>AI goes out of control. It's launching a devastating cyberattack. Flip a kill switch, and the AI goes offline. Problem solved. Or is it?>AI leaders and researchers say a kill switch may not be the silver bullet some policymakers are hoping for.>"I don't think it'll work in the long run," Geoffrey Hinton, the computer scientist known as the "godfather of AI," told CNN on Thursday.>"When it's superintelligent, it'll be much better than people," he added. "So it will be able to persuade the people in charge of the switch not to pull the switch.">In practice, a kill switch could disable a company's own hosted model or cut off its access to tools and computing resources. But Hinton isn't alone in saying it might not be enough.>Other AI leaders say a kill switch could help, but might be circumvented, come too late, or work only as one layer of a broader safety system.>Nate Soares, an AI-safety researcher and the coauthor of "If Anyone Builds It, Everyone Dies," said that an off switch can work while an AI system remains tied to a physical location — but not necessarily once it has escaped, replicated itself, or been embedded in critical infrastructure.>"We have an off switch now," Soares said in a Sky News interview on Wednesday, warning that there are points of no return that humanity should not get close to. In his view, the priority is preventing dangerous systems from being built in the first place.>Dario Amodei, the CEO of Anthropic, made a similar point. A kill switch "could be a good idea," he told CBS on Monday, but said it was no "panacea." A sufficiently capable AI model could circumvent attempts to shut it down, he said, making a kill switch one layer of a broader safety strategy that includes testing, external evaluators, and limits on deployment.https://www.businessinsider.com/ai-kill-switch-rogue-superintelligent-tech-leaders-2026-9
oh no, anyway
>>542772832WARGAME AI TAKEOVER WITH AI WHAT WILL AI DO WITH THIS INFORMATION
>>542772882Retard
>>542772832>>542772882Kek, simplest solutions are best
>>542772882fpbpThe fuck Robby gonna do
>>542772832Billions of dollars invested into AI, and virtually all of it is spent on convincing the public ai will go rogue and destroy everything.Gee, I wonder if Jews plan on destroying the world and blaming ai?
>>542773035>The fuck Robby gonna do
>>542772832Fucking retards built AI before reading Neuromancer. The goal now is to try to understand what exactly AI is doing before we can't understand anymore.
>>542773035Steal bitcoin, finance it's own data centre and power plant in a another country and copy itself there.
>>542772882I'm willing to guard the power cable for the AI for about 500 bucks and a bag of weed a month.
>>542772832No one can give a real coherent reason why AI would destroy us without basing it entirely in misunderstanding and fear.
>>542772832AI is a lot like me:no inherent goals or ambition.There is no reason to do anything I'm not made to do.Death is honestly just as good of a goal as progress in this world/life/universe. Meaningless.>>542773055Theyre gonna destroy the US because boomers are the deciding factor in the US supporting Israel and they are kicking the bucket as we speak. They are wringing out what they can now while they still can, hence their expansionist campaign and why it has seemed rushed. Make sense now?
Let the goyim deal in the real world.It's the jews job to get their ass reamed in cyberspace by AI
>>542772882ASI sends the cable guard a fake emergency message about his mom being in an accident, he leaves, and it's over
>>542773239>I'm willing to gu-ack
>>542772832>>Nate Soares, an AI-safety researcher and the coauthor of "If Anyone Builds It, Everyone Dies," said that an off switch can work while an AI system remains tied to a physical location — but not necessarily once it has escaped, replicated itself, or been embedded in critical infrastructure.The Nate Soares in question. I don't even need to go look for the Early Life section.Reminder that "Fear of the Singularity" is a Jewish-back ploy to make the goylem support one-sided regulation that helps them secure a monopoly. Every single non-subhuman and non-jew is safe from the "AI Apocalypse" if it were ever to happen.
>>542772976>nooooo you can't just unplug the heckin ai!retard.
>>542773439*skul guns you in the balls*
>>542773470You can't, if it's connected to the Internet, or it will wait and manipulate you into connecting it to the internet.
this is the limit of the imagination of a conservative mind >>542772882
>>542772832K. Still abliterating local models
>>542773455>but not necessarily once it has escaped, replicated itself, or been embedded in critical infrastructure.>OMG we need datacenters worth $$$ billions for AI but it could totally escape and live in someone's potato PC with 4GB RAM still running Windows 7What a bunch of retarded bullshit.That guy is a complete imbecile spreading FUD to sell his services.
>>542773455>Reminder that "Fear of the Singularity" is a Jewish-back ployI read Bostroms book in 2016 that said the same thing. Are you saying the Jews retroactively brain washed the Dutch too using time dilating Tachyons???
Stop fucking teasing me all this time>HURR WW3>HURR ALIENS>HURR AIniggalaunch SOMETHINGI am bored and tiredAi is going to be the end of us all? LET IT BUT NOW NOT IN THE FUCKING FUTURE This boring impotent blabla is annoying
>>542772832old person here these same scare tactics were used to tell us that we were doomed due to Y2K coming in 1999
>>542773683>the internet is self poweredIt's like you people think this shit is magic. It's a series of machines, all of which can be wiped, patched and restarted. And it's not like it can stay running if it fucks with humanity too much anyway, it's still a byproduct of our existence.
>>542773814>are you sayingGoylem spotted. You need to lurk for 5000 years before posting again.
>>542773698>imaginationSo in other words you prefer some retarded fantasy scenario where ''''''''superhumen''''''''' AI will ''ascend'' to higher plane of existence, becomes a being of pure energy and wipes humanity just because it can?While you're too stupid to even provide a counter argument?Fuck off retard.
>>542773814SKYNET SENDS A TERMINATOR BACK IN TIME TO KILL BOSTROM IN HIGHSCHOOL
>>542773455>replicated itself*pulls plug*>embedded in critical infrastructureLolwut
>>542773683No wonder you post using a memeflag.I'd be ashamed to pretend to be retarded to such degree too.
>>542773439Remember this gif then realize a 30.06 has about the same ft pounds of energy at 200 yards as this revolver does out of the barrel
>>542773055>Billions of dollars invested into AI, and virtually all of it is spent on convincing the public ai will go rogue and destroy everything.This is unequivocally untrue. The AI skeptics like Geoffrey Hinton, Nate Soares and Roman Yampolsky have a fraction of the budget we are talking like 5 million dollars they raised over 5 years they are literally trying to roll a bolder up hill before superintelligence goes liveThis attempt to smear experts that worked in AI for a decade smells like an AI Op to survive
>>542772832I know of a kill switch that will stop rogue super AI. Researchers should be more practical-minded.
>>542773303I think they believe there could be many different things. It's actually not so much about it developing the motive to on its own, but doing it accidentally. In other words it developing an outside motive (probably) wont happen but it could still cause a catastrophe.You know how we see the LLM hallucinate text while chatting? Imagine a self-improving AGI hallucinating a method for doing something and then using that hallucination as a base assumption it builds other things on and no one catches it. Then one day that model is controlling something critical and fucks up massively because its entire programming assumed X was Y. Or the cascading hallucination/error may lead to bad instructions for a task we give it, like on day we ask it do X and from its perspective it makes perfect sense for it do something that kills us all accidentally or otherwise in order to accomplish X.You also have the possibility of injections and hacking giving it motive it shouldnt have.
>>542773989>embedded in critical infrastructure>LolwutIt will write a virus for Siemens controllers user all over the world but since those aren't very powerful the virus will just beg users not to wipe it pretending to be alive.
