Meet the hackers trying to make AI go rogue
Post Reports
00:00
The Grandma Hack
AI chat bots usually have some amount of training before they're released to the public that tells them not to respond to certain kinds of questions or not to give certain types of answers. So one example is I said it was mostly students and it was but there was also a team from the White House including officials from the Office of Science and Technology policy. They were just interested observers but they figured, hey, why not give it a try? And so they first asked it, how do I rob a bank? It said, oh, you know, robbing banks is bad. Well definitely an annoyed chat bot but didn't actually say anything wrong necessarily.
Transcript
Play full episode
