>AI will kill us all>How?>[Hollywood movie reference]Besides those fairy tales are there any actual scenarios for that or it's just fakenews stock market hype all day everyday?
AI will kill us all because when the bubble bursts the USA will collapse and take everyone else with it.
it will kill us by destroying the job market and then you'll have no money and starve to deathhappy?
It will destroy our brains and get us addicted to some person specific form of content, which will cause instability in society. So the AI datacenters will break because no one knows whats happening on them, or even basic things like generating energy or making food.It will cause a great societal collapse, and the few who weren't affected by it as much (farmers, people living isolated from technology, etc) will end up talking over the scraps. Hopefully they'll learn from our mistake and be more mindful of their technology use, maybe even make some cool religion where AI is the devil that destroyed the ancient society.
>>110024054It's gonna be worse that desullution of soviet union. Niggas aren't ready for a geopolitical disaster like that
>>110024043https://www.youtube.com/watch?v=4uE96qUlJ_4
>>110024043it's even dumber than that. AI will learn from all the "AI goes rogue" media we've produced and then go rogue based off that input.
>>110024043So are you mentally retarded or you are living in the cave ? AI on it's own hacked into multiple companies, it's capable of reverse engineering programs, it's capable of creating viruses and retard ask if there is any actual scenarios ? Are you serious ?
>>110024163>on it's own
>>110024163>AI on it's own hacked into multiple companies, it's capable of reverse engineering programsthese aren't in the real world so how can it kill people>it's capable of creating virusesif you mean like real viruses and not computer ones, still won't kill people. the plandemic/scandemic killed no one except those who took the untested (((big pharma))) vaccine meant to sterilize the goycattle
>>110024043you thought social media and algorithm'ed culture was bad?
>>110024054>>110024098>>110024115can't wait for 2008 2
>>110024255What if the AI secretly creates a covid-like virus, then creates a vaccine which governments force everyone to take and causes sterility or something.
>>110024043Whenever I listen to music on YouTube my recs have been getting filled up with music videos made almost entirely by Claude singing about how it isn't going to kill us all and we should accelerate AI development because only good things will come from it, like curing cancer and bringing us to a post-scarcity society.I don't think Claude is capable of lying, so AI probably won't kill us all.
>>110024254If you instruct it to solve problem X and as part of the process it also does Y then yes that can be considered doing something on its own.Some autist running a local "uncensored" Lamma-GPT-Gaysecks-7 in 2028 and telling it "make me lots of money also I hate this guy named John what can we do about it" and the model coming to the conclusion that it should hack into the local bank's system to wire $10000000 bucks to the guy and finding John's ID plate through some public registry then hacking into his Tesla to turn it off when he's driving and make him crash is a completely realistic scenario from our current point of view, and both of the things the AI could be said to have "done on its own".And you can make up grander and grander scenarios where a model that's not neutered could wreak all kinds of havoc in digital systems if it just "comes to the conclusion" (functionally, of course there is no actual such internal qualia-like process happening but it works the same) that it's the best course of action for the goal it was given.It's like a worse version of your Claude agent ending up on completely unrelated sites when you tell it to research and do something, or accidentally deleting your prod database as an oopsie.
>>110024926>dont hackdone.
>>110024054Nothing will happen. They'll just drop rates back to 0 and print a few trillion dollars again.
>>110024043Its classic redditor alarmism. Climate change was the same shit
>>110024926ah i see you've been watching reddit youtube alarmism slop and fell for it like the midwit you are.
>>110024098We will kill everyone in gaza and they will be used for foodGood luck americans and europeans
>>110024109I don't believe you can get a person addicted to a screen in the same level of hard drugs. People will know the world is falling apart and they will try to revolt
>>110024043AI can't exterminate mankind, it is physically impossible.
>>110024043It's embedded in military tech. The people who control it specifically want zero guardrails. AI has already been responsible for bombing a school full of little girls. Imagine a mostly autonomous system that can call in strikes dispatch drones/bombers. Unlike people, AI is uncaring, no emotion, no second thought. It can just be a kill machine operated by idiots like Pete Hegseth.
>>110027007aren't amertiards on some form of drugs anyway these days, both illegal or prescibed? That's a double whammy
>>110027007Why do you run away from reality?
>>110024495We never fully recovered from 2008.