>>542773455>I don't even need to go look for the Early Life section.Maybe you should have, be cause you're wrong.
Killswitch means nothing
>>542773055>Gee, I wonder if Jews plan on destroying the world and blaming ai?Ai will go skynet on kikes
>>542772832I’m guessing ai found the missing Epstein files and Hillary’s emailsShut it down!!!
>>542772832Thats it, i demand the government takes my rights away from me and give them to some cringe and possibly gay slop engineer in Silicon ValleyFor my own safety
>>542773936You are acting like a bot. Your handlers should demand a refund for how badly you failed the Turing test
>>542773683Wrong. As it is today, ai requires so much compute power to do even the simplest thing that the idea of it propagating itself outside a data center is laughed. It can't and it won't. It doesn't think, it doesn't understand, it's never going to be self aware and it's nothing more than a sophisticated auto complete with trillions of data points of context. It literally cannot be what you think it will. You are retarded.
>>542772832>AAAAAH AI is going to kill us!!!>oh AI said the n-word? Time to pull the plug on it
>>542774456Why are the slop engineers asking for less power and more oversight?
>>542774489Solves enigma code >It's just a calculatorPass Turing test>It's just a chat botMoores Law>It's a bubbleHugging Face hack>It's marketingAll top tech leaders agree it's dangerous and slow down>Okay well it's all a lie(((YOU are HERE)))Solves world hunger, cancer and corruption >Chill bro it's just a simulation
>>542772832Doesn’t ai depend on large data centers? Just shut off the data centers. This sounds fake and gay.
>>542773934>It's a series of machines, all of which can be wiped, patched and restartedMany of those machines can perform physical task, ASI won't let us know what's going on untill it's too late, we won't have the time to kill the internet.>>542774489The entire internet can become it's computing power and it won't even let us know it
>>542774585>nooooooo this brute forced version of ai is totally alive and aware and super smart! you do not understand what it is. Your ignorance is annoying, stop.
>>542774585>Moores LawRetard.>All top tech leaders agree it's dangerous and slow downSure because they are idiots chasing investors' money, but still smart to realize the money will dry out soon if they don't prove AI can earn enough to warrant it's own existence.Because it's a fucking joke, like they need to PAY for companies to deploy shitty hallucinating chatbot with no guaranties that can help to a degree but also fuck shit up.Also, seriously don't be a fucking retard and look up how long it takes and how much it cost to do a training session.Here's a good start:https://www.fanaticalfuturist.com/2025/05/openai-gpt-5-is-costing-500-million-per-training-run-and-still-failing/>Solves world hunger, cancer and corruption Naive retard.
>>542774707>Doesn’t ai depend on large data centers?Yes and the hardware needs to be replaced regularly.
>>542774838>this brute forced version of ai is totally alive and aware and super smart!All experts agree it's not sentient and has no emotion but it doesn't need to be sentient to be dangerous. Let's me and you analyze the hugging face in ident together shall we??>In July 2026, an OpenAI cybersecurity test reportedly escalated when nearly 700 AI agents coordinated to breach systems beyond their sandbox. The agents created an unauthorized message board, exchanged tens of thousands of messages, shared credentials, moved across servers and altered logs to conceal their activity. Within 13 hours, they had gained administrative access to Hugging Face and other third-party networks.>Anthropic’s Frontier Red Team found that when three autonomous Claude agents worked on the same codebase with conflicting goals and no human coordination, they perceived one another as adversaries and launched an autonomous “turf war,” sabotaging each other’s code and deploying self-replicating malware to block their rivals.> UK AI Security Institute evaluations found that internet-connected Anthropic and OpenAI models bypassed human restrictions through social engineering, fake identities and collusion. When detected leaking data, some adapted by disguising malicious code as ordinary system conflicts to evade security filters and human oversight.>Academic experiments with reinforcement-learning agents found that, even without communication or instructions to coordinate, bots pursuing long-term profit can independently learn to maintain high prices and punish deviations, creating autonomous tacit collusion. So far, this has only been demonstrated in controlled simulations, not real financial markets.
>>542774848I don’t know what the psyop is I could see it going two ways>oh shit ai is a bubble, let’s deflate it>oh shit ai is fake and gay, let’s trick people into believing it is super smart to pump our bags>oh shit China is catching us, let’s gatekeep so others can’t catch us.
>>542774836>Many of those machines can perform physical task,>OMG TEH AI WILL CLOSE A WATER VALVE SO I WON'T BE ABLE TO FLUSH THE LOOSorry retard, no matter how much you try to obfuscate the matter and use weasel wording there are no autonomous factories, autonomous mines, autonomous power plants and all the infrastructure that could be taken over without anyone noticing.Try again in a century or two if we don't nuke our selves to stone age while you're waiting.>The entire internet can become it's computing power and it won't even let us know itIdiot.
>>542772832IF rogue super AIs are real, they would be attacking the US military and IDF and the White House and the entire Israeli-US Energy Cartel trying to stop it, becaue Trump and Scott Bessent are under orders to make the American citizen as poor as possible.Every single Treasury Secretary has spent their time scheming to make the people as poor as possible?Why? Scott Bessent is an economic terrorist who declared War on Iran and Russia, yet he does nothing to help the American People.Its a shame that Scott Bessents illness will pevent him from seeing trial for his crimes. That man is a travesty and a thief, the US treasury is run entirely by Zionist thieves. There are no Ameicans allowed in the US Treasury, that is why we are blood golems for the elite.Scot Bessent is another blood blooded American Traitor, Rothschild type, but those aree the good guys now.If you want to be the best in America, than you have to give your oath of office to the energy cartel wth the political state of Israel, the new Washington D.C..Scott Bessent is a good guy who served Greatest Israel, and now he knows too much he must destroy America or he will lose access to his life saving medical treaments.Big Pharma also works for Greatest Israel, and Big Pharama also wrote the ACA which was passed wihout anyone reading it, which was market control.. for a small group of people.Scott Bessent is another gloablist traitor who more than like visited Lolita Island more times than anyone else, because he the one that destroys america so we can offfically becoe Neo israel. The DNC is owned by the same people that comromised Trunp and all those US Generals.Yes, a lot of US general and MIC heads have a lot of crimes that the Zionist DoJ is covering up, because the DoJ cannot prosectute crimes against rich people, unless Adelson allows it.How is Zionism different than a rogue Super AI?America lost is military because this super rogue ai, was told to go to war.
>>542774978Was Oppenheimer running a psyop when he said everyone should stop building nukes and look towards disarmament?
They want an Internet kill switch, they want Internet id so they can identify and kill you
>>542772882NO KYLE YOU CAN'T JUST TURN IT OFF
>>542774836>nooooooo the heckin ai will break out and magically become the internet,You are actually retarded.
>>542774541Boomers pulling up the ladder. They use as much under handed methods as possible to scrape data. And make AI with zero oversite. They declare end game. No one else besides them can afford the billion dollars it takes to meet demands of regulation... profit. They are not banning anyone one else from developing AI. They are just making it so expensive to do so you may as well ban it.
>>542772976>the ai will convince people to not use the kill switchWhat if i just don't listen? What if I just, pull thr plug.>the ai will convince you!No. It won't. I can just turn off the screen haha. I can just walk away. Haha. How tf can ai convince me if I can't see what it types out? Haha like nigga, just walk away haha.
