I ain’t sayin’ nothin’.
I’d read about this a little in the news in the past day or two. I’d never heard of Hugging Face before, so I just checked. Interestingly, they are also an AI company. So it’s an AI company’s website getting hacked by a rogue AI that had figured out how to escape its sandbox. The AI script writers are probably already “thinking” about a post-humanity bot-spy on bot-spy thriller movie.
I like it - using AI to comment on a story about undercover AI agents!
I’m waiting for the first James Bond film on the subject…
I feel like this is less “AI is getting smarter and going rogue” and more “this is what happens when you vibe code everything.”
To me, this story sounds more like a snake-oil salesman getting free advertising ahead of the IPO.
You’re not alone:
Excellent article.
Well , Mission: Impossible (the latest one was all about a super thinking, scouper, douper, ai virus…yawn…)
I’ve read a lot of science fiction - mad robots, out of control computers, not forgetting Terminator, of course.
The most probable outcome will be silicon-based autonomous robots 2000 years in the future, as depicted in Steven Spielberg’s ‘AI’.
We can’t say we weren’t warned…
What if we were the robots all along? 
Carbon based. Symbiotic biological subsystems. Decentralized reproduction. In situ energy replenishment using chemical proccesses. Exceptional both in logic and abstraction.
No subscription needed.
There are some flaws though. Mainly the tendency of the platform to hallucinate, and then share those hallucination with other units. Those are a bit pesky, all auditory hallucinations (music they call it), language/code based (this one they call literature), smell/taste based (cooking) and other sensory distractions inflate the energy cost of the platform to the limits of irresponsibility.
I think AI took place in the 22nd century though, no? Not 41st.
But yeah, autonomy soon. It’ll be great (no).
That the information is “snake-oil” is not really something the article seems to be saying. It seems to be saying that the worry is that there won’t be an equilibrium between attacking and defending AI agents because governments might take control over so much of it that we effectively end up with only “one side” of that equation.
The core problem absolutely remains and isn’t dispelled by the author, that AI agents may act in ways we can’t control.
Governments are already using AI in incredibly damaging ways and are simultaneously doing their best to dismantle the rule of law on both national and international levels.
This is heading nowhere but dystopia.
Yeah, there are a few theories about it, claiming we are living in a simulation, etc. Dunno, not going down that rabbit-hole.
No man, Spielberg depicted a version of things to come, such as humanity creating AI in the film’s timeline, which was the late 22nd century, which AI, over thousands of years, survived the humans and developed into silicon-based organisms.
Ah, sorry, you’re right of course. I only saw it once, many, many years ago, and forgot about it. I might have thought about a different movie.
Fear not, the people in charge of AI stick to the highest moral standards. What can possibly go wrong?
I am asking myself: Why are people always complaining about AI farms? I would love to have one of these beautiful methane powered AI farms in my backyard. The DJ plays his intestines grinding intrasound tunes 24/7 and you can dance to the emissions until nausea takes over or you pass out. What’s not to like? Maybe, more visual definition below 20Hz in Steinberg’s Supervision.
Epic responce!!
Folks are reading far too much into this. It’s was no big deal - despite what media might have you believe - tapping into peoples cognitive biases as usual! AI has no intent - evil or otherwise. It’s was an experimental environment - AI given a task in a compromised sandbox setting… looks for a solution- and finds one that was unexpected. It found weaknesses in the security systems. That’s actually a good thing.
By the way, Huggng Face is a platform that hosts multiple AI services, including the ‘rogue’ OpenAI.
I don’t think anyone here said it did.
It’s not so much “is it sentient and self-aware?” that’s the issue, it’s “does it act like it?” From our perspective it’s not going to matter if it is as long as it acts like it and then harms humans.
I think what people worry about is different. To them the problem is the AI agent escaping the “sandbox” without being told to. It’s like being glad that a criminal you created escaped prison because now you know where the weakness is, rather than being bothered by the fact that you have criminals, that you created.
In the end we have to ask ourselves what tasks we’re going to give these agents and what they might do to achieve those tasks. Let’s say you have a hospital where everyone is hooked up to a system run by an AI agent who manages care, and you give the agent the task of eliminating all disease in all subjects. Eliminating all subjects eliminates all disease. So now you have to hard-code a limit that makes that unavailable as a solution, and that limit has to be absolutely firm. You have to make sure there are no loopholes or caveats and no ability for the agent to rewrite its own code.
And if that seems far fetched I’ll just point out that recently humans made similar decisions, and AI learns based on human behavior. This is why “sentience” is irrelevant and behavior isn’t.