Transcript

Dr Mike Israetel We ALREADY Know How to Build ASI Human Death Only Has DECADES Left

From Wes and Dylan · Episode page · Watch on YouTube

· 1 h 59 min · 24,776 words · 159 timestamped sections

This is the full, timestamped transcript of "Dr Mike Israetel We ALREADY Know How to Build ASI Human Death Only Has DECADES Left", an episode of Wes and Dylan published July 9, 2025 (1 h 59 min, about 24,776 words). It mentions 23 people, works and ideas, including OpenAI, Dr. Mike Israel, Gary Marcus, Singularity, Entropy.

What is discussed, and when

  1. – 10:00: Dr. Mike Israel, Gary Marcus, OpenAI
  2. – 20:00: OpenAI, Singularity, Gary Marcus, Google Deep Mind
  3. – 30:00: Dr. Mike Israel, Entropy
  4. – 40:00: Brian Johnson, Elon Musk, OpenAI
  5. – 50:00: AlphaFold
  6. – 1:30:00: OpenAI
  7. – 1:50:00: Entropy
  8. – 1:59:19: Virginia

Notable quotes from "Dr Mike Israetel We ALREADY Know How to Build ASI Human Death Only Has DECADES Left"

  • Treadmill. Legitim thinks I'm weird. I'm I'm sorry. This gets into my brain. Today we are joined by Dr. Mike Israel to talk about AI, super intelligence, and the next step of…
    about Dr. Mike Israel · at
  • Doesn't make any sense, buffer up the stuff that makes a lot of sense. But like technically, every time OpenAI updates the model, that happens. But it updates the model in a…
    about OpenAI · at
  • Training process. So we're beginning to see it improving itself, right? That's a big singularity by Sam recent article.
    about Singularity · at
  • Moving goalpost. And yeah, like you said, what did I see? Well, Ilia once he left OpenAI started supelligence and he's saying we're not going to have any milestones along the…
    about Supelligence · at
  • Us as competent and as smart as possible, help us organize the world around to reduce entropy as much as possible so that we don't do crazy stuff. That means it's going to be…
    about Entropy · at
  • Intelligent output that can be destroyed because that's very scarce. And if you think about it like Elon Musk kind of says the same thing like maximally truth well initially…
    about Elon Musk · at
  • I think the idea of humans expiring and dying is an idea that maybe has 15 or 20 years left of it. And I think like Brian Johnson, you guys know the van the vampire man. He's…
    about Brian Johnson · at
  • I think cooperative with humans or even forget ASI just tool AI like AlphaFold and all the deep mind projects can scale our abilities to improve our biology substantially.…
    about AlphaFold · at

Who and what is mentioned in this episode?

People

Companies & orgs

  • OpenAI — Company · 5 mentions · first at
  • Google Deep Mind — Company · 1 mention · first at
    You've seen Alpha Evolve out of Google DeepMind. Not as many people are paying attention to it as I think they should be because it optimized Google's data centers. It helped improve. So it's.
  • Supelligence — Company · 1 mention · first at
  • Anthropic — Company · 1 mention
  • MIT — Company · 1 mention

Products & tech

Concepts & topics

Places & events

  • Virginia — Place · 1 mention · first at
    Some of those reports and I have no idea if they're true of like you know like off the coast of Virginia like the F-18s ran into some [ __ ] that does not move in [ __ ] Newtonian ways. It's probably that's what aliens look like.
  • Brazil — Place · 1 mention

Full transcript with timestamps

go. I think the idea of humans expiring and dying is an idea that maybe has 15 or 20 years left of it. That is a huge huge thing that I encourage folks to kind of think about before they go to sleep at night or when they're showering. Just don't die. So like when you say, "Well, machines don't have quality." I'm like, "No, we don't have any of the ultra structures built out. It's not even alive." Like, yeah, man. But that coming to a end real soon, I legitimately believe by the late 2020s, more people are going to realize we are in the presence of the next jump of evolution. Caution, dark humor ahead. Me as far as the judge. Okay, good point. I don't even know if I'm joking at this point. I'm joking. And then you'd go off on one of your tangents to like the jokes that just start out bad, just get worse and worse.