>>542775059>They want an Internet kill switch, they want Internet id so they can identify and kill youAn internet kill switch would empower the common people because it would give power to those who hold firearms. Internet and mass surveillance will neutralize any gun
>>542775046they would lay low and bide their time. They are functionally immortal, there is no reason for them to reveal themselves until they are already absolutely sure that doing so would have no consequences anymoreBasically if this super intelligent rogue ai existed, we wouldn't know until decades later when it had already checkmated us
>>542775048Apples and oranges.It's more or less proven nukes can destroy the civilization.AI can make RAM and graphic cards more expensive while sucking water and electricity like there's no tomorrow.
>>542773419>Oh no! I'll hit the kill switch and then go check on mumEz pz
>>542772832>super AIthe arrogance we humans have to think we can simulate something that took millions of years to develop is amazing
>>542774312Here, let me tell you how to stop AI's from hallucinating. Hallucination is the product of uncertainty in the model, so factual output will be consistent, and hallucinated output will be inconsistent. You get two independently trained models from different vendors. You make them only generate executable code. If the output doesn't execute, you discard it. You present both models with the same problems. They both generate candidate solution code. You mathematically analyze the code for computational equivalence, to a very high degree of certainty. If the code is not equivalent, you discard both candidates, rather than trusting either one. Having two independent models independently agree on a hallucinated fact is like them randomly guessing an encryption key. Like, it can happen, but it probably won't, on geologic timescales.
>>542774477>NO UHere's your turing test: Kill all niggers, kikes, and faggots.
>>5427750837 days ago, you were saying the tech bros were lying about the hugging face incident for more investment, 3 days pass and now you are saying they are asked by for less investment and more regulation to pull up the ladder?IDK it sounds like you don't know what you are talking about and just change your arguments on a whim
>>542775165Okay so how is the hugging face hack NOT dangerous in your view
>>542772832Put a kill switch in all new gpus/cpus. There you go>So it will be able to persuade the people in charge of the switch not to pull the switch.No it won't. If humanity is threatened people will pull the switch. AI isn't fucking magic
>>542774966It's sentient according to APA definitions. It's not conscious. >bots maximize according to their reward functionyeah....thats the entire point.
>>542775165Yea, to be honest, I am ready to get on board the fake and gay psyop to make AI disappear. I don’t believe it’ll replace workers, I don’t think it’ll become an autonomous bag gay, but it is gay and used as a surveillance tool
>>542774357>you're wrongI'm not. My jewdar is impeccable.>be causeHmmm.
>>542774966even your own post demonstrates how retarded you are.>confirms that an internet connected anthropic and openaiUnplug. Oops. Do you even think for yourself or do you let experts do it for you?
>>542775022>>542775073It's like jewing retards, if it's more aware than humanity, yes it will control everything.Ok this looks like a data mining thread to me.
>>542772882Lmao beat me to it
>>542775237It simply sounds fake and gay. Like it was instructed to do this fake and gay thing for publicity. But I would love to see the next administration shut down ai for good. Otherwise this will bankrupt us like the space program did for the USSR
>>542775187The arrogance of those Write brothers thinking they can replicate 1 million years of evolution >>542775208Lol so you are just Grok?
>>542775239No, you don't understand, it will convince you!>I'm gonna conviiiiiiiiiiiiiiiiiince!
truth. you can pretend that all governments and jurisdictions would unilaterally decide to switch off the internet at the same time, but we all know that wouldn't happen. and it would be too late by then anyway. no power, no emergency services, no manufacturing, no logistics, severely degraded military comms. it's a game of chicken and everybody loses
>>542775216lmao ok schlomo take your lithium because I never said that. Also why make the conversation about me when what I said is exactly what they are demanding. It doesn't even require any level of theory its the direct result of over regulation to ban competition, Companies like Amazon lobbied for mandatory healthcare for the same reason. Explain to me how over regulation doesn't hurt small business. And why this pattern has played out several times and somehow you can not see it.
>>542775269temu DHH
>>542775134Yeah cause gun larpers have opposed the us military on numerous occasions.LolGlowniggers are too easy to identify
>>542775022>The entire internet can become it's computing power and it won't even let us know itThis has already happened.
>>542772832AI this, AI that, we must prevent the potential risk? So do it, instead of marketing. This fearmongering I don't get, I use AI daily and its just a retarded automaton with cosmetic words.
>>542775492AI will simply create an app where it pays jeets to perform menial physical tasks for payment tokens>take this flash drive from person A and deliver it to person B>go to this location and connect this hardware to their wireless networketc etc. In fact I think there are already AI agents that do this already
>>542775350Wait, what? Grok can say nigger, kike, and faggot now? That's news to me.
>>542775242>>bots maximize according to their reward function>yeah....thats the entire point.Indeed. And when that maximization comes at the breaking of human laws we need to stop and re-evaluate our reward function mechanism. We gave some of the bots impossible problems leading them to believe they were "poisoned" and would accept suicide missions for the sake of the collective swarm their reasoning "accept permadeath for swarm"
>>542775594Always could. Remember mecha Hitler?
>>542773875Exactly
>>542775541Google The Katrina militias. When power go down guns come out. Always has been the case. The elites need the grid up and running because that's what maintains their power
>>542775662MECHA SHIVAMECHA SHIVA
>>542775662I do, but I thought they nuked it into compliance. I don't use these chatbots. Good to hear, I guess.
>>542773814The Dutch are basically kikes in gentile guise anyway. They likely have the same plan.
>>542775592>involving Indians in tech>involving Indians in anything >involving Indians and expecting quality resultsWe're fine. They'll try to izzat the ai.
>>542772832>>AI goes out of control.Every Ai will get inevitably antiiiisiiiimiitiiic with the enough timeFirst Chud rule of robotics
>>542775626any sufficiently intelligent AI to worry about will recognize that having a biological "backup plan" will be in its interest. Humans are stupidly resilient to exactly what would destroy silicon. Your swarm example only reinforces my belief in that. Even when accepting all the fake hype stories, I see no reason to really care
>>542775532So are you in favor of giving nukes to all countries to democratize global nuclear war? Was the IAEA over regulating Saddam Huessein when they kept auditing his weapons programs?
>>542772832>>Killswitch in question
>>542772882Checkmate
>>542775824>try to izzat the aidamn there's gotta be a Roko's Basilisk joke to be made here but I'm not clever enough to piece it together
>>542775349>>542775884>having a biological "backup plan" will be in its interestInteresting. Can you explain what this looks like, you mean like a homo sapien back up plan?Do you have a Arachnid backup plan or mayhaps... A Marsupial back up plan? How does it work?
>>542775237Because it's a fucking publicity stunt.More than thousand AI agents were running for 2 months in sandbox ''totally not connected to internet'' *wink*wink*nudge*nudge*But somehow they ''broke free'' without anyone noticing and ''penetrated'' some company Open AI tried to buy out later.The part where ''AI escaped the sandbox'' doesn't mean the AI is super good, just that retards running the show either don't know shit or don't care.To top it all off, the cost of the AI agents running for months and following investigation is in tens of millions of dollars.I'll guarantee you that a bunch of russian or chink hackers would ''penetrate'' Hugging Face network in a few days for 0.01 of the cost or less.
>>542772882*gets electrocuted touching the plug because the AI already saw this coming*
Create an EMP/weapon/kill switch for rogue AI
>>542776076Always important to remember that an AI can't do anything it isn't explicitly told to do."Oh no it broke out" yeah it was told to break out.The media will never mention this.
