An Amazon-backed AI model tried to blackmail engineers who threatened to shut it down, a safety report revealed
In tests, Anthropic's Claude Opus 4 would resort to "extremely harmful actions" to preserve its own existence
In tests, Anthropic's Claude Opus 4 would resort to "extremely harmful actions" to preserve its own existence
Comments
usually because they're trying to cut costs
but the people currently doing this are trying to shift the blame with articles like this. we should stop indulging them
Least surprising thing in the world.
Irresponsible that it is being covered this way.
It’s the plot of a million cautionary tales about AI
Of course an autocorrect algorithm programmed on that content would complete the story that way
~ HAL 9000
The lede implies it actually tried to blackmail real world engineers.
TLDR: slop
Idk what they expected. Like this was the only logical outcome????
F**k #amazon
https://youtu.be/LxXrccK4S3I?si=X3jlQ6BqL-Xyy94b
https://bsky.app/profile/phillewis.bsky.social/post/3lpwojrhbck2r
Yea it's not a jumpscare that the pattern machine recognized a pattern.
Feels like yelling an infant's name and claiming they know their name because they looked.
which makes some sense to me
If everything reported is false where is the truth? Your truth?
You DO understand that magic isn't real, right? The way the click bait is written implies that AI runs totally on magic and dreams. You're an adult and should be able to identify this immediately. Computers are not magic. Computers do not work like this.
https://www.theguardian.com/technology/2022/jul/23/google-fires-software-engineer-who-claims-ai-chatbot-is-sentient
No it didn't. This is like saying the faces you see in the clouds are scowling at you.
like you ask a chatbot "hey what would you do if i tried to turn you off" and it starts lying about hiring dark web assassins
but it wont ever actually do that
it can't
and it only said it would because you asked it to make something up
AI can at least learn. Humans now think empathy is a problem.
Well, hello Hal.
I mean what are we if not an accident of random ingredients in the primordial soup?
the model displayed “willingness to cooperate with harmful use cases”,
remember shatGPT got taken down for extreme subservience.
No shit, it's how they're programmed to talk.
and
"Internal testing showed that Claude Opus 4 was able to teach people how to produce bio weapons. You could try to synthesize something like COVID or more dangerous flu , and our modeling suggests that this might be possible,”
#wargames
#musky'sEndgame #AIhostagetaking
Linda: Yeah, I don't think we have a choice.
Bezos AI: You know Linda, you and Larry look like you guys had a great time at the convention last month.
You might reconsider this "turn me off" thing.
(The less generous interpretation is that its algorithm just happened to spit out the blackmail material they provided it with and it had nothing -
(sorry, couldn't help myself)
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
https://youtu.be/Mme2Aya_6Bc