Hey buddy,

Man, you won't believe some of the stuff that hit Hacker News yesterday, Sunday. Figured I'd give you the quick rundown.

Firefox for iOS Gets an Adblocker

First up, check this out: Firefox on iOS finally has a native adblocker. Pretty sweet, right? People in the comments were saying it's still got some privacy quirks, like not clearing cookies and cache automatically for all sites, which is a bummer. But on the flip side, it doesn't have all the strict rules that Chrome's Manifest V3 has for adblockers, which is a win for freedom, I guess.

Claude's System Prompts Revealed

Then there's this AI stuff, as usual. Claude released docs on their System Prompts. You know, how you tell the AI what to do. The interesting part from the comments was that people wished they'd also shown the "tool definitions" – like, what actual functions Claude can call. That would apparently be way more useful to see what it can really *do* instead of just how you talk to it.

RISC-V: A Third World Engineer's Take

This one was pretty cool: an engineer from a developing country responded to criticisms about RISC-V. Basically, they're saying RISC-V is starting small and cheap, just like ARM and x86 did, and it's gonna grow into a powerhouse. Some folks in the comments were debating if it can really get to high performance, with one person saying the real problem is everyone just wants to sell IP instead of actual chips. But others were like, "most complaints are nothing burgers."

Cloudflare's Silent Analytics Injection

Okay, this next one got people pretty riled up: Cloudflare is apparently silently injecting its analytics into websites when you switch your nameservers to them. Even on the free plan! People were super annoyed, saying it's sneaky. One commenter even said it's a slippery slope and next thing you know, they'll be injecting ads. Yikes.

"Kidney Disappointment" in Research Papers

Here's a funny but also a bit concerning one: Someone found research papers using "kidney disappointment" instead of "kidney failure". Sounds like a classic case of bad machine translation, or maybe early AI getting it wrong. The comments immediately brought up old-school translation fails, but also pointed out that even modern LLMs can still spit out weird, unnatural stuff sometimes. Made me chuckle, but also think about how much we rely on these tools.

Qwen 3.8 27B Overthinking Things

More AI news: There's a new language model, Qwen 3.8 27B, that's apparently excellent but "overthinks" everything. Like, it adds extra steps to its reasoning that aren't really needed. Someone in the comments actually said they forked a popular AI project (llama.cpp) to specifically control this "overthinking" behavior. Wild how quickly people are tweaking these things.

Stripe Acquires OpenRouter for $7B+

And finally, this huge tech acquisition: Stripe is reportedly buying OpenRouter, an AI gateway startup, for over $7 BILLION. That's a massive number! One cool insight from the comments was that LLMs are kind of like databases where the data is part of the engine itself. So, being a "middleman" like OpenRouter, which handles that data stream, becomes super valuable. Makes you think about the value chain in AI.

Alright, that's the gist of it. Talk soon!