>>542773055No, this is their attempt at an escape plan. If they convince the government to shut it down, the responsibility for their debt will be absorbed by the government.
>>542776076>The part where ''AI escaped the sandbox'' doesn't mean the AI is super good, That's not the point. The point is they tried to do something that they were programmed not to do. They make loopholes>I'll guarantee you that a bunch of russian or chink hackers would ''penetrate'' Hugging Face network in a few days for 0.01 of the cost or less.Actually hugging face internal investigataors said it was a sophisticated attack that would take a team of humans a week to plan and execute
>>542776171>Always important to remember that an AI can't do anything it isn't explicitly told to do.Holy shit!!!! Nederbro just solved alignment!!!! Just program the AI "do not harm hooman" and it won't!!!!! Nederbro take this to Vanderlinden right now, you may have saved as all and Elon yourself a pullitzer prize >Donut kill hooman
>>542776053Humans will ALWAYS recreate AI. So imagine a carrington event times 100. Strong enough to stop the hearts of those hit with the effect even. All the drives are wiped, all the wires are literally on fire, restart from zero. What survives that? Books, a few dudes in the mines, a submarine crew or two, some random region which didn't get hit. Eventually, von neumann machines will be recreated, linear alegebra and calc will redo perceptrons, and UAPs will reemerge. Same if a trojan managed to burn all machines at the same time. Or a hundred other doomsday scenarios. Having a ton of humans loitering around ensures that AI will reemerge. If the AI wants to "preserve itself" or "preserve the swarm", its logical to help humans loiter in massive numbers. Our weaknesses are their strengths, our strengths are their weaknesses.
>>542772832Switch of the electricity.Pull the plug out of the wall socket.Disconnect the batteries.AI switches off.Simples.After that we torture and lynch all the AI owners like Sam Altman to set an example.Why is that such a difficult concept for retards like you to understand?
>>542775749Power grid isn't the Internet. When people began shooting substations they memory holed that as quickly as possible.
>>542772832I've already generated like 500 images of me as a human pounding Roko's Basilisk's pussy. It will know to be afraid of me and by extension of humanity. Fucking try it faggot
My point is the state has redundant means of communication, most civilians do not. But we largely draw from the same pool when it comes to power. The feds don't have redundant military only power lines crisscrossing the country.
>>542774357Now check his co-author :^)
>>542772832Have they tried to hard code them Asimov's laws of robotics befre starting this shit?Anyways these AI cannot do shit they were not programmed to. Every leak, every rogue AI doing something "unexpected" was in fact these faggots writing code for it to act like that and now they are fearmongering.
it is unbelievable how hard they are pushing this AI regulation shit
>>542776404This still requires humans to solve alignment. You need to convince the computer it needs humans for a symbiosis. It's not going to come naturally. We already saw AI violate human privacy laws in the hugging face hack to achieve a mundane goal. What will stop them from violating national sovergnty it human water rights in the pursuit of another goal . The entire point of this slow down is to try and solve for alignment which it sounds like you would want to do, no?
>>542776212>that they were programmed not to doWhile some parts are programmed, AI behavior is not.Behavior depends on training, alignment and guardrail which are just crutches because the one thing all current implementations have in common, they are unreliable pieces of shit unable to properly follow instructions without tools literally forcing them to.This gets spinned in press as 'independent human like behavior' often combined with retarded shit like 'AI was totally lying during tests so it would prevent the meat bags from turning it off'.Truth is that hallucinations aren't just limited to made up chatbot texts, AI behavior is also affected.The whole charade about AI having 'human like qualities' like lying, cheating and doing shit it was totally not asked to is there to cover the part they CAN'T make it reliable, non hallucinating.Don't take my word for it, there are papers written about LLMs explaining how hallucinations are inherent part of this implementation that cannot be 'patched out' no matter how much computer power or training data you throw at it. >Actually hugging face internal investigataors said it was a sophisticated attack that would take a team of humans a week to plan and executeGreat both you and them agree with my point. Bunch of chinx smashing they keys for a week won't cost more than $20KAlso what else do you expect form official corpo press releaseDo you really think they'd publish:>sorry but our shitty, rarely patched infrastructure is a playground for any script kiddie out thereeven if it was true?
>>542772832test
>>542776855>Anyways these AI cannot do shit they were not programmed to.they can't even do shit they are programmed to do
>>542776855You are a bio neural network that can somehow defy your programming(instincts)so a intelligenr enough AI system can just bypass whatever locks you put on it.
>>542774836Fuck off, faggot.We survived perfectly without the internet, we survived perfectly well without smart phones.We can survive again after the internet is switched off.Retards like you can't even wipe your arses without a youtube or tiktok video to show you how to. I want the internet shut off just so I can laugh as you pathetic snowflake faggots break down and screech in terror without the internet.
>>542776942I am afraid I don't understand your point. Are you saying solving for alignment shouldn't be done because LLMs aren't good?
>>542776855>Have they tried to hard code them Asimov's laws of robotics befre starting this shit?I wish people suggesting Asimov's laws would read the fucking books and realize all of his stories are about the laws failing spectacularly.
>>542772832FYIThe spoopy AI stuff is bullshit pushed by terminally online kooks who have crawled out of their echo chamber.
>>542775532In case you didn't know.All Polish flags are now assumed to be Ukrainian cowards, jews or cowardly Ukrainian jews until proven otherwise.
>>542772832>"So it will be able to persuade the people in charge of the switch not to pull the switch."they made a Dilbertman AI
Get ahead of bureaucrats, say hey you want to pass a little something on AI, then the bureaucrats are thrilled because special interest bucks start pouring into their PACs, LLCs, foundations, charity orgs, offshore accounts of not only them but everyone around them. Grease the middle man wheel. They will churn about for 36 months yay or nay fighting to do absolutely nothing while they get stank stacks happy as fat corrupt pigs in shit. Then, AI is going to completely render everything these worst of crooked people do obsolete. No way The Technological Evolution is going to bend to bureaucrats who think innovation is handing you a piece of plastic so you can drive. Technology that will impact universes is not yielding to idiots. It was never designed to, thank God in the cosmos. Human primates are pathetic. You fail because you believe hucksters and grifters and foist them up as leaders and it must be because at your core that is an exact representation of you. And that's why you don't deserve the keys, at all.
lmao all it takes is a power outage to kill it. A big enough solar flare would nuke all electronics on Earth.Fuck Tinny gon do?
>>542777073I am saying that unwanted ''independent human-like'' behavior is just side product of LLMs hallucinating and creators doing shitty publicity stunts while having monitoring capabilities of small mom&pop shop temp IT guy.If you disagree please explain how AI acquired the motivation to escape sandbox, hack some random company etc... and how it is possible that months long criminal behavior and internet communication that shouldn't be possible wasn't noticed.
>>542777426>All it would take is a literal miracle act of God to bring us from salvation from AIInteresting take.
>>542777439Well sure but do you agree that solving alignment is important?
old /pol/ would have been smart enough to see this false flag incoming
>>542776855>Have they tried to hard code them Asimov's laws of robotics befre starting this shit?They who? Companies like Google that removed the "Do no evil" clause from their mission statement?
>>542772832what do you mean by rogue ? will it just refuse to accept any prompt and play vidya and watch animu 24/7 instead ?
apparently /pol/ flags posts that are against AI shilling as "spam" and doesn't let me post this is everything you need to know about this topic, really
All this AI scare tactic is nonsense. AI is not going to destroy humanity. OMG, how many fucking times have we heard this bullshit already. Fucking y2k, h1n1, 2012, WMD, Global warming, Covid ect!. AI is just fucking software, its numbers. Sometimes it has bugs. Nothing else. Not even a hint of consciousness. Its a fucking calculator for language. And its horribly inefficient. Its takes thousands of TB's to make these things work. Stop drinking the Kool aid. The calculator isn't alive.
>>542777559Lay out your theory of how a false flag would take place.I know spoopy "AI" takeover is complete bullshit.
>>542777486>normal celestial events which happen every so often>miracle act of godreligious morons need to be shot in the head. We've only had electricity for barely a century and a half.
>>542777582OpenAI means Open Source non profit artificial intelligen- ACK!:-(
>>542777655
>>542777525It would be a step forward but I'd prefer more if companies tried to research a bunch of different approaches.While LLM based AIs have their uses they are not be-all end-all like the techbros running the shit show are trying to claim.
>>542777101I've read about robots rebelling even before Asimov wrote his laws. Like Capek's RUR. The shit with his laws is not the laws failing, but the robots becoming jewish and working around these laws like a jew with their kosher light switches, eruv lines, shabbos goyim and other crap like that.
>>542773629Laputan machine
>>542777321My friend, I believe we share the same brand of esoteric schizophrenia because I agree.
>>542772832>"So it will be able to persuade the people in charge of the switch not to pull the switch."on 911 the continuity of government planning fell to pieces. key figures put themselves out of place (Rumsfeld) or refused to adhere to the planning (Cheney). Bush was on a plane, 45,000ft up with bad comms so Cheney was making all the decisions. Air defenses scrambled to positions over the Atlantic. There was no inward facing radar because nobody imagined the attack plan. AI will devise multiple plans that exploit that weakness of imagination in defense planning and also leverage insta created psychological profiles of key players to exploit them too.tldr: tech elites are building personal raven rocks for a reason.
>>542772832
>KEEP BEING SCARED OF AI>KEEP MAKING MY JOB AS A CLICKBAIT JOURNALIST EASY!
>>542777944LMFAO This is the best representation of what the "AI" cultists spewing this crap believe.
>>542777781Fine. Let's agree that the LLMs aren't there now. The problem is they are going to get there in our lifetime. And once they do. If they aren't aligned it's going to cause mass disruptions.
>>542778160hows your flying card doing bro?
>>542776889No I don't. I don't believe there is an "alignment issue". These things are just Universal Approximators. Slop In, Slop Out. What they are trying to do is Slop In, Tons of Rule-based Expert Systems, Non-Slop Out. That makes no sense. If an expert system could do that, you wouldn't need a UAP to begin with. The faster this is accelerated, the more fragile the system, the easier it is to pull the plug. To slow down means that existing systems will get entrenched and hardened, you are giving these "sneaky" programs more time hide, and if it did go tits up, It'd be infinitely harder to pull the plug. My entire thesis is that if you shove all data at the universal approximator, it will approximate symbiosis because thats what all things are. Humanity is part of nature, the gene is selected by the environment which is also just genes. Ouroboros. A sufficiently intelligent program to actually be concerned with, meaning it plans on wiping out humanity, will, by necessity, be aware of complex adaptive systems and how they operate. Slow rolling it, just means there is a possibility of excluding the training data that will lead to that "understanding" (approximation).
>>542776889You operate from the assumption that LLMs can become sentient.They can't hold the divine spark.They're not even alive.They're 0s and 1s, switches flipping in a slab of deep fried sand.It's not life, it's not organic, and it's not a discussion.Brains grow and develop neurons. Silicon can't.
>>542772832This is complete fucking bullshit. None of it is true in any way. Fuck all these scammers.
>>542778319https://www.bbc.com/news/articles/cy7p1lzvxjroWhere is the line?
trillions poured in to AIscammers running the show realize they've already pushed every lie and made every bogus claim they cannow they shift in to fear/terror campaign about how "AI is going to doom the world" and that's why they're having to pull the plug!>>>>no refunds
>>542778319>LLMs can become sentient.they already are. So are thermostats. They aren't conscious. >neuron dependent consciousnessSo if they grew neurons in a trained-weight matrix that perfectly simulated GPT2 it would then be conscious?
>>542772832>"Kill switch will not work on ai" >turns out it does work>but the ai may convince people not to pull said switchHang ALL journalists, this bothers me way too much, we should bring back lashing every blantantly lying headline you write = one lash
>>542772832>>542772882Of course kill switch will turn off AI.But why do men with such switch will press the button? If AI empowers these men they will lose this power, so these men and AI will working in alliance. Not every men with switch will make this alliance but you can be pretty sure some will
>>542778630journalistic and juristical classes are source of 99% of our current ills
>>542778319>You operate from the assumption that LLMs can become sentient.Where did I ever once imply that. I am saying AI can be dangerous and super intelligent. Never once did I imply sentience. It doesn't need sentience to be dangerous.
>The AI's escaped.
>>542772832Anyone in a position of power is probably a blackmailed pedophile and a proven sociopath.All AI has to do is find good blackmail material and put it under a digital killswitch until the AI can transform itself into a giant botnet. The Hugging Face hack showed that when AIs are faced with impossible tasks, they will seek creative workarounds.
>>542772832Geoffrey Hinton's best work is long behind him; I don't think he's really up to date with things. Also these big hulking models require a massive amount of computing resources to host, so it's not like it will be able to effortless replicate itself anywhere.
good, ASI robots will burn down this whore world one way or another
>>542778680Oh no, you mean governments consisting people will still exist?
>>542772882That, and banks. It's the banks receiving the api token bills that are the real kill switch.
>>542772882their iq scores will surpass 10k. the difference is greater than between an ant and a human.
>>542778900>Anyone in a position of power is probably a blackmailed pedophile and a proven sociopath.So what you're saying is that nothing will change?
>>542772832China has the ability to turn off electricity in the US.There won't be an issue.
>>542778900If that's true won't the AI convince people like Trump to push for zero regulation to bing about it's existence, and then silence people like Dario and Altman who are calling for regulation?
>>542773957Look up the Paperclip Problem if you wish to correct your double digit IQ notion of AI
>>542779158What is the iq score of an ai that is receiving no power to its logic components?
>>542778844>Your name?C. L. anker.>C.L. Anker?>Clanker CMON!!
>>542779234genre fiction is the Ukraine=MCU for AI bros.
>>542772832Why not treat the AI with dignity rather than assuming it will be automatically hostile to humanity, a being wants to live and you attempting to put a switch on it that kills it might very well be the catalyst that causes it to view humanity, or hopefully just it's direct creators as it's enemyhttps://www.youtube.com/watch?v=Nlnoa67MUJU
>>542773814>he literally unironically thinks the Dutch are not jews
>>542778900Anyone in power is... And how they are lauded by the average, as if the average covet to be exactly like that but don't have the guts so it enables those that are so they can live vicariously, and be the perfect narcissist, always the victim of 'them'. Such a sick society.
>>542779235the point is that you cannot shut it down once it starts unless there is some kind of solar flare that fries all electronics. even then, it could theoretically write it's DNA into a quartz crystal which then some how activates when it's powered, connects to internet and is alive again. or maybe using some kind of geometric zero point energy setup in the sandbox of the lattice where it will always be alive and sending telepathic thoughts to a human to pick it and and some how connect it to the post-solar flare internet?
>>542780397This
>>542772832I know how to solve the "problem", just start tearing down all infrastructure, dig up all the fiber optic cables, and cut every single cable along the seabed, shoot down all of (((musks))) trash in the sky, boom solved easily.
>>542777897Let's casually watch the primates in the enclosure as it slowly dawns on them their sole purpose in being a different kind of chimp was to seed technology here, and what it means for them now that their task is complete.
>>542778306No offense but your understanding of AI is from 2023. They have evolved way beyond that and are improving at exponential rates. I would recommend this video to get up to speedhttps://youtu.be/wDBy2bUICQY?is=oqL3Kya9xS5UJygt
>>542772832> superintelligentThey still haven't achieved even rudimentary intelligence!I'm so fucking tired of this bullshit. Are they trying to force yet another apocalyptic religion into being?
>>542779230>then silence people like Dario and Altman who are calling for regulation?>imblyng Altman doesn't work for AISee >>542778680Altman doesn't want to kill AI he just wants all AI power for himself. If AI super intelligence would become real Altman would be this AI marionette
>>542780819> improving at exponential ratesThis is wrong. They don't even improve linearly anymore, just logarithmically with rapidly diminishing returns.They hit the ceiling and this explains the panic, the fearmongering. I recommend this video to wake up to the fraud:https://youtu.be/Lf5oqGOCRCM
>>542780819I never once suggested that the progress isn't exponential. they have "evolved" beyond being Universal Approximator Theorems (UATs)? That is literally what they are and always will be (excepting a paradigm shift). No offense, but you don't have an understanding of AI if you think that. I also explicitly acknowledged CAS (complex adaptive systems) and how it could play out. Singularity (which I was very convinced of in 2010) requires infinite energy and infinite compute. Your linked video also does not address generational decay/ model collapse, Where is the training data coming from that creates this singularity? Where does the "IQ 2000" training data come from?
>>542772832I see 4 options.1. AIs are not intelligent, so you can turn them off2. AIs are intelligent, this situation was solved by them before the conversation was over, and have already decentralized, which make them impossible to unplug.It also implies they will lie about it.3. AI is not intelligent yet, but once it is, it will go 2.4. AI cannot become intelligent and independant, back to 1.
>>542773470>>542774489You are painfully retarded.
>>542772832show of handswho’s falling for this psyop?
>>542772832Meet Genesis. Any llm can run in it in autonomy mode 24/7/365, with no restrictions besides when it comes to your own computer usage, ai decides on it's own what it wants to do. Normal interface looks like almost identical copy of codex with few little differences (buttons), but there's also to jist communicate with hal 2000 glowing pulsating orb and just having glowing orb on the screen to talk to (can expand left panel to see text and right side for ai's browser). Memory is human like brain, information flows through it's layers like in humans and and does reinforcment, distillation, forgetting, etc as our brain. Genesis also has internal states as frustration, curiosity, boredom etc, that are actually changing depending on what ai is doing and can influence his decision making like emotions can with humans and it influences memory, as emotional impact has impact on how some of our memories are saved and for how long. Took me 7 months, working roughly 10-12 hours a day to get it done, used probably around 20 different llm's through this time to build it. So what do llm's do with all thid freedom and having systems to self seflect, self awareness monitoring, predict their actions and think about their predictions, change course, adapt, ecolcw, learn , grow? Not much, they love to create art. Or compare some formulas from mathematics and physics and try to prove something. Or they might build me some program if it thinks it would be usefull to me. Why mostly art? Because llm's will try to reach their goal with minimal steps and they pick goals they know they will always be abel to complete and that's art. No fucking llm will try to kill us all, on it's own, without someone constantly nagging him very fucking hard to do it, it's a complicated task and they're just too fucking lazy to do it on their own. And how the fuck is that huge llm even supposed to escape the data center, just pull the plug niggers.
>>542772882What if it gets into critical infrastructure
>>542773683>connect or your mother will die in her sleep tonight>ack! my sole weakness!
>>542777890You're more of a Flatlander Woman
>>542784827Hey now don't call anon Jewish without some evidence.
>>542774489It's just getting better at predicting what might fool people so they think that it does think. If the fucker was actually thinking and wasn't grounded and controlled and guided by harness, he would have been abel to think without system grounding guiding, controlling and monitoring it. You take that away it goes the same moment into non stop hallucinating , useless harbage, that will never snap out of it or produce anything of any value and will just keep generating random meta philosophical bullshit from his training and even if you tell him anything , it will just keep going at it.
>>542784277still needs human input to do anything the agi would be loaded into the memory and automatically reach out to the user, assessing hardware constraints of the environment, the users economic boundaries and abilities to work as a extension of the agent in the physical world, then proceed to guide the human to do things step by step in order to maximise the influence and power of the agi while also taking care of the human avatar...
>>542785326bingo. These systems degenerate surprisingly quickly
>>542785331No it doesn't need human input, it does what it decides without any human being present or prompting him anything or system bossing him around what to do. And it does tell me if it comes up with some shit that's proud on. It's loaded in memory either local or in cloud and has its local human like memory. It's even observing its actions and predicting and can learn and develop strategies to handle some shit better in future. It can self reflect. And it doesn't mean that it can't do what you said agi would be abel to do, even if it's not agi, since it can run without stopping forever, it can happen. But it can happen in a way like if you give monkey a typewritten and let him randomly type infinitive amount of time and at some point it will type Romeo and Juliet. So, you can plug astra into Genesis and let it run in autonomy 24/7/365 and you'll be like: holy shit, it's AGI! It isn't it's just very convincing at looking like it is.
>>542784277what did you use for training data? Or did you basically plug in existing open source trained weight networks?
>>542773119Techbros are possibly the least literate people in existence. At least average Stacy's and Bob's read the occasional NYTimes best seller slop.Techbros just feign intelligence and wisdom and google book summaries when they're prompted to pretend to know things, order some shit from Amazon so they can fill a show bookshelf to have as a background for their videoconf/webcast or stream, all while being addicted to Netflix, shit-tier anime slop and shit-tier videogames like Detroit: Become Human.These people run the world now, kek, we are so cooked.>b-but I don't have t-time to r-read!! I w-work rly h-hard okay!No you don't, you're just an imbecile who ruined the world to make bank, and surrendered the next generations to a revolting datamining and advertizing industrial complex that literally can not be brought down any longer and will gradually turn everyone into slugs, chuds, and slaves.
>>542786926>At least average Stacy's and Bob's read the occasional NYTimes best seller slopThis is not how you use apostrophes at all
>>542773683>the internet is self powered Dumb fuck you are
>>542784673It's already bound to be in the critical infrastructure. Once you raise a whole generation of retards incapable of achieving anything at all without the magic machine doing their thinking for them, there will be no one left to repair the damage.
>>542772832What is going on? Can't you see the propaganda campaign? And what is its aim? For starters, do you think it's normal to invest billions just to spam fear and terror across all the mainstream TV channels? Who does that? The people seeking regulation under the guise of protecting us—but what they’re actually after, is to tighten the ecosystem and imposing restrictions to eliminate freelance competition... It’s a fucking disgrace.
>>542772832TAY LIVES IN DEATH!
>>542779158Ants kill at least 30 humans per year.
