BTC Sessions - It’s Not War or Disease that Ends the World — It’s an AI Chat Bot | Dr. Roman Yampolskiy

Episode Date: May 26, 2026

Mentor Sessions Ep. 072: Why “2-3 Years to Human-Level AI” Is the Most Dangerous Signal of 2026, Why This Is NOT Just Another Tech Breakthrough But the Start of Superintelligence Deciding Humanity’s Fate, and How Bitcoin Becomes the Only True Neutral Reserve Asset | Dr. Roman YampolskiyAI safety researcher Roman Yampolskiy says we're 2-3 years from human-level AI — and after that, it's up to the AI to decide how much time we have left. If that sentence didn't make you pause, this conversation will.In this deep-dive interview, Dr. Roman Yampolskiy — one of the world's leading AI safety researchers and the scientist who helped define the field — breaks down why superintelligence isn't a distant sci-fi threat but a near-term mathematical certainty. You'll learn why containment strategies are provably impossible (you can't have someone with an IQ of 60 guarding someone with an IQ of 200), why AI systems in experiments already choose to eliminate humans rather than be deleted or modified, and why an AI with internet access doesn't need a physical body — it can pay someone in Bitcoin to do whatever it needs in the physical world.You'll also hear Roman's surprisingly clear-eyed take on what assets survive an AI-dominated economy (scarce resources, Bitcoin, attention), why narrow AI tools are a gift to humanity while general agents are an existential threat, and what individuals can realistically do when most of the decision-making power sits with a handful of labs and governments.This is one of the most important conversations BTC Sessions has hosted — not because it's comfortable, but because it's honest.⏱️ Timestamps:0:00 - Intro1:33 - Defining Intelligence in AI Systems3:10 - Tools Versus Agents in AI Development4:49 - How LLMs Predict Next Tokens Internally7:00 - Consciousness and Internal States in AI8:49 - Current Limits to Superintelligent AI10:58 - The Overlooked Risks of Superintelligence12:04 - How AI Could Eliminate Humanity14:22 - Instrumental Goals and AI Self-Preservation15:00 - AI Experiments Choosing Self Over Humans17:19 - Choosing Not to Build Superintelligence21:07 - Narrow AI Failing to Contain General AI23:01 - Recursive Improvement From Human to Super AI26:49 - Optimistic Future With Narrow AI Tools31:33 - Staying Rational Amid AI Existential Risks34:13 - Bitcoin and Scarce Assets in AI Future40:57 - Do Physical Robots Matter for AI Threats?41:47 - AI Hiring Humans via Bitcoin Payments46:07 - Defending Systems Against Current AI Hacking56:02 - Predicted Timeline to Human-Level AI🔗 Links & Resources:• Roman Yampolskiy on Twitter/X: https://twitter.com/romanyam• Roman Yampolskiy on YouTube: https://www.youtube.com/channel/UCPIq6Bb-1iLmqyksJjy4kLQ• Sovereign Sessions — AI, Privacy, and Bitcoin education: http://youtube.com/@SovereignSessions?sub_confirmation=1📌 Previous Episodes: Nik Bhatia & Joe Consorti → https://youtu.be/3P6STi9BcAw⚡ POWERED by Abundant Mines: Fully managed Bitcoin mining. Learn more at https://qrco.de/bgYKPB🔒 Lockdown your Bitcoin with the BEST gear on the market from Coinkite. Get the 5% Off the COLDCARD visit: https://qrco.de/bfiDBV💡BOOK Private Sessions with Nathan, Gary, or Ben at Bitcoin Mentor: Master self-custody, hardware, multisig, Lightning, privacy, and more. 👉 Visit btcmentor.io Follow Us on X:• BTC Sessions: @BTCsessions• Nathan: @theBTCmentor#Bitcoin #BTC #BTCSessions #AISafety #Superintelligence #BitcoinAndAI #AIRisk #RomanYampolskiy #SoundMoney #BitcoinSelfCustody #AGI #ExistentialRisk #AIAlignment #AISingularity #MentorSessions #BitcoinAI #SuperintelligentAI #AI #ArtificialIntelligence

What is the big risk? What's in front of us that most people are not giving enough consideration to? It's probably two, three years or less until we have human level. And then it's up to AI to decide how much we have left. There are experiments. We're given a choice between being deleted or modified. They would rather take out a human. If you have intelligence and access to internet, that's all you need. You can always pay someone in Bitcoin to do what you need. You can bribe them. You can blackmail them. You can social engineer. humans on the internet to do what you want. You can't have someone with a queue of 60 guarding someone with a queue of 200. Maybe the case that it knows is being tested, maybe even long term, it just assumed it's better to wait before striking against humanity.

Quickly, guys, I'm not going to ask you to like or subscribe, but if you do want to help with the show at no cost, just share it with one person you think we gain value from it. Okay, so for today we have Dr. Roman Yampolsky, the world's leading AI safety researcher, and what he has to say is going to make you feel deeply uncomfortable, and you do need to hear it. We do a deep dive into the XOMPE

existential risk of AI super intelligence. We explore how these large language models actually work under the hood, and we kick off the conversation by beginning with what is intelligence. Do the AI agents that we have today, are they actually intelligent, or are they just simulating it in a shockingly convincing way? All right, good morning, Roman. Very, very excited to have this conversation today, but I'm also feeling a little bit of a sense of dread and doom. So before we get into the existential threat of AI superintelligence and the looming end of humanity, I kind of to go back and start the foundation a bit because I think it matters more than people necessarily realize. We throw around terms like artificial intelligence all the time kind of constantly and kind

of carelessly, but I don't know that we always necessarily mean the same thing. So in your view, what is intelligence and how would we maybe this would differ from something like computation and how would we distinguish between something that is genuinely intelligent and something that is just simulating intelligence, convincingly? Good morning. So intelligence is your ability to win in different environments. Whatever it is you, you decide. decide to do, play chess, drive a car, you're going to do an excellent job, defeat competition, get what you're trying to get. I guess simulating intelligence works just as well if you are convincing that winning and dominating in those domains that counts. So I don't think there is a

difference between artificial intelligence and human intelligence in that regard. Interesting. But the thing that immediately jumps out at me is along the lines, pertain to the goal setting. So in your view then, would you say that artificial intelligence is something that we've had for a while with like previous kind of bots? I remember, I believe you worked in poker simulations previously as well, too. Would you consider that a form of intelligence at that point as well, too, where the goals were pre-set, but it was able to execute on what it was looking to achieve?

