Onihikage

You’re entirely correct, but in theory they can give it a pretty good go, it just requires a lot more computation, developer time, and non-LLM data structures than these companies are willing to spend money on. For any single query, they’d have to get dozens if not hundreds of separate responses from additional LLM instances spun up on the side, many of which would be customized for specific subjects, as well as specialty engines such as Wolfram Alpha for anything directly requiring math.

LLMs in such a system would be used only as modules in a handcrafted algorithm, modules which do exactly what they’re good at in a way that is useful. To give an example, if you pass a specific context to an LLM with the right format of instructions, and then ask it a yes-or-no question, even very small and lightweight models often give the same answer a human would. Like this, human-readable text can be converted into binary switches for an algorithmic state machine with thousands of branches of pre-written logic.

Not only would this probably use an even more insane amount of electricity than the current approach of “build a huge LLM and let it handle everything directly”, it would take much longer to generate responses to novel queries.

Up in the Hardware Information section of hyfetch, on the left.

Webtoon is still shitty in other ways. When they adapt a property, they want it their way, regardless of the author’s original vision. I’ve seen several stories that originated on Royal Road get Webtoon adaptations, and the adaptations always seem to change or leave out important parts of the story, making characters look stupid or just completely replacing entire sets of characters, forcing the story to diverge substantially when inevitably something they got rid of turns out to have been critically important to where the author was taking things. They turn great stories into middling slop every single time.

Not them, but I do! https://youtu.be/s1fxZ-VWs2U

Unfortunately I can’t even test Llama 3.1 in Alpaca because it refuses to download, showing some error message with the important bits cut off.

That said, the Alpaca download interface seems much more robust, allowing me to select a model and then select any version of it for download, not just apparently picking whatever version it thinks I should use. That’s an improvement for sure. On GPT4All I basically have to download the model manually if I want one that’s not the default, and when I do that there’s a decent chance it doesn’t run on GPU.

However, GPT4All allows me to plainly see how I can edit the system prompt and many other parameters the model is run with, and even configure multiple sets of parameters for the same model. That allows me to effectively pre-configure a model in much more creative ways, such as programming it to be a specific character with a specific background and mindset. I can get the Mistral model from earlier to act like anything from a very curt and emotionally neutral virtual intelligence named Jarvis to a grumpy fantasy monster whose behavior is transcribed by a narrator. GPT4All can even present an API endpoint to localhost for other programs to use.

Alpaca seems to have some degree of model customization, but I can’t tell how well it compares, probably because I’m not familiar with using ollama and I don’t feel like tinkering with it since it doesn’t want to use my GPU. The one thing I can see that’s better in it is the use of multiple models at the same time; right now GPT4All will unload one model before it loads another.

I have a fairly substantial 16gb AMD GPU, and when I load in Llama 3.1 8B Instruct 128k (Q4_0), it gives me about 12 tokens per second. That’s reasonably fast enough for me, but only 50% faster than CPU (which I test by loading mlabonne’s abliterated Q4_K_M version, which runs on CPU in GPT4All, though I have no idea if that’s actually meant to be comparable in performance).

Then I load in Nous Hermes 2 Mistral 7B DPO (also Q4_0) and it blazes through at 50+ tokens per second. So I don’t really know what’s going on there. Seems like performance varies a lot from model to model, but I don’t know enough to speculate why. I can’t even try Gemma2 models, GPT4All just crashes with them. I should probably test Alpaca to see if these perform any different there…

I actually found GPT4ALL through looking into Kompute (Vulkan Compute), and it led me to question why anyone would bother with ROCm or OpenCL at all.

Rather than replace, I’d guess it could be an alternative. The best part of it will be the ability to natively disable the top tab bar instead of needing a janky userchrome css hack that doesn’t even fully work.

PCIe gen 5 is for the PCIe slots and NVMe storage slots, but they’re backwards compatible; you can put a gen 3 component in a gen 5 slot and it will work at gen 3 speeds. Similarly, if you put a gen 5 component in a gen 4 slot, it will be limited to gen 4 speeds. Right now there’s very little appreciable difference between gen 4 and gen 5 unless you’re spending a lot of money on the component (GPU/storage). Another thing to note is that Gen 5 requires that both the CPU and motherboard support it; a CPU with gen 4 support in a gen 5 motherboard will limit all the slots to gen 4 speeds.

RAM is a totally different standard that must be matched exactly for what the motherboard has; if it’s a DDR5 motherboard then you have to use DDR5 RAM or it won’t even fit in the slots. You can get a PCIe gen 5 motherboard and just use gen 4 SSDs or GPUs, that’s perfectly fine and leaves you room to upgrade later.

Seems mostly fine to me, I game all the time on Linux (Bazzite gang 🤘) with a 3900X + 7900GRE, haven’t had any significant issues aside from needing to make sure clock speeds were configured correctly on the GPU. Two ram sticks is the way to go with these systems as sometimes they don’t support 4 sticks at full speed.

You’re right that GPU passthrough is definitely more for tinkering or advanced users with very specific needs (usually professionals who need Windows/Nvidia and choose to run it in a VM rather than dual-boot), with a budget to match. For a gamer couple, having fully separate systems is going to be much less hassle and more resilient against failure.

The one thing I would recommend changing is the power supply, it’s unironically the most important component in the computer because if it fails it can kill everything else, and the System Power 10 is known enough for being low-quality that discussions of that come up in web searches. Poor quality power supplies can damage your hardware and otherwise cause weird, intermittent issues even if everything seems to work fine most of the time, and will fail and shut off the computer when a good power supply would have just kept on chugging. Seasonic and Corsair are considered the best brands and have 10 year warranties - they’re more expensive, but they’re worth it. You want 80+ Gold or better these days, this is a buy once, cry once component.

If you don’t have a UPS, I would also recommend getting one at some point, either one big shared unit (if they’ll be close together) or two individual units. Having backup power will allow you to shut down the computers gracefully during a power outage, and prevents the worst-case scenario where the power goes out while the computer is installing updates and it turns into a brick.

At least their username is accurate!

I mainly recommend Universal Blue distros to newbies, like Bazzite or Aurora. The immutable nature more or less means users don’t have to worry about performing maintenance of system apps like they might on some distros, mostly don’t have to worry about dependencies, and are less likely to irreversibly break the system themselves or in an update.

That said, these distros are Fedora-based, and I think that’s fine. No idea who out there is recommending Arch of all things.

He did at the beginning, but he helped them get what they wanted in the end, and I think that counts for something.

“We’re thankful that the Biden administration played the long game on sick days and stuck with us for months after Congress imposed our updated national agreement,” Russo said. “Without making a big show of it, Joe Biden and members of his administration in the Transportation and Labor departments have been working continuously to get guaranteed paid sick days for all railroad workers.

“We know that many of our members weren’t happy with our original agreement,” Russo said, “but through it all, we had faith that our friends in the White House and Congress would keep up the pressure on our railroad employers to get us the sick day benefits we deserve. Until we negotiated these new individual agreements with these carriers, an IBEW member who called out sick was not compensated.”

Counterpoint: Scumbag companies ninja-editing their timestamped warranty page such that the only way you know they edited it after you bought the product is because it was archived previously.

Archives are ideal for identifying sneaky behavior like that. You never know when an admin might have the ability to delete or edit something without anyone noticing.

The ELI5 for Fedora’s atomic desktops is that if Windows had an Atomic Desktop version, Program Files and most of the Windows folder would be read only, and each program you installed yourself would go into its own folder in your user directory. That’s the basic idea. It’s harder to screw up an Atomic system as long as you stick to containerized app formats like flatpak/appimage whenever possible. It makes it easier for everyone to diagnose problems, and easier for users to roll back if an update has problems. Even if you were to install it right now, you could use one simple command to “roll back” to any image from the last three months.

The benefit of Bazzite is you have all of the above, plus a lot of gaming-related stuff preinstalled which, if you were to install them yourself in a normal Fedora environment, you’d likely have to spend a lot of time just learning how they’re supposed to be configured, how they interact, which versions have problems, and how to troubleshoot problems when an update to one app breaks a prerequisite for something else; eventually you end up in config hell instead of actually using your computer. With Bazzite, the image maintainers are the ones in config hell - they work out the kinks, app versioning, communicate with upstream to fix issues, all that, so your system should be in the most functional state that a Linux system can be, so you only have to think about using your apps.

tl;dr

• Atomic Desktops are more resilient to randomly breaking from updates or user error, and are easier to revert to a prior state if problems do arise
• Bazzite is a custom Atomic image with lots of gaming stuff preinstalled and preconfigured to work properly out of the box
• If you’re a gamer and wanting to try out Linux, Bazzite is going to be the least painful way to get your feet wet.
• Immutable distros are excellent for daily driving. I daily drive one myself!

MPV is a video player.

To be fair, even in my family it’s not a full streaming replacement. We have Discovery+, Nebula, and (free) YouTube. Live TV from the Roku player is the main thing I want to replace through IPTV, either Jellyfin or maybe Kodi, but both the metadata and functionality of free sources is a crapshoot. If I could replace the Roku live TV use with some inexpensive paid IPTV source, then I could easily switch to any streaming box brand, like ONN or some other generic Android TV.

My setup is not recommended, honestly. Old gaming PC from about 14 years ago with a couple extra hard drives, thrown in the closet with stripped-down Windows 10 on an old SSD, desktop version of Jellyfin, and an external drive for backups. Not even running in a Docker container because the CMOS battery is dead and getting to it is way too much of a hassle on that particular motherboard, so virtualization defaults to off whenever it completely loses power. Which it unfortunately does on occasion like winter storms, or summer heat, or if the wind is blowing.

But hey, for the movies and shows we have on DVD/BD, as well as the music we’ve bought over the years, it does work for access from PCs and phones on the local network (Finamp + Jellyfin Media Player). I dabbled with IPTV for live TV replacement but found that only using totally free IPTV+metadata would take either much more work on no-virtualization Windows 10 than I’m willing to put up with, or have much more jank than my family is willing to put up with.

If “nearly every app” that people already use suddenly has a big warning on it, people will quickly decide the warnings are meaningless and start ignoring them, like Prop 65 warnings. Congratulations, we’ve moved the needle backwards.

You have to meet people where they’re at. I finally switched to Linux when MS introduced a feature I wanted no part in (Recall AI), but I would have given up within a day or two if the transition hadn’t been basically seamless. I was able to pick up right where I left off, using all the same apps I did on Windows except MusicBee RIP, but now I’m in a better position than before, on an open-source OS instead of closed-source. Now there’s a little less friction between me and better, freer software.

When I look at Firefox in Discover, it only shows the list of permissions the flatpak will be given out of the box, with no warning of it being “potentially unsafe.” This certainly does seem like the better way to handle it.

Also, the warning on the Flathub website is clickable - it expands into the full permissions list. Why it defaults to “no information except maybe dangerous” is beyond me.