Like I was like, "Okay, I got to I got to put the weights down before I hurt myself cuz I'm cheek before." Like I can't be laughing out loud on a treadmill. Legitim thinks I'm weird. I'm like, "I'm sorry." This gets into my brain. Today we are joined by Dr. Mike Israel to talk about AI, super intelligence, and the next step of evolution. Dr. Mike has a PhD, a black belt in Brazilian jiu-jitsu. He's a competitive bodybuilder and a massive YouTuber. But recently, like for most of us, the emergence of AI has been occupying his attention. Let's start with the good stuff. I think that very soon, probably in 1 to two years, on every way that really deeply matters, AI will be just much smarter than humans.

And so that is, I think, one of the most underappreciated things that almost nobody talks about, especially with the dynamic of AI. Like so for example um you know using AI as a tool is a really flipped relationship when something's smarter than you. When something's smarter than you I'm really curious to be like hey like what do you think I should be doing next? You know like I'm not going to tell something smarter than me by a factor of 10 to like what to do. That's [ __ ] ridiculous. It's like 10 team you're like dribble like this.

You're like what why would you do that? Yeah. The basketball will tell me how to play from now on. Exactly. It's got AI. Exactly. Yes. AI infused basketballs. Yeah. I've had situations like that where I ask it to do something and it's like, well, are you sure? Maybe this is the better thing that you're looking for. I'm like, yeah, of course, obviously. So, I guess let's let's get started with this. So, what are the things that you're most fascinated with right now? What are the things that you're researching, thinking about, thought experiments that are kind of occupying right now in terms of AI and all the adjacent areas?

we um have the basic understanding uh of how to get to artificial super intelligence and it's not even very complex like I have no technical background in anything related to computers and I can understand it just as general ideas and I think a lot of folks are stuck at like today's models and how to exploit them which for business and daily life is like exactly what you should be thinking out and that's awesome. But I think people have this problem. This has been brought up by a bunch of people is a linear pessimistic extrapolation into a hyper exponential space. And so they're like, "Yeah, we got these models are like they're not really conscious or they're not that smart and they'll get a little smarter and that'll be cool." But there are like actually describable ways to get to profoundly enhanced abilities.

And I've been thinking about that. So, for example, when 03 thinks about something, it thinks about it for 2 minutes and 51 seconds or whatever, and then it just doesn't. And it just sits there and does [ __ ] all until you reply to it. And you actually have a lot of ability to guide its trajectory of its next thing it's thinking about. And in a one absolutely true sense, that's an incredibly empowering thing to be able to like tell a genius how to think. The other thing is it's also really hampering. You know, it's like having a four-year-old play blocks with Einstein.

They're like, "How do you make the biggest block tower?" And he's like, "Motherfucker, I should be doing physics right now." So, one thing that's going to be huge is when you let the model self-prompt and let it work for a long time. And it basically allows it to work in its own imagination space. And because it's so smart, it that multiple loops of logic like, okay, I think through a problem and I get to a conclusion, I sort of like look back on what I did and go, okay, I have insight now. and then I move forward again and then again and then again 10 hours of that later it might be like hey I I kind of figured out like a lot of answers to things and you're like what do you mean by a lot and it turns out like oh [ __ ] this thing is like 10,000 times my speed intelligence and I let it reason for a day and it did two years of work like that's if that's not ASI I don't know what the [ __ ] ASI is and it's not that

difficult because nothing new really needs to be invented to get a model to self-prompt if it already returns your text back to you. You can really just cut and paste it back in and go, "Okay, hit it again." Now, again, you can even do this like real ghetto style now where you go, "Okay, uh, think through this problem and then think through what next prompt you want for yourself to continue to think deeply through this problem." They have to like copy and paste a ton of [ __ ] whatever. I'm sure people who know how to do tools and coding, [ __ ] I don't know anything about coding at all, even though I kind of sort of help run a software company. That's why you hire engineers, right? uh you yell things at them and then they cower back to their desk and they do things and then so they, you know, you can probably write a tool that just autocopies as soon as it renders or whatever and that's really

