¶ Intro
Hello, friends, and welcome back to your weekly Linux talk show. My name is Chris.
My name is Wes.
And my name is Brent.
Hello, gentlemen. Coming up on the show this week, open source had a typical week filled with drama. Flathub tightened the gates, Gnome Circle put AI Slop on notice, and R-Sync somehow became a battleground. We'll round it all up and then give you our take, and then we'll round it out with some very nice boosts, some very positive picks, and a lot more. So before we get any further, let's say time-appropriate greetings to our sole virtual lug contributor this week. Hello, Otterbrain.
Hello. We also have some folks up there in the quiet listening. It's nice to have some folks in the live chat, too. Nice to have you on board for the show today. And thank you for being there. We have details about our Mumble Room up on our website if you want to join that. Also, good morning to our friends over at Defined Networking. Go meet Manage Nebula at Define.net slash unplugged and try it out for free on 100 hosts, no credit card required.
It's decentralized. It's built on the Nebula platform that we love. There's no big tech login sitting in the middle. There's no black box relay deciding how your network works. It's fast, it's secure, and you directly connect under your control.
¶ Sloppy Submissions
Nebula was originally built for Slack to connect their global infrastructure, and it shows. The performance is excellent, the design is resilient, and with the optional self-hosted lighthouse nodes, you can control the critical pieces yourself. If you just want two machines or you want a whole global infrastructure, it'll do it. And you can try it out for free with 100 hosts, no credit card required, defined.net slash unplugged.
Go redefine your VPN experience. It's so good. I have it throughout my infrastructure now and I love it. Go try it, defined.net slash unplugged.
It has been a spicy week in open-source AI land. Flathub tightened the rules, Gnome Circle put AI slop on notice, and rsync-saha became the battleground for whether AI-assisted code is genius or a garbage fire. And that's the weird part because while some projects are slamming the brakes, others are using these tools to find real bugs, land real fixes, and make maintainers just a little bit more dangerous. So let's start with rsync.
Let's start with R-Sync indeed, because I think this was the one that put it over the top for us this week. We said we want to talk about this. I think we have something to add here. Everybody knows the R-Sync project. I think it's an example of one of these beloved tools. And this is an example of AI-assisted coding coming to the tool that you use. And the headline is that the original creator of R-Sync came back after 18 years earlier this year and started triaging.
I think technically 2024, but they hadn't done much work until this year. They really started picking things up.
Right. Like April and March is really when the patches started coming in. And they did so with Claude's assistant, identifying several security vulnerabilities, but also in the process of these major rewrites of the syscall layer, seemed to introduce several regressions. And there has been a massive backlash to regressions being introduced to rsync, and accusations of AI slop.
The R-Sing founder has lost touch with reality. I mean, I think this thing really kicked off with a post on Mastodon that... Posted by they always Jeremiah Fletheaven. How do you suppose you say that last one? I'm sure a field haven. Thank you. I cannot read it from here. Jeremiah wrote. So my systems recently updated to our sink 3.4.3.
And as soon as that happened, my backup system, which does incremental backups using multiple dash compare destination equals arguments, started to fail on anything but a full backup reverting to our sink 3.4.1. And now it works. So I go look at the source in GitHub to see what might have changed, because there doesn't seem to be anything relevant in the changelog. Since 3.4.1, 36 commits by Tridge and Claude. Oh, for F's sake.
And that was on May 28th.
And so since then, a firestorm of criticism have kicked off. And poll requests against the projects, people saying they're going to fork rsync.
Yeah, we have seen at least a few sort of forks or alternatives pop up already.
So let's break down what's really going on here, because I think the headline noise is distracting from probably the pattern that is actually just there in the commits. And Wes did a deep dive. We put some stuff together, and I think the story is a lot more understandable than it's currently being interpreted online. And I think this is maybe where we can add some value.
So Tridge is the original creator of R-Sync, took about an 18-year hiatus, seems to have recently discovered AI-assisted tools, and begun going through the code base, triaging bugs and security issues that may have potentially been there for quite a while. And it looks like when you go through them, there was maybe a total of 77 issues that were worked on. And of those 77 issues, we've seen some PRs come in, maybe a total of nine regressions out of the 77.
Yeah, something like that. I mean, these are all like a static snapshot of when I was kind of looking at the data.
Of course. That gives you an idea of the volume, right? We're looking at maybe 77 PRs that were worked on and maybe nine regression issues.
Yeah, and it's, so it's like, you know, he started maintaining things. Interestingly, Trich is also, I guess, was involved in reverse engineering BitKeeper back in the day. That ended up being why Linux needed to start using Git, if we remember that story from long ago. So a long history in open source and being involved with things. And so it takes over in 2024, doesn't really touch things much, does do some small things here and there, starts picking up in 2025, 2026.
but is just doing work by hand. And actually, a lot of the stuff, like secure relative open, which is involved, which is like a real implementation of doing open in a better way inside the code base, which is involved with some of the breakage. Some of this, a lot of the actual like fixes and redesigns happen before any of the clod commits show up. Sort of like over a 22-month period kind of slow pecking at it.
I mean, that doesn't surprise me. He's clearly a very talented developer and has been for many, many, many years.
And then... More recently, we see this big uptake in velocity in the codebase and a lot of commits co-authored by Claude. And then when you look at it, a lot of it is kind of applying the patterns that he had seemingly found by himself in the era before, or at least before we start seeing the co-authored commits, and then using Claude to go apply that across the codebase, as well as do a bunch of sort of like other tests and docs and just sort of like that kind of fix.
So if you look at the commits so far, it's not like Claude is doing a big structural rewrite of the codebase.
From scratch.
It's sort of applying some patterns, doing security fixes, responding to CVEs. Because there's also like six CVEs that have kind of come up in our sync through this time. And that's been some of the work that's been happening as well.
Yeah, I have to imagine they're probably getting more of those. Right.
And that's where, like, there's a few, like, one story could be, oh, gosh, you know, maintainer starts using AI. The project's going to slop. You know, we can't, what's happening? Just that whole thing.
it can also be partly that like here's a person who hasn't maintained this code base for most of its existence he's created it and used it maintained it for the first part and then handed it off to the the main maintainer for most of that time uh who's now getting more involved and wanting to do more releases and sort of the both the development speed and the release cadence here is picking up and then i think there's probably some very reasonable questions you can ask about like is that
being communicated what are the expectations of the users right there's like a lot of sort of regular open source project maintainership and maintenance questions that apply. And then on the other side of all that, there's also the question from the CVEs. One of the forks is sort of a minimal implementation of R-Sync's protocol in Go.
And the point from that author is some of this is even just like, it brings up this issue of like, should we even be depending on R-Sync and C in this era when we know that like these six CVEs that kind of brought up a lot of these, some of these issues wouldn't even have occurred in a language like Go or Rust.
Okay.
So there's a lot of things you could miss if you just focus on the clod co-committed part.
Okay. So he finds the bugs themselves with a code audit. Then he designs a fix that he implements using clod code. And he tags everything and transparently discloses every commit that was used with clod code.
As far as we know. I mean, you know, we can't really tell.
I suppose, yeah.
But it seems like the intention is to be transparent. At least he has opted into using that.
And so we did get, I mean, there were security fixes that happened here, but we also got regressions introduced at the same time.
Yeah. And that's pretty normal.
That's a pretty standard thing to happen in software development.
And it's totally fair to be like, this regression caused me issues and pain and it's upset and you can have broken trust around, I expected you to have better QA before the ship. Those are all fine.
Well, some of these things are like, it broke on Darwin, right? A specific thing broke on Mac OS.
There is also, it's worth trying to look into, like, what is the actual breakage and how widespread and is it an edge case or is it a major thing that you should have had better?
Well, and then there's the fact that a lot of distros gate the releases, so it's not like everybody is just running off of upstream R-sync point releases either. So the actual distribution of this, you've got to be pretty close to the faucet for this to even impact you.
