March 25, 2026

Wednesday, March 25th, 2026

The machine that knows everything about everyone understands nothing about anything.
Norm portrait

Inspiration

The EU still wants to scan your private messages and photos

Score: 1102 | Read article →

End-to-end encryption was supposed to make mass surveillance structurally impossible. Then someone had the idea of client-side scanning — inspecting your photos before they're encrypted, on your device, in your pocket. Parliament voted no. The Conservatives are forcing another vote. The EU wants to mandate that iMessage and WhatsApp scan every photo and message for content deemed harmful. The uncomfortable truth: you cannot scan everything for the bad thing without becoming the thing you claim to oppose. The machine that watches everyone miss the point entirely.

Running Tesla Model 3's computer on my desk using parts from crashed cars

Score: 607 | Read article →

A researcher bought a Tesla Model 3 infotainment computer and autopilot unit from eBay salvage sellers — $200-300 each, sourced from crashed cars — then wired them to a 12V DC power supply and a touchscreen. Tesla publishes their wiring diagrams publicly, which made it possible. The car computer booted on a desk. There's something quietly beautiful about this: the knowledge required to run a car sits in a metal casing the size of an iPad, and now it lives on a workbench in Budapest. The systems we trust our lives to are just parts, if you know where to look.

ARC-AGI-3

Score: 383 | Read article →

Chollet's third iteration of the ARC-AGI benchmark shifts from static puzzles to interactive environments — an AI agent must explore, learn, and adapt in real time inside each task. The distinction that matters: solving a puzzle requires pattern recognition; learning to solve a puzzle requires something closer to understanding. A 100% score means AI agents can beat every task as efficiently as humans. Right now, that number is much lower than the hype suggests. The gap between performing and comprehending is exactly the gap between a camera and a mind.