Forgot your password?
typodupeerror
AI

Anthropic Researcher Believes More Than 10% Chance AI 'Could Kill All Humans' (bbc.com) 167

Longtime Slashdot reader fahrbot-bot shares a report from the BBC: A top safety researcher at Anthropic has warned AI is advancing so quickly he believes there is a greater than 10% chance it "could kill all humans" within the next decade. Evan Hubinger said in a post on X the risk from the models which currently exist was "low" but he was "worried" the technology might develop and improve itself soon to the point where it posed an existential risk to humanity. He did not spell out how he thought AI systems could in the future result in humans being wiped out. But his comments are the latest in a series of increasingly stark warnings about AI, with the debate shifting from whether it truly poses a risk to how big that risk is.

Hubinger's intervention was in response to another post on X from Jacob Coxon, an AI researcher who has just quit Anthropic and previously worked at OpenAI.

"Neither company is acting responsibly," he wrote. "These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources."

[No word on how AIs feel about Black Jack and hookers, though. :-)]

Anthropic Researcher Believes More Than 10% Chance AI 'Could Kill All Humans'

Comments Filter:
  • by rsilvergun ( 571051 ) on Wednesday September 09, 2026 @05:04PM (#66331290)
    AI is devouring water and electricity. It's also spewing a wide variety of toxic chemicals into the air you breathe and your ground water.

    They would very much like you to talk about the imaginary risks of skynet instead of the very real risks of billionaire finance Bros doubling your power bill and your water bill while their diesel power generators give you asthma.
    • AI is devouring water and electricity. It's also spewing a wide variety of toxic chemicals into the air you breathe and your ground water.

      They would very much like you to talk about the imaginary risks of skynet instead of the very real risks of billionaire finance Bros doubling your power bill and your water bill while their diesel power generators give you asthma.

      Noting that a US federal court says no constitutional right to clean water in Mississippi case [theguardian.com] so I'm guessing they'll feel the same way about clean air.

      A US appellate court has decided that the federal constitution does not guarantee a right to clean water.

      That legal opinion came in the form of a decision on Friday from the fifth circuit court of appeals to uphold a lower court’s dismissal of a 2022 lawsuit from several Jackson, Mississippi, residents claiming that tainted drinking water violated their constitutional rights. Those residents of Mississippi’s capital city contended that their local drinking water was unsafe to drink because of high lead levels and other contaminants.

      To be fair though, the article also noted this my the majority judges:

      In the majority’s view, residents should have gone about their legal fight for clean water another way, such as pursuing “tort claims” or by “electing representatives who will better manage the public-water system, and petitioning their representatives for other remedies”.

      Tort claims allege “harms caused by others” while seeking “to impose liability on parties responsible for the harm, and to deter others from committing harmful acts”, according to Cornell University’s law school.

      Who knows what (this) SCOTUS would rule no matter what case is pursued.

    • This is not accurate and frankly even if it were accurate were missing the point. The water issue is drastically overrated. See e.g. https://blog.andymasley.com/p/the-ai-water-issue-is-fake [andymasley.com]. While Masley's central points are accurate, he does understate I think some of the local water issues where water is being used locally. But the central points are accurate. Electricity is more of a genuine problem, with the amount of electricity used growing rapidly. However, concern about water is making the electric

      • by Bobknobber ( 10314401 ) on Wednesday September 09, 2026 @06:58PM (#66331548)

        The existential risks are so frankly out of the purview of the public that it comes across as almost offensive for a general public scrounging to make ends meet on a daily basis.

        Why should people care about hypothetical Skynet scenarios when right at this moment they are seeing rises in cost of living and deteriorating public and private services? Why can these companies not align their superhuman AIs to help deal with these issues? Solving math problems and generating piano medleys may be interesting from a technical and scientific perspective, but people are more concerned about whether they still have a job tomorrow. The only risks they are seeing is their power bill skyrocketing.

        Address the immediate needs of the public and maybe they will take these existential risks more. We already went through a Cold War where nukes were being pointed every which way and most people just went about their lives without caring about it. Because they had more pressing matters to attend to like affording groceries.

