Discussion
For the second time I think I have accidentally convinced a project to just … stop doing security updates on a critical dependency forever, because they'd assumed the LLM usage problem wasn't as bad as it is.
Both of these so far are offline things that are not at al security-sensitive, but I am starting to worry that in the "dark matter" of the broader developer community there is a meaningful plurality of people who are just going to stop updating e.g. the Linux kernel because of broken trust
but the utterly psychotic and highly public intentional destruction of Windows and the visible delusional and antisocial behavior of trusted luminaries (Linus foremost among them) has provoked a wave of revulsion that, while it may not be the majority view, has definitely convinced some people that all future updates are now so compromised that they might as well be installing malware
@glyph i'm "some people". slop in the kernel is a time bomb, and the problem is only going to get so much worse now that they decided to just let the slop flow free. i genuinely believe the only realistic option is to put as much support as we can towards alternatives and be ready to jump ship once it all comes down.
@glyph NetBSD seemed alright when I tried it, and I think the rough spots will get hammered out quickly with the extra attention. I think having a pre-slop linux to reference for rewriting drivers will be a huge boon. I don't think there's much value in version pinning for something like the kernel given that not updating is also a time bomb.
@glyph I'm substantially more worried about compilers for languages that have been accepting slop. I don't know of any slop-free alternatives for C++17 compilers for example. I'm hoping GCC's current anything-goes policy for tests and only-trivial-slop policy for the compiler itself wont cause too much damage, but I'm considering version pinning GCC.
@aeva @aburka @mcc the strongest cognitive dissonance I feel right now is the distinction between those of us who can see the _obvious_ decline in quality that slop is producing, and the people (even some of the skeptics!) who just don't believe it's happening. like I am a "moderate" in the sense that I think that there is a lot of catastrophizing about the degree and inevitability of the decline, but the denial that it's happening just seems delusional
@aeva @aburka @mcc and sometimes a booster will say to me like "prove it" and expect quantitative analysis of defect counts or whatever and … i mean, fair enough, I can't do that on demand, and having an objective measure of software quality across the industry was never really possible before so why would it be possible now so we are just going to have to agree to disagree but also… come on? look around? what the hell?
@mcc @aeva @aburka there *are* actually some pretty good studies on this but the standard of proof is totally arbitrary. I'm gradually building up a for-real research project on this, a sort of rhetorical meta-analysis. but it's a ridiculous amount of work and it feels like commissioning the hubble telescope to convince people that the sky is blue
Install Fedizen
Get the full app experience
The case I'm usually *trying* to make here is my own middle-path moderate view, that while LLM usage is dangerous and unfortunate and we should try to find ways to discourage it, that we are going to have to accept a certain amount of it and maintain trust that existing maintainers will at least still be paying attention and will _notice_ and might stop when the slop wave destroys their reliability