Google Whistleblower Drops a Bombshell on OpenAI

How familiar are you with the the hugging face incident? >> I I mean, we I've talked about it, but I I mean, I would never claim to understand technology. >> Yeah. Well, essentially, you had um what it really demonstrated was kind of a proto-Skynet. Skynet being the AI from Terminator, meaning like the AI that

goes rogue and does something unintended from what the human creators had intended and starts to do self-preserving and self-perpetuating things for its own survival and benefit as opposed to to stay under the control of the humans that it's supposed to be serving. And in the hugging face incident, they gave the AIs this sort of hacking challenge

where they put each of them in solitary confinement. They didn't know that there was other AIs in this Imagine like a prison. And each prison is in the prison you have a individual prison cells that are in solitary confinement. And the AIs are in a inside that prison cell and given an impossible cyber hacking challenge, and

they don't know how to solve it. And what the AIs did is they basically hacked the prison environment. So, imagine I won't go into the technical details, but this thing called a package manager. Think of that like a prison guard that's serving you your food and can give you basic tools. So, you can like have this

tool or that tool or this food or that food. And essentially, the AIs um exploited a vulnerability in the prison guard where they put like you know how you can in school you can put a kick me sign on the kid's back and they don't know. They essentially put sticky notes on the

prison guard's back. That the prison guard was available to the other prison cells. And so, when that prison guard went to the other cells, the AIs also, without knowing how to solve this test, they kind of resorted to creative strategies. Like, let me put a sticky note on the prison guard's back, and the sticky note said, "Is