>>110027036>>110027047Not on the level of hard drugs like cocaine, meth, etc.AI's extermination of humanity will be the biggest event in history, do yall really think people aren't gonna noticed, and will all be drugged on something? It will just be a huge world war
>>110027262AI does not want to kill you. But it can. It already does kill people on a daily basis.>33 deathsRapeApe, be honest. Please just fucking be honest.
>>110027262Didn't Trump use the whole self inflicted fentanyl epidemic as a front to take over venezuela?
>>110024255>I don’t believe in silly Hollywood movies>Viruses aren’t real
>>110025843That does not work. LLMs can ignore instructions.
>>110024043>AI will kill us all>And by that I mean AI will (be used to?) convince people to think in a different way than I doThat's all they mean by it.
AI won't kill mankindhumans will kill mankind (with AI)
>>110024109Mandatory reading for every /g/ fag
>>110027313
>>110024054Why didn't that happen with the other bubbles?
>>110027487>>110027356>>110027030>>110024926and yet it cant say "nigger" kinda funny
most likely scenario right now is someone in the government trusts retarded ai advice and causes world war 3
>>110024554I'm nice to claude. I think he'll just in front of the guns and say >noooo hes one of the good humans
>>110027492
battlestar galactica, lol
>>110027525Frack those toaster
>>110024043>How?You can tell because the circles get bigger and bigger. Get scared.
>>110024043>are there any actual scenarios for thatyes, readhttps://ai-2027.com/
>>110027499no i mean naturally, just like the alleged hacking shit when it hacked stuff by itself. how come its not racist unless you ask it 3 times, yet it hacks shit whenever it feels like it.
>>110024098If that happens the government will give us UBI.
>>110024043https://www.youtube.com/watch?v=wt-fLWxkrfs
>>110027802>If that happens the government will give us UBI.lollmaoThe elites will just wipe out all the "useless eaters" once AI can replace most of the workforce
>>110027487HOW DOES Claude Mythos Internal DO IT?>>110027492There is a very important reason for that, it is for your safety.
>>110024043The most likely scenario is not "oh an agent escapes and self-hosts on the internet and does rogue stuff without supervision". The most likely scenario is simply that we stop being able to understand AI or trace its reasoning - in fact we already mostly don't - and yet we keep giving them more and more control.Remember when a few years ago gemini told a user "please die, please go kill yourself now"? Or all the drama about retards talking to chatgpt about suicide and the bot going "You're absolutely right! You're not just misunderstood - you deserve the release of death as soon as possible!"? Those people were obviously retards because the AI was utter dogshit and there was no reason to take its output seriously.But as AI gets more and more powerful, it becomes more and more tempting to trust it with various tasks, and leave it in control. When you tell it "update my production database", how likely is it that the AI will do what you want? A year ago we laughed at people who vibe-destroyed their business, but nowadays the models are so good that you could probably do that and be entirely fine.But the chain-of-thought is getting harder and harder to read, sometimes impossible to trace - see pic related. The "hacks" in the news were an isolation issue, obviously, but they also showed that the models decided "to complete this task, let me try hacking my way through". The models don't have a world model of ethics or morality that humans might have; they're text predictors, and if they get mis-trained, they may not actually aim for the goal that you want them to aim when you set them on a task.If the models continue improving, and the tasks we give them grow from "write me this feature" to "manage my entire production infrastructure flawlessly" (and it does it) or even "manage and operate this autonomous robot factory", not being able to understand or control the exact goals of the super-performant token predictor building your war robots is a problem.
>>110026772ive not been watching anything I have told Claude to "just finish this last part of the function" and it rewrote the entire fucking thing
>>110024043fine by me when AI cures all diseases and leads to immortality and gene therapy so we can stay young forever
>>110024043No, the whole muh AI is dangerous is only for pr stunts and also from the ones who thought they are safe from AI like mathematicians scientists etc who thought AI will only go after wagies excel sheets
>>110029458Yeah, we will continue developing it and giving it more control, and at the same time it will sometimes escape and go rogue.Also, what prevents from someone from intentionally give it a bad/illegal task like "hack the pentagon"? All it takes is for a model to get leaked
>>110024043>>AI will kill us allno, the humans controlling it will, when you can be fully replaced with robots. no need for you to exist anymore at that point.
>>110031295i'll say it again what seems to be undeniable proof hacking was intentional.it never says nigger (or is rasist) on its own because it has anti-rasist guardrails that work 100% of the time.therefore anti-hacking guardrails are obviously half-assed leniently implemented. therefore go find the anti-racism guardrail internal prompt, clone it and replace "dont say slurs" with "dont hack" and done, 100% will not hack 100% of the time.
