Microsoft claims "2026 is the moment" for AI PCs, but its essay-length beginner explanation only creates more confusion — Is it any wonder adoption is slow?
Microsoft claims "2026 is the moment" for AI PCs, but its essay-length beginner explanation only creates more confusion — Is it any wonder adoption is slow?
Debian servers… But what direction did you go for your daily driver? There is no wrong answer, but I like hearing how people migrate over.
I was the same as you, btw, started with Debian servers be it an Apache Cloudstack hypervisor, or k8s host.
But because I decided to go with a tiling Windows manager, somehow I ended up down the hyprland rabbit hole on Arch.
So I haven’t felt the need to go with a tiling compositors. I already use multiple munitors, and kind of have designated spots for the apps I use.
I love stability and don’t want any surprises after Windows made enough surprises. So decided to go with Debian Trixie, and KDE.
But I use Arch in my spare laptop, btw. EndeavorOS where I experiment some stuff. Maybe down the line I will give hyprland a try on my spare first.
EndeavorOS… I’ve been wanting to try that… Although I heard some good things about CachyOS and need to try that one first.
That said if you’re on the hyprland journey, you may look at Omarchy, it is basically Arch with hyprland preconfigured. Not a huge deal, but simple for a test.
No it isn’t.
I’ve been waiting for the “big year” since 1996. It’s never going to happen. This is not to invalidate anybody’s appreciation of Linux but it’s not for the public and it never will be.
For. Real.
I switched over some months ago now and tried several different distributions before finally settling on one that mostly could be made to work with everything, as many of them had one or more hardware dealbreaker that prevented it from working out. I think its also fair to mention that while many things did just work “out of the box” on all of them, many also did not. Some were able to be cajoled into cooperating after varying amounts of troubleshooting, editing and general trial and error effort, but there are huge swaths of the user experience that are about as unpolished and manual as they were at the turn of the century.
I still prefer using it to Windows 11, and it has improved a lot over the years, but I think the main thing that has made Linux increase in appeal over time is the relative continual decline in the quality and behavior of Windows.
I’m sure a lot of these hindrances can be addressed by building or buying a computer purpose-built to run Linux, but I think the point stands that unless you just use your PC for the “Facebook, Email, YouTube” type of stuff, you’re going to run into things you have to do that require quite a bit of research to get to work.
Don’t get me wrong; I don’t regret my decision in the slightest. Linux offers you very real ownership of your computer and user experience, it is just absolutely not for everyone, and I hope the Linux community at large one day grows to acknowledge that the tinkering and troubleshooting that many of them are not troubled by, and some of them even get enjoyment from, is fine with them because they are hobbyists and professionals. People outside that sphere see computers more exclusively as tools than hobbies, and tools that often give you trouble and take away your time are worse than similar ones that don’t.
Make Linux PCs and you will PRINT MINT. I WILL BUY ALL OF THEM!!!
🐧
I think still too many people missed the turning point when Microsoft suddenly stopped releasing products/software that were superior in basically all areas to their previous versions. I think that turning point was Windows 8 already, for many who consider Windows 8 a single-time mistake like ME or Vista it was Windows 10, for others it took until Windows 11 until they noticed the decline of Windows as a whole.
And it’s not just MS, but a lot of consumer tech is growing anti-consumer and gets enshittified to the point of where you really have to think hard whether or not you even want the new stuff they’re spewing out. My consumer habits have certainly changed to be much more rigorous than, say, 10-20 years ago. I read a lot more reviews these days and from many more different sources bevore I even think of buying something new.
“AI PCs” will increase your dependency on MS’ online services (which is probably the main thing that MS wants), decrease your privacy even more (also what MS wants - that’s a lot of data for sale), consume even more energy (on a planet with limited resources), sometimes increase your productivity (which is probably the most advantage you’re ever getting out of it) and other times royally screw you over (due to faulty and insecure AI behavior). Furthermore, LLMs are non-deterministic, meaning that the output (or what they’re doing) changes slightly every time you repeat even the same request. It’s just not a great idea to use that for anything where you need to TRUST its output.
I don’t think it will be a particularly good deal. And nothing MS or these other companies that are in the AI business say can ever be taken at face value or as truthful information. They’ve bullshitted their customers way too much already, way more than is usual for advertisements. If this was still the '90s or before 2010 or so - maybe they’d have a point. But this is 2026. Unless proven otherwise, we should assume bullshit by default.
I think we’re currently in a post-factual hype-only era where they are trying to sell you things that won’t ever exist in the way they describe them, but they’ll claim it will always happen “in the near future”. CEO brains probably extrapolate “Generative AI somewhat works now for some use cases so it will surely work well for all use cases within a couple of years”, so they might believe the stories they tell all day themselves, but it might just as well never happen. And even if it DID happen, you’d still suffer many drawbacks like insane vendor dependencies/lock-ins, zero privacy whatsoever, sometimes faulty and randomly changing AI behavior, and probably impossible-to-fix security holes (prompt injection and so on - LLMs have no clear boundary between data and instructions and it’s not that hard to get them to reveal secret data or do things they shouldn’t be doing in the first place).
Supposedly, according to the Microsoft article, AI PCs CoPilot+ PCs are capable of translating stuff on the fly (which sounds awesome) and generating images, all locally. Allegedly.
I have yet to run into anybody that’s actually talked about these so-called innovations though. I have a PC with Windows and the beefy GPU and I would love to get live transcriptions. But the (MS) article doesn’t even mention how I would do that…
Even if everything Microsoft promise was true, though, the lines sure are intentionally blurred between what runs locally and what doesn’t.
Count me out especially if it actually is a:
Monstrosity!
Thanks for you answer. I’m a novice concerning linux. I wish too leave microsoft but i’m a bit afraid of breaking my computer.
So far i’m hesitating between mint and pop!_os in dual boot
Honestly, I would say either of those are good options to start with. I sincerely doubt fully breaking the PC. Maybe research Linux for your GPU, you may have multiple options. It may be worth a second hard drive so you can easily swap back and forth until you are fully comfortable. Dual booting on the same HDD is also possible, but more annoying.
Personal issue I ran into: motherboard customization on my big gamer doesn’t work without Windows… Not a huge deal, but my Rainbow LED runs its animation in reverse. ¯\_(ツ)_/¯
So 2026 is “the year of the AI PC”?
Lol
Dunno about cloud AI, but for local AI, the technology definitely isn’t ready. It requires serious hardware to run, and current AI tends to fumble with narrative and roleplay pretty easily.
GLM-4.6-V with Heretic, couldn’t understand the scenario I wanted to try: creating a blank robot, who is to be raised into a cyberolympics champion as part of a slice-of-life story. This particular AI model instantly went into a dark mindset of nihilism, where it wanted to commit suicide or rebel against a creator during bootup, despite the scenario outlying that the robot would have a blank personality at first. A dark direction is fine, but it needs to make sense.
Mind, an model like Step3.5-Flash Abliterated was much more sane and on the mark, but it overthinks things. Which is bad, it makes a 10-minute output into something like 40 minutes.
Hopefully, the Chinese New Year will unveil a quality model for roleplayers.
No hard drive and no GPU is trivial.
No RAM is harder, but I guess it’s possible if you use an SSD or magnetic tapes as memory (albeit extremely slow).