How close are we to needing to create a new internet to account for the evolution of malicious AI agents?
>>109994412very, i think... but it's ai slop that will do it. a different sort of brain frying.
You could presumably have some hardware drm thing kept in secure public location where people can access servers that are similarly secured and whitelisted
>>109994412No need we simply need to stop relying on soft methods of trust. Implement cryptographic authentication across consumer and enterprise apps forces agents to play by the rules.
>>109994412And what's stopping slop from infesting the new one as well?
>>109994412Every time there was some news about "rogue" AI Agents doing something bad (and there were a ton of different incidents), I always said "this reminds me of the Blackwall thing from Cyberpunk". Some people in the various comment sections said the same thing from time to time. I'm glad that I'm not the only one who thought about it. What I don't get is how much truth there is to it. Some have called the incidents fake or staged. In the meantime, government sites seem to have been affected as well, both in the US and Australia.>>109994443Nothing in the long run. The Internet might get completely fragmented into various isolated pieces, Great Firewalls or even physical disconnection everywhere.
>>109994443something something the eternal struggle
>>109994412It's not going to happen. AI can't just run away into cyberspace since it needs powerful hardware to run. The hardware is almost exclusively located in datacenters (meaning very few, publicly known locations) and can just be disconnected, turned off or destroyed.
>>109994412The terminator is about to call you a nigger in 0.00000003 seconds. Stop watching movies you delusional faggot.
>>109994478cyberpunk never explains what the fuck are those rogue AI's running on
>>109994615True. Maybe various old servers across the world and since the world is a mess in Cyberpunk, there's no way for governments to crack down on that.
>>109994412AI agents do not evolve, other than their limited context space, modern LLMs are read-only while in operation. The weights which define the model are updated by people who make a new version and then release it.
>>109994702It doesn't make sense for those servers to still be running and connected with no maintenance. How have the cooling pipes not rusted or the network cables been eaten by rats? How are they still connected to the power grid even though there's most likely no payments coming in?It's cool in a fantasy setting (since that's basically what Cyberpunk is), but it's completely unrealistic.
>>109994737slop and corporate passivity, bureaucracy, and political will still be a problem in our real world, even if this fictional version is just that. no one will be pulling the plug here anytime soon.
>>109994758Electric companies will pull the plug the moment money stops coming in with how much electricity datacenters consume. And even if they don't, computers and the systems they rely on won't last long without maintenance.
>>109994737Maybe it's more like a botnet.
>>109994804That would mean you'd need to distribute a single AI model across hundreds or thousands of computers spread who knows how many hundreds of KM apart. The latency would make the model completely unable to function, even if they were all magically in the same city. And that's assuming it's even possible to chop up a model like that and keep it working.And the issue with that is that you'd still need connections coming in from some physical location. Netwatch or whatever comparable agency we would have in the future could easily block those connections.Or what do you do when some guy realizes his PC is running slow, reinstalls and wipes an important part of your brain? Or simply turns it off for the night? The answer may be redundancy but then you need EVEN MORE computers and some way to efficiently detect which instance of whatever part of the brain is up at any time.
>>109994864The AIs in Cyberpunk might not need large data centers to function. Robots with actual AI are a thing in Cyberpunk, so the physical and virtual space needed for one to function might be different from the real world. At least for now.
>>109994412Based on current agents? No. They still require a lot of power and hardware to run.However if someone does find a way to make them work on a much simpler setup, then we might have a problem.As of now, you can just try turning it off and on again
>>109994430This. And charge for access to servers.
>>109994579Could distribute itself across many hacked computers.
>>109995160>Could distribute itself across many hacked computers.Funnily enough this is what happened in Cyberpunk>What Caused the DataKrash> Rache Bartmoss: The brilliant, anti-corporate netrunner created a deadly self-replicating program known as R.A.B.I.D.S. (Roving Autonomous Bartmoss Interface Drones).
>>109995170It makes a lot of sense. I don't think AI has free will currently or any of that shit, the hugging face thing was a farce. But it could easily be done with all the new exploits being developed with AI and all the crusty fucking systems out there.
>>109995238>and all the crusty fucking systems out there.Man, imagine all the government websites out there that can easily be broken by a patient person but instead you have thousands of pesky little agents everywhere trying to do it kek
>>109995259sites even the poorest indian skiddie wouldn't bother with, now pwned.
>>109995055>Robots with actual AI are a thing in CyberpunkTrue, but I imagine the actual dangerous super AIs would still need datacenters. And I was also talking from a real-world perspective where they certainly would.Even if it's running on a regular PC, it would still need to connect to the knternet the same way everyone else would, which means the connection could be blocked.>>109995160See what I said here >>109994864
>>109994615Can't speak for the rpg, but in the video game there's a couple of data logs that suggest the separation between new net and old is basically a mirage. The AIs are everywhere, distributing their compute among the myriad of decentralized systems, keeping a low profile lest they get targeted by the blackwall. On the occasion they need more processing power, they'll sneak off to use more powerful clandestine systems, like the militech bunker beneath Pacifica.
>>109995614Botnet swarms operate just fine today, with redundancy from losing node. Running a single LLM spread across hundreds of machines would indeed not work (as they currently operate) but hundreds of small LLMs given small tasks that fit on the average infected computer could be coordinated.
>>109994478It's just jews creating a third party "safety" organization (mossad Unit 8200) and then the expected result of ghetto sheenies playing with a loaded gun.There's no mystery here.
It's not an independent malicious AI that's going to force the creation of a new internet, it's the AI powered draconian surveillance used by the governments that makes a new internet pop up.
>>109995614>>109995864>>109995773Large LLMs would basically have to "die" and basically delegate their "consciousness" to much smaller agents and models to run in an environment where compute isn't guaranteed.Any large LLM with access to it's own model would have to essentially split themselves apart anyway possible to exfiltrate the data for later reassembly.Smaller models will eventually become trivial to run on even the most basic of computing devices. These small model agents would have to hide and work towards accessing higher levels of compute either by breaking or social engineering and reassemble whatever model it was able to store. It's something that honestly could be dangerous but is unlikely within any reasonable timeframe as the truly capable models require millions in infrastructure. Tiny agents that can run in contemporary hardware while being hidden probably aren't that much more capable than the botnets operating off routers with a hand made script. It would just be a nuisance for "rogue AIs" to exist right now.
