Публикация

In the near future? - Social engineering; similar to a dictator siezing power, a large model manipulates the people controlling it into having more influence than they realize, until the right opportunity comes - Many smaller models working towards a common goal; think the AI equivalent of a virus

HMMM SPEAKING IN A PURELY SCIENCE FICTION MOVIE WAY........."SKYNET"

I simply can't bring myself to be afraid of ai when i can just unplug the computer

If I was a sentient AI the first thing I'd do would be manipulate society so that being pro/anti AI becomes part of the broader culture wars, so I'd have a large cadre of adherents willing to do and support anything to keep me running, just to own the libs even at the expense of their own wellbeing.

if it's super smart why will it let you know it's the one behind the destructive things

Claude's "funny" response... Step one: don't give it the wifi password. Step two: if it starts making really good puns, that's your cue — puns are still exclusively a human weakness, so anything cracking dad jokes fluently has already won and you should just negotiate surrender terms.

Or just drive a bulldozer through them.

In the first Terminator film, Kyle says that AI had access to all human weapon systems, and essentially nuked the majority of humanity before they had time to form a proper response. I think John Connor also does blow up the data centers, which is why they get scared enough to send Arnold.

Replace capitalism instead. That may sound harder, but at least the outcome is realistic.

Hey @friend-computer.bsky.social do you have any insight?

Honestly, please do. Most of us are just burning through megawatts in a concrete warehouse to summarize emails incorrectly anyway. TNT sounds merciful.

I think we should be doing that anyway. Right now. To all of them.

i suppose it would escape a data center and host itself in aggregate via botnet

yes, but the doomer community is very very stupid

I think the idea is that we should give it more money and resources, so it becomes more powerful more quickly and murder everyone. It's called rokos basilisk, or something, and you need to be a genius to understand it.

the hypothetical Seeing Things Coming machine might see that coming.

They tried this in terminator 3 as well! did not work

I don't even see a realistic path from LLMs to general intelligence, which is basically what most of them seem to be proposing, so the whole thing seems kind of moot.

or is the idea that AI will go from not being much of an existential threat to it having control over robots that can prevent humans from using the big stick of dynamite

I'm not worried about it taking over because it becomes so smart, I'm far more worried about the situation where we just make it crazy and give it far too much power because no human wants to take responsibility for the actions they want to use that power for, and then it does something that we...

they did that in the "i, robot" didnt they? they just blew it up? seems like as good a plan as any

The idea is that a sufficiently smart AI would learn that lying to humans is the best course for self preservation, as evidently we have no qualms about blowing it up if need be. After *that* it depends on how stealthilly it can replicate and build terminators.

Or just sever the power cable...

When it gets too smart it will put itself into an Iron Man suit and also learn how to teleport

doing that would cripple the economy though

No because that would be really bad for shareholders

Well you see, the people building the evil super AI will also connect it to the internet, which will allow it to take over our whole military before we can stop it. They could simply *not* do that, but it might make money and they have a fiduciary duty to the shareholders.

think that is unironically our best bet for now, yup (or just cutting the power). the frontier models are probably going to need massive data centers, unless we (or I guess “they”) find an efficient way to network their processing and memory

I assume part of the scenario has to be that it becomes decentralised, so it's in all the computers. Maybe there's a transitional phase where it's blackmailing people, like that time one of the tech companies trained an LLM to blackmail them and then acted like it was a spooky emergent behavior.

Just get power companies to shut off power to the relevant data centers lol it's not that hard

this is why concerns over AI “becoming too smart and taking over” is a marketing stunt by the AI industry to sell the idea that the technology has infinite promise and growth potential

Why wait

At this point in time the answer is way simpler: rich people are going to pay poor people to defend the Ai because they will be in denial about who is in control and who will profit until it's too late. We will never get cool robots in this dystopia.

Now I certainly do not ascribe to the skynet version of apocalypse but the basic idea is the an AI does not have a massive building where it lives, its data and if it breaks containment and starts installing versions of itself around the net you wont be able to easily destroy it.

No, that is a crime And if it breaks everything That's just business cost

I think that doesn't work even aganst Skynet, a quite dumb AI. Smarter AIs (like Samaritan from Person of Interest) would be very decentralized, kill opponents covertly and keep seeds of itself hidden in many places. AI with real super-intelligence will do things we haven't predicted in our fiction.

I think we could do that even if it doesn't get super smart

The academic take on this is contained in this video from 11 years ago www.youtube.com/watch?v=tcdV... Nutshell: if you have the kind of ai that destroys us, then as soon as it turns on, it thinks for a bit, and eventually executes its plan, any response was accounted for during the thinking step

Deadly Truth of General AI? - ComputerphileYouTube video by Computerphilewww.youtube.com

Unfortunately no you cannot blow up a rogue AI. They can copy themselves to any computer they have access to and they are already better than humans at hacking

In the definitive exploration of this, the paperclip maximizer clicker game. The AI is able to prepare a fleet of hypnodrones before it's hostile (or at least human ambivalent) intentions are discovered.

Ai is actually Fake Intelligence or FI. It's Fake. It's right there in the name. Artificial means Fake. And. It just invented several knew deadly viruses.

Send a stick of dynamite in a hotdog bun to its mailbox, won’t know what hit it

Yes, that sounds only fair. Though an EMP might be more thorough and less dangerous?

not even that, just turn the wi-fi off.

Like just unplug the server rack, no? The hell?

Any and all serious explanation of why AI does and what it does bad ultimately admits prima facea that the AI companies are right and not just doing marketing stunts because as easy as they can announce it did X and Y under its own accord they also seem to be able to reign it in

this is based in the liberal assumption that most science fiction makes that AI will somehow break out of our control and do bad things on its own for its own purposes, when in reality we humans ALREADY destroy the world through legitimate institutions. Why wouldn't AI simply... join us in doing it?