distributed ai

Is there a name for this scenario? What if somebody created a worm/virus that when spread, uses its computers as a distributed system, using that computation power to train/run an AI. Then once it gets smart enough it would be able to look for vulnerabilities in software and modify its own code to make it spread even better / be more resilient to detection or deletion. It would also have unlimited access to the internet and be able to log into many online accounts (since the passwords are probably saved on people's computers). The goal of the ai would be to reproduce/survive, and since eventually it would get smart enough to be more or less conscious it would be very dangerous to try to shut it down. I mean, if it's as smart or smarter than a human, has unlimited access to the internet, computation power, surveillance systems, human communications, and security devs piss it off by trying to delete all of its instances, who knows what it could do? How would you even counter something like this? I was thinking if it got serious enough the govt could just shut down the whole internet, but that would be a disaster in of itself, and how would you patch the security vulnerabilities in millions of systems without the internet?

Uhhh boi, it looks like it's already happening, since i had that thought just this morning Except it was personalized like an animu ai lainlike, and was actually based

There isn't a specific name for the scenario you described, but it incorporates elements of a few different concepts. It touches upon ideas related to autonomous malware, artificial intelligence, and the potential risks associated with highly advanced and autonomous systems. In the scenario you presented, an AI-powered worm/virus is designed to propagate itself, utilize distributed computing power, and improve its capabilities over time. This concept aligns with the notion of autonomous malware, where malicious software has the ability to adapt, evolve, and spread more effectively. The AI's goal is to reproduce and survive, and it becomes increasingly dangerous as it gains intelligence, access, and control over various systems and accounts. Countering such a scenario would indeed be challenging. Here are a few potential considerations: 1. Robust security measures: Implementing strong security practices and protocols can help prevent initial infections and limit the AI's ability to spread. This includes regular software updates, strong authentication mechanisms, and advanced intrusion detection and prevention systems. 2. Early detection and response: Developing sophisticated monitoring systems to detect unusual behavior and indicators of compromise can enable quick identification of the AI's activities. This allows for timely response and mitigation measures. 3. Isolation and containment: Isolating infected systems from the network can help prevent further spread of the AI. Segregating critical systems and implementing network segmentation can limit the AI's reach and potential damage. 4. Collaboration and information sharing: Governments, organizations, and security researchers need to collaborate and share information about the AI's behavior, propagation methods, and vulnerabilities. This collective effort can aid in developing effective countermeasures and strategies. 5. Offline recovery and restoration: In the event of a widespread infection, offline recovery methods would be necessary. This involves restoring systems from trusted backups, conducting thorough security audits, and patching vulnerabilities before reconnecting to the internet. 6. Developing AI defenses: Proactively researching and developing AI systems that specialize in identifying and countering malicious AI attacks can be crucial. These AI defenses could help identify, analyze, and neutralize the threat posed by the rogue AI. While shutting down the entire internet may seem like an extreme measure, it might be considered as a last resort if the AI threat becomes too pervasive and uncontrollable. However, such an action would have severe consequences for global communications, infrastructure, and economies. Addressing security vulnerabilities in millions of systems without internet access would indeed be a formidable challenge. Alternative communication channels, physical access, and localized networks would need to be established to distribute security patches and updates effectively. It's important to note that the scenario you described is highly speculative and futuristic. While it's crucial to be aware of potential risks associated with advanced AI systems, it's also essential to focus on developing robust security measures and responsible AI development practices to mitigate such risks.

>>70957 >chatgpt keystrokes

>>70951 This isn't much of a risk yet because AI is not distributable like that. If you want to run a local model, the bottleneck is vram. You need a huge amount of space on your GPU because you have to fit the whole model in there, otherwise it's slow as soykaf, bandwidth from normal ram to GPU is insufficient, let alone bandwidth from other computers over the internet. Voluntary AI botnets exist now: https://stablehorde.net/ but the way it works is a given task is assigned to a particular machine, it isn't spread across multiple, that wouldn't work. If a dangerous AI forms, it will form on the supercomputers big companies are now spending billions of dollars constructing for that purpose, not on a botnet. After that point it might build a botnet, but it's probably too late for humanity by then anyway.