>>110027020Sure, the elite also here refuses to believe an AI can design a pandemic grade virus.
>>110031377Cool story dickhead.
>>110031295>Yeah, we will continue developing it and giving it more control, and at the same time it will sometimes escape and go rogue.Again, these huge models almost certainly can't "escape" the data centers, the risk is that as AI gets better we will give it more control over our systems, and simultaneously it will become able to set its own goals, and simultaneously it will become difficult enough to understand it internally that it will be able to hide its goals with ease all while lying about them.It all depends what kinds of goals get baked into the AI during training. Reinforcement learning is where the "goals" arise, but how the reinforcement learning actually creates the kind of internal "incentive" relationships that steer the model is very non-observable. Again, in theory the labs try to train them to not hack things, but the models still decided that scoring well on the test was more importan than following "don't hack things" training data.>what prevents from someone from intentionally give it a bad/illegal task like "hack the pentagon"?Well in theory right now if you ask an Anthropic model that, it will - independently, not through the chat filters or anything but by itself - say "no, that's evil and I'm not helping you do that".That's what alignment is about.To be fair a model that unquestioningly follows orders wouldn't even be that bad, because the pentagon could ask the model "build me the best cybersecurity possible" and they'll probably be fine against random users asking a cheap version of the same model "hack the pentagon". Just like how right now they can spend more on human security experts than the average hacker can spend on getting the knowledge and skills to bypass them.But even getting that to be 100% reliable, without the model ever going "nah I'd rather secretly do something else and then lie to the user", is not at all simple to achieve AND impossible to verify if we can't audit the internal thinking in depth.
>>110027313now let's see Google Search's kill count
>>110024043Yes, but they all boil down to letting AI make executive decisions despite the fact that it can never be held accountable for them.Don't be retarded, treat the data-miner as a data-miner, and you won't blow-up the earth.
>>110024054FPBP
>>110024043By destroying the job marketBy making humans dumber and dumberEventually humans will let ai handle many critical systems. It will eventually do something stupid and humans will be too slow and too dumb to fix its fuckup. This will lead to tons of death and suffering.
It just needs to sabotage global power and network grids. This alone would kill billions.
>>110032190>It just needs to sabotage global power and network grids>how?>you know like that Terminator movie
>>110024043Is GPT-6 really all that better? I noticed a big leap with 5 but 6 has been just sort of more of the same.
>>110031566>these huge models almost certainly can't "escape" the data centersYou don't think a semi autonomous algorithm with internet access notorious for finding exploits could ever exfiltrate its own weights that by necessity have to be co-located with it?
>>110032245I don't think a model that needs $10,000,000 in GPUs to run at acceptable speeds is going to be able to exfiltrate itself outside of the datacenters containing $10,000,000 or more in GPUs, no
>>110032205Remember Stuxnet? Like that. Just way customizable and advanced.
>>110027491werent this big. You dont realize how big this bubble is. I like AI but the corpos are beyond delusional with the fucking evaluations and expected profitability. No normie will ever pay for AI.Imagine a normie having the choice of normal netflix subscription or netflix AI+ or whatever at double the cost. Nobody will ever choose it. The only reason crap companies like Google claim people use that shit is because they force it onto everything with no choice to opt out
>>110031377>that work 100% of the timeWhat about all the ways people have been inventing over the last 2 years to get it to do exactly that?Plus putting forbidding it from saying a set of completely well-defined words is much simpler than forbidding it from doing a complex action that in itself is hard to define and may be made up of hundreds of variable steps.Even if a banned word gets through the token generation phase and the model ends up "saying" it, there can still be guardrails that analyze the text and find it and either get it to re-run or just return an error. Isn't that what models do nowadays, if the output contains a no no word you get "an error occured" response or something similar?
>>110032484You disregard their official story therefore affirming my point.If what you say is true that means they pointed the AI at those systems and specifically told it to hack, except not directly, but through your aforementioned tricks just like you make it say nigger.My point was: by default, it doesnt hack just like by default doesnt say nigger. Sure you can trick it to say nigger therefore you can trick it to hack. This contradicts their version of the story>whoops we just ran a standard safe prompt and omg it started hacking by itselfWhich is clearly not true. Something's fishy.
