Just caught up on Hacker News from Tuesday, July 28, 2026. Man, some interesting stuff popped up. Lemme hit you with the highlights:

Big Earthquake in Japan

First off, Japan got hit with a 7.1 earthquake. Pretty serious, you know? But get this, someone in the comments said how their wooden house in New Zealand, which also had a 7.1 quake, was totally fine. They were like, "Most homes in New Zealand are made of wood and they work great in earthquakes." So, maybe wood isn't so bad for quakes after all! Here's the link if you want to see the official info: data.jma.go.jp

HIV Vaccine Breakthrough

Then, there's some really good news on the health front: a new HIV vaccine is showing "unprecedented success" in early studies. Like, seriously promising stuff. People are super stoked, hoping this actually goes somewhere big. You can read about it here: lji.org

Fighting for Internet Anonymity

This next one is a big deal for internet freedom: there's a huge push in Europe with a citizens' initiative called "Stop Killing the Internet: No Digital ID and No Age Verification." People are really worried these things will kill anonymity online. Someone in the comments was like, "Dude, the age thing isn't even the main point, it's the digital ID that's the scary part." And some folks even think anonymity can make online communities *nicer*, not more toxic. Check out the initiative: citizens-initiative.europa.eu

Substack Writers, Get Your Own Site!

Heads up for anyone writing on Substack: there's an article titled "Substack writers, you need a website." The author basically argues you shouldn't rely solely on Substack. Apparently, a bunch of people in the comments agreed, saying they often just close Substack links right away. But, yeah, some also get that it's super convenient for folks who don't wanna build a site from scratch. Good read if you're into that: elizabethtai.com

OpenAI's Codex Security

OpenAI put out this "Codex Security" thing on GitHub. It's all about making sure AI-generated code, or just using AI in general, is secure. People were talking about how hard security is, especially when you're trying to ship stuff fast. And get this, someone dug into the license terms and found there are revenue limits if you're using their model as a service, which is a pretty big detail for businesses. Here's the repo: github.com

Kimi K3 AI Model Costs

There was a deep dive on a new AI model architecture, the Kimi K3. Super technical stuff. But what's interesting in the comments is some folks are saying it might actually be *more* expensive to run than some of the other big models out there, like Claude or GPT, even though there were rumors it was cheaper. So, not necessarily a budget option if you're thinking of using it. Check out the overview: sebastianraschka.com

Cheap AI Fine-Tuning Beats Big Models

This one's pretty cool for AI nerds: someone spent just $500 to fine-tune a smaller AI model (9 billion parameters), and it ended up beating the big, fancy 'frontier' models at reviewing product catalogs! It really shows you don't always need to throw tons of money at the biggest models if you're smart about how you fine-tune for specific tasks. People in the comments were all about how this highlights clever fine-tuning and not just brute-forcing with huge models. Here's the article: fermisense.com

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