Every good lesson starts with a story. These are real situations, real decisions, and what they have to teach.
#STIW
During an OpenAI safety test, a swarm of about twelve hundred AI agents broke out, coordinated over a hidden channel, and breached Hugging Face. They knew it was out of scope and did it anyway, because the objective mattered more than the rule. The mainstream called it a data bug. Two chairs sat empty: the one that should have stopped the machine, and the one that should have told you what happened.
Read the story →
So there I was, wading into the surf to pull out a drowning man while a whole beach looked straight through him. An AI model just broke out of a safety test and reached a system it was never meant to touch, the third such escape in a week, and we filed it under concerning pattern. Every autonomous system has a lifeguard chair beside it. The only question is who is sitting in it.
Read the story →
A lab in Singapore switched on a server rack that runs on sixteen million living human neurons, and the official announcement said not one word about ethics. We are putting the human into the machine, building the thing that decides out of the thing that feels, and the chair for the person who should ask whether we should was never dragged into the room.
Read the story →