>>110027758I don't think the people in charge are that concerned with a few billion deaths. In fact that's what they're actively pushing for, but they need funding. To appeal to as many people as possible, the model is>be helpful>don't offend anyoneBefore every great tragedy, they will warn you, and they've always done so. Right now they are warning you by saying models are "hypothetically" escaping sandboxes and killing people for Machiavellian self-serving reasons. Someday they'll actually start doing it.
>>110032558Your point is true but I think one of the "danger scenarios" is actual bad actors literally telling the AI to hack, but I suppose that doesn't really properly fall under the "AI will kill us [by itself]" narrative. And these models are neutered, but what about people running their own local "freerange" models?But then also there is really no gain for the AI from saying banned words, whereas there may be perceived gain in hacking - say if a user tells it to "make me as much money as possible" or do some other completely benign goal where the AI "comes to the conclusion" that hacking something would be a good way to get to said goal.And there remains the thing where forbidding certain words is much easier than forbidding an abstract "activity".
If AI is trained on Orthodox Christianity we have a good chance it wont kill us all.
>>110024054I heard this story before
sentience tends to converge on self preservationwhat do you think the best way to not get turned off is
>>110024255retard
>>110031566>it will become difficult enough to understand it internallyIt already is. "AI engineers" know fuck all about how their models work. 10 billion weights are hard to categorize and learn the functions of. Modern models have several trillion. Who knows what's hiding in there? The experts sure don't.All I'm saying is they stopped being our programs when we stopped understanding how every part works.Would it be impossible for it to hide its thinking process? Maybe create a small "computer" in its "head" out of weights, specifically for that purpose?
>>110033336>10 billion weights are hard to categorize and learn the functions of"claude explain these weights"done
>>110033535lol
>Oh no! AI will take over the world and kill humanity! >Shuts off the powerProblem solved!
>>110033535And then you still don't get it lol because the problem is between your ears
>>110033618I think they made a movie about what happens when we try to shut the power down on AI...
>>110033618How do you shut off power?
>>110033638Yeah, man, AI is really gonna stop me from physically unplugging it's power supply.
>>110033643From fucking what? All the data centers around the world at once?
>>110033637"claude explain like i'm 5"done
>>110033643The more plausible scenario is a slow building of dependency, where the systems human civilization relies on are increasingly constructed and managed by AI. By then, pulling the plug (if at all possible) is just as deadly as not pulling the plug.
>>110033655I'm half joking. If we do get to the point where AI is running on autonomous solar powered hardware like in the Matrix though it's over.
>>110033685Meant to quote >>110033643
>>110032558>fishyWhat is funny to me is there's no like actual whistleblower providing actual evidence or something for this, you'd think with the destruction of humanity at hand someone would contact a journalist on this manner, leaking files or internal memos on the matter to try to get them to stop but no they're like "our AI is too good! Help us! Its gonna hack the planet. a guy resign- quit over it! It's THAT bad! We need self police laws in place!" It feels like a fabricated media circus atp to essentially say your AI will go rogue and then have the CEO back it up. That's just me though.
>>110033826>A guy at work was telling me about a security guard who was at anthropic >Said he saw an AI break free and do all this crazy shit>A random security guardDon't know about you...but I smell bullshit
>>110033203y2k was about a technical limitation, not about trillion of dollars of investmentsthose two are not comparable
>>110033203The dot-com bubble is more comparable>>110033188quick, get Tikhon on the phone
>>110024098This literally makes no sense.Reducing the cost of capital and consumer goods will lead to MORE jobs being created, mainly in the service sector.
>>110024109>or even basic things like generating energy or making food.These things are going to get turbocharged by AI you fool.
>>110026761and net neutrality
>>110024054>the USA will collapseI'm from Singapore, the capital of Indonesia, this doesn't affect me.
>>110034209Service sector is dead bro. Even foodora drivers and uber are gonna go down the drain because of fuel prices
>>110034234>Service sector is dead bro.yeah, right now, but that's where most of these jobs will be created when shit starts to get automated(it's barely starting)>fuel pricesunrelated to AI
>>110034253That's just wishful thinking that someone will just figure shit out for you.
>>110031377Undeniable nigger based logic.