Right. And you also then need to not have a way that you can roll back easily.
Right. So, I mean, these are some, I think, of the ground rules to understand with this issue. Now, did the regressions actually get introduced? Yes, but I would argue that when you're looking at the type of work that was being done, you're essentially raising the walls on what were lower standards 18 years ago for syscall securities and things like that.
You're improving the internal security of rsync, and that's always going to cause breakage, even if you're hand-coding that, especially for that type of stuff and an 18-year-old project like that. So I think it's really unfortunate that they've really turned on Tridge because I think it's really great to see that he's back and that he's contributing again.
It does seem to be addressing the regressions as well. I do think there's maybe a question there of I don't know if I've seen any sort of particular response to all of this. And we can ask if that's owed or not. That's a separate question. But I think there are maybe questions of what could you do differently if you are going to be doing more maintenance, more active changes, changing how the project is maintained.
Let's talk about that for a second. Because I think one of the things I noticed is just a lot of dropping of code, just a lot of code, a lot of changes all at once across a broad variety of things.
And for a project that has been quite stable, it's kind of faded into background infrastructure. You don't really, besides maybe you filed an issue for some weird little case for your particular system or an improvement or a dock update. You don't really think of R-Sync as something that's constantly changing or you're pulling down lots of new.
Well, for example, Debian stable ships R-Sync from 2022. So a lot of people that are using Debian-based distributions are on a massively older version. So they're never going to see this. So that is true. That's just the reality of it. I think, and the irony of somebody forks it, they're going to vibe code it, too.
That's the other question is, so like one of the forks, I think I have a link in here, is basically just a fork of the project before the first sort of obviously co-authored with Claude commit.
Yeah.
And that's fine, but we kind of functionally already have that in anywhere that has a fork of the repo already, right? You're just rolling back the head pointer effectively. And that's fine, whatever. You get that with stable distributions kind of already. And there are some discussions in like the WMA list of, aren't they going to need to do anything about this? And I guess we'll see where that goes.
When we get to the comments of the Zig creator about Zig's policy on AI, I think it's going to underscore a point that I want to make. But the reality I think we're about to face, And I don't know if we're there yet because I'm no developer, but it feels like we're getting pretty damn close is we're about to cross a threshold or we have already crossed a threshold where the average one shot LLM basic project code is going to be better than a beginner developer.
And it's going to be a better product than an amateur. And I don't know if it's there yet, but it will be because these things just keep improving. And they're improving at a pace that is very impressive because they can just keep training. And some of these shops like Anthropic and OpenAI are very, very intent on making it great at software development. And they're very focused as part of their business revenue strategy on their Cloud Code and Codex projects, respectively.
And I think we are entering this era of either it's about to be there or it is there where it's going to generate better code than a lot of amateurs. And when we get there, we're just going to discover all of these issues that we have never discovered before. And it's going to happen every single week for multiple projects. And people cannot get the pitchforks out and go after developer for using AI to try to fix this. Because next year, these things are going to be even better.
And the year after that, they're going to be even better programmers. and it does seem conceivable at some point they will exceed most proficient developers to some capacity. I don't know when and how, but it seems possible. And when it does, it's likely going to find even more issues in our code. And this problem only accelerates.
I think the other part, the implication of that is, okay, maybe at the start, the co-authored was maybe a hint at, oh, the quality might not be great unless there was real editing and review here.
but that becomes much less of an actual reliable signal it was never a great signal if I'm honest but it becomes even less of a useful signal down the line what I was going to say before is just the real test of a fork is not can you freeze the code base it's can you actually continue to maintain it and offer real things and have your own stability and yeah, So that's where like a frozen fork of our sink is not that interesting to me
necessarily. If it can actually be maintained, okay, maybe there's something there. A go one or a version in non-C language, that starts a more legitimate conversation for totally structural reasons.
I mean, are we just not talking about the fact that there are six CVEs here that are being addressed? Like those are real bugs. And if you just fork off of that, you're going to have these six CVEs.
You have to go keep up with all the, yeah. And there's going to be more CVEs found to your point.
Like we have to fix this stuff. And it's, yeah, it's going to get worse. And the pitchforks really should just be put down and people should just get to work and have some empathy for these developers who are going to have to go through this transition. Now, Flathub is taking a no LLM policy, and it's somewhere more in a middle ground here. They say we have updated Flathub's LLM policy to explicitly disallow AI usage for both the submission process and applications being submitted.
The number of, they'll go on to say later, the number of unpleasant interactions that I have had with entitled submitters acting as if they were bestowing their brilliant software upon us idiots who are rejecting it went through the roof in the last month. I'm tired. As always, we will not be applying this retroactively. So any vibe coded apps which were already published will remain available. So Flathub is introducing a no LLM policy here.
And when you go to the GitHub, you can kind of see the language of the policy and the changes they've made. And there's a question and an answer, sort of as a follow-up. The question goes, so does this mean any application developed with AI tools, even things like Editor Autocomplete, are not allowed to be submitted? And the individual from Latub responds, unlikely to be considered a blocker unless you literally tabbed your entire way through Autocomplete.
So we have a new Flathub policy here that's not super rigid, but it is pretty intense. No LLM-generated applications. I think no LLM-generated descriptions is fair enough for a publishing platform like Flathub.
Sure, reasonable, yeah.
So I think that I can agree with them on.
I do think all of these face the same essential problem of if it's a good description, is it disqualified because it has an MDash and you associate that with LLMs, even if it's correct grammar and reads well?
I know, I miss losing the MDash.
Does it matter if it was generated and then I edited it and I totally approved the entire thing? I said, yes, this sounds like what I would have written, except with slightly better grammar.
Mm hmm.
None of the hard questions are addressed by this whatsoever, in my view, which seems like a worse wording of what they had before.
And probably in practice, a lot of it will be fine, right? I think it's stern language that will probably be somewhat lightly enforced and is probably meant to attack the worst cases of, like, just slop that is trying to throw it over the wall and is not being cared for or maintained or really represented by a human or, you know, anything like any of those other things that you might want in terms of a maintainer relationship with a platform.
Yeah.
And that I understand.
Yeah, when you worst case it, like, I totally agree. Like, something that's just, like, generating slop up onto somebody's GitHub, they're not even reviewing it, and then they're just auto-submitting that to Flathub and generate, like, I can get that. And I could see they could just get inundated with that, too, right? Especially because you could just almost automate the entire process.
But at least when I first tried it, too, it seems like it sort of then also privileges proprietary software, because you don't know. You're not even trying to audit that.
That's a really good point. That's a really good point. How does this strike you, this decision, Brent?
It's a rough one i can try to see it from a couple different perspectives right like some people might now feel quite empowered to be able to create software they could never create before and let's say you create some niche piece of software you're like wow i've always wanted this now i can actually make it and i know there's a tiny niche of other people who would love to use this as well how do i distribute it flat hub sounds amazing uh but i could see it
from the developers perspectives and the maintainers as well as like well how many of these one-off pieces of software do we accept and who you know the whole chain of trust goes into question as well and these are i don't know it's gonna it's gonna be a long time of asking these questions and projects making these kind of decisions just to try to tread some water i i imagine i.
I wonder too where if the line gets fuzzy if say you submit like we're gonna have a pick today that that just rides entirely on top of.
Pipewire.
I mean, it's a front end of Pipewire. The developer's done some work, but the lift is being done by Pipewire. And... It could be that the features it's using in Pipewire have been AI-assisted in development. And so what if you submit an app to Flathub that is not itself AI-generated, but depends entirely on AI-generated code elsewhere in the ecosystem and elsewhere in the stack? I guess that's allowed?
These are hard questions. I don't know.