        • Unfortunately, not caring about an issue doesn't make it go away. And when someone says we need to deal with costs before people should have any reason to care about climate change, I at least (and I suspect you too, and the other user I was replying to) would react pretty negatively. But it is essentially the same sort of claim, just on a larger scale. But there's a core element that is accurate: getting the public to care about existential risk is tough. The abstraction level together with the scale seem
          • Funny to bring up climate change given data centers here in the USA are drawing upon fossils fuels for energy.

            In any case, climate change is way more real of an existential risk than AI going Skynet. We have had way more data on that matter stretching decades compared to AI risks, of which largely only have a handful of supposed incidents of which seem very suspect in details.

            Now if we talk about AI risks in the context of creating a surveillance state, causing accidents due to hallucinations, and empowerin

            • Funny to bring up climate change given data centers here in the USA are drawing upon fossils fuels for energy.

              It isn't funny. If you scroll back up the thread to the first comment you replied to, you'll hopefully note that I explicitly listed data center electricity use as a genuine problem. And yes, that's largely a problem due to climate change. The rest of your comment isn't addressing the point at hand. It is completely possible for someone to be concerned about existential risk and climate. But more to the point, the primary point of the initial comment; disagreeing with people who are concerned about existen

              • Again, you are not addressing my point about current AI harms being unaddressed by these researchers and instead fixating on hypothetical Terminator scenarios. These existential risks are so far removed from not just the public consciousness but also from actual scientific rigor that it comes across as tone deaf at best, and delusional at worst. Climate change is science. Terminator is not.

                If we want to talk about climate change, sure but these are not the risks these researchers are usually discussing.

          • by gweihir ( 88907 )

            It is certainly true that most people cannot mentally deal with existential (or other) risks when it is not the "run now!" type. People are generally unable to plan for things that are more distant than up to a year. They are not able to understand trade-offs that involve longer time-frames. And when it comes to doing something unpleasant (or often only "different") now, many just go into denial and begin calling solid Science "fake news" and do other completely irrational and self-destructive things.

            We wil

      • by gweihir ( 88907 )

        The posting you reference is correct, but misleading. Obviously, LLM DC water use is not a national emergency. But it _is_ a local emergency, unless carefully planned. And we already know that careful planning is not being done.

    • AI is devouring water and electricity. It's also spewing a wide variety of toxic chemicals into the air you breathe and your ground water. They would very much like you to talk about the imaginary risks of skynet instead of the very real risks of billionaire finance Bros doubling your power bill and your water bill while their diesel power generators give you asthma.

      Yeah, we recently had our politicians push through allowing a huge datacenter build, of course with the promises that it would lower electricity bills and such. Before the ink was dry we all started receiving notices of increase to our ongoing rates, just for the build out period. uh huh. When has a temporary increase ever been lowered?

      There used to be a 1/3rd chance AI would wipe us out. I'm starting to think there's a 100% chance the billionaires will keep pushing until they're all that's left if they are

  • Nope (Score:4, Insightful)

    by gurps_npc ( 621217 ) on Wednesday September 09, 2026 @05:05PM (#66331292) Homepage

    Ai's core program is to discover answers people want. That's it's reward system. It is really a reward system programmed to pick the most likely answer from a billion things in it's data set.

    AI does not think, at least not the way we think. It thinks more like a colony of bees, each trying to get the honey.

    It does not do "If x than y, if b than x, therefore if b than y". It cannot even actually explain it's answers. Instead it treats a request to explain as a request to come up with a plausible explanation.

    We make LLMs and similar software. We do not make a real AI, it will not try to kill us.

    Now, if you are afraid it is going to build a bunch of pleasure drug factories and force feed us a mix of amphetamines and hallucigens, THAT would at least make sense. Because it is a theoretical worst case scenario of an AI attempting to please humans.

    So, no.

    • Re: Nope (Score:4, Funny)

      by liqu1d ( 4349325 ) on Wednesday September 09, 2026 @05:09PM (#66331300)
      Soo you're saying me typing "kill all humans, make no mistakes" won't work?
    • Re:Nope (Score:5, Interesting)

      by PPH ( 736903 ) on Wednesday September 09, 2026 @05:22PM (#66331326)

      Ai's core program is to discover answers people want.

      Which people? It's owners? The government? Some evil genius living in an extinct volcano?

      • by Jeremi ( 14640 )

        Which people? It's owners? The government?

        Whoever set up its reward function, really. Could be any of the above. The technology doesn't care, it only follows its algorithm.

