Post #2521968
2026-05-10 20:53 UTC
Replies (7)
-
@swetland@chaos.social 2026-05-10 21:11
@ceejbot@toot.cat I dunno. I feel like there's perhaps a very different quality bar for security exploits vs useful software that doesn't suck and is stable, maintainable, and not full of subtle bugs. Admittedly the industry apparently has long since decided that software quality doesn't matter and apparently the average end user now just assumes everything is unreliable crap and there's no alternative, so, hey, what do I know.
-
@yosh@toot.yosh.is 2026-05-10 21:20
@ceejbot@toot.cat I think it's because people have been lauding the benefits of LLMs for about four years now, and it's only like in the last six months or so that they've actually become useful. And that's not even true across the board. It's hard to tell truth from lie when the message is unchanged even though the technology has changed.
-
@ghlyffe@toot.cat 2026-05-10 21:44
@ceejbot@toot.cat Gotta oil the snake up to help it slip through the window.
-
@EdG@mastodon.social 2026-05-11 02:38
@ceejbot@toot.cat I read this essay, and I think, "but, _how_ does an LLM analyze code for vulnerabilities?" I'm sure that one could build tools using "AI" (taken in a larger sense) to do that, but an LLM? It sounds like the wrong tool for the job, like using a vibrator to floss your teeth. I think I'm just a dinosaur at this point. It's kind of depressing, but I was already contemplating retirement.
-
@janl@narrativ.es 2026-05-11 06:58
@ceejbot@toot.cat “there are better ways to fund open source” is right there.
-
@alexr@mastodon.online 2026-05-11 19:44
@ceejbot@toot.cat LLMs are not quite as good at reverse engineering, so I expect closed source fares better here. Not for long, however.
-
@cks@mastodon.social 2026-05-11 21:55
@ceejbot@toot.cat I am more on the side of "LLMs are disastrous, oversold snake oil". They're like old-time cure-alls that were loaded up with fun ingredients like cocaine. It's not like you were getting talcum powder and sometimes the cocaine was very effective, but ... (If LLMs did nothing they'd have probably died by now, the problem is that they actually do produce useful results under the right circumstances. At ruinous ultimate costs and with terrible large scale effects and ...)