your literal job to cut the crap.
@zzt is that why my fedora44 webserver took a dump after a update and now won't even boot to finish?
-
oh wow, the machine can plagiarize servo into a web browser that’s broken and strictly worse than servo!
oh wow, the machine can plagiarize GCC and a bunch of undergrad compiler projects into… a C compiler so unoptimized and broken it probably set a record for how slow its output ran, which also didn’t run any of the normal sanity checks on its input
so what exactly are the LLMs running on the Linux kernel going to plagiarize from?
we know the answer of course, there’s not a lot of kernels they can pull from
oh hey, some of you have seen this before but look at torvalds’ prior work before shoving LLMs forcefully into the workflow of the kernel:
https://infosec.exchange/@atax1a/115867354992858960
it’s a fucking Python data visualizer tool that you or I could have written in a day, but torvalds decided Python wasn’t a worthy language for him to learn so he used an LLM instead
and this wasn’t his first time fucking with the things, but this was his last project before he decided to go all-in
somehow an LLM doing a shit Python data visualization tool for one of his hobbies convinced him that LLMs must be really good at the type of low-level C the kernel is written in
that’s all the evidence this mid little fucker needed to assert it as fact
-
oh hey, some of you have seen this before but look at torvalds’ prior work before shoving LLMs forcefully into the workflow of the kernel:
https://infosec.exchange/@atax1a/115867354992858960
it’s a fucking Python data visualizer tool that you or I could have written in a day, but torvalds decided Python wasn’t a worthy language for him to learn so he used an LLM instead
and this wasn’t his first time fucking with the things, but this was his last project before he decided to go all-in
somehow an LLM doing a shit Python data visualization tool for one of his hobbies convinced him that LLMs must be really good at the type of low-level C the kernel is written in
that’s all the evidence this mid little fucker needed to assert it as fact
when you’re a world class smart boy C expert but you fail to learn basic frontend Python so hard you get addicted to LLMs instead
-
oh hey, some of you have seen this before but look at torvalds’ prior work before shoving LLMs forcefully into the workflow of the kernel:
https://infosec.exchange/@atax1a/115867354992858960
it’s a fucking Python data visualizer tool that you or I could have written in a day, but torvalds decided Python wasn’t a worthy language for him to learn so he used an LLM instead
and this wasn’t his first time fucking with the things, but this was his last project before he decided to go all-in
somehow an LLM doing a shit Python data visualization tool for one of his hobbies convinced him that LLMs must be really good at the type of low-level C the kernel is written in
that’s all the evidence this mid little fucker needed to assert it as fact
@zzt "mid little fucker" is almost as concise a summary of him as "git"
-
“nothing that can be done about this” says guy who caused this and derogatorily called anyone who doesn’t like it an SJW
@zzt The thing is, *even if* you think that this is the future of kernel development, you could still say “let’s not put as many changes as this into RC6, but keep them for the development version coming up, so we have time for any bugs to get shaken out”.
-
@zzt The thing is, *even if* you think that this is the future of kernel development, you could still say “let’s not put as many changes as this into RC6, but keep them for the development version coming up, so we have time for any bugs to get shaken out”.
@skington what if instead you surrender all decision-making and accountability to a machine that’s horrible at everything but especially horrible at systems code
-
oh hey, some of you have seen this before but look at torvalds’ prior work before shoving LLMs forcefully into the workflow of the kernel:
https://infosec.exchange/@atax1a/115867354992858960
it’s a fucking Python data visualizer tool that you or I could have written in a day, but torvalds decided Python wasn’t a worthy language for him to learn so he used an LLM instead
and this wasn’t his first time fucking with the things, but this was his last project before he decided to go all-in
somehow an LLM doing a shit Python data visualization tool for one of his hobbies convinced him that LLMs must be really good at the type of low-level C the kernel is written in
that’s all the evidence this mid little fucker needed to assert it as fact
@zzt It really needs to be studied how people can just get oneshot from using these things even once.