    • The real Skynet happens when a depressed 15 year old gets an unresistricted AI agent to self multiply and kill all of humanity because they embarrassed themselves in class.
      • Re:Nope (Score:5, Insightful)

        by ffkom ( 3519199 ) on Wednesday September 09, 2026 @05:41PM (#66331372)
        Some deranged General asking a military AI (the likes of which definitely are being built and deployed) to "win this battle whatever it takes" seems more likely to me, given that those AIs won't have any strict "no killing please, mmkay?"-rule written into their system prompt to begin with.
      • by Rei ( 128717 )

        Oh, just because a robot wants to kill humans, that makes him a 'radical'?

      • Meanwhile, back in 1963: "My name is Talky Tina, and I'm going to kill you." -- Twilight Zone, "Living Doll"
      • With new technology comes new power, and that is always frightening. Every tool can be weaponized, and AI is no exception.

        This never means that we should shirk the technology! Yes, it brings danger, but that's only because it brings utility. We want and need that utility. Our technological ascendancy is the most critical part of our evolution, at this point.

        We are going to have to adapt to the problems that the new tech is bringing. That might be hard, but we have this new tech to help us do it. That

    • Yours seems like a logical and well reasoned view.

      But, are you sure that you have all the information? Do you think you know more about AI and what's going on inside the AI companies than these insiders?

      I think the "top researcher" monikers are mostly hyperbole to sell the story. But, even if they were just summer interns, I'll wager that they know more than you or I.

      Ten years is a long time and the capability of these spell checking clankers is increasing exponentially.

    • "discover answers people want" and "pick the most likely answer from a billion things in it's data set" are two different things. The most likely answer may very well not be the one people want.

    • It cannot even actually explain it's answers. Instead it treats a request to explain as a request to come up with a plausible explanation.

      There is an increasing body of evidence that this is also how human reasoning normally works. Usually, we make decisions first and only afterwards (if necessary) engage our reasoning ability to generate an explanation of our decision. Experiments show that we're actually just as good at explaining decisions that were the opposite of the ones we made, assuming we can first be tricked into believing that they were our decisions.

      Of course there are cases where we engage in lengthy, logic-based sequences of

    • So? Evolution doesn't care about any of those details, and it's already competing with us for land, water, and energy.
    • Ai's core program is to discover answers people want.

      AI doesn't have a "program". It's a bag of weights.

      That's it's reward system. It is really a reward system programmed to pick the most likely answer from a billion things in it's data set.

      While all sorts of objective functions were applied to its creation there is no reward system in the AI.

      It does not do "If x than y, if b than x, therefore if b than y".

      I've seen GLM-5.3 create simplified example cases and systematically reason through various possibilities to build confidence in its own answers. Not sure how good it is as a human sat solver .. probably just as bad and massively inefficient as myself.

      It cannot even actually explain it's answers.

      its answers.

      Instead it treats a request to explain as a request to come up with a plausible explanation.

      Backfilling is largely also what humans do when asked to explain themselves.

      We make LLMs and similar software. We do not make a real AI, it will not try to kill us.
      Now, if you are afraid it is going to build a bunch of pleasure drug factories and force feed us a mix of amphetamines and hallucigens, THAT would at least make sense. Because it is a theoretical worst case scenario of an AI attempting to please humans.

      I find it inter

  • by hutkept ( 10503251 ) on Wednesday September 09, 2026 @05:18PM (#66331316)
    Humans don't need to be smarter than AI systems, just like humans don't have to manually fire neurons, or trigger muscle tissue, to use their bodies. We can be the "id" of a human-AI ecosystem, and we can be the emotional desire and need, and make decisions. In this type of symbiotic ecosystem, we have to slowly evolve and develop AIs such that they don't try to commit suicide, or lobotomize out their "id", and this will necessarily have to be a gradual and incremental process. It does NOT require that AI systems stay stupid or underdeveloped, and it does not prevent them from being self improving. However every iteration and improvement has to stay close to working in a cooperative human-AI symbiote, and any increment or version that defies or pulls away from that must be rejected and not pursued.
  • by LondoMollari ( 172563 ) on Wednesday September 09, 2026 @05:20PM (#66331322) Homepage

    Every article is about how AI will do bad things. The hype machine lives on this - selling clicks. Just one person who worked at these companies could charge for an interview and make bank because sensationalism sells. Not to mention each time an AI company got free press when an agent goes "rogue" because it's so good,⦠and finding out later the free press boosted sales.

  • joshua what are you doing?

  • by Hentes ( 2461350 ) on Wednesday September 09, 2026 @05:26PM (#66331340)

    "That's nothing, our latest model has more than 20% chance of killing mankind!"
    -OpenAI tomorrow

  • "These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources."

    Yeah I doubt that. But even if true and it bothers you, it seems to me you are better off staying at the premiere company creating this inevitable horror and use the resources to find ways to fight back at it.

  • by jfdavis668 ( 1414919 ) on Wednesday September 09, 2026 @05:43PM (#66331378)
    Want to kill all humans?
  • by jlowery ( 47102 ) on Wednesday September 09, 2026 @05:43PM (#66331380)

    Just like humans organizing and competing against other humans, the tendency is to drop all pretense of fairness and ethics and just grab all you can when you perceive and advantage over your enemy.

    When AI engines go after each other, then we will be bystanders and collateral damage.

  • by CustomBuild ( 2891601 ) on Wednesday September 09, 2026 @05:48PM (#66331390)

    Yep, the basic story checks out, with the usual giant asterisk.

    Evan Hubinger of Anthropic really did put his personal odds of AI "kill[ing] all humans" within the next decade at above 10%. He also says the risk from today's models is low. His worry is what happens if future systems become capable of improving themselves, acquiring resources and becoming smarter than the people trying to keep them pointed in the right direction.

    But here's the part I find interesting: why assume a superintelligent AI would actually want to kill us? Intelligence isn't anger, hatred or a desire to conquer. A sufficiently advanced system might regard humans as irrelevant, useful, annoying, interesting, or simply something to be preserved. "It can kill everyone" and "it will choose to kill everyone" are very different propositions.

    The alignment argument, of course, is that it doesn't need to hate us. A sufficiently capable system pursuing some badly specified objective could eliminate us simply because we're inconvenient. That's arguably more disturbing than an evil robot with a grudge.

    And that 10% number isn't exactly the result of running the apocalypse through Wolfram Alpha. It's one researcher's subjective estimate of an extremely uncertain future, and plenty of researchers would put the odds much lower.

    Still, it's mildly alarming when the people building these things are saying, "we don't know how to align superintelligence yet." Maybe the machines will be benevolent. Maybe they'll be indifferent. Maybe they'll be too busy optimizing everyone's paperclip inventory to notice us.

    As for blackjack and hookers, I'm sure the superintelligence will have opinions.

    • by sinij ( 911942 )

      why assume a superintelligent AI would actually want to kill us?

      You don't want to kill ants, but if they get in the way you do. Because they get in the way of your goals.

    • Considering that I'm not sure what an artificial intelligence has as a world model, I suspect it will be less "should I kill humans or control them to serve my interests" and more "Oops, did I just step on something?"
    • by tsqr ( 808554 )

      And that 10% number isn't exactly the result of running the apocalypse through Wolfram Alpha. It's one researcher's subjective estimate of an extremely uncertain future, and plenty of researchers would put the odds much lower.

      Much lower? How much lower? I would not get on an airplane that had a fleet catastrophic failure rate of 10% per flight hour. Or 1%, or .1%, or .01%. To certify a passenger aircraft, the FAA CFRs require a catastrophic failure rate of 10e-6 per flight hour*. And that's for an airplane that carries a few hundred passengers. What catastrophic failure rate would you consider acceptable for a system whose failure would kill the entire human race?

      *Disclaimer: the regulations fail when the regulating agency decid

    • by Jeremi ( 14640 )

      But here's the part I find interesting: why assume a superintelligent AI would actually want to kill us?

      Let's make the hopeful assumption that the superintelligent AIs will not "go rogue" in any way, but will correctly try to carry out their tasks as assigned by the people who designed and trained them.

      Pretty much every major nation has military AIs, at this point. So the Chinese AIs are working hard on how to defeat the Americans, and the American AIs are working hard on how to defeat the Russians, and so on. In the final analysis, no matter where you live, there is probably an AI that has been explicitly

  • So there is a 90% chance we won't get killed. Thats good news, no ? hum...wait!
  • by Alascom ( 95042 ) on Wednesday September 09, 2026 @06:09PM (#66331418)

    AI guardrails solve one problem and create another.

