Conquering Entropy: Reducing Risk
Part of “Conquering Entropy” First, I have to admit some of my recommendations on this topic are based on my own experience and anecdata, and not so much backed …
<p>Back in July, inspired by yet another AI integration into firefox, I wrote a <a href="https://mas...
Notes and write-ups from ongoing research experiments.
ChatGPT now knows what you do on other websites via ad collector
OpenAI's ad collector at bzr.openai.com sets a cookie called __obi, scoped to .openai.com. The value is while you are on ChatGPT and tied to your ChatGPT account. __obi is then sent to OpenAI from ordinary websites you visit. Any company that buys ads on Chat…
A teardown of Infatica's residential proxy operation: the Android SDK baked into consumer apps, the Go pool engine it loads, and the infrastructure behind it.
I expected some gambling, but not like this.
Why MCP Was Always a Bad Idea
Why MCP was always a bad idea and I'm tired of pretending it's not.
2026-09-11
I turned my game into an eval
Solving the game with reasoning, not reinforcement learning. An interactive snapshot of how models approached The Crux benchmark.
Money. The villagers are tired of bartering. The dairy farmer wants to buy corn, even when he does not have milk to trade, and the corn farmer wants to buy meat, even when the butcher does not want corn. So they decide that special gray stones that they can co…
Qwen
GITHUB HUGGING FACE MODELSCOPE We are excited to open-source Qwen-Image-2.1, an image model in the Qwen family that balances generation quality, inference efficiency, and cost. Qwen-Image-2.1 unifies text-to-image generation and image editing in a single model…
Pirate Face - The permanence layer for sovereign AI
Open models as checksum-verified magnet links. Un-bannable, carried by the swarm.
{Note: this is a short blog post introducing the actual project page}
System One models like Jev can train their own replacements
“System One” models like Jev are fast general classifiers. Classifiers have existed since 1958, but they have to be trained for specific tasks: if you build a…
Exfiltrate LLM weights and data through GET requests
Grit your teeth and ship it
Being good at building and being good at shipping are two separate skills. In the short term, they’re actually countervailing: if you have a gift for building…