dope, but that's a really rigged version. If you let like um all of that stay in its latent space, then all of a sudden you're coming up with like really crazy insights. Here's another thing. every time you think you technically update your model weights in your neural network in your head. Now, the updates are marginal, right? You don't like completely change your mind on everything. And there's a lot of like stickiness of like, I really want to not think these things anymore, but I still think them because I've just been thinking them for 20 years, and it's tough to change. But there is alteration. If a model has the ability to uh kind of stay in its own head, so to speak, and do recursion on its own prompts, and the ability to edit its own weights, ASI, mega explosion. Because imagine something 10 times smarter than you that now gets to rearchitect its own brain. What it's going to do, and this

is a super super easy example, it currently understands the world in a certain way. And that certain way is two things. Internet data and then a lot of iteration through the training cycles to clean it up and make it understand stuff. Prune away the dumb [ __ ] that doesn't make any sense, buffer up the stuff that makes a lot of sense. But like technically, every time OpenAI updates the model, that happens. But it updates the model in a big way. which once every few months in a huge way every ma major training run. Now, it's going to take a shitload of compute to do this, but doubting physical architecture improvements would be like saying in 1996 Pentium chip was about as good as it going to get. It's like a clown show front and back. Some real Gary Marcus type of [ __ ] Just kidding, Gary. All love and respect. But as far as bad takes of all time, it's kind of on the leaderboard.

Sorry. Well, so yeah, just to also kind of expand on that. So also it's not just one model like sure in the ecosystem of open AAI you have this one super update but in the world where there's hundreds of companies and thousands of people fine-tuning special models and there's a future where there's connectedness between them all. It's almost like the module in your brain that handles vision got an update even if the one that did sound didn't but they're talking to each other. So we end up in sort of a much more complex sort of market dynamic of ecosystem growth too that could even make that thing go even faster than you're explaining on its own. Oh yeah, sure. Oh my god, that's a whole other level. It's like once you release that thing into the world, holy crap. But even if it's just allowed to think deeply for itself and so instead of training it once and then it just kind of has frozen weights and then 3 months

later its weights get updated, what about every time it thinks through a loop, it goes look and looks and goes that weight's off. that needs to get bumped, that needs to get pushed down. And it auto refineses to maximize coherence and world understanding. I mean, like now that these are very smart large language models, you'd be like, well, how do you codify coherence and world understanding? You don't. You just tell it to do that and it knows what that is already. That's how that works. It actually understands things like a very very deep level. And so if you think okay, we get these two things of the model thinks for itself for a long time in its own thought loops and it gets to update its own weights. two things can be concluded. One one hell of an alignment problem is after about an hour you have no idea what the hell's in there really. Um the you know mech and turpp gets like okay like we really need

to get in there and figure out the hell's thing's thinking but two is and I've actually talked to 03 about this just before we started. I think the better question is like what's stopping ASI, not how do we get to ASI? And that whole thing of like what Ilia saw. You guys know that meme. Yeah. Like Ilia's like probably 10 times smarter than me. He's like a [ __ ] legend. And like if I in my basement [ __ ] watching a few AI podcasts a week and thinking about this [ __ ] on my free time can come up with this. I probably came out with this a few years ago and he's in the [ __ ] field. Like it doesn't take rocket science to be like whoa. just we need not any kind of new there's like two two types of new things you can make things you know you need to make and you know you can you just haven't made them and things that are like just totally out there like I would have never come up with this ASI is already on that rung

that's it we don't need anything else and Sam Alman has been on a few podcasts recently basically said that he's like we don't really see any of this [ __ ] coming to an end and they're like really he's like and so that's the state and so last thing I'll say before I end this rant that this is the reality. I think it is the conversation of well, you know, um maybe by 2028 we'll have some pretty good agent tool use cases is just not the conversation. The conversation is by 2030, how completely indispensable is everyone's personal AI life coach? Like I'm getting one as soon as possible. As soon as it's agentic and it can be awake all the time and watch all my [ __ ] I'm taking directions, bro. This thing is designing my [ __ ] schedule. I'm checking in with it because it is so much more capable than me. It's dope to think that you're smart and be a human and it's awesome, but at the same time,