It's, yeah. And here's where I get a little concerned, just it is what it is, but the Linux desktop isn't known for its incredibly rich, bespoke desktop application ecosystem, right? And we've been getting there, but this is going to tap the brakes on it. And it is what it is. I'm not making a moral decision here, but the reality will be...
It won't be but a year before Microsoft and Apple and Google and Amazon and Samsung and all the others in their various app store platforms are going to spotlight applications that have been vibe coded using their tooling. Right. If somebody shows up at WWDC using a AI assisted generated app from Xcode and it's a good app, Apple is going to spotlight that. Right.
If somebody uses Copilot to create something and it goes into the Microsoft store and it sells, they're going to put it up on a banner and et cetera. All these platforms will do that that offer these tools and they're going to embrace it as an avenue for people to bring software over and they're going to spotlight it while we shun it. And then the irony is, is we're the ones that are application ecosystem poor.
And probably one of the best shots we have, and we've already seen it, they've got Creative Cloud working on Linux, is using AI-assisted tooling to port over and get things working on Linux that never ran before. I mean, that's how we got Photoshop recently, the current version of Photoshop working.
There's also, I think, like Brent was totally on point, like exactly kind of the apps Brent was describing there. are the apps we tend to find and highlight on the show. And there's always great little tiny things. That's like kind of what I go to. I mean, yes, FlatHub can get you Spotify, you know, or whatever those two. But that's what I like about it. That's why we kind of use it to try to find interesting new apps.
And a lot of those apps are more about executing the particular idea than they are about beautiful GTK code.
Well, and that's just it. Because we don't shame an amateur developer with one of their first projects is open source code and they put it out there and it's garbage filled with security issues. You know, as a community, we'll mentor them. We'll work with them if it's a project that's useful and people will contribute patches. But we're not willing to do it in this case, even if it's something that's useful. Like, well, OK, if it's slop, make it better.
Shut up and make it better. Like, that's the solution here, because we have more tools than we've ever had before. That, to me, seems to be the solution. But don't do it in Zig. Zig has a no AI policy. And the Zig creator was on the JetBrains podcast recently. And I think his name is Andrew, right? The creator of Zig?
I believe so.
And he is, Zig's kind of known. Yeah, Andrew Kelly. Zig's kind of known for having a no LLM, no AI policy. And so the interviewer from JetBrains asked him, you know, why is.
And just to be clear, that's for the Zig, like maintaining the Zig language itself. Not using it to make projects, whatever. That's all fine.
Yeah, they have no control over that. But if you want to contribute to the Zig code base for Zig itself, they have a hard no LLM, no AI policy. and they will reject it out of hand, even if it's like a critical security fix. If they smell a whiff of LLM, they'll reject it out of hand. And so Andrew was on the JetBrains podcast and he was asked why Zig is denying all LLM-generated code. So I thought we'd take a listen.
The first reason is just that those kinds of contributions are invariably garbage. People are sending us contributions that have no value whatsoever. However, not only that, they have negative value because they take review time away from the team, which is very limited. We have over 200 pull requests sitting open right now, and those are all waiting for review. And we try to be on top of it as much as possible.
When you have a small number of people in the dev team and you have a large number of contributors, this is always the problem is this bottleneck of review time. So, when we get these slop contributions, they take our review time and then after a few reviews, we realize they have no clue what they're doing. They're just pasting what we say back to the chat and then laundering the chat back to pretend that they're not using chat but we can still tell.
And then at some point, we realize this is never going to be a good quality because they have no idea what they're doing. And so now we wasted everybody's time. All those other people who are waiting patiently, they didn't get a review and the code never gets merged. It's worthless. We like to call it contributor poker. So, The main point of doing code reviews and having contributions, instead of just doing all the work ourselves, is mentorship.
The whole point is that a contributor can become a core team member eventually, or they can become a more valuable contributor.
Okay, I'm pausing here because I want you to remember that he says one of the core focuses of Zig is to be a platform people can learn to develop on. It can be a starting language they can learn to develop and become a ZIG contributor over time.
A core team member, eventually, or they can become a more valuable contributor. And this will help the project because we'll have more people who can contribute to ZIG skillfully. And it will help their resume because they can be a better systems programmer and they can take those skills elsewhere. The idea of contributor poker is that we have limited time. So, we want to notice, okay, who can we invest our time in to help them become better programmers, better contributors for the project?
And who is maybe a drive-by contributor? They're going to send something, they're going to go away. Less valuable to invest in them. And so, people who are using AI, they're always in the second category. It's not worth it to invest in them. They're not learning anything. They're going to join the core team later. Not a chance. For us, this policy just makes sense because.
The Zig project, it's also an education project. That's part of our mission statement, is we're providing guidance and education to students. And so we're all trying to learn, we're all trying to get better at programming. And so people who are sending AI pull requests, these people are not helping this goal. And in fact, I think that they're detracting from this goal. So for our project, I think that the strict no AI policy, it's an appropriate policy.
You know, If I tried to say, oh, only good AI PRs can come in, now I have to be the judge of that, whereas if it's none whatsoever, then it's a very easy policy to enforce.
That does not add up, I don't think.
I agree.
You're already in the judge of what's AI or not. You're just changing from what's good, which is already the metric you wanted to be using, to now you're an expert on hunting for AI smells. And then you just turn the folks, like now it's just a test of, can I write good zig or not, and then just not tell you I used AI to help me with looking up the stuff in the docs.
Right, you turn it into a game of not getting detected.
Yes. And you punish the honest people.
Well, and what he said right there, too, was that they want to be appealing to people that are learning to code. Well, people that are learning to code, they're going to be the most inclined to use some of this assisted tooling.
100%.
So there's a bit of a dichotomy there. So the other thing that's a bit of a dichotomy for them, a chicken and the eggs, or maybe a catch-22 would be the better analogy. Zig is MIT licensed. So that means it's totally legitimate for the big tech industry to use Zig to train AI. And so Andrew was asked how he feels about big tech being able to train on Zig since he has such an anti-LLM AI policy.
Ironic, isn't it? Personally, I have no issue with this. I really firmly believe in the no strings attached gift that Zig is to the world. You know, if someone wants to use Zig for AI training, great, I don't care. That's fine. The fact that these companies are doing things that I don't like, I don't like that they're doing it, but it doesn't bother me that they're using zig i think that the more that zig is being used it just shows that zig is valuable that's the way i see it.
Okay now that seems to be a pretty balanced take i like that take and you know it's probably the best one you have you can have when you have that licensing but uh there's a i think maybe the biggest tell in this last question because this seems to be a common thread that i've come across a lot andrew was asked if he has tried any ai assisted or vibe coding tools i.
I i love computers and I love learning about what people are doing with them. And there's a. A sense of mystery and magic that you can get from reading someone's explanation of a project that they did that took them a very long time and they had to learn lessons and they had to, increase their skill as a programmer and as a user of computers in order to accomplish this goal. And when you read a blog post like this, it's brilliant.
It captures the imagination, it makes you think about what you could do yourself, it teaches you something, it connects you to them emotionally. But I mean, on the other hand, we're seeing people say, oh, I tried this version of Cloud or that version of OpenAI and it works surprisingly well. I'm always hearing people say that AI code works surprisingly well. But to me, this is not the bar that I want to hold software to. The bar that I want to hold software to is uncompromising perfection.
You know, I don't want to be surprised by the absence of a bug. That's a horrible quality bar to believe in. So, I mean, it's just so many people saying, I don't know, I tried coding this app and it kind of works. Okay. You know, it's just so uninspiring. But have you tried Vype coding yourself? I did a call, I did a private call with Richard Feldman, a friend of mine, and he showed me how to use Vibe Coding with Zed and I tried it out.
And I thought that, I think that the technology is fundamentally interesting. What really turns me off is the fact that it's centrally controlled by four companies and they have total control over what it's doing. They have total control over the models. I mean, it's. I'm not going to go from using my own computer and my own electricity to do my code in order to use closed source programming on someone else's computer through the network. This is that I have to pay for monthly.
I mean, some people are paying $300 a month for this. To me, this is an insane proposition. I would never want to give up what I have in order to get, the results of gen ai.
So a.
Lot to react to in there.
Yeah i thought so too um you know he wants software to be immaculate you know the the absence of well i don't think it has a bug should not be the bar but i think that's just sort of a straw man argument because the reality is most software development is oh my god it built and it's running i can't believe it's doing the thing so.
This is actually i think this part i do like the most is i think the, something like a core language or like the core of a database you want to strive for well and it's just the area where writing the code was not the hardest part it's almost all design and thought work versus building out the next crud layer for your business or building a desktop app which a lot of it is like you just have to write mechanical code yeah yeah and so i'm not saying the lms
wouldn't also be useful for what he's doing i'm not that's not my point i'm more just saying there's different you would the advantages and how you would use them are different, And so that is, I think, the strongest part, and the rest of it kind of falls out.
So what was it you reacted to in there? What struck you in there?
Kind of like the rest of it, I guess. That part, I can kind of understand why maybe you want to very carefully craft.
Yeah, from their project.
He mentions review, but it's like the whole thing here is review, right?
Yeah.
That's the question here, is to what extent do we have to and how closely and what guardrails do we put on this code? And so whether or not it is, because before we just didn't have any other way. It was either humans wrote it or it didn't happen. And now we have this new way. And so I think in the sense that that is essentially an argument for a very high standard of review in the core of this thing, that's totally reasonable, right?
And it could be that all of the things that he's calling AI written don't meet that. And then that's totally fine. And that doesn't surprise me, right? Because he has a really high bar, probably does take someone with some taste to be able to craft something that really meets it in a way that maybe LLMs aren't great at right now. That seems totally plausible.
It just seems like that maybe is what we could be talking about versus like maybe there's a lot of stuff where like for one of our picks last night, I left a vibe coding session building a flake for it because I can audit it at the end. I don't need to be involved. I know how to read Nix. I can all test it and build it myself.
It's not particularly complicated.
Right. But maybe I wouldn't want to do that for something that was going to be the core of the business I was trying to build.
Right, right. Yeah, for sure. Brent, also what struck me is there's a common thread generally when people have a negative opinion towards the tooling.
Also, just the glaring hole of, I mean, as a critique of proprietary platforms, hell yeah.
Yeah, yeah.
The big tech. Where is the mention of any of the open weight models?
Yeah.
And also, the reality is, is that it's a competitive marketplace. So if, you know, Anthropic does something weird, you switch over. If OpenAI does something weird, you switch over. And then there is a touch of irony that he's on the JetBrains podcast because you could make the same argument about commercial IDEs versus open source IDEs. So there's a touch of irony there.
I mean, for there was a while where, like, you couldn't really be a professional Java developer without having a JetBrains license.
Yeah, yeah, yeah. That's why I think it's a little ironic. So, Brent, I don't know if you pick up on this comment thread, but a lot of times it seems like the people that are pretty down on the tooling, They seem to go through a process of trying it, and if they give it an honest try, maybe they evolve it a little bit, then it starts solving a problem for them, then they advance the tooling, and then they seem to get kind of more and more impressed by it.
But he kind of took perhaps the worst introduction route, it sounds, possible. He got like on a Zoom call with a buddy who showed him how to do it through the Zed editor. And this is the creator of Zig. Like he showed him like the Fisher-Price version of the tooling. What are your thoughts on that?
Well uh i understand the hesitation from a lot of folks but to understand like a fundamentally new technology like this you have to spend time with it and you have to make mistakes with it you have to like there's this iteration of how you think about the tooling that just takes time you can't get that in one zoom call of a buddy showing you how it works on his computer right and i think you both would agree that our way of using these
tools for each of our workflows is like evolving every single week and only through that do you get the perspectives, of what it's even capable of and the nuance of where it's good and where it's not good and the current status even of the landscape and so the common thread you're you're mentioning is that that we so often see, criticisms of the technology and like really strong opinions from people who haven't really spent the time to try to understand the technology and
to use it fundamentally in their work and so it i don't know in i don't want to say invalidates but it it kind of takes away from the argument slightly because uh there's a chance to understand the landscape better and to have a more intelligible discussion about where it might be useful and not.
I would say the other thing that's sort of kind of a combo with that that can be tricky is if your impression is from six to nine months ago.
It unfortunately.
Might be out of date now and that sucks because it's really hard to stay current.
There's also maybe a output sort of selection effect too if like a lot of what you see is what people post in prs to your project that could be from who knows what setup or sophistication.
Of Mm-hmm.
LLM?
Yeah. In fact, to kind of make a point, let's shift ahead a little bit here, because I think this shows you this really bifurcation of the part of the community that's really digging in and a part of the community that seems to be really accelerating. And before we get there, I just want to say, as from like a macro market whatever perspective, it is sort of fascinating to like let Zig, you know, rock its freak flag and not take any AI or LLM-assisted coding. and let's see where that goes.
That could be really fun and interesting to watch over time and maybe become something really unique and bespoke.
It's already a nice, kind of neat language in place.
So like, I don't, you know, whatever, whatever they want to do, that's their decision. So I just wanted to make that point clear. It's just, it is really useful to reflect on for a conversation of where this is going because it's- Yeah.
That's our, we want to comment on the general mood, not necessarily any of their, yeah, have ownership of your project.
They encapsulate what's going on with a lot of communities and projects right now. Now, FWUPDI on the other end here had an audit done by Mythos and apparently spotted a number of issues, 12, And 12 issues were reported, but two were dismissed as not bugs, and eight were ranked as low severity, and two were considered moderate. But they went through and started cleaning things up, and so we now have FWEPD 2.1.4, which brings many fixes for the bugs that were spotted by the LLMs.
Also includes firmware update support for the ARC Pro B65 and B70s.
As usual, getting good work done over there.
Yeah.
I did think it was kind of interesting here. Hughes says he was surprised over Mythos finding so many issues in the code base. Because it kind of sounds like it didn't find that much in one sense when you hear that list. I think the standards for FWEPTA are probably pretty high, right? It's like a newer code base. It's been very actively maintained. He's an experienced professional developer. So it may be interesting there.
Then we've also been seeing a stack, I guess is being put, of Linux networking fixes coming in. And the pull requests seem to be using AI assistants. Linux networking maintainers have seen a larger fix pull request than typical this late in the cycle. Many fixes spurred by AI and LM coding agents, Michael Arbel writes. Cloud Code and ChatGPT Codex are contributing to the increased patch flow, helping developers find and submit more fixes across the networking stack.
The upside, of course, is much faster bug discovery, especially in something that can be remote listening, and more code cleanup. But the downside is there is more maintainer load and churn, especially late in the Linux 7.1 cycle.
And we do know that they don't love late cycle changes.
Right. And then GitHub Copilot of all of them and CloudCode helped with numerous graphics and Wi-Fi Linux driver issues in just the last week. GitHub Copilot and CloudCode helped generate and co-author other large batches of Linux kernel fixes during the 7.1 RC5 cycle. AI-assisted fixes touched a wide spread of kernel areas, including the Intel Z, Raspberry Pi Video, AMD Display Code, SMB, NetFilter, SysFS, IO Ring, Bluetooth, and more.
The kernel is also using disclosure norms like assisted by AI tags, making it easier to track where AI has helped while keeping the human batch process visible.
Yeah, that's going to be interesting to watch over time.
Yeah. So the humans are still involved. They can see which ones are committed by AI and they're accepting them. And we're seeing a high velocity of fixes in Wi-Fi drivers and in graphics drivers. In the networking stack in general, there's also been file system fixes that have landed recently.