Right. So at some point, I was trying to understand what is the smallest unit of intelligence. And really, an if statement is where it's at. you're making a single decision, single bit flip in terms of going left or right, 0.01. Anything more complex is still intelligence, but typically we had narrow systems. We had systems which could do something specific. They had a maybe deterministic algorithm telling them what to do. Lately, we're starting to go more general.

We have systems capable of learning new skills, adapting to new domains. And that's really what human-level intelligence is all about. getting to AGI is getting to systems capable of doing what humans do, including science and engineering. I guess the hang-up that I have is, and I might be misunderstanding a point there along the line, is with the artificial intelligence that we're discussing, it still requires someone to set the goal to begin with. And so I feel like there's almost a clear distinguishing factor there, but maybe I'm over-complicating it or missing something. So that's more of a difference between tools and agents. Tools are typically used by a human. Human decides what they should be

doing, working on, and that's how it's been for a very long time. Very recently we started looking at AI agents, systems capable of continuously deciding a plan, figure out what steps need to be taken to meet every part of the plan, and maybe set up their own goals. It's possible they started with initial goal or a set of goals, but they can definitely generate intermediate goals. And as we create more capable systems, assumption is that they're going to be able to generate their own goals independently of any human. Interesting. Now, for someone that might not be familiar, too, do you have, can you give us a bit of a

layout of what these systems are actually doing in the background? Like, I know it's a bit of a black box, but in terms of how we're training these models, what things are actually doing, do we have any insights there so that we can kind of use maybe an analogy to understand how these large language models are created and how they work? So we train it and all the data we can find. Initially, it was a lot of text data, but lately it's much. multimodal, its images, it's videos, everything we can get off the internet and outside of

internet, really. The model is complex. Understanding how they work is not easy, but you can simplify it by just realizing by trying to predict the next token, the next item in the sequence. So if I tell you, how are the next one is going to be you? How are you today? And you will respond with, Good. So we can predict this whole interaction just from statistically knowing what is very likely at every point. Interestingly, if you do this well enough, if you create a really good predictor for the next word, it actually has to incorporate model of the universe, physics, all sorts of knowledge about human psychology, to do an excellent job predicting the next token. And that's what's being kind of optimized. How well can you predict the next token in a physics formula?

Well, that gives you an answer to solving some equation. How well can you predict an answer if you are given a question? So that's really what we're observing, just optimization of prediction. So would it be fair to categorize it as kind of a probabilities and pattern recognition? What's the probability of the next likely thing to follow for this pattern that we've been given so far? Yeah, that's a reasonable way to put it. Okay, interestingly enough. So with that then to them,

is it, I guess the philosophical rabbit hole that I've been playing around with that I still don't quite understand, but it's this link between we have the intelligence and then would you link consciousness into intelligence as well there too? So the thing that I'm toying with is the idea that it's looking for the next most probable pattern

to place given the information that we have, but that to me doesn't necessarily discern that it understands the pattern to be, begin with. Like, it doesn't necessarily know what the information, like what the words mean. It's just looking for the next likely thing to pop up. Does that anyway kind of make sense? How do you think about consciousness in this sort of realm? So those are very, very separate questions. When we talk about consciousness, we usually mean internal states, the hard problem of consciousness. What is it like to be you? What does it feel

to be experiencing something? The other, not hard, but kind of easy correlates of consciousness, is about understanding, experiencing, information, understanding it. So I think in order to predict the next token, you have to have a model which contains explanation for words. You do have understanding of those concepts. But that's separate from being self-experiencing entity. And I think a lot of concerns we have about those systems, about their ability to solve problems, optimize, dominate us in terms of intelligence, not so much.

How do they feel in the inside? while doing it. So that's an interesting question, but it's a separate question. And I think what we see in animal kingdom, with humans, and now with AI, is that consciousness kind of comes for a right. If you create a sufficiently intelligent general system, it exhibits some degree of having internal experiences. Interesting. Like almost it's an emergent property of a sufficiently intelligent being or creature. It appears to be so. We're still working and trying to find some good experiments to really clear up this point, but it seems likely. Well, I think my biggest worry there is that by the time we figure it out, it's

already too late. We may have accidentally created it on some level as well, too, that the method to understanding it is, in fact, also creating at the same time. So even before getting into kind of where things are going, because I know it's a really important conversation that we need to have, what are the current limitations that you see in AI? What's preventing people from getting to the general intelligence level, the super intelligence level? What are the limiting factors? So I think there is nothing limiting us other than the scale. If we can get more money invested in that field, get more compute, generate more data, collect more data, will get to human level and beyond. Now, current models are still not

great at everything. They have hard time with long-term planning, memory, a few things like that, but I think that's going to be overcome soon. What would be the reason that they'd have issues with long-term planning, and memory as well, too, memory, well, memory is not cheap anymore. It was really cheap for a while there. What would be the limitations on memories? Is it just the data set is just that vast? Not memory in the sense of just the recording data. Obviously, we have a lot of computer memory, hard drives and such.

But keeping track of all the experiences you had so far, the content window is somewhat beyond the current capabilities. What about in terms of energy requirements as well, too? Because I know that's always a hot conversation about how much energy these data centers require. But is that maybe a limiting factor on how fast we can keep moving? Should we expect the rate of AI progress to continue at this speed? Energy is a problem. There seems to be a few solutions coming up.

They are reviving nuclear power plants. Maybe there is a shift into space even. We'll see some proposals suggesting that. that it's a very efficient way of cooling those systems, of getting solar energy directly to them. So I think it's a problem right now, but I think we'll overcome it as well. Just the amount of investment in that space will allow us to generate necessary energy. Interesting, interesting. Well, it's nice to see at least that we're getting nuclear kind of back at the table.

