Sent out today as a request for comments (RFC) by a Linux kernel engineer employed by IBM is a machine learning library for the Linux kernel. The intent is on plugging in running ML models to the Linux kernel that could be used for system performance optimizations and various other purposes.
Viacheslav Dubeyko who is employed by IBM as a Linux kernel developer sent out an RFC patch series today that aims to introduce a machine learning library for the Linux kernel.
I hate how LLMs have made any talk about machine learning immediately cause a bad reaction for most people. ML isn’t inherently bad. LLMs are a neat technology, but it is hard for people to understand that when capitalism does what capitalism does best (extract and pervert). Now no one is excited to see anything related to ML.
I remember when GAN images started to be shown around the internet (which looked like acid trips lol) and I was so excited to see the technology improve. Now no one is excited to see anything ML because it is being used to do the worst things: spy, replace workers, cheat, propagandize, steal from the commons, lie, destroy the environment, destroy the economy, and (of course) kill people.
I fully agree but I can’t help myself and have to make this nitpick:
Replacing workers could be the best thing ever if we had a better system to distribute the results of work being done.
This. This is how i feel about ml/llm/ai tech.
Im going to give a lecture this year about mcp, what it is and what it does, with a live demo of letting an llm see what an application does ( via an reverse engineering api with an mcp endpoint ) and recreate it in .net ( via the build in vscode mcp ).
That, or i might make an old api to access data, attach some mcp endpoints and let a lmm design some basic ( = aka to be validated and build upon by an actual engineer!! ) structures of the data for migration purposes and maybe even let it migrate some data depending on how my poc goes.The tech is cool as hell, but what its used for and how companies throw it against everything, even if it has no added value, is making me puke and hate it so bad…
Even the way llm’s are trained make me puke. Fuck you all companies out thereEven without considering LLMs, I don’t think this is a good idea since it will make it very difficult to know for sure what the kernel will do in certain situations.
The headline and/or article is confusing. It says “machine learning library for the Linux kernel” but goes on to say it’s going to do ML in userspace.
It sounds like it’s going to be a process that lives outside the kernel and tunes the kernel via some interface. In which case there would be no need for any ML-specific bit in the kernel. Rather a generic interface to collect stats, tune thing. Processes would use those however they want, using ML or astrology or classic algorithms for their internal logic.
Can we like… just not?
This is literally how killer computers come into existence.
If this increases inference performance for local LLMs, could be cool
It’s wild that one of the problems is that the kernel doesn’t use floating point numbers.
