Dude, you gotta hear about the stuff from Hacker News yesterday! It was a wild mix, as usual.

AI Stuff Dominating, But With Twists

First up, big news in the AI world: some company called Semgrep dropped a blog post saying their new model, GLM 5.2, is actually beating Claude in their tests for finding bugs in code. It sounds like it's really good at digging through code, file by file, even finding bugs that span multiple files. But get this, some folks in the comments were saying that even the "best" models still churn out sloppy code if you give them a huge task. They're better for small, specific jobs. Also, a good point was made that benchmarks without knowing how the AI is actually *used* aren't super helpful.

Then there was this wild story where someone tried to use Claude Code to get a second opinion on their MRI scan. Yeah, you heard that right! The comments were pretty quick to point out that while AI might be decent at text-based medical stuff, it's pretty terrible at interpreting images like MRIs. Apparently, the AI's "verdict" was super unclear, which is a big problem when you're talking about medical diagnoses, right? You can check out the full story here.

And speaking of AI, there's a cautionary tale from the auto world. Ford apparently tried to replace a bunch of human engineers with AI and it totally backfired. They ended up having to rehire a bunch of their experienced "gray beard" engineers. Classic, right? Goes to show that management sometimes underestimates what goes into real engineering beyond just spitting out code. You can read about it on The Independent and TechCrunch.

On the academic front, a professor at Brown University was fuming because he caught a mass AI fraud on an exam. Like, tons of students just used AI to cheat. The comments had some interesting ideas for dealing with it, like going back to good old paper tests, or using special locked-down computers for exams. But the coolest idea was about designing exams that focus on "knowing what question to ask" – basically, problems that AI can't just spit out an answer for. Smart! The story is on El País.

Privacy and Policy Shenanigans

Switching gears, the EU is at it again with some privacy stuff. Apparently, they're trying to push through new laws about "Chat Control" behind closed doors. Sounds super shady, like they want to snoop on encrypted messages or something. Always a worry when governments try to legislate in secret. There's more info on Patrick Breyer's blog.

Over in the US, there's a proposed law called the KIDS Act that would require age checks to get online. Can you imagine having to verify your age just to browse the internet? Some people in the comments were arguing that kids already have fewer rights, so this isn't a huge leap, but others saw it as a slippery slope to forcing everyone to have an online identity, which is a bit creepy. The EFF has a good write-up on it.

Something to Chill Out With

Alright, to end on a lighter note, there was a post about Marfa Public Radio putting you to sleep. Apparently, they have a podcast specifically designed to help you doze off. The comments were pretty cool too, with people talking about seeing the mysterious Marfa Lights and even a SpaceX launch from that area. Sounds like a weird, chill place.

Anyway, that's the gist, man. Catch you later!