This article reads like AI slop.

Tuxedo is continuing work on a Snapdragon X-Elite laptop but haven't released it yet because Linux support is lacking around the SoC. They've made some device tree commits around it to the kernel. That's it. Saved you a click.

Get out your popcorn because this should be fun to watch. They're already vibe coding all of the value and stability out of their OS.

As someone who only still has a Window install because Wine can't handle the CAD tools I rely on, I look forward to the day when Linux becomes a more attractive platform to release professional software for. I'm not holding my breath for the Year of the Linux Desktop but I can certainly enjoy the ride of MS's self sabotage to get there.

You might be able to get by with the power supply but 750+ watts is recommended for a 3090. Any B550 motherboard should be totally fine.

For what it's worth I'm running a 5950X, 3090, a pair of water pumps, and 11 fans all off a 750 watt Corsair PSU and I've never experienced any instability or signs of an insufficient PSU.

If it were me in your situation, I'd just pick up a decent 750-800 watt PSU, and upgrade the memory to a decent 32 gig memory kit while I'm at it just to have a little better time. 3600mhz with decent timings seems to be the sweet spot. Otherwise, as long as the GPU fits the case, you should be in great shape to run it.

Thanks for all the work you're putting in on this and for taking the time to honestly express your views on AI and development.

I'll be the first to admit that in general I'm an AI hater, and think that in the big picture this current cycle has done a lot of damage to society in terms of wasted resources (natural and financial) and accelerated us all into a place where what we see and hear simply can't be trusted any longer. It's gross to think of the hundreds of billions being spent on it while people are starving and homeless.

At the same time, I also see that some of the latest models are incredible tools for software development, and have made it possible to get some things done that are sorely needed in OSS. It's my understanding that a lot of the open source models that can be run on affordable hardware are not that far behind the big expensive frontier models. While I really do believe that the current OpenAI/Athropic rise is a bubble that needs to pop sooner than later to prevent greater damage to society, I also think there's no sense in wasting a good tool.

The fact is that Pandora's Box has been opened. Even if the big players all shut down development tomorrow, the tools are still out there and will still be used. For better or worse, outright bans on LLM-generated code will only cause the people using the tools to try to hide what they've done.

My sense is that there's a growing distinction between "slop" and useful output (for lack of a better term). It's easy to churn out slop that might get the job done while making a mess of things, but a capable human in the loop directing things and working to achieve the goal with top quality will likely always make the difference. Software development is particularly well suited to this because of its iterative nature.

In our increasingly polarized society with increasingly short attention spans, we all want simple black-and-white answers to things. I hope that developers, maintainers, and enthusiasts can come to appreciate the nuance of the situation. And I think your post describes that nuance well.

Take care of yourself, and I wish you the best in your endeavors.

While I totally agree that there’s no way a family is living in a tiny one bedroom home and the term is deceptive, “single family home” is the zoning term.

I tried out Lightyear Frontier and Homeworld 3.

Lightyear Frontier made for a generally pleasant low-stakes low-stress first person farming game. The mech aspect had some minor jank in places, but nothing I think the developer can't smooth out with some minimal adjustments. Seems like a good game to relax with.

Homeworld 3 looks beautiful. I played through the tutorial and it's got me hyped for the full game. Only complaint was that there isn't a proper single player level to the demo. It appears to just be a tutorial and skirmish/online, and I'm not a competitive multiplayer guy.

Least corrupt thing about Clarence Tomas.

That's what baffles me in this kind of situation. The dev basically did their job for them, and now they're leaving happy paying Linux subscribers out in the cold. Just hire the clearly passionate and talented developer and make the product offering that much better!

Now when I look for Qobuz apps for Linux, all I'm seeing are weird little download utilities to essentially steal music from the service. Gabe Newell sure had it right.

I think the best part of this is that there's no apology or even the slightest acknowledgment of what a shitshow Windows has become. Just "listening to feedback" and "improving the experience".

He was already fired because of the incident. I'm not holding my breath for meaningful legal consequences, but at least he's not getting paid leave during the trial.

I’ve got an R3 at home which generally works well. Flashing mainline OpenWRT was pretty smooth and easy. It’s been a while since I did the bring up, but I do remember having to jump through some hoops to get a partition layout that would utilize the onboard storage properly. By default it only left 10mb to install additional packages which seemed to defeat the purpose of having all of that emmc available. That may have changed in the more recent releases.

One bug I encounter regularly is that some (maybe older?) Apple devices seem to be able to lock up the router. Adding watchcat can get the thing rebooted in less than a minute in the event that it does hang, which makes it barely noticeable, but it’s not an ideal fix.

Depending on the devices you have in your house that might be a showstopper or of no consequence at all. Otherwise WiFi speeds and signal are great, as are general performance and reliability except for that bug I mentioned. Haven’t used VLANs but it’s all there and the flexibility of OpenWRT is great.

