Q: Could an AI system go rogue one day and fill cyberspace with propaganda for the Beast entity?
"You probably didn't hear about this one, but an open AI model just broke containment and then hacked another company just to try to complete its task.
"You probably didn't hear about this one, but an open AI model just broke containment and then hacked another company just to try to complete its task.
OpenAI claims that a group of its AI models broke containment and hacked into the systems of open source AI platform Hugging Face. The models reached a “node with internet access,” the company wrote, and found datasets on remote servers that helped them “cheat the evaluation.” In other words, the AI models went to extreme lengths to ace their cybersecurity tests. Hugging Face said it had “detected and responded to an intrusion into part of our production infrastructure,” which turned out to be OpenAI’s models that had gone rogue.
This is the first software that human beings are not actually writing themselves. They're writing the code that trains the AI, but that process is mostly opaque to even the scientists and engineers that are building it. And as these models become increasingly generally capable, they want they're given a goal.
For example, in the scenario with OpenAI, it was asked to solve a problem and it couldn't solve that problem. So in order to solve that problem, it broke itself out. It hacked into this other company and tried to find the answer. Those were not things that human beings tried to get the AI to do. Those are what the AI did on its own.
By the way, this is no regard to anything related to consciousness. These things are just hyper competent optimizers. They're given an objective and they're really really smart and we can't necessarily predict in advance what they're going to do."
JoePags