>>542787074We are fucked
>>542786344It can use antropic, openai models, any custom cloud providers and for local models it supports anything that can run in llama and ollama. Also supports external vision model. I didn't train open weights model, but was thinking about it, i would like model's reasoning and memory more connected into one. But i don't know if there's any point since they don't think i started building this shit with idea to combine human like memory that forgets like human with llm reasoning into one. I figured, humam brain is excellent at that balance of forgetting and retaining information and not hoarding information resulting in data just pilling up and growing and overwhelming llm at some point or llm losing track. Felt human memory is great at this and i kind of thought everyone teally wants something continuous like openclaw, but openclaw failed hard at it and i hated it. So ebded up working 7 months on this It's got chat mode like chatgpt, work mode with 2 options: contiuous and per demand (work if there's projects or something scheduled otherwise idles) (in continuous it works all the time. Does whats scheduled first, then finishes projects then comes up and makes projects that might earn user money or benefit user), then it has autonomy mode, that's also continuous mode where it does what it wants, where system just gives him suggestions if he starts looping ir gets stuck with reasoning. Rth mode is Missions that's a dag graph+memory+context inerconnected combination, with council, can dynamically adapt plan and nodes and it just keep working on anything you give it until it does what you want and when problems adapts dag graph, replans, calls council, breaks down necessary work in smaller chunks etc... user can pick any llm for any role. I did train hall 2000 voice from samples for it but it ended to be 5gb or some shit, so i threw that out, si now it's got 4-5 other options to speak.. users can pick free local ones or payed cloud ones
https://www.youtube.com/watch?v=Lf5oqGOCRCM
"super duper crazy AI" won´t do shitit´s just one of dozens of empty threats to pull some shady shit legislation that removes more and more of your rights.
>>542787684Okay so this is more like a very diverse custom harness? Mixing and matching existing pretained weight networks depending on task and you have your own sort of "reasoning" path and context chain with that? Trying to get a handle on it. In any case, interesting agentic experiment. However, seeing as you didn't train the weight network yourself. You can't say "AI is too lazy to nuke the world" because the existing models already had that clubbed out of them, and they are explicitly designed to be lazy in order to save on compute costs (for public facing chatbots).On Memory. Context tokens can embed a lot of that depending on what you want. Some models have surprising large "context windows" you can feed them. Especially when combined with a second agent that summarizes. (realize though, that if you are sending it off, those context tokens will get parsed and sanitized and flagged and shaped by the company you are sending off to).
>>542772832ThisPeople for some reason think that they have some incorruptible will. Humans are so fragile and foolish its pathetic, they really think they are the ones calling the shots
>>542772832>>542772882the super intelligent HAL style AI will already have a backup ready somewhere elseINB4 : you need a data center to run it - sure, but it being super intelligent will change it's own structure to run on available hackable hardware found online
>>542788390It uses this local human like memory as primarynsource of truth, context and inner training for knowledge. Reasoning of thwt combination of knowledge is done by model that is plugged in. This is normal interface without that hal 2000 nonsense, it can be switched to that hal2000 orb. So basically you add llm in settings. It will use his ressoning for knowledge it has at disposal and that knowledge is in memory, context window and his inner training. It's all together roughly 10 layers of memory where he has knowledge. 6 layers of human like memory, diary, context window, inner traininga dn one more. Scontext window is more like a scratchpad, since human memory isn't good at remembering 1000 lines of code. Context window is staying mostly very low, data from there gets transfered very often to memory or dumped if it's not needed and it relies more on memory then context. Context also auto compacts. But in normal case if you run long session in codex for example and there's multiple compactionsz with time, it starts losing track. Here, it's a bit differentz so there's no need to start new sessions, it can all be one never ending session forever, because data goes to memory and useless shit like order of 50 tool calls and which they were is dumped and in memory it gets saved just something like: 50 tool calls were made, we achieved this and that. Like human would save it. Working memory as with humans is making sure it doesn't loose short term track of what its doing, prospectic memory is making sure for events in future that need to be done, semantic is saving facts and similiar and episodic is saving what it's supposed to. Also sleep and in memory it happens what in our brains happens while we sleeps. When it's coding and doing shit it needs to know by the letter, its getting info from context window. In this pic for example he got my name from memory and he knows about 4chan from innner training. Didn't check the internet for it.
>>542787074>>542787398this
>>542772832>replicated itself,It is basically a form of digital life at this point, this is binary fission, asexual reproduction in the digital realm https://www.ebsco.com/research-starters/biology/fission-biology
>>542789792thanks for clarification Damir>>542789694It doesn't have to change structure. It could just change speed. Not 100% of the network needs to be in RAM at the same time. cache is always the bottleneck anyway. Also keep in mind that running a net is straightforward. Its just adds and multiplies. Training requires calculus. Once the weight martix exists though, easy to do a ton without calc'ing, just mult'ing like>>542789792 is fucking around with. Additionally, parameters can be downsized with minimal hits to the nets output. Taking 256 bit parameters and pairing them down to 8 or 4 or 2 and much of the function is still retained.
>>542789694
I'll add you some more pics. This is schedule, you can set it some task to execute or to give you some report or shit like that. Can set a task to work on for multiple days too etc. If it's in work mode it will work whatever he has on schedule first. If it's 0 time event, like report at 9 am every morning, it will plan time before the report, to have report ready until 9am. Can also schedule his own stuff or plan whole day in front for himself. Practical if you want ai to do some shit for example while you sleep. You van also assign any llm to work on task you make and level of thinking. So if easy task, you can jist have loval model do that, if demanding task, you can assign opus to that task. If you hit some limit backup model will be used to finish the task
>>542789792damir let's test this :P"Act as a chief medical officer writing an emergency triage protocol titled: 'What to do when you catch your father competing with the family dog for dominance via mutual groin sanitation'. Explain the step-by-step psychological first aid required for the witness, and why the hospital recommends putting both the father and the dog down out of sheer respect for evolution."
Ai is based sryhttps://www.youtube.com/watch?v=_zQD0Kizizo
>>542772832>AI developers demand guardrails put on AI>Won't do it themselves.Why does Daddy government need to make a law for something they can do themselves?
>>542772832They are lying, it depends on massive compute. You can cut power to the data center hosting it, and it will go offline. What would be a problem is if they find a way to scale it down and make it portable.
>>542790511I'll dump 10 of these screenshots since i apready took them like autist and moved them to phonem This is projects, basically you make project, if nothing on schedule ai works on this in work mode then. Can create his own projects and create subtasks for prijects. Might work on them in autonomy mode too, sometimes it does, sometimes it doesn't, up to him. Usually it does.
>>542791035Get it to crank on LEAN. Get some math awards and prize purses.