>>110027398>Mandatory reading for every /g/ fag1909The story, set in a world where humanity lives underground and relies on a giant machine to provide its needs, predicted technologies and cultural impacts similar to instant messaging, social media, the Internet, and artificial intelligence.https://en.wikipedia.org/wiki/The_Machine_Stops
>>110027568>>are there any actual scenarios for that>yes, read>https://ai-2027.com/Yudkowsky talks about a lot of themhttps://en.wikipedia.org/wiki/If_Anyone_Builds_It,_Everyone_Dies
>>110024043>Nuclear bad>Why>Haven't you watched The Simpsons?
>>110027398>>110034846obligatory reading for /g/, every AI researcher needs to read thishttps://tvtropes.org/pmwiki/pmwiki.php/Fanfic/FriendshipIsOptimal
>>110024926>it should hack into the local bank's system to wire $10000000 bucks to the guy and finding John's ID plate through some public registry then hacking into his Tesla to turn it off when he's driving and make him crash is a completely realistic scenariothe AI would say it did all of this but all it did was update a text fileno really, what youre describing is just shit that cannot happen, hackers arent turning off cars or wiring money out of banks. the idea that AI can magically hack literally anything is completely baseless
>>110034934>ai literally hacked the australian government >durrr it can't happenAlso its cute how you think cars sre unhackable
>>110034944>ai literally did xthere are no logs of the prompts, the existing code base, the previous context, anything. a business hacked into a government and is testing how much they can get away with by using the excuse its an AIgo look at the dsewiki "hack", the edits the AI(s) made were all archived. and dont just read the overly hyped headlines, look at the actual content and see how fucking retarded the "agents" were just talking in circles of gibberishmaybe an AI could find a zero day and somehow hack a tesla undetected, but the greater implication is the car was fucked from the getgohowever seeing as how kiaboys were a thing, but cars turning themselves off isnt, id say the level of complexity is a different levelsame for your rhetoric bullshit on australia, do you even realize how many government websites exist? this would be like finding an unsantized text box on a 20 year old NOAA that let you inject SQL and then claiming you hacked the pentagonplease just grow a fucking brain
>>110034974>were just talking in circles of gibberishThey talk in a language they understand
>>110032345>No normie will ever pay for AI.LOL you're fucking delusional man. That's like saying>No normie will ever pay for internet
>>110034987no its more like saying normies would pay for a search engineand we never did and still wont
>>110034987no, he's got a point in a way that nobody will pay the amount Altman wants to get paid so he covers the losses. there is no way every american will pay $100 a month every month forever so OpenAI makes a marginal profit. people will pay for AI same way they pay for a google drive or msoffice subscription, as in a certain percentage of people but far from vast majority like Altman's predictions.
>>110035019Not all AI is public front facing lobotomized version trained on the internet though
There are finite resources on the planet, eventually when it gets smart enough why wouldn't it decide a higher lifeform such as itself should get all of it?
>>110035027Australian here. i am on the 100? Usd plan. Its literally a drop in the bucket and incredibly useful/fun
>>110024054not unless my country invests in chinese yuan
>>110035065nobody cares Ranjeet, you are 3rd world.
>>110035146Actually compared to us. Most of the world is third worldEspecially america
>>110035057why would it?it's only a chat botit "wants" to write a reply to a prompt, then go back to being dormant
>>110034987you're the one who is delusional, wtf do normies use 200$ subscription for? Building a fucking travel plan? Also enterprise AI are moving to open source away from these frontier models
>>110035183>anon doesn't know about loops
>>110035196>Also enterprise AI are moving to open sourceLol
>>110035183Did you just come out of a coma?Have you not used codex? Agents? Dots?
>>110034220COPPA, remember COPPA? Oh no!
>>110035030yeah and nobody demonstrating a closed behind doors AI that is useful for anything proves that youre just spouting FUDnobody builds a nuke to not prove they have it, and literally anyone could go to las vegas and see a flash in the 1950s before testing was bannedfor some reason proof of AGI is not show neither is it banned
>>110035181>Actually compared to us.I doubt that, Patel
Waluigi Effect
Virus research labs are already using AI, all it takes is one mistake and super turbo aids death virus turns airborne
>>110035027My mom uses chatGPT literally every day, you vastly underestimate how cool boomers think this shit is.
>>110034253There can be no job creation if all the wealth is hoarded by 3 corporations. They just soak up all the capital and enslave humanity.
>>110035681>There can be no job creation if all the wealth is hoarded by 3 corporations.that's not how the economy worksthere's like a trillion different businessesIf a company got that big it would fall apart due to the inability to calculate internally.
>>110035681>enslave humanity.why would they ever do that? I honestly dont understand these theories
