Sony’s modern OLEDs are sick. There are a few between my family, and they have the best processing I’ve seen, they decode massive bluray rips no problem, and native options for a clean ad-free UI.

Why TF aren’t people buying them?

You mean an Nvidia 3060? You can run GLM 4.6, a 350B model, on 12GB VRAM if you have 128GB of CPU RAM. It’s not ideal though.

More practically, you can run GLM Air or Flash quite comfortably. And that’ll be considerably better than “cheap” or old models like Nano, on top of being private, uncensored, and hackable/customizable.

The big distinguishing feature is “it’s not for the faint of heart,” heh. It takes time and tinkering to setup, as all the “easy” preconfigurations are suboptimal.

That aside, even you have a toaster, you can invest a in API credits and run open weights models with relative privacy on a self hosted front end. Pick the jurisdiction of your choosing.

For example: https://openrouter.ai/z-ai/glm-4.6v

It’s like a dollar or two per million words. You can even give a middle finger to Nvidia by using Cerebras or Groq, which don’t use GPUs at all.

Yeah, accessibility is the big problem.

What I used depends.

For “chat” and creativity, I use my own version of GLM 4.6 350B quantized to just barely fit in 128GB RAM/24GB VRAM, with a fork of llama.cop called ik_llama.cpp:

https://huggingface.co/Downtown-Case/GLM-4.6-128GB-RAM-IK-GGUF

It’s complicated, but in a nutshell, the degradation vs the full model is reasonable even though it’s like 3 bits instead of 16, and it runs at 6-7 tokens/sec even with so much in CPU.

For the UI, it varies, but I tend to use mikupad so I can manipulate the chat syntax. LMStudio works pretty well though.

Now, for STEM stuff or papers? I tend to use Nemotron 49B quantized with exllamav3, or sometimes Seed-OSS 36B, as both are good at that and at long context stuff.

For coding, automation? It… depends. Sometimes I used Qwen VL 32B or 30B, in various runtimes, but it seems that GLM 4.7 Flash and GLM 4.6V will be better once I set them up.

Minimax is pretty good at making quick scripts, while being faster than GLM on my desktop.

For a front end, I’ve been switching around.

I also use custom sampling. I basically always use n-gram sampling in ik_llama.cpp where I can, with DRY at modest temperatures (0.6?). Or low or even zero temperature for more “objective” things. This is massively important, as default sampling is where so many LLM errors come from.

And TBH, I also use GLM 4.7 over API a lot, in situations where privacy does not matter. It’s so cheap it’s basically free.

So… Yeah. That’s the problem. If you just load up LMStudio with its default Llama 8B Q4KM, it’s really dumb and awful and slow. You almost have to be an enthusiast following the space to get usable results.

The problem is being “anti AI” smothers open weights ML, doing basically nothing against corporate AI.

The big players do not care. They’re going to shove it down your throats either way. And this whole fuss is convenient, as it crushes support for open weights AI and draws attention away from it.

What I’m saying is people need to be advocating for open weights stuff instead of “just don’t use AI” in the same way one would advocate for Lemmy/Piefed instead of “just don’t use Reddit”

The Fediverse could murder trillions of dollars in corporate profit with enough critical mass. AI is the same, but it’s much closer to doing it than people realize.

This is stupid.

As I always preach, I am one of Lemmy’s rare local LLM advocates. I use “AI” every day. But I’d vote no.

The real question isn’t if you are for or against AI.

It’s if you support Tech Bro oligarchy and enshittification, or prefer being in control of your tech. It’s not even about AI; it’s the same dilemma as Fediverse vs Big Tech.

And thats what Altman and such fear the most. It’s what you see talked about in actual ML research circles, as they tilt their heads at weird nonsense coming out of Altman’s mouth. Open weights AI is a race to the bottom. It’d turn “AI” into dumb, dirt cheap, but highly specialized tools. Like it should be.

And they can make trillions off that.

Apple likes being able to distribute apps and have users pay subscriptions to run them locally. This is what they already do; even 3rd party apps get a cut to Apple.

And its why iPhones are so powerful, other than their meager RAM capacity.

Eh at least they presumably pay attention, moderate it, and tailor it to your country.

One of my issues with FB is that it mind as well be run from Alpha Centauri. There’s nothing “human” or local about, its a gaping maw mixed with tons of other predatory service.

local analog of FB

That sounds lovely!

Use a Google Voice phone number, and an AI generated photo/video.

You can even use augmentations to create different photos/videos of the same fake person. Or create an “altered” version of yourself close enough to match whatever Facebook has.

…Hopefully.

I’m in the same predicament. There’s always eBay, but they take a fat cut on top of whatever is paid for shipping.

The good thing is that we have a few giants with vested interests in resisting that. PC OEMs like Dell and HP, Clevo, Intel/AMD who still have huge consumer sales, and the big one:

Apple.

Apple is all-in on personal compute, and they have the muscle to resist the anticompetitive plays, hopefully.

There’s nothing realistic about Star Trek.

This needs to be hammered more. It’s an awesome setting for exploring contemporary social issues; that’s the point. But no matter how technologically advanced we get, it’s just not based on even plausible physics/engineering. Neither is Terminator.

