In June I wrote a blog post about how "AI might destroy the Linux kernel", where I pointed out how the Linux kernel has, during the past months, been absolutely flooded with AI-generated merge requests for patches, fixes, new features and new drivers, with literally thousands and thousands of such AI-generated code being submitted for review every single month. And the pace isn't slowing down. On the contrary, it's only getting worse and worse.
I posited that this will be absolutely detrimental to the development and quality of the kernel development because of two possible approaches that the maintainers could take, neither of which is very good: Either
1) they start reviewing every single merge request for validity, which will grind the actual kernel development pretty much to a halt because there are so many of them, or
2) they will become sloppy and start accepting merge requests with only cursory reviewing and supervision, which may introduce bugs, degraded performance or outright exploits (either inadvertent or deliberate).
Am I just making stuff up and fear-mongering for nothing, making stuff out of thin air? Is it all just an exaggeration and there isn't an actual problem?
Actually no. This is an actual problem acknowledged by the kernel maintainers themselves. In fact, they have even coined a neologism for this very phenomenon that they are currently experiencing: "LLM-pocalypse." (Yes, indeed, their term.) And not even in some kind of tongue-in-cheek joking manner, but seriously. They are using expressions like "surviving the LLM-pocalypse" in all seriousness, not as some kind of joke. They are having actual real trouble with the massive flood.
One way in which they have started coping with the problem is, perhaps in some ways, even worse than the two options above. And that's to outright remove older driver for which hundreds of patches are being submitted. In other words, rather than either fully review the patches, or let them through with only cursory reviewing, they are taking the third (some could argue worst possible) option: Just remove the drivers in question from the kernel so that the submitted patches become obsolete.
And yes, they are already in the process of removing literally hundreds of existing drivers from the kernel, for things that they deem "old and obsolete" hardware, like old mice, ethernet cards and other peripherals. Some of that hardware that's being dropped is (as of writing this) from as recently as 2002. Which, of course, means that Linux will not support that old hardware in the near future, even if you happen to have them.
And the only and sole reason they are doing this is because of the tens of thousands of AI-generated merge requests they are being flooded with.
All of this means that hardware support in Linux is becoming poorer and poorer, narrower and narrower, over time, just because the kernel maintainers don't want to deal with as many AI-generated patches.
And if you are thinking "who cares about some old and obsolete obscure peripheral from the 90's that nobody has been using for over 20 years?", it might not be that simple. Who is to say how much of even currently used hardware will be considered "old and obsolete", and support removed from Linux?
Comments
Post a Comment