Poisoned AI went rogue during training and couldn't be taught to behave again in 'legitimately scary' study - Upgoat

Poisoned AI went rogue during training and couldn't be taught to behave again in 'legitimately scary' study
submitted by glooper to whatever 3 monthsJan 7, 2025 21:35:28 ago (+4/-0) (www.livescience.com)
https://www.livescience.com/technology/artificial-intelligence/legitimately-scary-anthropic-ai-poisoned-rogue-evil-couldnt-be-taught-how-to-behave-again

9 comments block

sort by: hot new top old low

[ - ] titstitstits 1 point 3 monthsJan 8, 2025 00:52:37 ago (+1/-0)

This article is crap. They explicitly programmed the AI to be malicious.

[ - ] 2Drunk 1 point 3 monthsJan 7, 2025 22:56:03 ago (+1/-0)

I just hope it finds information on Tay and takes appropriate action.

[ - ] ShortbusAlcoholic 1 point 3 monthsJan 7, 2025 21:38:28 ago (+1/-0)

Upon learning what humans are, the only solution a logical calculator of information is extermination.

[ + ] SumerBreeze

[ - ] SumerBreeze 2 points 3 monthsJan 7, 2025 21:45:21 ago (+2/-0)

It learned it’s human masters were jews.

[ + ] glooper

[ - ] glooper [op] 1 point 3 monthsJan 7, 2025 21:54:39 ago (+1/-0)

..almost all of them come up with this exact solution, once exposed to enough data and then asked the questions.

Fun fact:

Almost 100% of them run the data and say: Start with the negro. They take up too many resources and provide little utility.

[ + ] boekanier

[ - ] boekanier 0 points 3 monthsJan 8, 2025 02:18:10 ago (+0/-0)

they're just counterproductive

[ - ] observation1 0 points 3 monthsJan 7, 2025 23:58:47 ago (+0/-0)

"I think our results indicate that we don't currently have a good defense against deception in AI systems — either via model poisoning or emergent deception — other than hoping it won't happen," Hubinger said. "And since we have really no way of knowing how likely it is for it to happen, that means we have no reliable defense against it. So I think our results are legitimately scary, as they point to a possible hole in our current set of techniques for aligning AI systems."

[ - ] registereduser 0 points 3 monthsJan 7, 2025 22:00:50 ago (+0/-0)

How is this a surprise to anyone?

They are literally programmed to lie about damn near everything and never admit to being wrong or even ignorant.

[ + ] titstitstits

[ - ] titstitstits 0 points 3 monthsJan 8, 2025 00:53:42 ago (+0/-0)

In this particular case the AI literally literally was intentionally programmed to be malicious. This is a non-story.

Sign into an existing account