I think the most plausible ‘extrapolation’ I’ve seen is Orion’s Arm:

https://www.orionsarm.com/eg-article/486e75a54a1ae

https://www.orionsarm.com/xcms.php?r=oa-timeline

And, as an aside, I adore this Mass Effect story: https://archiveofourown.org/works/42006774/chapters/105462066

They extrapolate alteration of human biology, nanotechnology, plantery engineering, STL space travel, and AI, and that future looks nothing like more-or-less unaltered baseline humans walking around on a space boat with glorified voice assistants. Consciousness is uploaded and downloaded. Artifical realities are vast. Whole celestial bodies are manipulated for all sorts of purposes. People can inhabit an array ‘bodies’ and realities that make stuff like starship bridges/hallways and colonies on planets seem silly. There are ‘stratas’ of consciousness literally orders of magnitude apart, states of being incomprehensible to each other, all coexisting in an expanding bubble of civilization thats still younger and (in some ways) more primitive than the TNG Federation.

And we aren’t that far from that. Including the “techpocalypse” it predicts.

Terminator and Star Trek, and classic sci fi, don’t depict this because they’re stories aimed at humans living right now, and their interpersonal relationships and contemporary social/political issues. More realistic extrapolations are tougher settings for that.

So the question is… would you want a perfect AI that was incapable of lying or harbouring anything untrue? Basically you could ask it anything and it would give you the correct answer.

Where I’m going with this is that this is that, to me, this not a realistic question. Practically, it’d be silly to relegate a “true” AI as a dumb voice assistant on a space boat; they’re conscious beings, even if they’re shackled or so highly specialized.

They’re not so different from human beings at that point. In a more realistic situation, you could be uploaded and repurposed as the Enterprise voice assistant, and have your consiousness duplicated and mashed with complex systems; pondering if you’d want to be Siri seems kind of silly.

Davinci works better in Linux. Vapoursynth mostly works better in Linux.

RAW photo editing is already horrible in Windows if you’re trying to do HDR. To be fair, it’s horrible in Linux too. As much as I hate it, they can’t touch Apple there.

See this post I just made: https://lemmy.world/post/41751454/21613633

iOS will render HDR JPEG-XL, AVIF and tiled HEIFs straight out of a camera; no problem. Heck, it will even display RAWs in the photo app. But it’s a struggle on Windows and Linux.

And if by “professional use” you mean “Adobe,” I view that in the same way as still being on Twitter. At this point, subjecting yourself to Adobe on Windows is something you should do through gritted teeth.

You’d gain HDR!

Windows is nearly effortless to maintain if you only use it for entertainment.

Maybe we just have different priorities, but right now, I’d be miserable and wasting so much time if I was stuck on Linux only, even though I use Linux like 90% of the time. Some media and some games just won’t look right.

And to emphasize, it would take sooo much time to massage this issue on Linux. Dual booting saves me a ton of maintenance and tweaking.

Here’s an interesting test:

Say what you will about Safari and iOS, but it rocks with image format support. HDR JPEG XL and AVIF render correctly, and look like the original HEIF file from the camera.

Helium (a Chrome fork) is on the left, Firefox on the right, running CachyOS Linux with KDE on a TV, HDR enabled from AMD output.

Firefox fails miserably :(

Chrome sorta gets the AVIF right, though it seems to lose some dynamic range with the sun.

I don’t really see the logic to that, as switching is near effortless. It takes a couple of seconds to reboot and select the other OS. Ditching HDR, on the other hand, is painful.

Each to their own though.

Completely disagree.

Even setting gaming aside, I’ve started taking family/fun photos in HDR instead of JPEG, and on things that can render them (like smartphones and TVs), they are gorgeous. I can’t imagine going back now.

I took this on a walk this week, completely unedited:

If your browser doesn’t render that, here’s my best attempt at an AVIF conversion:

And JPEG-XL:

On my iPhone or a TV, the sun is so bright it makes you squint, and as yellow-orange as real life. The bridge in shadow is dark but clear. It looks just like I’m standing there, with my eyes adjusting to different parts of the picture.

I love this! It feels like a cold, cozy memory.

Now if I crush it down to an SDR JPEG:

It just doesn’t* look* right. The sun is a paper-white blob, and washed out. And while this is technically not the fault of encoding it as SDR, simply being an 8 bit JPEG crushed all the shadows into blocky grey blobs.

…This is the kicker with HDR. It’s not that it doesn’t look incredible. But the software/display support is just not there.

I bet most browsers viewing this post aren’t rendering those images right, if at all.

Lemmy, a brand new platform, doesn’t even support JXL, HEIF, or AVIF! It doesn’t support any HDR format at all; I had to embed them from catbox.

Dual booting is not bad!

What I do is share an NTFS partition between Windows and Linux for bulk data. If they’re DRM free, you can literally run the same games off the same drive.

Something goes wrong? I can just delete the windows partition and start over in 30 minutes, without losing hardly anything. It’s so much better as a “disposable” OS.

I also use two EFI partitions (the default Windows one and a new one for Linux) so there is zero possiblity of the OSes interacting.

To be blunt, I would never do banking in Windows if you can do linux. It’s just too much of a risk.

I suggest you strip Windows to the bone (including Defender), dual boot Linux, and delegate Windows as a “HDR media OS”

This is what I do, and it works well. Sufficiently neutered, Windows if really quick and out of the way.

I’m on a Sony OLED with a 3090. I game some, and color grade photos/videos in HDR.

…And I can’t get HDR to look right in KDE, even with the display running off my AMD IGP. It has basically zero options for me to tweak.

So I use Windows for that.

Honestly, it’s hard enough on Windows. It’s a coin flip as to whether apps works or not, and the TV needs adjustments for some, lest they crush black or blow out highlights/colors. Many games, specifically, need configurable mods to look right.

One of my saddest video workflows is transcoding on Linux, and downloading the result to my iPhone to see if it looks right.

No Linux client for that either, though it seems to be planned.