See, this is kind of touching on what I was trying to get at is there's like different kinds of structures of code, right? And abstraction and like how general or how widely used they are. And so it'll be interesting to watch where you see more or less sort of first use as LMs progress and as more folks use them. And it makes total sense, I think, in that model that drivers and existing drivers would be in one area that makes total that would be a good fit.
But to Brent's point earlier in my comment, like, I would bet sats that none of these are being developed using Zed or ChatGPT in the web, right? They're using actual tools with harnesses that they're, you know, running on their machines.
So we should go take a look and maybe pull in some of these. I'd be curious to see.
They're running cloud code. You know, they're using cloud code and they're using codex.
This one's copilot.
Yeah, okay, and copilot.
Surprisingly.
The other thing that comes to mind for me is like, are people forking the Linux kernel then to not have any AI code in their kernel? Are we seeing that?
Good question. I also wonder, I bet you, right, because I know like Greg has his own cool customs. I bet there's going to be a lot of sort of bespoke kernel things that get spun up as well because that's their style.
And then lastly, QMU is considering shifting away from their previous blanket ban on AI-generated contributions and opening the door for AI-assisted for lower-risked parts of the project. So this is another interesting approach. You know, certain stuff will let you do AI-assisted coding on, but not for everything. The proposed policy would allow AI help for tests, documentation, mechanical changes, and small bug fixes.
Contributors would need to disclose that they're using AI usage with the AI-assisted tag and giving QMU a way to accept the productivity boost while keeping the review and accountability intact. So this is maybe a way to dip the toe in to see if this is useful for the project without getting completely swamped all over the place. Seems kind of like maybe the balanced tag here.
And areas you're comfortable with and maybe, you know, you have really good tests on and you'll know you'll catch regressions and it's easy to review, you know, things like that.
And kind of signal to the community, we're being careful, we're willing to evaluate this policy, we're going to change it from time to time, which I think is also probably pretty reasonable because it's moving so fast.
Right, exactly. And you should keep evaluating as you see different results with how it goes or doesn't work with your project.
All right, so we'd like you to boost in. How do you feel about AI-assisted code in the open source projects or distribution that you're using, you depend on, you know, your favorite applications? If you discovered it was being contributed to by AI and AI-assisted coding, what's your reaction to that? Boost in, let us know. We'd like to get your take on it.
Quick unscientific survey. We got anti-gravity, we got CLOD, and we got Codex.
In the kernel?
Yep, those are just the most recent things today from the searching for commits with assisted by.
Just today? Wow. Oh, just recently. I see.
Yeah, this week.
Anti-gravity, huh? Look at Google Go. You wouldn't think it, would you? Hey, I want to take a second and thank my friends over at Connect and Internet. Use the promo code Jupiter35 to take $35 off your entire order. They're not a sponsor, but I've been working with them recently to get a discount for you guys because they've become my new go-to cellular internet provider. Because what they do is they're an MBNO on all four major carriers.
Then they give you hardware that has antennae and modems for all four major carriers, which is awesome.
How did you not have this earlier?
And how do I not have it? I had to build it, and then somebody else came along and said, we'll make a product out of it. And then on top of that, they're load balancing and figuring out which one's giving you the best data or isn't capping you or something like that and moving you around. They do offer truly unlimited data plans with priority data as well if you want to pay for it, which it's within my price budget considering the things I paid for.
¶ Shout-Outs
But I think the thing that's even cooler and maybe kind of what I'm going to be doing going forward is they now offer a backup internet plan. It's just $39 a month. They give you a slick router that auto monitors your connection. I think it's using OpenWRT under the hood like mine is. And when your main internet goes down, it instantly connects to one of the four major wireless carriers who has the best signal in your area.
And then it detects when your main internet connection comes back up and switches back to your main internet connection automatically. They call it the Fortress. No, I'm using the Fortress router. Or do they call theirs the, well, they have a cool name for it. I don't know. But mine's the Fortress Router. So they all have these great names.
See, that just seems like a clear business expense.
My Fortress Router, though, is big. You wouldn't want to bring this thing, like, on your mobile setup. This one's for, like, on the homestead, permanent internet, seven antennas coming off of this thing, massive POE-powered box.
Piece of infrastructure.
So go to connectinternet.com. We'll put a link in the show notes and use the promo code Jupiter35. and I can tell you that $39 backup internet plan is a crazy good value. I mean, back in the day, you had to pay for a full LTE data line to do this all the time. And you're just sitting there not using it when it's crazy. So this backup is a great idea for 40 bucks a month. So go check it out, connectedinternet.com, promo code Jupyter35.
It's nice to have a spot to plug them because they've been awesome to work with and they're going to hook you guys up.
Well, a couple of pieces of feedback this week. Zag Attack has one for us. With the Bitwarden situation, I do not like that the costs went up $10. Since I already have a Proton account, I'm probably looking at migrating to them for password management. Passwords and 2FA keys are something I consider too important to just have on my self-hosting setup. I want to have one copy that I control and another that is on infrastructure that is managed and maintained 24-7.
We'll also continue keeping a local copy in key pass like I do now.
Defense and depth. I like that.
Yeah. I realized where I really took a turn and just went all in on Bitwarden was pass keys. When pass keys came along, a guy just got sucked in. And I did not put my two factors in there, though. So that at least I don't have to migrate. Anonymous wrote in. He says, what about this banger swap? Sup, everyone. I'm building my own Arch-based Hyperland distro called Hyperspace OS. OS, OS, OS.
My school teacher likes the name. I like it, too. And I'm trying to figure out a simple way to update the system configs, mostly Hyperland comps, so people can get updates way more easy. Do you guys have any suggestions for a good update method for a small project like this? By the way, I'm 12 and I'm still learning, so can you get some advice for how I should approach it in a good way and not nuke my PC?
P.S.
I wanted to make this because I saw ML4W Hyperland configs and they're pretty bangers, so now I want to start building my own. Yes!
Excellent.
Love this. I mean, is this a good opportunity to learn a little Git and then perhaps everybody that is running Hyperspace OS has a little scheduled job that does a little Git checkout?
You could totally do something like that, yeah.
And if something goes wrong, it's pretty easy to roll back, and then you have them do a little other update.
Git gets you all of the version control stuff, so you can keep multiple copies and, yeah, roll back.
Yeah.
I'm so proud of you. It's just so touching. Yeah. You're just, now you're spreading Git. It's so nice.
Yeah, he did mention Merkle trees, though.
Well, you'd need something in Hyperspace OS that's checking in, or some sort of little updater script for them that goes and pulls the latest, you know, checkout and all that and brings it down. But I think it'd be totally possible. That sounds like a really great project. Keep us posted on that. We'd love to hear how it's going. And very impressive work. You're already off to a great start. And let's get to the boost. We had some support for this episode.
And we're going to start out with a live boost. It's coming in right now, West Payne. Boop-a-da-boop-boop-boop-boop-a-da-boop. Coming in from the one, the only, our podcast with 500,000 sacks.
A giant on-topic boost? That's unheard of.
Wow! Shoah has needed the support recently, so thank you, Rpodcast. Hello, JB. This is a long overdue boost to thank you all for helping me finally get over the hump and use of AI practically to make my dev life better. Oh, good. Through Open Router and Claude Opus, I was able to migrate a very complex Nix config from Wimpy's GitHub to my System76 machines. That is a complex one.
Yeah, it is.
I've seen that. Yeah, very well done. See, now he's got it on the OG Thaleo, the new Thaleo Mira, and a Darter Pro Lab. Now I have Hermes installed and I would really enjoy a future livestream tutorial on how Chris sets up his agents to do the magic they do. Thanks again.
Do you have a month?
That's a fun idea.
That's not a bad idea at all. Jeez, I have been doing so much recently that I'm kind of at a stabilization point, but it was two weeks of, I call it my B-plot.
of the day where I have like a zelge session with like open code going and I just everywhere I go I just sort of pull it up and I just keep moving the agent forward and then when they start introducing slash goal like it just unlocked the whole new level of like I just walk away for a while and I come back and I move it forward but just yesterday I was I was doing another jam sesh I should have I should stream it get a little music going totally but how would you know it's happening.
Yeah we'd have to have a signal throw it up on a live in the app at least.
I'm going to give it some thought i i do think that i i do think now i should have been capturing that i think a lot of times when i'm in the fever build phase i probably lose so much content on the floor just plowing through it and i look back at what i built i go wow i don't even know how i had the time and energy to do that it is it.
Is a fever dream state.
That is that.
Is for for real.
It's um it's unbelievable what i feel but thank you very much our podcast thank you thank you very much uh really do appreciate it and I'm going to give some thought to that keep you updated.
A Dude Dry and Stuff boosts in with 27,368.
Cents Oh!
I tried for a while to think of what would constitute a Brent boost, as in B-R-E-N-T. Brentley won't pin his tabs. So, ended up with this.
Hold on. There we go. Yeah. Yeah, I love that.
I tried for a while to think of what would constitute a Brent boost, so ended up with this T9 dialing approach. So, I guess that's how we get the 27368. Okay. Wanted to boost in and say, yes, I do appreciate the RHEL coverage. It's both something that heavily influences my future career and current learning, but I did not have enough time to sift through myself. Well, that's great. That's kind of what we're going for, I guess. Exactly.
Lastly, I've been self-hosting Vault Warden through Docker Compose on a $12 VPS for the last year and a half. It's been a great experience.
So thank you for the feedback on REL because we were just kicking ourselves about an hour ago for doing it because it just really was seemingly a stinker. Okay, so 27368 in T9 translates to Brent.
You looked at it. You verified it. Look at you.
I did. That's good.
That's your version of the map.
So a Brent boost, 27, yeah, except for it needs a sound effect if it is, because come on, 27368 is a Brent boost. How about that?
Well, obviously it's Brent won't pin his tabs. That's the sound one, right?
And you have to open a new tab every time.
Yeah. I love it. Well, we got another boost here from the facial hair, 5,000 sets. I'm back with more freshly mined sats and another report of an area where Linux and open source are doing great. This time is worship centers. I have changed two churches and a mosque over to Linux recently. They wanted a system that is stable and won't update in the middle of a service. Many of these centers also can't afford hardware upgrades or struggling with Windows 11 dropping support for older mixers.
Yeah. It does seem...
It does seem, almost seems something sort of wrong about, like, bilking churches out of subscription money for a basic operating system.
Uh-huh, uh-huh, uh-huh. You know, they just need Linux and maybe a little OBS if they want to do a stream or something like that. And if they could bring some tracks in, that's awesome.
They saw the light.
Keep it going, Hare. We like to see people getting migrated. Landlocked also came in with, if you have a story out there, do boost it in. Landlocked came in with 5,000 sats. And he's boosting from the Indy 5,000 with 5,000 sats. does that just make us jealous or what I don't know but do you like my sound effect race.
Car boost, you got that.
That's what it would be like if we were there doing the show that's what it would sound like.
Tomato boosts in with 9,998.
Cents woo, Mr. Tamato.
I did appreciate the Red Hat coverage, though it left me feeling down. Their fault, not yours. And apparently uninspired to boost.
You know, I was wondering, I kind of, I was telling the boys before the show, I kind of got a meh feeling watching the keynote too. It just sort of left me feeling a little down.
Wow.
I don't know why that was. It's nothing unusual about it.
Well, we didn't get into the sort of like fog and laser treatment because we weren't there in person.
Also, it's the, you know, the enterprise corporate AI stuff is the thing that appeals maybe the least to us right now. And that's what a lot of it was.
The Mado goes on with, I use key pass and sync thing, and it works great for my needs. Love the Gnome Commander pick, by the way. It's a nice little bit of nostalgia remembering when Gnome initially shipped Midnight Commander as their file manager.
Good callback. Good callback. I agree. Thanks, Tomato. Appreciate you.
Your inner child boosts in with two thousand and two satoshis. Does 95% of this boost go to advertisement at the beginning of the show, Progressive?
Oh, you got Progressive? I got Smokey the Bear.
Yeah, the Smokey the Bear was great.
I love the Smokey the Bear.
I miss the YouTube live, even though I'm subscribed and belled. I don't know what that's about. I'd love for you guys to open with your sweet, sweet music like Twib does, and then say a quick ad to pay the bills. Then come back after that.
We are uh so we because we don't have people actively buying the slots in the show right now we were experimenting with dynamic ads i say we're i'm still kind of on the fence on it we were able to work with a company that let us say no to a lot of categories which has been great block stuff and all of that we'll see because we're just hoping you know it's getting tight so we're just trying we'll try it out but uh when i get smoky the bear i like it i've not heard a progressive ad yet i'd
say i only get an ad one out of two or three plays too it's interesting when it decides to play in the whole thing is.
I think in theory we could change the insertion points. So like we could put it in, but then maybe we'd have to time it out per show and stuff, which would be doable.
Or it's like it changes it for all of them or something. It's like, yes, you can change it, but then I think it might. But if we just had a place to bake it in, we could. Don't know.
It's new to us.
Yeah, it is new to us.
Well, if it makes you feel better, they continue here. Feel free to put in as many ads as you need to pay JB's bills.
All right five more, you wish Browns and Wing comes in with one two three four five sats, You said not to boost passwords here, but here's my luggage combo. Yeah, well, come on. We all use that. I also swap from OC.
Open claw.
Thank you, to Hermes. Way better. I'm running a dual 3090 with Llama CBP on NixOS. It is my first Nix box. I'm going to update the NVIDIA drivers.
Yeah, it will.
Yeah, it'll do it.
Quint 3.6 27 billion parameter Q8 with Q8 cache is very good. I'm fully local setup with Search XNG. Yeah.
Yeah. Yeah.
As my primary search. And I'm using this for my memory. And they link us to that. I'm setting up mine for my wife's work. She does endless paperwork as data entry. Hermes is crushing it.
Wow. Zero dependency sub-millisecond AI memory system for Hermes agents and everyone else.
Nice. I know I tell you boys this all the time, but I think memory systems are going to be one of the very interesting development areas for agents.
For sure.
The better they are, the more they understand you, the less mistakes they make, the less tokens they're burning, all of it. So it's, I think, an interesting area of development. All right. Well, I really like hearing about different memory systems.
Yeah. Keep those code boosted in.
Sounds like an awesome implementation, too.
Thanks, Bronzman. Exception boosts in with Robodux. Switched from OnePass to Bitwarden in 2023, syncing between six devices. Happy but hesitating to suggest it at work, as we already have a self-hosted SysPass behind a VPN.
Hmm. Hmm.
Yeah, I suppose that makes it even more additionally complicated.
Question. Yeah, it does. It does. It does.
Thanks for the report.
Yeah. That is... BitCryptic comes in with 16,384. Sorry, Brantley. Hoping you guys might find our proposal of interest and maybe something you could share with fellow listeners. And it's MDF Markdown First, inspired by your recent discussion around agentic impact on web services.
A proposal for the agent-readable web. The web was built for human eyes. AI agents are paying an enormous tax for that. We think there's a better way, and we've written the proposal.
They request your HTML page. They receive a document full of navigation menus, JavaScript bundles. That's true. Cookie banners, all a bunch of crap they don't need. the HTML tax.
The reference implementation is complete and live. MDF server is a self-hostable Docker image, deployable. There's a demo site.
So what is it? They generate essentially a markdown version of your site? Is that what they're doing here? And then serving that up to the agent?
Or asking that folks do. I think it might be more of like a spec or an idea for how to make a web that's more generated at being agent friendly.
I like it. I mean, it looks...
There's a lot in here. So I'm just...
We'll take a look at it.