So it's not a total loss then?

Nvidia GPUs seem pretty solid with the latest driver developments around Wayland explicit sync now. I haven't had the experience with a modern AMD GPU to compare, so hopefully others can chime in there. I think AMD stability depends on which driver you use.

As for AMD CPUs, I don't think you can really go wrong if you go with the x6xx class or above. The X3D variants are supposed to have a bit of an edge in games but a very slight penalty in other general computing tasks. Doubtful it would be anything perceptible other than in benchmarks or compile times. If your system is primarily being built for gaming, I'd say opt for an X3D part if all other things are equal. If it's more of a workstation kind of system, or you plan to leverage a lot of virtualization tech, I'd say spend as much as you're willing to stuff as many CPU cores into that machine as you can.

As for memory, more is always better. I've got 32 GB on my main system and never felt like there was anything it couldn't handle, and that's even being somewhat sloppy leaving other fairly memory intensive programs running in the background while I game. In that department, I'd just go with 32 gigs and call it a day, unless you're doing video capture/editing.

Nvidia Gsync shouldn't be an issue in Linux, at least not with an Nvidia card. If the monitor doesn't also support Freesync, then that might cause a hassle with an AMD card.

Good luck and have fun!

I’m not much of a programmer and I don’t host any public sites, but how feasible would it be to build an equivalent of Night Shade but for LLMs that site operators could run?

I’m thinking strategies akin to embedding loads of unrendered links to pages full of junk text. Possibly have the junk text generated by LLMs and worsened via creative scripting.

It would certainly cost more bandwidth but might also reveal more bad actors. Are modern scrapers sophisticated enough to not be fooled into pulling in that sort of junk data? Are there any existing projects doing this sort of thing?

The Litter Robot 3 is a well built and well designed workhorse. I think they cost a little more new than your budget, but they may have a model right around 500.

Depending on your comfort level with the idea and where you live, you can sometimes find them on the used market for a couple hundred dollars, or even occasionally a free non-working unit. Replacement parts are easy to get and install.

They have a WiFi model that can use an app to alert you when the tray is full or something unexpected happens, but it works just fine without the app or internet.

Not a definition of intelligence, but something I often come back to is what it does at rest.

To the best of my understanding of how LLMs work, if you're not feeding the thing inputs, the program is doing absolutely nothing. It isn't curious and doesn't seek new information or stimuli. It effectively is just on pause until the next prompt. There is nothing akin to an inner monologue or thought process happening in absence of what is fed to it.

I'm not up on all of the epistemology or theory of mind type stuff, but I'd imagine some might argue about that distinction in humans; we are always processing input in some form as the input from our senses is never shut off until we're dead. But I think the "at rest" case is a relevant distinction that does get at motivation and real thought born of real intelligence.

In addition to what others have said, it can also indicate different issues on different printers. Is it a core-xy or bedslinger?

If it's not just an issue with a dished bed, there could be twist along your x-axis which are moving the nozzle (or probe) slightly further from the bed in the middle. Could be wire/bowden routing putting a slight strain on the toolhead.

Do the mesh results look roughly the same when cold versus, say, a 30 minute heat soak?

Check out the Banana Pi BPI-R3. It’s getting mainline OpenWRT support with the 2023 release (RC builds available now). Amazing performance to price ratio. The only thing on your list that might be an issue is the 5Gb SFP. It has two SFP cages but I believe they max out at 2.5Gb. Also recommended to get the bundle with case and antennas included. Amazing price for what you get.

Heard this discussed on Knowledge Fight but haven't seen the twitter space call myself. It's not something I particularly want to seek out. But with Knowledge Fight being a credible source, I don't doubt it.

Assuming they did in fact appear on the same space together while talking to and sometimes over each other: either the Muskrat previously put on an elaborate performance to give his sock puppet credibility, or his biggest stan sounds almost just like him.

I don't think it's out of the realm of possibility that one of his worshipers would go so far as to adopt his speech patterns and try to sound like him. Or it could be a sock puppet run by his brother or cousin.

Every possibility is extremely pathetic, and I'm curious to see what shakes out.

Helium just used the same radio technology as Meshtastic (LoRa) and that's about it. Helium was some lame attempt to develop a mesh network tied to a cryptocurrency as incentive for users to deploy nodes. AFAIK Helium is basically over.

Meshtastic is free and open. The hardware and tiny amount of electricity are your only costs to get involved. A lot of devices are efficient enough to run on a 1 watt solar cell and single 18650 battery.

The reliability will depend on where you're located and how many nodes are running in your area. If you happen to be in a particularly dense area, you might also look into MeshCore, which is essentially the same idea but designed to scale a lot better.