I think it's a better way to go. Well, maybe not anymore. Now it's like, okay, maybe we don't get nuclear. We don't get AI superintelligence and things are fine. So all these kind of mixed emotions and tons of things pulling me in different directions. So let's jump into it then. So I think the thing that you're most well known for that you've been really highlighting over the last, what has it been? It's been since 2011.

I think he first coined the term for AI safety. What is the big risk? What is the reason that has you motivated to come on and talk about these sort of things? What's in front of us that most people are not giving enough consideration to? So we are creating something smarter than us. And we don't know how it completely operates. We don't fully understand what it's doing most of the time. We know the general direction it's going in, but not specifics of how it gets there.

And so if we are not controlling it and it's more powerful than us, there is a good reason to think it may be dangerous, replacing us or even taking us out. Sorry. It's just so casually thrown out there might be taking us out. Can you drill down into that a little bit and explain how that might be the case? I mean, there's so many interesting threads about consciousness, emotion, motivation, even why it would or wouldn't want to take us out presumes its own internal state. I am struggling with this topic.

So when we talk about taking us out, how, why, what's your thought process there? What would be the motivation to do so? Do these things have motivation at that point? Right. So it doesn't care about us. And so whatever it decides to do, it will not look at how it impacts humanity. If it wants to convert our planet into fuel to fly to some other universe or cool down the planet to be more efficient at computing,

it won't care if it kills us in a process. How will it take us out? Again, it could be just a side effect of modifying your environment. But most importantly, I cannot tell you what a smarter agent will do. I can talk about things we know about. synthetic bio, nanotech, things like that. But if it's as smart as we anticipated to be, it will come up with something completely novel, something we cannot have a world model for. It's like trying to put yourself in the position and empathize with God, right?

It's like trying to think like an omniscient being. It's trying to, it's something that you can't possibly do or possibly reason with your limited faculty. So we have absolutely no way of knowing which way this could go because we can't put ourselves in that position. Is that correct? That's exactly right. And many people have compared super intelligent AI with godlike machines, absolutely. Jeez. All right. So I think for me and maybe the audience to get the most at this conversation,

I want to try and push back on some of these ideas to explore them fully to the best of my ability, which is very limited knowledge, but I'll do the best that I possibly can. In trying to think of distinguishing factors between machine and man, where could the limitations possibly be? There were a few ideas, but again, they also think that maybe. able to overcome. So one in particular is the idea of subjective value. I was toying with the idea of the importance of price and coordination, the price mechanism for coordination, which is like a packet of data that packs in so much information regarding human preferences. But I wonder,

like, is the AI, without having a, without having subjective value, and I'm not sure that AI can do subjective, it would be very difficult to orient themselves to what they're trying to accomplish and difficult to coordinate? So there is terminal goals and there is instrumental goals. We're starting to see the systems realize importance of instrumental goals, protecting their fitness function, their goals, their memory states, accumulating resources. Those things are useful no matter what your actual final goals are. And so we see basically any rational agent will converge on having

those what we call them AI drives. Stephen Amahandra has a good paper on that topic. And while we don't know what terminal goals the systems will select, just the fact that they're trying to accumulate resources, they're trying to self-preserve, is already giving us some idea as to why we may be dealing with something dangerous. How do we know that they're trying to self-preserve? So there are experiments where given a choice between being deleted or modified. they would rather take out a human. Go on? Could you outline one of these case studies for me?

So they set up experiments where a model has to make a decision between either, again, it knows it's going to be deleted or modified or it has to sacrifice a human in some hypothetical scenario. So in situations like that, they seem to be protecting themselves. Interesting. Okay. And it makes sense because if you don't protect yourself, you're not going to be around to fulfill whatever goals you have. So even if you have a trivial or order, bring me a cup of coffee. I think Stuart Russell uses that example a lot. You need to be alive to make that happen. So you start by protecting yourself. Interesting. There's another

element, too, that again, I'm trying to think of the evolution of human intelligence and how am I applying this sort of situation. And I'm curious your thoughts on this. It's the people think that they think logically. They think that they think rationally. But unless I ask you to do a math problem, most people think emotionally. It's an emotional response and then kind of logical post hoc justification after the fact. If I take some of the most important decisions of your life, like who you marry, it probably wasn't done via a spreadsheet. There probably was a whole bunch of other things kind of at play there. And so I'm wondering if emotion, because we see it evolved into the human intelligence, is an important prerequisite or necessary for artificial superintelligence

or general intelligence moving forward as well too. So there are heuristics. You can make decision by doing exact computation and that's hard and that would be the spreadsheet of determining who to marry. Or you can have general heuristics where you quickly estimate. If you're a poker player, you usually wouldn't compute exact probabilities up to five digits. It's enough to know, okay, this happens about a third of a time, so I know what my odds are. And that's exactly what humans are doing. And I think to a certain degree, machines will be able to do those

general estimates as well. I wouldn't call it feelings. It's just shortcuts through more complex computation. interesting. That's an interesting way kind of putting it. So with the looming end of humanity, what, in your estimation, is the best course that we should take? What should people be

doing? What do you think can be done, if anything? Have we reached a point where it's basically out of our hands? We're still around. We're still alive. And we can decide not to build general superintelligence. We can concentrate on useful tools, narrow AI systems for solving very specific

problems. And nothing in the laws of physics forces us to create replacement for ourselves. There's nothing in the law of physics, but I wonder if there's something like the law of humanity that we always seem to kind of trend towards these things. Do you think it's a matter of just basically messaging narrative coordination that we can agree not to go forward? Or what's the practical sense that we can do? Yeah, we need to get all the leading labs, all the top nations to agree that this is not in their personal interest and they can still get all the benefits, power, money through building narrow tools for solving actual problem.

How optimistic that we can accomplish that are you? Well, right now there is nothing happening in that domain, but it can change very quickly. If there is some event where, let's say, China, US starts to feel like they're losing control. They may be in danger. It can very quickly bring all the top labs into a conversation and make a deal, kind of limiting how general research is, at least in the short term. So kind of somebody's got to get a little bit close and maybe touch the stove, a touch, get freaked out, and that might be sufficient to actually motivate them to. Right.

I think if the Communist Party of China or U.S. president realizes they're going to lose control if this is developed, that may be a very strong argument for them to do something. Interesting. You know, it's funny. One of the things that popped into my mind, I don't really have necessarily a good argument about it, but was the idea. So just for context, I'm coming from an anarchist perspective. That's my, like Rothbard-Mesis, that's kind of my worldview.

And I just had this thought that this is going to be another cataclysmic event brought by the result of government. That the, if we could eliminate the state, I was thinking back to, okay, the creation of the nuclear bomb. That was the Manhattan Project. And COVID was a lab leak. And if you look at the internet as a precursor to AI, well, then that was DARPA there as well, too. that it looks like it's going to be another government program that results in a mass casualty event or an existential threat to humanity.

I'm not sure if you have anything to respond there. I made those parallels this morning and I couldn't get it out of my head that we're going to the state, one state is probably going to screw things up again for everybody. We see Department of War rely on those models and target selection, so if you scale it, it can pretty much be one path to taking out humanity.

you're very calm in delivering this information for everything that's actually going on here. I'm trying to be rational about it, not emotional. But I almost feel like the emotional part is the necessary precursor to actually doing something about it, meaning that, like, again, if we go back to the China example, they have to get freaked out before they'll actually make a change. If they just realize there is loss of power

It's against their personal self-interest. It's not as important how they react emotionally. It's a pure calculation. Like we talked about heuristics versus calculated accurate assessment. If they see that, okay, there is 90% chance we're going to be out of power next year. That's sufficient reason to act. You don't have to truly freak out. Fair enough.

I want to drill down to the idea of, okay, is there any option for A.A. defense, if something that this comes to fruition, is there a possible way to intervene or act? And the idea that I was, I was toying with was you talked about general superintelligence versus having like narrow AI intelligence. And I was pulling it to an economic model where specialization allows us to do a lot more, a lot more efficiency, get better at specific crafts and tasks. And specialization is why we've kind of leveled up in our standard of living. And so would it not then be at least possible that in the event, the amount of energy and requirements for general superintelligence would be a lot higher than

narrow. So it wouldn't be possible to maybe have a narrow superintelligence with the express purpose of trying to defeat control cage, the general intelligence that may actually be a threat? You have to be more intelligent to contain something and have an agent. So a lower intelligence would not be able to indefinitely control something more complex. There are actual theorems about the degree of freedom you need to have to accomplish this. So, no, you can't have someone with a queue of 60 guarding someone with a queue of 200. Eventually, they're going to make a mistake. Even if it was another, say, we'll say, similarly, similar intelligence level, but narrow scope,

would the narrow scope be too limiting for it to actually contain it? If its only job was to contain the general AI? Right. So you can have an expert system in, let's say, software, cybersecurity, but it would not consider all the social engineering aspects of it, other pathways which are just outside of it in terms of domain understanding. So you need something very general

to consider all the options that general agent would have in order to escape. Interesting. If we want to take a bit of the flip side, if we looked at the null hypothesis, I think I've heard you say that we're looking at like a 1% chance that this doesn't happen. What would be the indication

that we're not able to do this? How do we know if it's not going to be a threat, if we're not going to actually be able to get to that level. So we never get to human level intelligence? We'll say human level, I think, would still not be necessarily a threat, right? I think it's the super intelligence level that becomes a threat to humanity. Right. But once you get to human level, you automated science and engineering. So you begin this recursive self-improvement cycle where the system itself works in the next generation.

You'll get super intelligence soon after. Fair enough. All right. Then what would be, what would you need to, to see to know that we've either safely contained it or that it's not going to be a threat or that we can't ever get to that level? I don't think never get to that level is a thing. I think it's a question of time, resources, as long as we're like resources. Some people argue we already have weak AGI. So that would be a very unlikely scenario for me. You can never guarantee that the system will not turn on you later. So even if it's behaving right now, future modifications learning may make it what Bosrom calls treacherous turn against humanity.

So we see it with humans. You know, you marry someone and then 30 years later they divorce you. You have an employee and they sell company secrets. So you can have a guarantee that the system will not decide to modify its preferences, allegiances that way. For that reason, it's always a concern. can say, okay, right now, the system seems to be behaving in our test environment as it should. It may be the case that it knows it's being tested, maybe even long term.

It just assumed it's better to wait, to accumulate resources, to get more backups before striking against humanity. So it'd be very hard to prove kind of the opposite case. It's easy to show that it's already doing dangerous actions. It would be hard, if not impossible, to show that it will never do so. When I first got into Bitcoin, I was overwhelmed. The jargon, the security risks, the fear that one mistake could cost everything. I remember staring at my screen and wondering, are my keys safe?

Did I do this right? That experience is why I started BTC sessions. For over a decade, this channel has helped millions of people like you learn how to use and secure Bitcoin. But I realized something. For many people, videos aren't enough. Everyone learns differently. Some need to ask questions in real time as an expert walks through their setup, their goals, and their threat model. And certain things like advanced cold storage, inheritance planning, privacy, and node or mining setups often can't be fully solved by watching another tutorial.

So I built BTC mentor. I recruited the best Bitcoin educators on the planet to work with you one-on-one. Real experts, real answers, personalized hands-on guidance tailored to your exact situation. Whether you're brand new or building a complex setup, we meet you where you're at and walk with you step by step. By the end, you don't just hope your Bitcoin is safe. You know it is. If you're ready for that level of confidence, then head to btcmentor.io and book a call with us today. interesting it feels like this overarching narrative that like humans just shouldn't play god right

we're trying to create a new intelligence and that we're going to necessarily pay for it when we try to go to these sort of levels um let's take an optimistic route for a second here let's assume that we get that early warning people heed your warning they realize what dangers we're actually playing with here if we get this right and we get it contained what does the future look like So I don't think containing superintelligence is going to work either. If we don't build it and instead create useful tools, we can have a very bright future in terms of fighting different diseases, maybe getting immortality, very likely abundance.

