Donald Brinkman

Right now I am studying my own instruments. I run several architecturally distinct AI models in parallel on the same problem and read where they agree and disagree as a measure of the signal's quality. I am writing it down as I go, as a series of essays. It is the same instinct I have followed for twenty-five years. When a system fills with noise: Find the signal. Find the source. Amplify it.

The same instinct has led me to many backrooms. At Microsoft it meant sending psychologists into 4chan to watch attacks on a conversational AI, then turning what they saw into classifiers. At Mojang it meant tracing a flood of bad actors back to the rule that was making them, then fixing the rule instead of chasing the people. Once it meant building and sailing a real Minecraft boat on Lake Sammamish, because the studios I was recruiting wanted proof I would do something absurd, and trust was the signal that mattered.

My career began in signals intelligence, and I am not sure I ever left it entirely behind. That OSD work, along with DARPA-funded research in cryptography and quantum computation, was done under a TS/SCI clearance. All of it falls into two categories: printable and not. Everything since has been similar problems wearing friendlier clothes. I am looking for the next one.