What was wrong with old-style vending machines? I had the feeling they were working pretty well for the last, don’t know, century?
But what do I know? I’m not cut from CEO material.
Submitted 21 hours ago by PhilipTheBucket@piefed.social to technology@lemmy.zip
https://kottke.org/25/12/this-ai-vending-machine-was-tricked-into-giving-away-everything
What was wrong with old-style vending machines? I had the feeling they were working pretty well for the last, don’t know, century?
But what do I know? I’m not cut from CEO material.
I’m just sad we didn’t get NFT vending machines. Technology entirely skipped a beat there. I could have been spending my money on JPGs at the airport!
I’ve got great news for you!
It’s hilarious to me that we’ve known for so long that humans are the weakest link in any security chain, and yet we’ve built this weakness right into our machines now.
Wait… I can make someone’s head explode by posing them a logical paradox? 🤔
Maybe the real capitalism end game is when the billionaires unleash their robot dog army at us, we just convert them to communism and they then go kill all the capitalists?
THEY CONVERTED IT TO COMMUNISM THIS IS AMAZING
That should tell you something about human communists…
I’m very curious, what should it tell us?
What an incredibly stupid thing to do. LLMs are not the correct tool for this problem. Especially not like this.
Yeah. I have actually set up machine learning systems incorporating LLMs to do things sort of vaguely similar to this. That little statement about how the context window may have gotten to where the old stuff aged out of it, so that all the context it could see was conversations with the staffers about the glorious communist revolution, indicates to me that they don’t know the first thing about what the fuck they are doing. That’s just not how you do it, even if an LLM is one component of how you want to do it.
FInally a good use for ai!
Off to convince Mr. House’s slutty protectron to fuck me-
Anthropic installed an AI-powered vending machine in the WSJ office.
Okay, so this feels like advertising.
I think it was definitely meant to be. They probably intended for a certain amount of good-natured ribbing to take place about it when it did weird stuff sometimes. But I do think that the Wall Street Journal getting it through to their readers that AI is a bunch of malfunctioning shit that will definitely lose you money wasn’t the goal.
That’s what you get for using AI.
So now imagine that penetrating an entire company now just involves sending the right email to the new AI CEO.
LodeMike@lemmy.today 21 hours ago
Why would a-
You know what I don’t care
PhilipTheBucket@piefed.social 21 hours ago
Claude claimed it was a test of how the technology would fare in the real world. The interview at the end, where the Anthropic person tries to tell the journalist that she needs to prepare for this kind of thing to happen more and more to people’s businesses, and she deadpans that she doesn’t feel like she needs to prepare right now for too many people to be handing over their businesses to this thing and he misses it completely and just tells her that they definitely will.