"Furores are fermenting in the forums"
*undamentally *ed-up with *requently saying *uck-off.
Both Ubuntu and Fedora have made it official: support is coming soon for running local generative AI instances. An epic and still-growing thread in the Fedora forums states one of the goals for the next version: the Fedora AI Developer Desktop Objective. It is causing some discontent, and at least one Fedora contributor, …
But are they actually talking about "baking it in" (whatever that might mean)? Or just offering a framework for running local generative models if you want to. If it's the latter (it's not entirely clear to me if this is the case or not, but the quote "AI tools will not be added to Fedora’s existing system images, Editions, etc, by the AI Desktop initiative" suggests it is), then what's the big deal? If you have no use for it, then you simply ignore it.
El Reg, this is not elegant... Why is the quotes' typeface so much bigger than the text's? Traditionally bigger fonts are used for headlines. Italics would be enough to make quotes stick out, a mix of typeface sizes inside the body of the article breaks visual homogeneity and looks simply bad... IMHO at least.
/rant
> is not available on the *BSDs
Such as?
Allegedly, the FreeBSD Linuxulator is good enough that these days you can just run server-side stuff using Podman's docker commands, and it should just work. I have not tested this as so far I do not have FreeBSD in daily desktop use. (But I'm considering it ever more strongly...)
Matlab. It's a curious beast that expects certain things to be in certain places, and ships with a lot of its own versions of various system libraries, glibc, Java, OpenGL and whatnot. It is not uncommon to run into library versioning issues even on Linux, especially when used in conjunction with external software and/or C plugins (which is something I need to be able to do).
There was a time when it was just about possible via the Linuxulator. The last time I tried that it was a world of pain and I gave up. My recollection is that the Java-based installer generally barfs, requiring arcane workarounds; and if you get past that initial hurdle, the Matlab licensing system (which has undergone many changes over the years) will almost certainly refuse to cooperate on unsupported platforms. Then the library issues kick in. That was about ten years ago, since which time it's apparently only got harder, as Matlab's dependencies have increased in complexity, outstripping the capabilities of the Linuxulator.
Virtualisation (via Bhyve) is another option; I may give that a go when I find the time (or retire1, whichever happens first :-/ although at that point I'd probably switch to Julia anyway). If it weren't for Matlab I may well have jumped ship to FreeBSD many years ago.
1A.k.a. carrying on working without getting paid.
> Enquiring minds want to know which one.
For the past few months I have been entirely satisfied wih MX Linux (Booted via SysV)
Pros: Included Snapshot for quck-&-easy backup, pretty much Debian standard, and a REALLY helpful and knowledgeable web forum.
Cons: Dunno, still waiting...
Not very 'snazzy' but it works,
Mike
P.S. @Liam -- Follow your articles - good stuff!
If this crap gets hard baked into any distro that I use. (Rocky and Alma Linux)
AI can sit in the DO NOT USE EVER folder like SNAP and even FLATPACK.
I use servers the size of Intel NUC's and around an i3 in performance. They are also RAM limited. Hardly the right place for a local LLM especially in these days of having to take out a 2nd, 3rd or even 4th Mortgage to be able to buy enough RAM for an LLM.
- Offer the AI package, yes, but don't preinstall it.
- Don´t shove everything on it. Don´t shove it on everything.
- Let the users decide what to do with them.
- Don´t run the Patch Tuesday for your distro through an AI.
I think those are acceptable terms for an AI.
But there is one snag: What about the source code? Does ANY AI even HAVE a source code FOR THE TRAINED MODEL WEIGHTS? A Neural network doesn't generate a language-descripted code per se, so how do you plan to show the SOURCE CODE of something that DOESN'T HAVE a formal one? Or will an AI spit a compilable code in... (C? C#?) if you ask nicely?
I think the trained model weights of all AI violate the very principle of Copyleft, and gnu-Linux distribution in general. They don't have a source code, are not subject to review, and you can't ensure they will even work the same. They are black-boxes BY DEFINITION, and defeat the very cornerstone of Linux.
I don't know how people will even consider to distribute non-auditable code on a Linux distro without a series of caveat emptors and warnings.
You can publish the formal code that creates an AI and grow your own model weight, sure, why not, but distributing one grown AI model that violates every Linux definition, ARE YOU OUT OF YOUR DAMN MINDS?
Color me confused for people creating rules for Linux, and then tossing them aside like an inconvenience.
> "But there is one snag: What about the source code? Does ANY AI even HAVE a source code FOR THE TRAINED MODEL WEIGHTS?"
Are they distributing the models themselves, or just giving users who bring-their-own-model a convenient set of hooks to integrate it with existing functionality?
Article reads more like the latter, and if that's the case, it's certainly a step forward in giving users choice on their systems.
If we learned anything from the editor wars, it's that the correct choice is whatever argument you supply to apt.
Oh yeah I had to proofread that. You are right. They are NOT 'copiloting' any AI on the distro, indeed! They are just offering the tools for people to fiddle around with their own models, and that's okay.
Sorry about the rant, (which doesn't stop being true by the way).
Good job bringing it up, have a good one.
As one commenter noted above, it's a problem looking for a solution. They want to add a framework for allow generative AI systems to be integrated into the distro? Have I got that? Wouldn't that be a problem for the AI vendor to solve? I mean, you wouldn't add a word-processing API/Framework to the distro so support word processing software, or browsers, or whatever. So why would you do it with AI? All that so you just say "Hey, AI agent, move any personal files that are more than three years old to my "old files" directory".
Odd.
Back in October 2012, Mark Shuttleworth/Canonical introduced Ubuntu 12.10, "Quantal Quetzal" - and one of the most distinguishing and least popular features was the "Shopping Lens", part of the "Unity Desktop Interface" - an agent that would literally "phone home" to Canonical to let the mothership know what search terms the user was entering. It took 4 years - until 16.04, Xenial Xerus to be precise - but after serious backlash from the community and the EFF and others, Shuttleworth abandoned the idea.
All they are doing now is introducing the same scam but via AI.
In other words, Shuttleworth is looking to use the work product of everyone running ubuntu to harvest data and potentially generate training data for AI models that Canonical might control.
Privacy Be Damned, of course.
We might have hoped that they would have learned their lesson 14 years ago.
Apparently not.
When is Shuttleworth going to realise that "No" means "No"?
Perhaps we might need to set up our own "Movement" - Not just "Me Too" but "My Computer Too".
And no, I do not mean to "make light" of the harms inflicted on women that prompted the "Me Too" movement, but rather to point out that the driving thought process - the "I/we can do what I want and get away with it" seems to be present in both cases.
No. Absolutely no ####ing way. With bells on.
I only use Fedora on one machine (mainly due to hardware support) but I for one won't be sticking around if they start stuffing it full of AI codswallop.
Just because Mr Spaleta doesn't have it written down in front of him doesn't mean that there isn't already a move away from Red Hat products to other distros.
Same goes for Ubuntu and anything based on it, although I hope Clement Lefebvre continues the good fight with Linux Mint and doesn't allow the AI stuff in.
Is this evidence enough?
https://anticapitalist.party/@thezerobit/116556892342101308
From OP on Mastodon:
"Am I jumping on NetBSD just because they are refusing to accept LLM-slop in their codebase?
Yes. Frankly, that's enough.
Is it a really neat project for a lot of other reasons?
Yes, absolutely."