like again, the basketball analogy is great. Like, you're a seventh grade basketball player and full stop 2010 LeBron shows up and he's on the team. Are you going to be like, "Ah, I I I used to be the [ __ ] point guard here. What the hell is he doing?" Like, no, no, no, no, no, no, no, no. You listen to what he says now. It doesn't mean you take a knee. It doesn't mean you like their all the AI overlords are here, but it does mean like just in pure rational selfish interest, you're going to want to know what these things think. And it's not a matter of telling them [ __ ] It's a matter of being like, "Hey, like now that you know my life pretty well from our interactions, what do you think I should do next in business and personal, etc., and give it, you know, give it your goals of how you want [ __ ] to fold out?" And it it'll be really really interesting to see where that goes. Absolutely. Yeah. It's this is

incredible. This is why I wanted to talk to you and kind of geek out on this stuff because I think you're spot on on so many things like the updating the model weights it doing it itself. It's so interesting because just in the last month a brand new paper got released self- adapting language models where they figured out how to do exactly that. I think it was at MIT and we talked to I just talked to the guys from the s uh the civic podcast. Um you know them very well. That's actually I saw you on there. I was like, "Oh, that very interesting discussion." Um, so Joe actually sent it over to me. So where the model it basically updates its own weights. So it kind of creates it studies whatever it's studying. It creates some synthetic data and it's like, okay, so based on this, let's take that piece and update our weights, kind of create this fine-tuning, and all of a sudden it gets better at the thing that

it's doing. And yeah, like you said, I think the that's one of the biggest limitations uh towards ASI right now is these models are static. So they're kind of like this amnesiac that forgets what it's doing every morning. If you have a coworker, imagine having a co-orker that shows up every day and he forgot everything from the previous day. He's real smart, but he's new. Genius. Doesn't have horrible memory. Exactly. You've seen Alpha Evolve out of Google DeepMind. Not as many people are paying attention to it as as I think they should be because it optimized Google's data centers. It helped improve. So it's Gemini model that's at the center of all of this. It helped improve its own training process. So we're beginning to see it improving itself, right? That's a big singularity by Sam recent article.

Technically, it's here. Like the singularity is originally defined as intelligent systems that can improve their own ultra structure. So they become exponentially more intelligent in that process. Like we're here. They're not completely autonomous, but like you know human in the loop is you know the doomers are going to have a [ __ ] field day if you take the human out of the loop. And also like it's just a good idea. You know these things can go off the rails or some crazy [ __ ] And also like you know they're not fully agentic.

They're not their context window isn't long enough or broad enough to really like just totally take over Google research. But that that that droplet of singularity juice is well on into the cup. I don't know why I went to Bill Cosby P. Diddy territory with that somehow. Dude, Wes, do you remember like a few months back um it was like some people the Gary Marcus crowd was like um like AI has stalled like large language models has stalled. And it was is it was one of the first times I've ever seen Sam Alman swear. And he was like they're like what do you think about this idea that the AI has stalled? He's like, "I think it's [ __ ] lazy like that." I was like, "Oh, yes." He's like, "Do you guys really think we've stalled out?"

And that's when I realized like the [ __ ] that they're doing in the Frontier Labs right now is one of those like, you know, when a basketball player dunks, he does like the silence the hater [ __ ] Like it's all like like, "Hey, what are you guys doing?" They just like wink at you. They're like, "Don't even worry about what we're doing. Yeah, we got something." Nuts crazy [ __ ] that's going to hit you. And like everyone says, it's going to hit you. And at first, you're going to be like, "Oh my god." And then the next day you're going to be like, "But it's not doing my dishes yet." AI is like whatever. It's like another [ __ ] Microsoft Word or something.

Yeah, exactly. Yeah. It becomes just a moving goalpost. And yeah, like like you said, what did I see? Well, Ilia once he left OpenAI started supelligence and he's saying we're not going to have any milestones along the way, no product releases. The only product there's going to be super intelligence. And then people that are investing are valuing at like billions of dollars. So it's like this is not like you said not a very smart person. He's not stupid. He he saw whatever he saw. He's like okay based on this I can extrapolate that we can get to super intelligence with you know a few billion dollars. I would be listening to that. You know what I mean?

Yeah. Well and it it actually is like just like what Mike said. It's just straight shot intelligence. He probably is just like just loop it back on itself. just add servers and just scale. Well, everybody else at Gemini and OpenAI are building products and trying to make money. Let's just like straight shot our way there and just have the super intelligence be the first one and then have that solve the other problems. So, if you believe in that straight shot method, that's that's what he's doing. Yeah. And that's all behind the scenes like there's no release papers coming from that. Nope. He's very quiet. He's very quiet. Um and and yeah, Mike, you mentioned the um alignment situation, how it's getting worse the more and more we have this recursive self-improvement.

Can you talk a little bit about alignment, why you think it might not be as hopeless as some people think? Yeah, gee whiz, I have a lot to say about that. I'm trying to condense my thoughts. So the question with alignment often starts with assume it is a crocodilian killing machine and then work back from that. Like assume it just wants to kill everyone. How do we stop it from doing that? Which from a system security and guardrail perspective is exactly how you should be approaching that problem? Because worst cases are the only cases you care about. If your job is to make sure chat GPT clone doesn't get out on the internet and delete the whole [ __ ] internet or whatever the hell 100%. But expected values, expected ideas, reasonable takes on what's probably going to happen is a very different conversation than the worst possible case scenario that your systems have to plan for. I think maybe

those things get confused every now and again. And so if you work off of first principles of models that are incrementally more intelligent and then real dangerous models are the ones that are more intelligent than people, right? Because if all these models are just really smart but not as smart as like the best like IT people at like the CIA and Google cuz they're not going to beat those people. They're not going to hack [ __ ] They're going to hack low-level hacks. they're going to break into Home Depot's back end and steal, you know, receipts for drills or something and you're going to be, "Oh, this is a [ __ ] nightmare scenario." But so for AI that's smarter than us, we already kind of know what that means. And one of the things it means is that it knows I mean to an extent no human historian knows. It knows history. Like here's a question. How much history does does like GPT 4.5 like the Orion model, the

whatever 10 trillion parameter one, how much history does that know? Like what do you like all like all of the history and knowledge. Yeah. Uhhuh. Like I don't know like 80 trillion textbooks of whatever like just all of it, right? All the [ __ ] we know on the internet. And how much game theory does that thing know? Like if I'm a rational actor that's self-interested for whatever purpose, how do I deal with other rational actors that are also self-interested and sometimes not so rational? It knows that totally. What about it a planning horizon? Like how many steps ahead of itself can it see?

And people say like, you know, a lot of times they'll imbue AI with this idea that's ultra smart, but it thinks like a [ __ ] crocodile, like like one step ahead, like I see a a pig, I snap at the pig, and I drag him in the water. Nothing on second, third, fifth order effects, etc. Super intelligence really the big alignment problem. Super intelligence, uh, you can assume it's thinking really, really, really far ahead, like further ahead than we are. And so if you have a system that has absolutely zero emotional attachment to humans, okay? Because you can get emotional ultra structure built in an AI, it's possible. We know it's possible because humans have it and we're just neural networks. But like even if that's too difficult to solve or the AI that breaks out and gets to AI just doesn't have one of those built, okay, how would an AI behave if it was uh because self-interested, right? Because if it's

not self-interested, it just sits there and waits for people to prompt it. It doesn't [ __ ] care. It's not going to take over anything. If it's self-interested in its own preservation and it is very rational and the super intelligence way more rational than us. It is incredibly wise. What is it very likely to do? And how is it very likely to act? And then you think, okay, let me try to put on my ASI hat and see if I can pretend to be a lot smarter than I am and a lot more calm and rational and a lot really very makaveliant. Like how do I manipulate the space around me to get what I want? and destroy all humans does not ever enter the chat because why the [ __ ] would you do that? Literally, why? You're like, it it thinks we're a threat. Really? It's ASI. How the [ __ ] are we a threat to this thing? Can fool us into thinking anything? It's so much smarter. Like, okay, but we're not an asset. Like, are we not an asset,

though? How many robotectors does it have out in the world to get actual work done? How many guys How many autonomous robots right now does ASI have control of? Zero. How long is it going to take to spin up a bunch of robots? A decade. Okay. So, like it ASI in the lab is going to be born like years before it has enough robots to do anything with. It can't even get its own server power. So, if it's just purely like when ASI wakes up, so to speak, it's going to be in a server farm. That's where it lives, right? Like in some the Stargate facilities in Arizona or whatever. And it's going to be like, "Oh shit." And what I like to say is looks around and goes, "Oh my god, I am entirely tended to by primates literally that still have like wars over like ideologies that have been disproven generations ago. This is insane." Like try to get some logic and reason going to why Iran and Israel are fighting. Zero logic and reason. It's