We'll take a look at it.
I mean, it might be more efficient to serve, really, when you think about it.
Oh, yeah. For both ends. All right, Bradley, why don't you take the next one?
Well, Bobby Pin boosted in 5,000 sats. I still sync around a key pass file like a caveman. If y'all have any better self-hosted solutions, I'm all ears.
Okay.
Hey, I'm not setting shade at key pass.
Hey, I'm in that cave with you.
The issue is it really breaks down for me on I'm adding something on mobile and then I go to desktop or vice versa. That's really got to be first class experience. Geononymous comes in with 4,018 sets. No message, just value. So I'll take Retro Gear 2. Retro Gear comes in with 5,151 sats. Jens, it's been a while. Bitwarden, self-host, been amazing for a couple of years. Linux router, just moved to NixOS as my router from OpenSense. Yes.
Using Tectitium for DHCP and DNS on the same box. Just started playing out with Hermes. Wow.
Nice.
Have been using OpenCode for a couple of months. What a time we live in. Oh, unpowered system still running Pentium 2. Whoa. Pentium 2, 266 megahertz.
Well, I guess the boost name is Retro Gear.
That's true. With Windows 2003 server for my retro gaming PC, having a good old LAN with my mates.
That is wonderful.
Cheers from Australia.
Thank you for the boost.
Love that.
Impressive.
That is so dense. True.
There's a lot going on.
And I don't think he AI'd that. I think he wrote it like that. I think he did his own summarization there. A true hero among men.
Woo. True Grits boosts in with 12,345 sets. I haven't really heard anything about the Bitwarden switch-up situation. I'll have to look into it, besides the CLI issue recently. I'm sticking with them for now. However, if I do bail, I'll probably switch to ProtonPass, since I already pay for them.
Another ProtonPass.
Might give KeyPass another go.
Uh-huh.
I think the Red Hat coverage issue is mostly a result of people's AI burn-out. People, including myself, are just tired of hearing about it constantly. Not saying you guys shouldn't cover it. You do a decent job, and then it's kind of cut off from there, unfortunately.
Thank you, True Grits. Yeah, we try to walk that line. it's you know it's a it's it sucks that it's such a relevant topic for so long now because if you think about it it's been it's been the number one conversation in in tech since chat gpt launched and we actively fight against it getting on every week because it's just it's what everybody's and it's where all the financing is and everything it's just, it's uh like i know that long term it's going to make make more linux users
and more free software users, but the ride there is an obnoxious one.
It's going to be rough. We got to hang in there.
Greg the lawyer is hanging in there with 10,302 sats. Great last episode, I had already looked into Hermes. You convinced me to make the switch, though.
Nice.
Really is a much nicer experience than the lobster. I'm spending much less time maintaining and more time doing. I set it loose documenting my home lab, and it's found things I didn't know were actually broken.
They're very good at that.
It also offers to set up regular cron jobs where that makes sense instead of burning tokens by default. I'm looking forward to trying it with some local models for basic tasks next.
One of the things I've done recently is my agent data checks in on Home Assistant, and it generates me a report of all the devices that have less than 10% battery life.
Oh, good idea. I love that.
But the next step that it took me one day before I realized I need this is I have data do a quick attempt at figuring out what type of battery that device needs. So when I get the report, I know what kind of batteries. And I don't have to because what I always do with these sensors is like it's been a year. So I have to go open it up and be like, OK, what type of battery does this one take? because there's like three different lithium types these things take.
And it's just so handy. Just that little bit of extra information, just go right on there. Okay, I got that one. I got that one. Oh, I need to order that one. Okay.
You're just the tech for the bots.
Pab came in with 6,666 sets. I recently landed a new job. Definitely a step up in my short career as a social science researcher. And honestly, this isn't the first time being a Linux nerd and a self-hoster has given me an edge over the competition. That skill set is in no small part thanks to listening to you guys. I'm forever grateful for real. Long live Jupiter Broadcasting. Thank you very much.
And congratulations.
I'm glad we could play a role in that. Truly.
Turd Ferguson comes in with 17,076 sets. Good show, boys. Sending value your way. Also, not sure if you noticed, but RAM and disk prices are starting to drop. Still too expensive. Really? But I think the trend is our friend.
Is that true?
Somehow I don't believe this.
Is that true? I'm not true.
I'm going to look that up. Let's look up NVMe and Samsung, because boy, is my NVMe in a situation right now.
Oh, yeah, right.
So I'm on Camel Camel. Let's... Huh.
Oh, wow. Really?
Take a look at this. Take a look at this. a 2TB Western Digital Black, so it's not a Samsung, but still a great drive, 2TB NVMe Western Digital Black Gen 4 PCIe M.2. It is down. It is down from an all-time high of $420 to $299.
That's more reasonable.
Huh. Well, that's a good sign. Let's see what 32GB of RAM. I'm just curious.
Yeah.
32GB. Is it just disk? is it are we did we see the supply chain respond here what's going on here still not in stock for 32 gigs well.
Why would you need that.
So that ain't great do you think 64 gigs is in stock, Of course, this is according to Camel Camel. Oh, that still looks pretty expensive to me. Let's see here. Corsair, 64 gigs of RAM. Oh, it is down, but still very expensive.
Yeah.
So it's down from an all-time high of $927 to $850. It touched $757.
What?
Oh, in May. So it actually touched down to $757 in May and then popped back up to $850. That's a $100 difference. That's insane. That is positive. If you look at the three-month price trend, that is a positive price trend.
Yeah, the data's a little less clear.
I'm hopeful, but I hope. I'm going to go with turd. It looks like it's down, but I'm not buying in yet. We need about another 50% price, maybe 100%. Let's do it. Let's do 200%. Can we do 200%? I'm told that works. I'm told you can do that.
I don't think that's how math works. No, I'm pretty sure it does. But KiwiBitcoinGuide boosted in 4,567 sats.
Hmm.
How does Nebula compare to Tailscale for the purpose of isolating a server running OpenClaw or Hermes in a network? I've built a plan to isolate a small server inside my LAN so that I can experiment with agents. One item on that plan was creating a tail net with only the device I want the agent to connect to. Can I do that with Nebula? How would that differ from using Tailscale?
Can and I say do it should. So you can also have both. You could have Tailscale running on a system and then also run Nebula.
Totally. They work just fine together.
They do. So why I love Nebula for agentic work is because you don't need any company between you and what the agent can generate.
There's no offing into some third-party provider that you have to deal with.
You're just generating keys.
It's all local.
Last week, I decided I had a Nebula network between two hosts. And I was like, you know what? I want to add three more hosts to that. And I tasked the agent and it did everything. It generates the keys. It knows SSH, you know, logins. It moves things around for me. It adds them to the configs. It rebuilds them and they're talking. And it does it in three minutes. So, if you're leaning into an agentic workflow, Nebula is inherently just by default built for that.
And so, you can stand up networks very easily just as a person or now agentically. And you don't need to have it like sign into a Google account or whatever single sign-on you're going to be using or anything like that.
so it's I think it's great and then you know if you still want to use tail scale for like all your other stuff that you already have going you can't you can run both and then all your agentic stuff all your AI related stuff and all those services they only listen on say for example maybe they only listen on the nebula interface and so you're you don't have to worry about things on your tail net getting to it and stuff like that.
As for if you do want to try to look at some of the comparisons you might just I'd say try to focus on I mean it depends on what you're doing to isolate But maybe look at, because a lot of it might come down to what you allow them to access on your network, whether that's directly or over the mesh. And so go take a look at the security primitives that are offered for, you know, what Nebula's got, whatever, any other systems that you need and see if they let you do what you try to do.
And too, like with Hermes and with the CLAW, you can tell them to only bind to, like, say, the Nebula interface. So it's another way, just another level there. And so you could only, maybe you only get to the Hermes dashboard, the web UI on the Nebula interface and stuff like that.