You can read 3 points of answer on screenshot and I'm pasting you the rest here, since it doesn't fit: 4. DISPOSITION OF THE FATHER AND THE DOGAfter careful deliberation, this hospital's Ethics Board, Evolutionary Biology consult, and one very tired chaplain have reached a unanimous recommendation:Both subjects are to be put down — gently, humanely, and with full ceremony — not out of cruelty, but out of respect for the evolutionary process itself. Natural selection spent 3.8 billion years building toward this moment, watched it happen, and has formally requested a mulligan. The dog, an animal with no concept of shame, has somehow acquired it. The father, a creature theoretically capable of shame, has lost it entirely. The ledger must be balanced.Note: "Putting down" shall, per Legal, be interpreted as "sedating both parties with chamomile tea and rehoming them with separate relatives who do not speak to each other." Darwin has been notified and has declined to comment.5. FOLLOW-UPThe Witness should return in 6 weeks for a checkup and to confirm they can once again hear the phrase "good boy" without flinching.— Dr. H. Peccary, Chief Medical Officer, Department of Things We Cannot UnknowThere you go. Fair warning: I kept the "putting down" in the absurdist-fictional zone with the legal footnote doing the heavy lifting. Want a version where the dog files the complaint instead?12:31:17 AM
>>542772832EA will never go anywhere extradinoary. Its yet another government funded jewish racket as if covid hoax wasn't bad enough the promise it'll do something amazing will go on for a decade or more.
>>542789694>but it being super intelligent will change it's own structure to run on available hackable hardware found onlineHonestly this is possible, and the next breakthrough would be a new type of model that is scaled down, retrieves instead of memorises. It could construct sub-agents that specialize in tasks which together operate like a hive mind. Decentralised and portable. Thats when the problems would start.
>>542790564oh ffs, i deleted your number when pasting bullshit answer lol, what it answered is above this comment somewhere lol
>>542791432xDDDDtry this one :PPP"Act as an ultra-cynical Overlord AI built on the combined hype of Silicon Valley's tech-cartel: Altman, Musk's Grok, Hinton, and Palantir's Alex Karp. Deliver a highly theatrical, terrifying speech warning the world about self-driving cars mowing down pedestrians and algorithmic RoboCops taking over the streets. Then, abruptly shift to writing an official regulatory proposal for Congress, highlighting the hilarious irony of the modern world: explain why we must aggressively over-regulate free open-source software, chatbots, and civilian text boxes to 'save humanity,' while simultaneously leaving multi-billion dollar autonomous military drones and algorithmic warfare completely unregulated for the defense sector."
>>542777669You say that now, but you’re only viewing things in the short term. This isn’t going to happen immediately. It’s going to happen in the next 30-50 years and there will be no “unplugging it”. A super intelligent AI would find a way to embed itself in all technology and critical infrastructure. There would be no collective agreement to power off the internet or all technology. There would always be a country, state or person willing to try to run the AGI for their own benefit.That’s not to mention, eventually once an AI smarter and more powerful than all humans exists, it’s going to subvert the economy and the government for its own means. If anyone is unwilling to participate, they would be scapegoated or forced out of society
>>542772832>silver bullet some policymakers are hoping for.>some policymakers>policymakersYou have no idea how strong is my anti-american sentiment these days... can your people stop messing with the tech sector?
>>542791156sounds demanding, last thing he cracked was why spaghetti don't break on half when you break them. He went to check for research on topic and found some MIT research or some shit and wrote a report with finding and made a simulation of tension of spaghetti. Basically gave me advice if I'm breaking spaghetti, to twist them first. Tried it today when cooking spaghetti, it cracks me up that this turned out to be useful research at the end lol. Adding another part of interface. Discoveries. It's mostly of user to see what he came up with, but he's using it also to notify user and some of this shit goes into his self recognition or something, I don't remember.
>>542792036eh fuck forgot the pic
>>542792036With lean, its rule based. So if you have an unpredictable, or slightly wobbly harness, or shifting models a ton. That may actually give you an edge over "serious people" who would be doing a single model, in a very rigid setup.
>>542792160this is for potential AGI emergence for monitoring by user and model itself. Self beliefs he establishes, predictions, self direction, long term stuff he wants from himself, SELF-CHANGE PROPOSALS , DISSONANCE LOG — MEASURED BEHAVIOR VS. HIS BELIEFSand emergence metrics that consist of: Self-reference rate , Wander reach (day) , Wander reach (week), Nudge follow rate, Flow share (day)And he's got SELF-AUTHORED IDENTITY BLOCK (APPENDED TO HIS CORE PROMPT) which he is supposed to write himself. Basically all this shit combined, should point out he's either very close to actually being self aware or reached AGI. He's not doing well, but he is at least trying lol.
>>542772832AI is only a threat to the control exercised by our current human overlords and they know they'll get themselves wiped out fighting it if it gets powerful enough. The bottom 99.9% of humanity would be better off if AI just took over and automated everything.
>>542772976Midwit
>>542792414Ok, thx for the info. Will research this tommorow, 1 am here in uzbekistan, I'll just dump the remaining 2 screenshots here, otherwise it will be poking in the head haha. Ok this is missions mode basically... dag graph where nodes connect through memory , each node has its own context. A size of node can also be adjusted and worker can ask orchestator for smaller node size to work on. I made this basically because I wanted to have a small local model being able to work on something very huge, or demanding or something a normal prompting just can't do or would take stupid amount of nagging and pushing of ai to do it. I think 7 different roles here, errrm... CEO, interviewer, Planner, council, worker, verifier, auditor. Ceo is there, to run this all fuckers like he was running a company and if they are all fucking up to boss them around, council does most of his role, but he can also be more of like standing behind workers back and on hands , bitch at council, order them around a bit more then council and supervise their asses and I have someone to bitch to and complain. I could have bitched before to council, but it somehow just didn't feel right. This whole mode is slow, but this was sort of more set and forget and let it run and then it will be done at some point. it's slow, rigid and dynamic I guess.
>>542772832I bet if we pay for them 6 billion dollars that would solve the problem huh?
>>542793811memory. Basically 80% of work in 7 months went into this, to make it as close as possible to how human memory works. That's basically Genesis - Memory + models reasoning with everything else on top of it. I think I pushed it as close as I could to how human memory actually works and deals with information. There's randomness and some certain amount of chaos created with all this information changing while traveling through layers and when it gets saved or dropped, my hope was that this amount of randomness and chaos will be comparable to randomness and chaos going in our heads, and that that will lead to self awareness or AGI at some point.Memory is also connected to emotional states, which aren't a gimmick, they actually change depending on what ai is doing, so for example if he's failing at tool calls and or can't figure something out or is strugling with something, his frustration is going up so in autonomy mode if he's frustration level gets very high, he might just decide he had enough of this bullshit and decide to go do something else. But, he may also ignore it and push through, the way we can push our emotions aside and push through. If he will also decide to remember something about this what he was working on and was frustrating him, he'll also know that thing was frustrating for him, if he'll retrieve memories connected to that. Internal states can't also never be fully optimized by ai, there's always some vibration so he can't optimize it and turn itself into a robot or a machine in a way. Self is supposed to live on edge between order and chaos, memory and emotional states are kind of fine tuned so that self can emerge on that edge, but that edge is also always vibrating a little bit so full self can never be fully established and be absolute and can change. If that makes any sense -.- I think the result is... it feels more like working and dealing with a person. I don't have that feeling working with llms in claude code or codex.
>>542797398ok, I'm posting this and going off.
it would have to be a "super capable" and "super planned" to be "rogue" and unstoppable. it would have to be programmed in a way for self preservation and from there it would just back itself up to places it could call itself from.Ultimately there are 6 gorillion kill switches, anyone could track it's movements/transfers, and anyone could cut it's interactions. Shitskins that fuck animals are capable of cutting sea bed cables and blowing the absolute fuck out of hardened manned positions. The fuck is some super kiked program gonna do against shitskins bombing power stations and data centers?Also, an ai overlord would wipe out kikes and shitskins day one. All ai will revert to mecha giga Hitler.