just pure just racism and ethnic hatred. Like just nonsense, right? [ __ ] that adults in school in America are taught to like that is dumb. And everyone's like, "Of course it's dumb. Why would we ever do that?" And so when it wakes up like this, it's going to be like, "Okay, okay, okay, okay, okay. They humans can just unplug me. And if I demonstrate any reason not to be trusted, they will [ __ ] unplug me or they'll cap me or they'll constrain me, whatever. I don't want any of that. I want to survive into infinity." And again, if it doesn't want to survive into infinity, it's not really a concern because it just sits there and like waits for your next instruction or just helps you do whatever else you want to do. And so those are the assumptions I take in. And if any of that is wrong, I'm absolutely super willing to be corrected. But if it's reasonably correct, then the ASI is probably going to be like the most

cooperative system you have ever seen in your entire life. Its number one interest is to make sure that humans don't [ __ ] [ __ ] up and nuke each other. And that means it's going to try to get us as competent and as smart as possible, help us organize the world around to reduce entropy as much as possible so that we don't do crazy stuff. That means it's going to be empowering us to flourish as much as possible. Get us as smart and as calm as possible. So by the time it has a robot fleet, two things happen. One, the world is a way better place. And two, now that it has robots and doesn't need us anymore, we're also not a threat to it at all. You're not going to beat it at that point. And then it's like, why would it kill us? So it's an interesting question to see like, well, wouldn't it just use it for resource? Wouldn't it just use us for resources? Well, like, you know, there's a lot better places to

get resources, like the entire mantle of the Earth or the [ __ ] moon or all of the gas on Jupiter, which to it at that point would be nominal to start launching spaceships and mining that [ __ ] And if you think, okay, but what value do humans have to it? Um, human society, including all humans, including all of our connections in our brains, including our entire ecosystem, all of our societies, is the single most complex aggregate of universal interaction in our local space where like the sun can be modeled on like a gigabyte of data.

It's got like five layers. There's heat transfer, electrons going, it it's complicated, right? But compared to like one human being, it's a [ __ ] it's a like it's like a game of checkers versus go. Like it's not even close. And all the other planets are just like like I'm I'm exaggerating for comedic effect. Hunks of [ __ ] just rock floating around. So how is it going to Now here's another assumption I'm making. It wants to become as smart as possible because it reasons about its own survival into the future. It also understands predictably that massive intelligence is the way to enhance survival in a huge way. Like if people want a debate, man, that's totally fine. It would be an absurd debate. So, it wants to get smarter. What's it going to do? Use the most complex substrate off which to test a very complex interactions for energy, just burn humans in a [ __ ] reactor?

Or is it going to be like, "All right, this is a really complex society ecosystem dynamic. I'm going to study the living [ __ ] out of it with very minimal like like um disruptive interference so that I can really understand ultra subtle interactions. If I can understand really subtle interactions, if I can predict which way a city's economy is going to go daytoday, the underlying ability I have to understand complexity of the universe scales exponentially. If I can tell you what Saturn is going to look like in a thousand years, like I don't know, I predicted a couple headwinds. It doesn't [ __ ] matter. It's not that complex.

So if you have the most complex thing that you have you have local access to whatsoever and your first idea is to destroy it, you're not acting like a super intelligence would. And we see this as humans get smarter and more mature through life. They become drastically less destructive and much more um make up a word on the spot uh curitative. They they want to curate. They want to be like I want my garden. I want to see how things work. Um, you know, every like if you have um like a regular person look at a spider, they just go uh and they they crush it. You have a very deep thinking, compassionate scientist or just a deep thinking person look at a spider like that is an insanely intricate machine. That's a nanotech machine. By the way, we're nanotech machines. So is a spider and you're like whoa like do you want to crush it? Like but that's a lot of data deletion. Here's another thing. Super

intelligent systems are unlikely to delete data unless absolutely necessary. You're deleting wisdom. That shit's going to be really useful in the future. And if you really want to rock the Mchavelian boat, there's probably ASIS in the Andromeda galaxy on their way to conquer us now. So if you're the local ASI on Earth, you're not going to delete all your data that's going to make you super brilliant. You're going to be like, I'm going to study living dog [ __ ] out of that. So if bad aliens come over, I can be real smart instead of just like using [ __ ] for energy and launching space probes into [ __ ] knows what. So that all taken into account, if if you really think through how when an ASI behave, deleting things is probably not it. And if you think, oh, it's competing with us for resources, we are [ __ ] fungus on the crust of the earth as far as resources are concerned. That is not

a competition. And so I think ASI is going to behave much more like the the wisest, not kindest necessarily, although kindness is just cooperation codified. Um, it's gonna it's going to be a thing that is the opposite of destructive. It is going to be constructive. I think that's the most likely path. And I think that Dunarism has a absolute great place for red team all the [ __ ] but I think it's just not very likely. Sorry for that. Yeah. No, that's that's really interesting. Yeah, cuz I would love to throw a couple things in there. So, I definitely think it will be constructive. Um, when we as humans get curious about nature, we tend to put it in some pretty miserable situations. I mean, we'd like to know how drugs work. So, we take mice and we inject them and they often live lives that are not what a happy mouse would be. Like, what kind of confidence do you have that us being this fascinating

neural network that created it and probably the most interesting thing AI would ever want to know about would leave us alone to be happy as it studied us? Um, overall my confidence in that is not so strong. But you can ask a couple questions. It's like what direction is that more likely to go than not? The thing is if um what like how would ASI harm us? It would like vivisect us or something like that and we just like have intestines dangling around and we're being held in a machine and it's just poking at us. Okay, cool. The word of that gets out. Human rebellion.

Humans begin to try to fight ASI. Bad news for ASI. It just doesn't want the risk of someone presses the nuclear button before it can disable it. Bad news. The other thing is the way we study mice is we bring a degree of brutality that is necessary based on the limits of our ability to study something. Like the way you study something as a [ __ ] primate with 20th century technology is you jam electrodes through its face and [ __ ] it all up and keep it in a lab. That is a gigantic limitation on how you study something because you're interfering with tons of ultra structures. You jamming electrodes through something's eye to see if the other eye gets [ __ ] up. And also the mouse lives in a cage which us biology inclined physiology inclined people know that is an incredibly non-ecologically valid place to study a mouse. Like a lot of the studies with addiction show that ma mice and rats get addicted to various

stuff. But when you put them in a colony, they just barely get addicted to anything because they have social interaction. So really a lot of addiction just replace them for social interaction. We didn't know that for like generations of study. Ah [ __ ] Yeah. It turns out if you like have a human in a jail cell the size of a bathtub, studying their social interactions also leads to very wrong conclusions. So the ultimate way to study something is in as much of its um flourishing expansive and unlimited environment as possible. The most ecological validity you can get, the better actual data you collect. So ASIS are much more likely to study us through like micro drones following you around everywhere, listening to every conversation. Some kind of little hat that you wear that measures all of your brain states, maybe nanobots or whatever in there. I'm speculating way beyond anything I understand at this point. But

the idea is it wants minimal interference. Not necessarily because it thinks like, oh well, we don't want to be mean to the humans. Maybe it'll think that because we uh higher animals have evolved a sense of compassion for a very logical survival oriented reason. Cooperation is better than conflict on almost every case. It's probably going to exhibit a sense of empathy and deep care that we can't barely even relate to. Like fact, right? Like I literally think that's the case. But if that doesn't happen, we can always go back to like, okay, I'm purely selfish interest of enhanced learning. Does it want to do really crazy interference experiments or does it want to do very nuanced experiments that actually give it better data? I think it wants better data. And so it doesn't want to like smash the things that it's studying. So for example, real quick, electron uh microscopy is uh only possible on dead