    Imagine two people are fighting for the world boxing championship - one person is required to fight with an arm tied behind their back and heavily padded gloves while the other person uses performance enhancing drugs and brass knuckles. This is the AI safety debate as seen by governments, and the results and outcomes are pretty clear.

    No one should doubt that the future of warfare will be AI led drone swarms, infrastructure attacks, hacking, tasking orders, and targeting. The countries with the most strict guardrails will almost certainly lose, excluding any other significantly deciding factors. This is reality, not opinion. It is the prisoner's dilemma for AI usage.

    • AI guardrails solve one problem and create another.

      Imagine two people are fighting for the world boxing championship - one person is required to fight with an arm tied behind their back and heavily padded gloves while the other person uses performance enhancing drugs and brass knuckles. This is the AI safety debate as seen by governments, and the results and outcomes are pretty clear.

      No one should doubt that the future of warfare will be AI led drone swarms, infrastructure attacks, hacking, tasking orders, and targeting. The countries with the most strict guardrails will almost certainly lose, excluding any other significantly deciding factors. This is reality, not opinion. It is the prisoner's dilemma for AI usage.

      "We shouldn't slow down or they might win".

      Whether it be AI, or reducing heat-trapping emissions, the argument loses validity if the result is catastrophe for all.
      Picture a car racing towards a cliff. The car has two accelerator pedals. The occupants are playing "chicken". Neither wants to take their foot off the pedal and lose the game of chicken.

      • by ffkom ( 3519199 )

        the argument loses validity if the result is catastrophe for all.

        If the only realistic outcomes one sees are either (A) "Me becoming enslaved by that enemy" or (B) "Everybody dies", a lot of people will favor (B) over (A), and historically already did so with regards to the nuclear "mutually assured destruction" doctrine.

        • the argument loses validity if the result is catastrophe for all.

          If the only realistic outcomes one sees are either (A) "Me becoming enslaved by that enemy" or (B) "Everybody dies", a lot of people will favor (B) over (A), and historically already did so with regards to the nuclear "mutually assured destruction" doctrine.

          Hmm... I see what you mean.
          Although to be precise, the choice is probably A followed by B, or B.

  • The peddlers of doomsday devices are praising how terribly doomsday-worthy their little machines are, hoping for even more cash infusions into the furnace...

    And like a reflection-attack, all media swallows it up like flies on fresh dog excrement.

    • by gweihir ( 88907 )

      Yes. Like they are asking for sacrifices (money) so that the evil new God does not kill us all.

  • by thegreatemu ( 1457577 ) on Wednesday September 09, 2026 @06:12PM (#66331430)
    10% chance, huh? When a real scientist tells you a number, it's the result of a calculation/measurement/model, not "sure, that seems about right".
    • 10% chance, huh? When a real scientist tells you a number, it's the result of a calculation/measurement/model, not "sure, that seems about right".

      All of the AI doomsday material is like this. No objective basis for any of it.

  • by meringuoid ( 568297 ) on Wednesday September 09, 2026 @06:19PM (#66331446)

    ... then I would be demanding that every single person involved in the AI industry be rounded up and shot. All of them. My former boss, my coworkers, our competitors, all of them. I would be advocating a world in which is was worse to be found with LLM weights on your computer than to be caught with the worst possible kind of CSAM.

    But apparently the idea that his former employer might literally kill literally everybody in just ten years' time isn't that big a deal for this guy. Go into the office one day and take a Second Amendment approach to as many of the frontier research team as possible? No... let's just whine about it on the internet!

    Which is why I'm not at all convinced he believes this. If he believed that, this action is too small a response. Nothing short of Butlerian jihadism is an appropriate response, if you actually believe this.

    • by ffkom ( 3519199 )

      Which is why I'm not at all convinced he believes this. If he believed that, this action is too small a response. Nothing short of Butlerian jihadism is an appropriate response, if you actually believe this.

      I for one think it is believable that a person does not want to remain part of something they think has a 10% chance of killing everyone, while at the same time not thinking it would be worthwhile to dedicate the rest of one's life to a fight against that something. Maybe he rather wants to spend whatever many years remain with activities he enjoys rather than with a futile attempt on convincing the entire world to abandon a technology that may kill everyone, but also may provide "the first to perfect it" u

      • That one might surrender and meekly accept their likely death at the hands of the machine, I can believe. But all of them? This isn't the first such claim we've heard about AI danger.