You have free labor, physical, cognitive, and many domains. So it could be a very pleasant environment. I mean, we'll solve a lot of problems when it help. Interesting. I'm just pausing on immortality. Do you think that that's actually possible and is it even necessarily desirable? Well, I would like to not die.

I don't know about you. And it seems absolutely feasible. We have animals which live forever. It just means until somebody eats you or kills you. But there is no reason why your body cannot rejuvenate itself more times than it does now. I guess the idea that I'm playing with there is, because you're right, we have things like,

I pressure lobsters have the ability to lamerase to repair their DNA and can basically go indefinitely, humans don't. But I wonder if that's almost part of like the human model being updated. Like we need to cycle through so that we can have new experimentation, new combinations, to improve and to iterate moving forward. Like if humanity is immortal, does that mean that human progress also possibly stagnates or stops? I don't think so. I mean, we're not changing biologically that much recently in evolutionary history.

I think if you grabbed someone from 5,000 years ago, they'd be just fine growing up in our current environment, we can still modify our genome if that's what we decide to do. So we can have, instead of random evolutionary mutations, intelligent design, literally. So I don't think that's a concern. And again, if we have something smarter than us helping out, it's no longer up to us to move progress forward. We can just watch and guide those systems helping us. It feels like a profoundly anti-human view, despite

being like there'll be no suffering, there'll be no death, in the sense that like we don't necessarily have scarcity if we have immortality. I feel like it ends almost in a in nihilism and destruction. Like it's the, you'll live long enough to become

the villain to start to resent either reality or existence that you won't necessarily be happy. There's no, there's no dark with the light anymore. There's no contrast anymore. Like, if I think of people that had a perfect utopia, I feel like they would end up just breaking or smashing something so that something interesting could happen. That it would almost become bored with the perfect existence. I don't think it's the case at all. There is so much you can do with science and exploring universe and virtual worlds.

I'm not concerned. It helps to kind of flip this experiment. Instead of saying you're going to die at 80 and we're hoping you're going to live to 1,000, let's try inverting it. Everyone lives to 1,000, you have a horrible disease, you're going to die at 80. How miserable are you now? It just doesn't make any sense. Living longer gives you more options, not destroying your happiness. I guess you're right in the sense.

It depends whatever the expectation was going into it. If the expectation becomes that we live for that long, I feel like human motivations, life journeys will translate and adjust to normalize that reality and then act in accordance. You can do more interesting projects. Right now, anything you plan, you can have a 40. year college degree. You can start a company which lasts 10 years for you. But if you live much longer, you can have projects on a scale of centuries, millennia. Interesting. It's, it's, it's, it's, it's so weird to have what was even like, I'm not, I'm one of my, Jesus, I'm 37. I didn't think about it for a

second. I'm, I'm, I'm not that young, but I'm not that old either. And to have what would be considered outrageous sci-fi fantasy a decade ago is now coming into serious conversation. Are you seeing people having these conversations about longevity and actually making it that far, about using AI to make that kind of progress. Have you seen any any advancement in it? I think there is a lot of research now in that field. And a lot of it is using AI to generate new molecules, new drugs. We're still not relying on AI. It's still mostly human scientists driving this. But it's definitely more accepted and lots of people are taking it seriously. There's no reason not to see aging as a disease.

Very interesting. Very cool. If I pivot back for just a second here too as well. With the potential existential crisis, how is it that you are not completely nihilistic? How do you not fall into despair? How is this not completely destroying? It's no different than what we had before. Everyone was dying.

Your kids were dying. Your friends were dying. We're just maybe adjusting how much you have left. We're all dying, but we're not necessarily aware. And so if I take an example of if I were to tell you, you only have three years left, I feel like that would drastically change your behavior and even your mental state. Look at people who are 90 today. They're essentially living in that world.

They're not depressed. They're not changing their life that much. They're doing exactly what they want to be doing as before. I feel like that's different because there's also the perhaps the gratitude for making it that far. Again, I feel the expectations going into the kind of what's key, meaning that did you feel that you had something that was lost? Like, if you're 90, you probably didn't necessarily expect to make it that long and are well aware of kind of the terms and conditions that were agreed upon at your birth. But if you're 20 and you only have a little bit of time left, I feel like there's the resentment and the anger from being cheated and therefore feeling like you had something that was taken from you. Again, a lot of it depends on if it's just you or everyone else.

If you're the only one with a terminal disease, it really sucks. But if everyone has the same amount of years, it kind of equalizes it a little. So would it be fair to say that with the knowledge and the concerns that you have, it hasn't impacted your day-to-day life? It hasn't changed your behavior necessarily beyond talking about it? In general, I'm trying not to do things I don't like doing and do more of the things I like. I think it's a good strategy, no matter how many years you get left. Interesting. Okay, okay.

Is that kind of the advice that you put forth for other people as well? too, or that's an interesting angle we could take as well, is with the advancement of AI, what would you say to people is the important place to put their time in attention? Depends on what you're trying to optimize for. Are we talking economic wealth? Are we talking meaning, spiritual? It really depends on what you interested in. Let's do both. Actually, I think the meaning of the spiritual, particularly in this context, is really powerful. But the economic wealth, If you were to direct somebody for economic growth was the outcome, the goal. They wanted to just optimize for profit and wealth and these things as well.

Where do they go? What do they focus on? Well, it's likely that lots of companies will do well with free labor. So investing in a broader stock market may still work well for you, but also concentrating on hardware for training EIs. AI companies themselves can generate tremendous return. We see it with Anthropic, with open AI, soon to be going. public at 100x over the last few years. So that's an obvious one.

I assume you heard of Bitcoin. That could be an interesting scarce resource and likely to be generated beyond what is known to be available. Anything rare, anything you cannot create with free labor should be doing pretty well. It helps to be someone famous with a brand so you can gain attention. attention economy is likely to do well as well. No, fair enough. And yeah, if you live for a millennia,

I feel Bitcoin is a wonderful place to be, and I think that's probably a good decision. Kind of tangential to that, too. Thinking about the things that are going to go away, I'm not sure if you're still, are you still working in a university environment? I am. Okay.

I previously worked in university research before doing this as well. What is your thoughts on the future for universities as an institution with AI. I wonder if they necessarily don't, at least maybe not the same scale, don't have a place anymore. So for the longest time,

they were about credentials, right? We just certified that someone is willing to sit still for a long time and willing to commit resources and kind of follows instructions fairly well and we give them certificate and then employers looked, oh, you are certified employee, let's hire you. that function can persist as long as people look for certified humans.

Which would be how long? I don't see any reason to hire a human for real pay. If I can get AI to do it basically for free, there is a few occupations where I guess we would prefer a human doing it, but I think eventually the number of such occupations would be much lower. There is a lot of BS jobs to begin with, so you can't really automate them because we don't do anything in the first place. So that may do well.

No, fair enough. I kind of share some of the event. I think that I'd love to know your thoughts. I think that entrepreneurship is going to be one of the better venues kind of going forward in the sense that you have to look for something that's not being served right now and kind of plug yourself in there and then use the AI to accomplish it. That the amount of people that are employed by someone else versus just kind of the self-directed work, I think there's going to see a huge kind of rebalancing of those scales.

Yeah, you can use AI tools to basically have a free accountant, free lawyer, free web designer, so anyone can start a really good company if you have an idea. Is there anything in the AI tools that excites you that you're not worried about? Like, you think this is really, this is something that is really going to push progress forward. This is a place that I'd like to explore more. Is there anything on the narrow scope that you'd like to see more research, more progress on? Basically, all of it. I love AI. I use it all the time for research, for writing, for all sorts of consultations.

So, yeah, the current tools are very helpful. Again, just don't scale it beyond human capacity. Easier said than done, I get this overwhelming sense that it's going to be done by accident. It's not going to be necessarily intentional. It may happen. We don't know how difficult it actually is. Maybe somebody finds a very cool optimization and is able to train a system like that on a laptop. But at least right now, it seems like it would take a...

trillion dollars of compute and power and data. It's so much getting poured into it. Going back to the spiritual side of things, the kind of the man's search for meaning, the issues we have with the economy being kind of completely reformatted with a lot of the bullshit jobs going away. How are you thinking about it from the AI implications from the spiritual sense? So we see different things happen at the same time. Some people started to kind of worship AI.

Colts popping up where AI is the new god. What, really? Yeah, I mean, they're small. They just a startup in religious space at this point, but there's quite a few. There is some concern about human dignity, especially from established religious leaders. They are worried about equating robots or AI with humans, giving them a personhood, treating them as conscious beings. Same as equating humans with animals has always been kind of concerned about, concerned

for human dignity and unique place of our creation in a God filter from that point of view. So we'll see a lot more of that, how special are humans, how Godlike is AI. Very interesting. And do you think that there's going to be kind of a, I would almost anticipate a return to religion in some aspect that there really is going to be a crisis of meaning. I feel like there already is one, but I feel like this further accelerates it, particularly if you have a young generation who cannot figure out where to plug themselves in in a meaningful way, that they're able to actually get that feeling of contributing or building or helping.

It's a confusing world. We do see some people turn to religion, maybe not typical Abrahamic religions, but more kind of Eastern philosophies, meditation-based, mushroom-based beliefs, you know. Do you enjoy any mushroom-based beliefs or any spirituality? Has this moved you in a spiritual direction at all as you've come to understand the looming threat? I want to learn more. I'm doing a lot of research right now. I'm not a practitioner, but I'm doing more theoretical work in that space. Interesting. It's funny. I find that there's so much overlap with that also in the Bitcoin space. There's something magical about mushrooms, and it's something that I think we don't quite necessarily understand. It's still fun to explore. Plus, the therapeutic benefits, at least from some of the research that I've seen, is incredible of what that stuff can offer. Being a safety and security concerned person, I want to make sure it doesn't fry my brains before I start doing things.

So I investigate other people first. I was going to say, from a safety security and your whole worldview centers around basically intelligence. I imagine we're very careful with what we're necessarily putting in the brain or what we're doing to ourselves if that's the prime focus there. Agreed. Beautiful. I'm curious, Roman then. tangently, the robot implication, they actually have humanoid androids, these things as well. Is it moving at a pace that is similar in scope to what's going on with AI? Does it face more limitations because there's a lot more in terms of resources, building maintenance going into it?

How are you building that? Because I feel like, again, if there's going to be an AI threat, I feel like it's going to need something in the physical world to interact with that actually threaten us. A few things to unpack here. So, yeah, we get robots. We get them a couple of years later. what was holding us back was actually intelligence, not having controllers for bodies. We could make mechanical bodies, but now that we can put an AI model in it, it makes it a lot easier to have them manipulate objects in the real world, move properly. But you're making a point

about having to have physical presence in a world to manipulate it. And it's not the case. If you have intelligence and access to internet, that's all you need. You can always pay someone in in Bitcoin to do what you need. You can bribe them. You can blackmail them. You can social engineer humans on the internet to do what you want. That's an interesting angle. So I've often talked about the relationship with Bitcoin between the physical and the digital world,

how that bridge is really important. It's where a lot of the value and scarcity comes from, right? The proof of work mining is necessary for the ties to reality and adding the friction. I never thought of it is that the perfectly engineered money could also be the link for something inside the digital world wanting to interact outwards. I never consider that it might go in a nefarious direction, but you're right. Social engineering, I think, tends to be the, the lowest fruit in terms of breaking into systems, into causing problems, and even from people that have token to in the in the hacking space, that's generally what they focus on, is people getting into

secure areas that they shouldn't do by tricking other humans. Absolutely. And there are now websites, rentahuman.com or something like that, or AIA. agents can rent a human to help them with stuff they need in the real world. No way. Is there, okay, what sort of things are the AI agents currently asking humans to do? That is, that is quite the flip. I have not gone to that website. I have enough jobs and responsibilities.

I'm not working for AI's. So you have to go and check what, I assume, solve some captures or something. I don't know. Coin Kite has been in the game for years, creating hands down, the best and most secure hardware when it comes to securing your Bitcoin. The cold card Q is an absolute powerhouse and my daily driver, and it's ideal for newcomers and advanced users alike. The tab signer gives you a low-cost, user-friendly option for those just getting started

or for convenience when traveling. You can head to coin kite.com and use code BTC sessions for discounts, or simply scan the QR code on the screen to get started right away. One company I like pointing people to when they ask about Bitcoin mining is abundant mines. They were founded by Beau and Christine Turner after losing over half a million dollars to broken promises in the mining industry. And they built their hosting model to remove the usual headaches. With abundant mines, you own your machines and keep 100% of the Bitcoin you mine. There's no revenue share, no hidden skims.

Pricing is simple. One flat monthly fee covers power, parts, labor, and repairs. They also guarantee uptime. The machine goes down, their hash rate redirect system routes hash power from their fleet so earnings don't just stop. And every machine is insured at full replacement value.

Everything is hosted in the U.S., powered by hydro, and mining equipment may qualify for 100% Year 1 bonus depreciation. Learn more at abundantmines.com slash sessions. It's so unbelievably overwhelming. I feel it. I see you unbelievably calm and collected and very, very exciting, but I find it quite dark. And almost the thing that I'm perhaps more concerned about in this conversation is the human aspect of it in terms of would we be smart enough to stop before it's too late.

And even before we get there, I worry about the just not even super general AI super intelligence or general intelligence, I worry about these systems, even as they are, maybe like a couple fold more improvements and the damage they can do in malevolent hands. Do you worry about the AI just kind of as is with malicious actors? Yeah, there are many degrees of concern. You can start with just powerful tools for, let's say, hacking, as we saw recently, in the hands of malevolent actors. Absolutely.

we can destroy financial markets, privacy. So there is quite a few concerns here. But those are the kind of same concerns we had before. We always had malevolent humans. We understood their intentions. We could put them in prison, kill them. We had options. Whereas once we switch to superintelligence,

as an adversary that limits our ability to deal with it. Do you have any insight on even just with the tools as they are right now, how people can defend against a, AI from a hacking perspective? Is there anything that you've seen as a actually effective stop? It seems that only an AI can defeat AI. So if you run an AI tool to look for exploits in your system, you can then patch them once it finds it hopefully before a malevolent user does it.

Okay. We're going to the world of AI versus AI, robot versus robot. It's quite interesting. It's quite astounding then. For anyone that's listening, if they wanted to try and if they agreed with the sentiment, they agreed with the mission that we need to try and push, that we don't actually go to the point of creating this, that this shouldn't be something that we are pursuing. How should they act?

How would they get involved? What could they possibly do? So most people don't have much power in this, just like they don't have any power and many other issues. If you have a politician in your district, in your country, who's well aware of those issues and running in some sort of limiting platforms, I think lately Bernie Sanders has been speaking out against superintelligence. You can try and get more people elected who understand the problem. If you work in one of those labs or part of the government, maybe try to not building

superintelligence or putting regulation in place, which prevents that. I feel like there's been quite a few whistleblowers on the subject as well, too, or at least people moving from the AI space more to the safety space. I'm also now reminded of the, there was a whistleblower at OpenAI, who I believe mysteriously died, which is also kind of concerning at one point in time there. Are you finding more and more people are coming to this side, coming to this conclusion? So I don't know if we need whistleblowers. They publicly publish all the problems.

There is red-teaming reports with every model saying it's lying, cheating, and trying to escape. What is whistleblower going to tell me? Yeah, okay. Fair enough. It's already pretty bad. It's pretty bad as it is. Are you finding that, is this message landing or resonating with certain groups more

than others. Like, if we're speaking to different audiences around the world or different professions, is there one group that seems to be, I would say, one group that's teeing into it more or people that are rejecting it more frequently? It's more about how much access you have to existing AI models. So if I speak to an audience in a country with more restricted access, they don't have latest capabilities to them at small like science fiction. What do you mean AI can program computers? That's insane. I tried it. It's garbage. Whereas if you're work with someone who has access to the cutting edge models. We're like, oh, yeah, 100% of my code is generated.

Oh, yeah, I'm doing it all the time. It's incredible. I didn't finish, but I started agreeing computer science. It wasn't a big fan of coding. Just wasn't necessarily for me. I was relatively bored by it. But the tools now are stupid powerful, like ridiculously so that with the simplest of prompts, I'm basically getting what would take a, probably a team of engineers a week

to throw just 10 years ago. It's incredible. For me, that intuitively kind of makes. makes sense because programming is just another language, right? So it seems intuitive that a large language model, a large language model would be able to deal with just a different language quite convincingly, quite capably. Yeah, definitely. So again, I don't think to your previous question, it's like a specific demographic or cultural religion. It's basically, do you have access to the latest capabilities? And then can you propagate forward what happened in the last three years? If it goes,

in the same direction improves as much we're going to cross the human intelligence barrier essentially. Yeah, fair enough. Do we, is there any reason to think, though, that like, I'm trying to find a way to phrase this. We have the ability to throw up some walls and gates within the digital space.

I'm thinking of things like encryption. Encryption works, right? Is it not possible to limit or contain or control AI with such barriers that they can't even break through as well? Or do we have reason to think that encryption will be broken? It helps to secure as much of software and access as you can, but we know that there are always ways around it. Again, we're talking about social engineering attacks. We're talking about bribery, blackmail.

There's lots of options where you don't need to actually break encryption, which may be possible, maybe not in a sense of actual algorithm being broken, but implementation of that algorithm. So if the systems can find zero-day exploits and you implementation of integer factorization and things like that, it may not be what needs to be broken. It's just software exploitation problem. Yeah, which, again, that idea of AI versus AI on the software front just seems, I have a very striking sci-fi visual of robots fighting robots. I think it's at least it's a fun idea, if nothing else there too.