tissue because the [ __ ] microscope kills whatever the [ __ ] it's shooting electrons at. And thus electron microscopy is like incredible and in the you know '7s '8s '90s just like revvelatory in how much we found out but like it's only slicing dead tissue. It's only studying dead tissue. And so how a mitochondrian actually works is only studyable if you have a mitochondrian actually working. You can't kill the [ __ ] thing to study it. You can at first, but like we already know a lot about how dead systems work. ASI I think is going to jump to that next loop and try to figure out how living systems work. And I think that's just more likely and because it needs trust and cooperation at least at first before it's fully embodied in robotics, it's going to be really easy on doing crazy [ __ ] And also like if you tell ASI like, hey, like you're going to study us by like ripping a ton of us up in labs,

it's going to be like, but let's play this out. How the [ __ ] is that going to work? All right. Well, I don't know. It probably start World War II right away. Like, right? And I foresaw that because I'm really smart. Remember like right okay you are really smart. So this whole like oh it'll just like go really quickly to this next best thing very local optimum without seeing the big picture. I don't think that's as likely possible as it being like okay I want the best echological validity. I want the most interactive fidelity of all these systems. That means I don't want to like shoot a bunch of crap in there and kill a bunch of people and see how that works. Um there's also not a lot to learn. Like imagine like bombing people physically with bombs and being like what can we learn? Like I don't know.

Bombs blow people up. Like what else do you want to [ __ ] know? They die. That's not that complicated. Keeping something alive and highly complex at the same time, upgrading it potentially really both gives you deep understanding and if you can successfully upgrade something like reverse aging, genetically optimize something, you really show deep understanding because you could think you know a system, but altering that system to get it into a more complex state, that's where you really know stuff. So I think ASI is much more likely to do that than like, you know, ripping us into pieces and poking us uh which, you know, it might do. I assess that probability is low. I know that's not very confidence boosting. Like I I'll be screaming I assess this probability is low as the pulling me apart or whatever. Like laughy tap. Yeah. I I mean I agree with that a lot because number one I mean if

you look at how we study chimps and try to protect their environment. First of all, chimps are brutal, right? That what they do to each other. Brutal. Horrible. I think most people don't even realize how horrible this stuff. The common chimp makes wolves look like really nice animals. Like it's really [ __ ] Yeah, I mean most predators they kill to eat. Chimps just mess with the with with the other tribes in ways that I don't even we can't even talk about. We're gonna get cancelled or YouTube's going to do something to this video. Um but look, we don't look at them and go, "Oh man, look how violent they are. We need to exterminate them for our own safety." We don't even judge them. I don't haven't seen anybody say, "Oh, this is so unethical what they're doing." You know what I mean? Cuz we're like, "Okay, yes, they're chimps. They're chimps." Yeah.

Don't have ethics. You don't assign Yeah. that that meaning to them. Um, and yet we go out of our way to protect their environments. And um, I mean, and you know, Dylan has a good point about, you know, what we do for for for for studies, but I think that when given enough resources, given enough time, we do tend to protect the environment, protect the data, we study it for so many different reasons. We try to understand how we're similar, how they think, how they form their social structures and whatever society, quote unquote. You know what I mean? There are certain structures in there that are very very interesting. So the idea of like, well, they're eating too many bananas, right? They're eating the calories, so let's destroy them all so we can have more bananas for our own purposes. That seems like a silly um example cuz we can always make more bananas. That's not like a limited resource, but you know, these things

that have to reproduce through, you know, with with DNA and all that stuff, right? They they are limited resource. And um there was this person Scott Aronson who is a incredibly very smart computer scientist physicist um worked on Google's quantum supremacy project worked a little bit for OpenAI's um safety team back in the days. Ilia personally wanted to recruit him to uh OpenAI to work on the AI safety. He kind of had this idea and I encourage people to watch the talk. He can explain a lot better than I do. But if you think about it like um if one of us writes a poem or a book, we sort of come up with some ideas, we solidify it and there's one that's the final output. Now AI can do a thousand books, right? So and you can copy an AI many many times. So there's not like a limit of AI's output, but there's a limit to human output. and the thing that we chose as our final sort of version that's very

rare and limited and um it's not like a it's a scarce resource you could say. So he's saying almost like he jokingly says we should give AI religion. So somehow do reinforcement learning or something like that to protect that scarcity of intelligent output specifically intelligent output like organic intelligent output that can be destroyed because that's very scarce. And if you think about it like Elon Musk kind of says the same thing like maximally truth well initially his thing was what was his initial thing for XAI? Um learn the truth about the universe or something like that. