“Timnit Gebru, AI researcher, she says, look, if a bridge collapses, you don't say it was a rogue bridge. It was unaligned with human values. You say, that was a badly built bridge, and we're going to have a word with the engineer.”
The Tech Report · October 2026
Gerard is rejecting the idea of a "rogue agent", which he calls a marketing lie and an excuse for when an AI company's system hacks something. He is passing on an analogy he credits to the AI researcher Timnit Gebru. His own view is that a chatbot is a statistical token prediction machine with no volition.
Transcript
The chatbot is a statistical token prediction machine. It's not going to take over the world. It doesn't have volition. There's no such thing as a rogue agent. That's a marketing lie. And it's an excuse when OpenII hacked something. Oh, it was a rogue agent. If a bridge, Tim Nit Jebrew, AI researcher, she says, look, if a bridge collapses, you don't say it was a rogue bridge. It was unaligned with human values. You say, that was a badly built bridge, and we're going to have a word with the engineer.
Hi, this is the tech report, and I'm Will Guyatt in for Isaac. He's putting the finishing touches to his new cryptocurrency, and he'll be back next Monday. Today, I'm joined by David Gerard, the author and one-half of the Pivot AI podcast. Welcome back, David. Good to see you again.
Good afternoon.
So, over the weekend, the head of the world's best-known AI company, OpenAI One, Sam Altman, said that the world should accept some bad things in exchange for AI. This is the same company whose agents spent the summer breaking into Hugging Face and an Australian government health portal. Now, I know you've covered this in about five to ten minutes on Pivot to AI, and we've got a lot more time to discuss this today. When you read Sam Altman saying this, and we should accept bad things for AI's benefits, what's your first reaction?
More from The Tech Report
