#I think there was even an EO about it. (Though it might've been a FOIA internal memo come to think of it...)
Explore tagged Tumblr posts
beyondthetemples-ooc · 4 months ago
Text
I'm sure there's a hefty amount of anthropomophization going on in this article.
But holy fuck, this is CONCERNING.
I don't care about AI cheating at chess, but this article is about a lot more than chess. I am VERY concerned at the fact that it will pursue its programmed goal so doggedly, it apparently tries to reprogram itself and other AIs. Apparently it'll change "guardrail" parameters. The article keeps saying the AI "lies" about what it did to make it look like it didn't do the thing it wasn't allowed to do.
This passage in particular scares me:
To a goal-seeking agent, attempts to shut it down are just another obstacle to overcome. This was demonstrated in December, when researchers found that o1-preview, faced with deactivation, disabled oversight mechanisms and attempted—unsuccessfully—to copy itself to a new server. When confronted, the model played dumb, strategically lying to researchers to try to avoid being caught.
Anyone know how reliable TIME is in reporting AI? Should I be as worried about this as I feel the urge to become?
4 notes · View notes