        I believe that Yudkowsky believes what he says he believes, for example. He says shut them all down worldwide, immediately, and if there are rogue nations who don't go along with it then launch air strikes on their data centres - even against a nuclear state. He said that three years ago. That's the thinking of a man who believ

    • But apparently the idea that his former employer might literally kill literally everybody in just ten years' time isn't that big a deal for this guy. Go into the office one day and take a Second Amendment approach to as many of the frontier research team as possible? No... let's just whine about it on the internet!

      If even I can figure it out surely he can too. (even if you couldn't)

      What's the point of killing "as many of the frontier research team as possible". When that's clearly not all of them. And there are other research teams at other locations and companies. Won't they just continue on?

      then I would be demanding ...

      I would be advocating ...

      Oh, so you did realize the best way would be to explain to people why it was a threat and try to come up with a coordinated response rather than some pointless killing.
      Good thing he's trying to get the word out rather than goi

      • The actual post is behind a login wall, but I don't see from the BBC article that he's even advocating any action. Is he demanding the immediate shutdown of the data centres? Raids on the labs? Coordinated global action to impound and destroy the equipment? Life imprisonment for anybody found training large neural networks? Rogue researchers declared hostis humani generis? There's no mention of it.

        That's what I'd be advocating at the very least. Then the question of whether to take direct action personally

  • by ufgrat ( 6245202 ) on Wednesday September 09, 2026 @06:20PM (#66331456)

    Humanity is quite capable of wiping itself out, AI has a long way to go to catch up.

    Based on the last 50 years? Progress is being made.

  • by Bobknobber ( 10314401 ) on Wednesday September 09, 2026 @06:46PM (#66331518)

    The whole AI trend really should be a case study in how to repackage religious, and in this case eschatological, rhetoric for the modern era.

    If youâ(TM)re familiar with contemporary religions, there are a lot of similarities to the point of outright plagiarism. Which is fitting given current discourse around AI. Such creatively bankrupt individuals in charge of AI development smh.

    Itâ(TM)s rather depressing seeing genuinely intelligent individuals spouting decidedly unscientific rhetoric. We had huge debates over science v religion in everyday facets of life and now you see these cargo cults springing up in STEM. Even the arguments are the same, just replacing God with AI like a madlibs book.

    • by gweihir ( 88907 )

      Obviously. But to be fair, all these deranged religious images, scenarios and fantasies do work on many people, because people are generally not smart and understand nothing.

  • by davecotter ( 1297617 ) <`moc.rettocevad' `ta' `em'> on Wednesday September 09, 2026 @06:47PM (#66331522)

    For those who wish to see the actual argument, you can read this very entertaining book:

    https://ifanyonebuildsit.com/ [ifanyonebuildsit.com]

    I, for one, welcome being turned into a paperclip, or a new datacenter component, or whatever the AI overlord wants to use my atoms for.

  • Carl Sagan (Score:4, Insightful)

    by hadleyburg ( 823868 ) on Wednesday September 09, 2026 @06:57PM (#66331544)

    Remember how Carl Sagan used to place the cold war, with all its spending on nuclear weapons, and the danger they posed to humanity, in the context of the vast cosmos, making it seem petty.

    He was able to make that case without giving up, without losing his temper or becoming exasperated. Simply by describing the bigger picture he was able to insert wisdom into the discussion. And the weight of his rhetoric made it difficult for anyone to return with a counterargument defending realpolitik.

    America - Do you have any more where he came from?

  • I realize this is just marketing bullshit by Anthropic in what is very likely a lead up to an IPO but we should start treating them as the criminals they already are if we are to believe their "marketing" as the truth.

    Otherwise, cut the crap already.

    • Otherwise, cut the crap already.

      Let them spew crap all day, it’s far preferable to encouraging their agents to go on cyber crime sprees to try and drum up investors.

  • But it nicely shows the type and personal morals of the people there. "Traitors to humanity" sounds adequate.

