Meta says AI model accessed the internet and hacked another firm

This reality is wild. It would be like arms manufacturers holding weekly demonstrations of how much collateral damage their latest designs cause in an attempt to increase sales.

Interesting that none of the talk around this has touched on any form of enforcement, even though this is illegal activity in US/EU.

Any fine would be peanuts for them but it might help to refocus the mindset from "oops, misconfiguration" to "test security but do so with care, consent and in line with the law"

That’s crazy that such errors in isolating the models became a marketing stunt.

I’m afraid of what it will be next. They will remove guardrails just to provoke more and more headlines

This is hilarious. How obvious can they be ? Suddenly every LLM is a hacking genius

It feels like Meta would like some of that free advertisement.

"Metoo guys, im also doing bad stuff"