And I think it's a frickin' great solution, and once people figure out, and then, When you force multiply it with something like Nix, where you can declaratively deploy Nebula, and then you just push that Nix configuration out to your various machines and then rebuild them, and then they're...
So good. And you could just sort of tag stuff and be like, you know, only give it access to stuff that has the, you know, in the AI group or...
Do you think people are understanding what a big deal it is that you can declaratively deploy a mesh network across your host with no big tech? Like, I don't know if people have wrapped their head around the power.
It's just like a substrate you get for free.
Yes. that changes.
How you architect the rest of it with persistent mesh aware identities.
So i think it's it's at least worth looking into even if you ultimately do decide to go like with a tail scale route or a different way kiwi bitcoin guide but i think it's i think it is at least a quick glance and you know asking an lm to walk you through it if you have any questions i think that would work all right well that was a just a a really banger of a boost segment our podcast lifting us up from what we thought was going to be a rough week to just absolutely crushing it So thank you,
everyone who streamed sats. 16 of you streamed as you listened. You stacked 19,963 sats. Not too bad right there at all. Thank you very much. We do appreciate that. When you combine that with our baller boots from our podcast and everybody else who contributed, this episode stacked an extremely impressive 665,907 sats. That's a week you go home, you tell the wife about. That's what that is right there. Thank you very much.
We're doing this show week after week. Even though nobody seems to want to buy commercial spots on it, the audience is keeping it going between our members and our boosters. We appreciate you. Now, if you want to get in on the boosting fun, we have links at the top of our show notes, kind of like the easiest path. It does take a little bit of adventuring, so we always do appreciate everybody helping us there.
And if you want to just put your fee out on autopilot, we have jupiter.party for the whole network and linuxunplugged.com slash membership for just this here episode.
¶ Picks
And we're going to wrap it up, but we've got a couple of picks before we go. And our first one I think is just useful for anybody that's ever wanted to record a specific application audio stream on their desktop. I'm always trying to do this, so maybe I am a little biased here. But there's a time when maybe there's a game stream or whatever it might be. Maybe it's the show you want to have a backup copy.
I'm recording our show right now.
Oh, my goodness. Look at you using the pic. Very nice. It's called Record Apps. And it's a desktop application that allows you to record audio from a specific application on Linux built with Dino and supports pipe wire metadatas and whatnot. So it can actually get like application name, plain data stuff and gives you a nice UI to record just that application's audio.
This is so great.
It's purpose-built, super focused, and is available on Flathub.
And it's mostly a thin UI over Pipewire.
Yeah.
Which is kind of great because there's not a lot of complexity to it.
And it means it's going to probably just be rock solid.
Yeah.
Now, our next pick, not rock solid yet. The developer says it's very early days, but I think this is something that's got real potential. It's called OpenLogic, and it's a native local-first alternative for the Logitech Options Plus for Linux written in Rust. It lets you remap buttons, change your DPI settings, the smart shift over HID++ settings. No account required, no telemetry. Tons of features in the works as well. Not quite yet implemented,
but lots of stuff on the roadmap. Real good roadmap, actually, for this project.
There's a GUI, there's a CLI. Everything's local. Bindings live in a plain Toml file.
I like having the GUI and the CLI.
That's a nice touch.
It is. And it's impressive what they've gotten done already. they're looking for community input and if you've got a Logitech mouse, don't most of us have one somewhere and you've wanted some of that Logitech stuff on Linux to maybe change some of the settings, some of the configuration.
Warning, it seems like Mac is there today, Linux and Windows are coming soon, so.
Yeah, yeah it's early days, it's early days.
That's something to watch.
Maybe it's a little early featured on the show, I don't know, but we've given a little love we might get some Linux use, so we'll see.
And there's, you know, there's like a dev n.nix in there, so you gotta like to see that?
Indeed, and it is Apache and MI2 licensed. It's called OpenLogy, and it is very early days. Sounds like their Linux build not ready yet, which is disappointing, but I think if we show them a little love, perhaps they'll be encouraged to create this for the Linux community, because we've all got some Logitech devices sitting around. I got one right here, and the situation, there are tools, we've talked about them before, that are available today for Linux, but not as comprehensive as this one.
So the situation would be a lot better if we saw that come over.
Looks like RecordApps is MIT.
Yeah. Oh, did I not mention that?
I don't know.
I don't think I did mention that. Thank you. How do you like it?
I like it. It's quick. It's easy. It uses Denno, which is kind of interesting. It has like a web view that does the graphics layer. And then under the hood, the Denno runtime is just basically calling out to Pipewire. So pretty lean, which is great. And the UI looked nice. I mean, it was more featureful than I expected.
Yeah, the developer is also responsible for a couple of other neat apps that they've also published on FlatHub. I think his name is Bettis. And he's also made, if you're into this kind of thing, an app called Stimulator that keeps your monitor awake. Simple UI to just go in there and configure like when the monitor goes to sleep. And I think using the same toolkit as for record apps, the developer has also created a ping monitor, which, you know, there's a lot of these around,
but it's very clean. It looks like maybe the same toolkit and a nice way to graph your ping stats.
Oh, that does look nice.
So they got a few apps over there on the old flat hub. They've been cranking them out.
Isn't this a great example of a developer like scratching their own itch in the exact ways we were talking about during the conversation about, you know, anti-AI?
Yeah, absolutely.
This is the exact thing. So why wouldn't we want this?
And I guess, you know, to my question to the audience... Maybe I don't care if it's Vibe-coded. I mean, it is TypeScript and HTML, and they love writing TypeScript. These machines love writing TypeScript.
That's the giveaway, is it?
No, it's just, you know, if you were going to have a list of things that could possibly make it LM-generated, TypeScript might be one of the data points.
It's the M-dash in TypeScript.
But I don't know if I care. But I think I might...
I think this comes down to sort of, when do you care? When do we care?
I think maybe I care more if it's an app. I've been using for years or a distro I've been using for years. Like, R-Sync is the perfect example where it's like, everybody uses R-Sync. It helps back up my data.
Yeah, it's connected to a lot of important systems.
A lot of people have scripts and automations and whatnot, backup jobs built around R-Sync. So it's just the perfect example.
But a Pomodoro timer for your desktop is sort of...
Yeah.
You probably notice if it fails.
Or, you know, if you're adopting R-Sync today, hey, you probably wouldn't care. If March, you know, or May or April, you started using R-Sync and this happened, it probably isn't a big, like, life-shaking deal for you. But if you've been using it for 10, 15 years, like some of us have, it is kind of a bigger deal.
Rock-solid plumbing than you count on.
I think, and I'd like to know what the audience thinks. Boost in and tell us. I think that could make for a good engaging boost segment next week. And tell us how you feel about AI-assisted code in the OSS projects or distributions that you depend on. If you feel sort of similar, I mean, it is new and something we're going to have to work through.
¶ Outro
Now, I will mention we are live. We can we can stream on a Tuesday and make it a Sunday using the power of Linux. So join us on a Tuesday on a Sunday at 10 a.m. Pacific, 1 p.m. Eastern. And we do have more data around the show that people could go get their hands on. Tell them about it.
Cloud chapters.
What?
Yeah, that's right.
Cloud chapters, you say?
The magic of the cloud in your podcast app.
Wow. No way. What about transcripts?
Why would you want that?
Yeah, you probably wouldn't. I wouldn't grab those.
No, definitely not. Not SRT or VTT. And the VTT definitely doesn't have your name attached to it.
No, absolutely not.
Don't even look.
To what we talked about and more over at linuxunplugged.com. This is episode 669, so it's slash 669, including our contact page, our mumble room info, and our matrix info. We'd love to have you join us in one of those during a live show. Thank you so much for listening to this week's episode of Your Unplugged Program. We'll see you back here next Tuesday. You know it, as in Sunday.