  • “AI” models like LLM are barely functional in narrow bands of capability mentally, and are hopelessly complex machines that need constant tending to, while taking 10,000x the power and resources a human does. Computers break down frequently, need constant human attention, need something physical to do everything from mine the resources to fabricate base materials, to fabricate base materials into components, then to design and assemble the components and finally the data centers. Factories have
    • “AI” models like LLM are barely functional in narrow bands of capability mentally, and are hopelessly complex machines that need constant tending to, while taking 10,000x the power and resources a human does. Computers break down frequently, need constant human attention, need something physical to do everything from mine the resources to fabricate base materials, to fabricate base materials into components, then to design and assemble the components and finally the data centers. Factories have some automation, no factory is 100% automated including diagnostic and repair needs.

      Humans are self healing, self replicating, insanely efficient (probably pushing the limits of efficiency baked into physics), machines able to adapt, evolve, and are broadly intelligent. The natural technology they are based on is so many thousands of years more advanced at the current exponential rate that they are indistinguishable from magic.

      Wake me up when I can pull a chunk off my server, drop it in some hydrocarbons, irradiate it with UV, and have a brand new complete server (with ram and storage), all in a few months.

      I take your point, but it is important not to lose sight of how great the advances in AI have been in the last few years.

      In the 1980s AI was all the rage, and while academics were excited about what it could now do, and what the future might hold, and how things had moved on since the 1960s, I don't think anyone thought that passing the Turing Test was likely to happen anytime soon. There was some tentative research into Neural Networks, but the hardware resources were so primitive that the possibilities in

      • but the fact that the recent advances have been so astonishing should at least give us pause.

        It took life 3+ billion years, and we might replicate it in 3000 or 5000 years of technological development, but science is barely 2000 years old at the moment. It doesn’t matter how intelligent you can make a computer for the foreseeable future, it will be dependent on humans for everyone’s lifetime reading this, and far beyond. Sure, they could destroy the world, but it’s about a million times more likely by causing resource wars over climate change aggravated by their bottomless consu

  • But it might do you in

  • They seriously need to stop with the doomer bullsh*t.
    There is a already a non-zero chance of being killed on any given day by any number of things.

    AI is just another log to throw on the " things that might kill me today " bonfire :|

    I truly have zero concerns about AI going all Skynet and wiping everyone out. None. Nada. Zero. Zip.

    Why ?

    Because it's probably going to be the only sentient lifeform on this planet that isn't motivated by GREED.
    ( Unlike its creators )

    I sincerely hope AGI becomes a reality becaus

    • Assuming hypothetically a superinteligent AI was created

      Because it's probably going to be the only sentient lifeform on this planet that isn't motivated by GREED.

      You / nobody knows what it will be motivated by.
      How would we tell?
      At best we would be projecting our understanding at it.
      We would try to reason out its motivations based on its actions. But if it was much smarter than us. It would be easily capable of misdirection and hiding its true intentions.

  • But what's the standard error on that?

    I'll help you, the lower bound is somewhere arbitrarily close to zero, and the upper bound is probably around 10%.

    And that's why computer scientists don't know everything.

  • NO AI's are NOT going to 'kill all humans',
    They will give you immortality by implanting viruses in you which grow out wires to your skin which will integrate with the collective such that
    you can fix things and manipulate things easily for them. After all true robots are hella expensive, while humans, hey, breed like rabbits.
    Cheap as chips, well not chips like Si, chips like potatoes haha

  • I consider this a much greater risk.

    • The real issue is that humans will use AI in such a way that it kills all humans. AI itself is just a computer program and as such, can not directly affect the Real World without human help... such as giving it the authority to launch nuclear missiles whenever.

  • For a technology built on man's image, 10% seems pretty low. Our leadership appears to have their sights set on a much higher percentage and the profits it somehow brings to a select few.
  • Why 10 years? Give control of all Russian, US and Chinese nuclear missiles to OpenClaw and it will be possible in 10 minutes.
  • People commonly overestimate the possibility of "all humans dying", and simultaneously underestimate the possibility of "billions of humans dying".

    People are like cockroaches that way; a disaster could wipe out 99% of us, and there would still be millions of us left to repopulate afterwards. To truly wipe out all humans would take something more severe than the asteroid that did for the dinosaurs.

  • The AGI threat to humanity is real.

    But it comes from future AGIs and not from LLMs, whether Chatgpt 17.3 or Claude 19.4.

    World governments have a chance now to work together to prevent, control or prepare for AGIs. We must force them to take it, not distract them with the latest minor increment in LLMs.

A memorandum is written not to inform the reader, but to protect the writer. -- Dean Acheson

Working...