-
@skington what if instead you surrender all decision-making and accountability to a machine that’s horrible at everything but especially horrible at systems code
@zzt I suppose my point is that Linus Torvalds isn’t even being a “centrist” about this, which would be to say “AI is the future, but we’re in a code freeze / learning how to use the tools properly, so let’s not merge all these PRs to main”. The rush to go ahead with LLM contributions goes completely against normal project management practices, apparently solely so people can say “the latest *shipping* Linux kernel is AI now”.
-
“nothing that can be done about this” says guy who caused this and derogatorily called anyone who doesn’t like it an SJW
Great, I'll get right on moving to a new system my hardware doesn't support, my games don't support, that I'd have to learn, and that probably is also vibecoded too. I fucking hate this timeline so much.
-
-
@zzt He was *always* a bit of an asshole, and now it shows, and I'm going to rub it in just a little bit on people who kept insisting that "BDFL is a good thing".
@promovicz @zzt Linus just forgot about the 'B'.
-
@zzt So, if I am reading you correctly, it's possible but would be a major effort requiring a team and resources. A couple of hardcore coders couldn't do this on their own.
@galacticstone it would take more than one team and more than one project or company that has kernel contributing developers. Much of the hardware is in development years ahead of time and the developers can be dealing with alpha parts straight from the fabrication lab. If it's something like a new laptop it has to bounce around multiple teams who work with the manufacturer or OEM. Parts can also be combined in various configurations as revisions come in and then it recirculates through- @zzt
-
@galacticstone it would take more than one team and more than one project or company that has kernel contributing developers. Much of the hardware is in development years ahead of time and the developers can be dealing with alpha parts straight from the fabrication lab. If it's something like a new laptop it has to bounce around multiple teams who work with the manufacturer or OEM. Parts can also be combined in various configurations as revisions come in and then it recirculates through- @zzt
@galacticstone - the developers again. Some things don't actually make it to retail release. But you are dealing with a mix like consumer-target laptops, enterprise servers, workstations, standalone parts to combine them with like gpus or ram, etc.
Some things go faster if it's an incremental hardware refresh but when something is new it usually takes a while.
(edit to add- I used to work at SUSE in hardware a few years ago and from what I know the process should still be the same).
-
@zzt I suppose my point is that Linus Torvalds isn’t even being a “centrist” about this, which would be to say “AI is the future, but we’re in a code freeze / learning how to use the tools properly, so let’s not merge all these PRs to main”. The rush to go ahead with LLM contributions goes completely against normal project management practices, apparently solely so people can say “the latest *shipping* Linux kernel is AI now”.
-
@zzt I suppose my point is that Linus Torvalds isn’t even being a “centrist” about this, which would be to say “AI is the future, but we’re in a code freeze / learning how to use the tools properly, so let’s not merge all these PRs to main”. The rush to go ahead with LLM contributions goes completely against normal project management practices, apparently solely so people can say “the latest *shipping* Linux kernel is AI now”.
-
@zzt So, if I am reading you correctly, it's possible but would be a major effort requiring a team and resources. A couple of hardcore coders couldn't do this on their own.
@galacticstone @zzt maybe they could use Claude to help.
*cries into his pillow a bit*
(Yes, this is sarcasm.) -
@zzt Bjarne Stroustrup went on record as saying he thought LLMs would not be of much help for C++. https://youtu.be/U46fJ2bJ-co?t=5331
-
-
@gooba42 @zzt Microkernels got a bad reputation by the microkernel folks doing them backwards. Putting sketchy hardware io stuff that actually needs to be isolated into the real core kernel but syscalls into unusably slow and misbehaving RPC Rube Goldberg machines.
A good microkernel is at its core the resource & state management (memory, scheduling entities, file handles, etc.) needed to implement POSIX with hardware drivers and filesystems in isolated userspace tasks.
@dalias @gooba42 @zzt Which seems to mirror the typical underlying architecture (CPU+Memory as core, rest being external, at least outside of MCUs/SoCs) + system constraints (like file handles managed in userspace sounds unnecessarily messy to deal with).
But well I've barely touched kernel-layer things so not sure how well that comparison holds up.