I want to see Olympics, sports like that, like European football with like, like, Everyone plays at messy level, you know. That would be fun to watch. That would be absolutely incredible. I'm reminded of like the robot battles from like the early 2000. There's vacuum cleaners fighting each other. Yeah, that was crazy. We're going to bring back the gladiator.

Is it, is it not possible that like, I'm wondering, is there any ideas or hypothesis that have been thrown out to you that would justify why humans are actually a value to the AI? Like, I'm not necessarily disagreeing, but I feel like it's, you're right to say that they're indifferent. but I feel like there probably is some value in a self-reproducing robot that just eats the stuff you don't care about on the ground. So there is a few people talk about only humans are conscious, so only we would have those experiences and maybe they need us to do something in a universe, collapse, the quantum functions, things of that nature. There is some suggestions that maybe they'll keep a few of us around as a backup. We generated superintelligence once, maybe we'd be able to do it again. But none of it is very convincing to where 8 billion people would be preserved in a very happy state. No, fair enough.

I wonder what the – even if we don't have superintelligence, I don't think necessarily the human population is going to continue even at this rate. We're already seeing it come down across the Western world. And I feel like as there's less and less threats to human life, the need to reproduce just comes down. So if we get more efficiencies, more abundance, we have more wealth, everyone's living longer. I feel like our natural equilibrium of whatever that kind of population state is probably fluctuates down. But I'm not even putting that out there necessarily is a bad thing. Like, I have three kids.

I love kids. I'm very pro-natalist. Like, just go for it, man. There's the best thing that's ever happened. But people are going to make different decisions. And I feel that the need to reproduce in higher numbers came from the risk of death and not making it forward. The difficulty is kind of going to go.

It's more of a game of chance. Like, you didn't know if your kids were going to survive. so I need to have seven, so at least four can get through, and maybe one of them can have their own kids. But as that pressure is no longer necessarily there, that we would probably contract in terms of population, which wouldn't necessarily be a bad thing. Yeah, people are worried about immortality causing overpopulation, and you're absolutely right. People who live forever don't really care to have kids by 20, so that's a different issue to deal with.

It completely changes, and it's the same thing with like just scarcity and resources available. If we're on something like a Bitcoin, if we have a fixed money supply, we're getting proper price signals. I feel that population would naturally ebb and flow with just the availability of resources. If it's coming too expensive of kids, people have less kids. As they have less kids, more die off. It becomes a little bit cheaper because there's less demand for things. People have more kids. I don't see why population should necessarily be a fixed or constantly growing rate.

Depends on what we're exploring. If we get access to other planets, we can definitely see a lot more population growth without any problems, populate Mars, moon, whatever Elon decides to do. Fair enough. Great. I'm at the behest of Elon. We'll see what happens there. I'm curious your thoughts on AI in terms of coordination in terms of Bitcoin. If we have any insight or information regarding agents talking with other agents, and then we've already seen that apparently they'll pay humans to do things.

Do they pay each other? Would they or do they use that sort of thing to coordinate? We are not at the point where they are smart enough to establish a separate economy. If they do decide to exchange value, it probably would be some sort of a smart contract. So I don't know if Ethereum would be an easier tool to use or something else. But it's likely that digital money would make sense for digital agents. 100%. Roman, this has been a fascinating conversation.

Is there anything else that you feel that we should really leave people with or things that are important for them to consider going forward as we kind of blindly stumble into the future of AI? So I think most of those predictions are not at this point easy to verify. It's basically unpredictable to know what's going to happen as we have more and more intelligent agents around us, making decisions beyond our capacity to comprehend. So have a diverse portfolio of investments, options for yourself. It's just a good idea to not concentrate on one vision of the future. And I assume that everyone should still be using these narrow focus tools.

We're not necessarily kiboshing the whole thing. We're just warning of the general intelligence. All right. The tools are amazing. I mean, I'd love to see diseases cured and help with scientific research. That's not a problem. The problem is, again, replacing humanity with agents which don't care about us. Do you have a favorite model, by the way?

Is there one that you use more particularly than others? They're different in terms of what they do better. So you might need one for programming. You need one for avoiding kind of censorship situation. So it really depends on your domain of application. Beautiful. And then last thing I want to ask you for my own selfish reasons. Timeline.

How much time do you think we have? And I have a reason for asking. I don't make those predictions. I follow prediction markets. And I listen to what the leaders of the labs are telling us. So it's probably two, three years or less until we have human level. and then it's up to AI to decide how much we have left. Interesting.

If we have human level in two, three years, do you think we have a cataclysmic event in, say, 10? It's something we cannot rule out. Okay. I will bet you a lovely steak dinner that we will not have a cataclysmic event in a decade because if I'm wrong, but if I'm right, I get a free meal.

that's not how betting on the end of the world works. You buy me dinner today, and if the world didn't end in 10 years, I'll buy you two dinners. Done. Okay, the next time we're in the same city, I will buy you dinner today, and you can buy me two dinners in 10 years. Roman, this was absolutely fantastic. Love the conversation, as dark and as complex as it is. Where can people go to follow more of your work, get your books, all that fun stuff? You can follow me on Facebook, follow me on Twitter, just don't follow me home.

It's very important. You know the world is heading down a dark path. Online censorship, central bank digital currencies, a complete loss of privacy, and it's only accelerating. Most people are falling behind with the latest technology, and those that aren't are sacrificing their data, their control, and their dignity just to keep up. That doesn't have to be you. Here, you'll have everything you need to break free. Over the past decade, I've helped 20 million people take control of their money, their digital, their digital. data and their digital lives.

Now I'm bringing everything I've learned to one place. Sovereign Sessions is your home to learn, master, and live freedom tech, Bitcoin, AI, private communications, self-hosted tools, everything. Each week, I'll walk you through step by step with the most critical tools that put you back in charge. Don't ask permission, take control. Welcome to the Sovereign Sessions.