AI, Burnout, and Everything After

**Jeff** (0:00)
Hello, and welcome into another episode of BitFlip. We're here to talk about the pragmatic side of infrastructure and all things nerds and whatever else we want to talk about. So this week, we're going to talk about AI. AI is what we're going to talk about. So why don't we start off by talking about the Linux kernel? And I think Linus changed his tune on AI. Does that, you know, he said in 2024, I think that AI was 90% hype. I think he's I think he said something very different recently, didn't he, Alex?

**Alex** (0:28)
Yeah, so in 2024, as you said, AI is 90% hype. That's Linus Torvalds' exact words. He then also told the kernel devs who don't like AI in 2024 to fork the kernel or leave. It's very dramatic, isn't it?
But recently, he has adopted a tool called Shashiko, which is an opt-in agentic AI code review tool for kernel patches. And this thing claims to catch 50% of bugs in patches that have passed human review with a 20-ish percent positive rate.
Now, I can only imagine the amount of pull requests and code reviews that the maintainer of the Linux kernel has to deal with on any given day. Particularly, if you've seen the charts about GitHub lately, with the volume increase, thanks to agentic coding that they're having to deal with. So, it's kind of like fighting fire with fire at this point, right?

**Jeff** (1:33)
Kind of. I mean, I thought it was... Because I think also recently there was a comment by the, you know, by the kernel dev saying, like, the security mailing group is just getting overwhelmed with AI, you know, prompted and AI-coded, you know, patches.

**Alex** (1:50)
You can't ignore them.

**Jeff** (1:51)
So, it's interesting to see. No, you can't, but, I mean, it's on the same side, you don't want to have yourself flooded with them, or you can't sort out what actually is useful and what's garbage.

**Alex** (2:03)
So, Torvalds this week has sided heavily against people who want to filter out Shoshiko's output before reaching the Linux kernel authors, citing Software Freedom Conservancy New AI Contribution Guidelines. Now, Linus has an exact quote here of, Linux is not one of those anti-AI projects. And if somebody has issues with that, they can do the open source thing, fork it and just walk away. A complete 180 in two years on its position previously. Keep your ethics where they belong in your personal life. Don't try to enforce your ethics on others.
Now, that is Linus Torvalds as we know him. I mean, it's not, there's nothing wrong in what he said. Ethics don't belong in the code base. Politics don't belong in the code base. That kind of thing. His stance here is simply mechanical. It's to say the quality of the code is what matters.
Your feelings, to put it frankly, your feelings don't matter. Am I wrong?

**Adam** (3:23)
Oh, and he's nothing if not pragmatic and honest. Like, that's the thing I love about Linus is that you don't have to wonder what he's thinking. And he's also, this is not the first time that he's changed his opinion on things, or kind of come back to the table to a question and said, you know what, that's changed now. So I really appreciate the fact that we're not pretending that this isn't the way that the world is moving to, you know, right now. And I think he's more than most people very aware of how ubiquitous these tools are being used at this point.

**Alex** (4:05)
The AI landscape of 2024 is, was very different to that of today. I'm in a hotel room right now, so hence my surroundings.
And I heard a stat today at Developer Relations Con from somebody saying, five years ago, there were approximately 400,000-ish software developers in the world. And now with AI, there's what? A hundred million? I don't know. Nobody really actually knows the number now, because anybody that can prompt Claude code or Codex or Py.dev or whatever, anybody that can type a prompt in is a developer now. And so you end up with people building bespoke software, which is great, but you also end up with a whole bunch of people now looking for finding and submitting bugs to major projects. I think we'll touch on this slightly in the Jellyfin news that we're going to come on to in a moment.
But there's this balance between gatekeeping of traditional software development being a craft, versus the pragmatism of dealing with the modern landscape. Like all the CVEs, I know Unraid have had to deal with a few, for example, lately, where you've had to light a fire, and light the beacons of Gondor to get things out in a timely fashion, because response with disclosure is kind of dead. And like, so without that pragmatism from the top, from Linus, from Torvalds, the Linus Colonel would be in a worse state right now, wouldn't it?