This wasn’t about integrating commercial AI, or even AI generated code, but supporting local, open weights AI.

Fedora has approved its own AI Developer Desktop initiative with a unanimous council vote. Three Atomic Desktop images are planned, two of them CUDA-enabled, and none of them would be phoning home to cloud services. source

Personally I saw it as a positive, when only criminals corpos have guns AI etc. Local only alternatives should be encouraged.

Was about to go ‘NO!, not Fedora’, but actually they’re doing the good local thing. Phew.

Sounds like all you need is an Ext4 USB drive with a LUKS key on it. Then add kernel parameters like

rd.luks.key=UUID=/.keys/TheKey:LABEL=KEYS rd.luks.options=discard,keyfile-timeout=10s

in GRUB and it’ll autoboot.

Pull the key and power down and you’re back to normal. I use it in a low threat model environment so I can hit reboot and go get a coffee and come back to a DE.

ETA: sorry, got the timeout format wrong, I don’t use it.

Ah, so it is vibe coded [I created] was a trigger. Let people know, it might not actually be bad if you know coding already, but it’s good to let people know, not real maintainable otherwise.

Didn’t all the junior dev roles get taken by ‘agentic AI’ leaving an entire generation of devs to the mercy of AI mentoring. That’s going to end well.

Historically this protection was the role of a competent project manager (Yeah, they existed, rare, but gold), a senior dev wrote code, a pleasing experience that made the slog uphill (both ways) worthwhile, much like art.

If OP got it from a snr dev, kudos to them both.

Pretty much on the low side, but you’ve not been up long. Using key based login you’re fine.

Nah, the fun is learning form others mistakes. Thanks for a fun read :}

It’s a public holiday in Australia, but on the 4th (we make sure it makes a long weekend).

slumped

JF, the EV market is going ballistic everywhere that isn’t the US ffs. Inventing reality FTW. Have you considered there might be something wrong with your market?

Seems pretty plausible, not 3-2-1 yet, but on the way, and should get the habits established well enough. Just having an offline backup is a huge step up from most. Consider a waterproof box (perhaps buried) in the back yard instead of just another room (in case of fire / flood).

If you have a friend with a similar setup, or who perhaps wants one, you can sync over internet and both get your offsite without the expense of online backups or the inconvenience of lugging HDDs around.

Eh, at a philosophic level I think it has validity. Pretty sure Soong dumped a fair bit of himself, and then augmented TF out of it. Not genetic, true, but does that actually matter in the long run?

“lost”, world keeps turning, some asshole makes less profit, less carbon burned.

I run podman containers on my bazzite machines, basically you convert a docker-compose file to a .container file, here’s a bunch of examples, nextcloud is there, drop it in ~/.config/containers and run systemctl daemon-reload and it’s now a systemd unit that you start stop etc like any other. Updates are with podman autoupdate.

You can use podlet to convert docker-compose files (90% it works, otherwise it gets you 90% of the way there). It’s basically the fedora (/redhat) way to run containers.

I have no idea where you got it not being recommended (but adding to the main image sure is discouraged), and it’s certainly better than adding a vm for containers, which pretty much defeats the purpose of containers (to run using your main kernel, but contained).

I’ve been running my arr stack (with gluetun in a pod) etc this way for years now, very trouble free. Here’s a immich example.

It’s a bit of a learning curve, but it pays off.

Given how long they’re viable for (obsolete GPUs in 2 years or less, high likelihood of needing significant retooling for the next gen) you can’t mothball them. Given how few exist in any way other than on paper there will be little to bulldoze.

On the plus side counties are scrambling to decouple their economies from the US and hopefully it’s just the trickle before the flood, so when the US economy tanks from the AI bubble pop it’ll have less effect on the rest of the world.

Just repeating what I read elsewhere, the reporting has been somewhat muddy.

Nope, not top, but apparently not awful (seems to be her husband coding). Some interesting things happening currently with memory systems for local AI which seem to truly enhance (make smarter) smaller thinking models, meaning you can use more, and more coherent context.

As with most of this stuff I’m waiting for a bit more maturity before looking deeply at it, but there’s definitely some excitement at the moment, and it has the potential to make models that fit on say a 16GB video card capable of many more use cases than previously.

Legal thieves, and their kin.

Can’t imagine how boring the world would be without stuff I don’t understand.

Don’t forget thinking…

Been using an quadlet podman arr stack for a year or two, pretty damn bulletproof once set up, easier to read, rootless, SELinux enabled, systemd controlled, update with podman auto-update. Worth the time to learn.

podlet can help you hit the ground running. It can create Quadlet files out of Podman commands or even (Docker) Compose files. 90% of the time it works every time ;}, but even the oopses get you most of the way there.

My arr stack is set up in a pod which means they all have their own gluetun network and come up as one, but you can just use Network=container:gluetun in container files.