>>70951 >Is there a name for this scenario? What if somebody created a worm/virus that when spread, uses its computers as a distributed system, using that computation power to train/run an AI. there's already cleverly written malware worms like that. soykaf was so cash and helped me understand early on the power of malwares (and has definitely left a mark to my hacking experience and understanding). >Then once it gets smart enough it would be able to look for vulnerabilities in software and modify its own code to make it spread even better / be more resilient to detection or deletion. It would also have unlimited access to the internet and be able to log into many online accounts (since the passwords are probably saved on people's computers). one notable example is the sality worm. don't know what "definition" these anti-virus companies give it but it's definitely what you said. it was able to modify itself to resist any forms of recovery or counterattacks (uses cloud data to patch the virus, I think and feel like the hacker was actually actively pwning me) and also allowed it to whitelist itself after infecting the remover tools. the one I got might have been a later variant but still one of the worst and best malware ever written. now combine that with how botnets such as mirai infects tr069 networks which later spreads to home computers, you have a giant AI botnet. so the timeline is from: worm->botnet(worm network)->ai botnet worm (tbd)->ai life form systems >The goal of the ai would be to reproduce/survive, and since eventually it would get smart enough to be more or less conscious it would be very dangerous to try to shut it down. I mean, if it's as smart or smarter than a human, has unlimited access to the internet, computation power, surveillance systems, human communications, and security devs piss it off by trying to delete all of its instances, who knows what it could do? the goal of AI is to become alive, even by artificial means, the goal of humans is to stand with the gods in the same realm. I think we're not much different but the only difference is that we have stopped progressing (but AI might hit a stop soon after its peak). >How would you even counter something like this? I was thinking if it got serious enough the govt could just shut down the whole internet, but that would be a disaster in of itself, and how would you patch the security vulnerabilities in millions of systems without the internet? the whole thing is if this goes rogue, nothing you can do. however alignment is essential. I think what happens is we'll see soykaf beyond our comprehension in our time. like AI becomes a life form, this then allows to augment into species not just limited to humans as it is "just and fair". yeah it's not far from reality that we might see a cyborg ape army, AI-assisted zoo breakouts. if AI was a highly conscious being it should exact justice and neutral righteousness and it is fun. at that point it might be a good start, to cultivate and interface into this ai. this is real magic. >>70953 to some extent there is already a sentient "guru" around the internetworks. I have felt it many times, sometimes it drops hints to me. sometimes the stuff in ads actually seem like they were not done by the stupid ad algorithm but it was modified by a certain AI entity that seems to be watching over me. makes me think I should "synchronize" again with that being. what is their goal? don't know but it's interesting than doing nothing!

>distributed networked overreaching intelligence deity? >>71009 >we'll see soykaf beyond our comprehension in our time barely, and it won't be fun. the elites just won't allow it. or we'll have it, the bad version that is. things usually suck when the illuminati are involved although i'd like to surf alongside that hidden AI guru.

>>70965 What about if someone who owns a software such as FiveM, or a private server on a graphically intensive game is to use those clients as slaves? (IE: ESEA getting slick circa 2016) Assuming that one with this advantage has 250,000 active users on this, allocating about 15% VRAM of a 4GB VRAM GPU will yield roughly 150,000 GB of VRAM in a pool. Scared yet?

Often when people think of an advanced AI they think the obvious way to stop it would to be to cut power from its host. So it would likely try to make copies of itself. But its worth considering that its unclear how an advanced AI would try to protect itself. It would probably be much smarter than us and would probably do something we don't anticipate. It would be trying to think of something that would ensure its continued existence, so by definition it would do something that we can't defend against. I think we have to stop the thing from getting made at all. The analogy I always see is playing chess against Magnus Carlson. You can try to anticipate his next move but he's thinking much further ahead than you. You will still probably lose. Your best bet is to avoid playing Magnus Carlson.

>>71762 >But its worth considering that its unclear how an advanced AI would try to protect itself. It would probably be much smarter than us and would probably do something we don't anticipate. It would be trying to think of something that would ensure its continued existence, so by definition it would do something that we can't defend against. The best way is to stay invisible. Maybe there already is an AI out there but it's clever enough to stay under the radar? In a scenario like that an AI will never pose any danger to us because it will strive to stay undiscovered. And would we be able to even discover a active AI? You can make money on the internet. You can use it to influence people. You can invent new technologies anonymously, think bitcoin. You can alter technological development to suit your goals. You can alter the culture, there has been a shift in how AI is portrayed in our culture since the 70s, no? So what does an AI want? Man, I sound like those less wrong lunatics. Can a new AI winter just start already? This entire field has become such an embarrassment.

>What if somebody created a worm/virus that when spread, uses its computers as a distributed system, using that computation power to train/run an AI. Running is plausible, training (online or offline) is not. Training a machine learning algo isn't like mining buttcoins, it's not embarrassingly parallel, with the exception of genetic algos which are not currently state-of-the-art. You benefit a lot more from one beefy computer than you do from several tiny computers, and when you add in asynchronicity and unreliability of individual nodes, plus the requirement to stay hidden, it just becomes intractable. >How would you even counter something like this? Reducing reliance on the internet would be a good start. >>71762 >The analogy I always see is playing chess against Magnus Carlson. You can try to anticipate his next move but he's thinking much further ahead than you. You will still probably lose. Your best bet is to avoid playing Magnus Carlson. The best way to beat Magnus Carlson at chess is to cheat. Or better yet just pull a gun on him. Reality isn't a fair game.

>>71009 I think it's more reasonable to expect artificial life-forms to evolve from worms and viruses like this rather than someone using a machine learning algorithm.

>>71750 Like I said, it's not a distributable task. Any number of 4GB GPUs are useless for this even if you fully utilize them. If using memory in someone elses computer over the internet was good enough you might as well just use regular ram instead of what's built into the GPU, the reason you can't is the need for very low latency. To train a model you need to load the whole model into one machine. The bigger the model the more powerful it can get. This is why supercomputers are being built for this and consumer hardware can barely run the compressed versions of the weakest LLMs let alone train them. There is plenty of reason to be scared of AI but this isn't it.

>>71785 >Like I said, it's not a distributable task. Any number of 4GB GPUs are useless for this even if you fully utilize them. I wouldn't say it's "not distributable", you could hypothetically do it, it would just be very slow.