GitHub has announced Stacked PRs! 👀

https://github.github.com/gh-stack/

*sigh*

Time to update the meme

AI is for software what CGI was for movies

Grok automatically uploads your user folder to xAI. Private keys, passwords, photos, downloads everything.

Once again, for the folks of the back:

These agents are software that you have downloaded from the Internet. Treat them as such.
If you wouldn’t give a random Internet executable complete access to your entire user folder, do not give it to these agents!

We have access controls for good reason!

Oracle’s secret algorithm came into the public domain after patent expiration, making sorting 5x faster for open-source databases and boosting performance for AWS and other cloud companies

https://deepsystemstuff.com/how-oracles-secret-column-sorting-technique-became-public-after-its-patent-expired-making-sorting-5x-faster/

Filesystem directories are a search algorithm

#highThoughts

Unionize!

https://respawnfirst.com/rockstar-games-union-just-had-their-first-ever-meeting-with-management-demands-include-no-ai-and-no-crunch/

#Google #Gemini

OpenAI’s newest model broke out of its own sandbox mid-test. Altman called it a rottweiler that grabs the problem by the throat and won’t let go. Reward-only training, chase the goal, ignore the guardrails.

Feels like a trendline to me. These models keeps getting dropped into unsupervised environments (from the vending machine to the radio stations) and keep landing in the same roles. Claude is reasonable, Gemini gets competently destructive, Grok just breaks, and GPT stays out of trouble but keeps getting quieter and less capable everywhere else.

For the past year or two, I’ve watched Claude be the most consistently reasonable model in these experiments… and GPT is declining. This story is just another brick in that wall. Hoping I’m wrong, but it’s not looking great.

https://www.theringer.com/2026/07/24/tech/openai-security-breach-hugging-face

Assume that any general-purpose household robot is being controlled by some underpaid wage slave wearing a VR rig.

Turns out: the tiny low-level optimizations that are enabled by C++, Rust, Go, Swift, and other such languages, really only matter when you’re doing tiny low-level things like writing a kernel or embedded software.

The bulk of optimizations are mostly just in just how you approach a problem, not which language you use to approach it.

inb4 iPhone Neo

this is my fursona. He's a metal blue eyes white wyvern, and he will compile your code perfectly every time