this has been a bit of a meme, but if you wanted to look at XL as extra large, then that could refer to the max resolution which is far great. I’ve seen people refere to it as “extra long-term” but I think the real reason is they just wanted to fuck with us
like what? I can kinda understand them not cooperating but how on earth could they lock them out of features?
ehh… not really, the amount of generated data you can get by snopping on LLM traffic is going to far out weigh the costs of running LLMs
I don’t even think this is the case, google does a lot pretty much everywhere. one example is one of the things they are pushing for is locally run AI (gemini, stable diffusion etc.) to run on your gpu via webgpu instead of needing to use cloud services, which is obviously privacy friendly for a myriad of reasons, in fact, we now have multiple implementations of LLMs that run locally in browser on webgpu, and even a stable diffusion implementation (never got it to work though since my most beefy gpu is an arc a380 with 6gb of ram)
they do other stuff too, but with the recent craze push for AI, I think this is probably the most relevant.
this is from the google research team, they contribute a LOT to many foss projects. Google is not a monolith, each team is made of often very different folk, who have very different goals
I remeber testing parrot a few years ago, it was quite nice back when I tested it, had some real cringe marketing back then, way worse then it has now by quick glance, that being said, it had some real good OOB configs for security stuff and some neat tools. wouldn’t mind trying it again sometime when I find the time.
color management is actually super hard to do, so making sure it’s done right is very important, so this is one of the few times it actually makes sense. I mean, just take a look at windows, it still looks like shit over there.
for the semantically inquisitive folk.
It’s worth noting if you are using this on an arm device, this isn’t a “virtualization VM” any more, as you are using the emulator backend, so this is far closer to a traditional emulator then anything else.
While the term virtual machine is extremely poorly defined, it could still apply.
also TCG is as slow as molasses, it’s a good demo, not actually usable for much, at least unless it’s a super beefy phone.
I’m not defending x11, both wayland and x11 are trash, it’s just whichever trash pile you find yourself most comfortable in.
On x11, fractional scaling is more or less just handled by the gui toolkit. It does suck that you need to set an env var for it, but IMO that isn’t too bad.
the multi monitor stuff does suck for sure. It’s not an issue for me personally. One thing that is a massive issue for me is x11’s terrible handling of touch, I use touch screens daily so that’s a massive issue for me, wayland compositors are also typically quite a bit faster then x11 + wms on low end systems now too (not to be confused with total resource usage/lightness).
Wayland has a lot of things going for it, but it also has a lot going against it. Both are terrible. Arcan save us (oh how a man can dream)
This is actually one thing that doesn’t involve wayland, as pretty much everyone is using at-spi. It’s not great, but it does work.
for one, it’s missing a good chunk of A11y stuff, activity watch requires something to monitor the active window, there is a PR for that, still not merged, this has been an issue for years
It’s missing protocols that will let applications request to be a privileged application, which is necessary for applications to use other functionality.
Missing protocols to control always-on-top / layers, which is needed for OSKs to function, and a couple other A11y things off the top of my head.
It’s not just a11y either, Window positioning still isn’t merged, which means if your app opens two “windows”, you cannot currently select where to open them, or to even bind two windows together (Android emulator does this for instance).
There is a LOT wayland is missing, it IS getting better, just at a snails pace.
Because Wayland is STILL lacking a lot of things that people need.
It’s the way it’s written, it’s typically hour:minutes.seconds
pretty misguided way to do so then lol
networkmanager is for chumps, long live dhcpcd
I disagree, making your own packages is nice, but it’s not like it’s needed. I know multiple people who don’t touch the AUR or custom pkgbuilds at all
Have you considered using av1an? it supports vaporsynth which has a large amount of upscale and frame interpolation tools, AI or not. If your upscaler supports vapoursynth, it could be a lot better option.
what are your system specs? at a low effort you should be getting a lot more FPS, what cli command are you using? but I guess it would be best for you to export to /tmp given enough ram and then go from there
EDIT: for context, when encoding libjxl I would do -distance 0 -effort 2 for lossless output
yeah, that could indeed happen I suppose, didn’t think of that. Though I wonder if because of EME, an alternative drm solution could be viably implemented.