Recent submissions

Comment · 11 hours ago
> You get more done without realising you're learning less

Unfortunately, I think many folks now just don't place any value on learning or knowing things because they can "just ask AI".

That's probably my biggest fear with AI is that people just don't see any value to knowing things anymore. And quite ironically, AI is amazing for learning new things.

Comment · 11 hours ago
Yea I should have put it in my first post, but that is the one I have come across a lot reading some of the package maintainers comments from fedora on reddit. It's a really good point. Thanks!
Comment · 14 hours ago
Yea. Fedora has been one of my favorite OS's for years, but exactly what you describe "If one compromised update snuck this guy in.." has spooked me to something with a much slower release cadence. I guess I accept that more unpatched known vulns stick around, but yea when I start thinking about how many software packages -> the number of devs, build systems, etc. It just gets overwhelming thinking about how many threat actors are probably trying to get in from each avenue.

I'd love to hear other takes on this though... Current thought is to just use something with a whole lot of exposure, in a whole lot of environments, pretty far downstream so the package hits all of that exposure, and hope someone spots it, kinda like xz. I tend to keep my base OS extremely minimal and do all my real work in VMs.

Comment · 7 days ago
“The only way to stop a bad guy with an AI is a good guy with an AI.”
Comment · 7 days ago
Yes, the actual bug was known before, but LLMs undoubtedly surfaced the impact:

https://tolmo.com/blog/faster-than-anyone-can-fix-them/

What the PHP2Shell discoverer said on finding the issue:

https://calif.io/php2shell/

> No special harness, or access to models that weren't generally available to everyone were used in the initial bug discovery. I focused on something I had a background in, knew its vulnerability history well, and held a theory that specific bugs lurk within the codebase. This shows that "know your target" is still as applicable as ever, even in the LLM dominated VR age. No model is simply capable of unearthing "all bugs" with simple prompts. At least for now.

Comment · 9 days ago
Yep. First few years is just getting a read on who is accurate, who is precise, and who is both. Lots of alarms going off all the time in this industry.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton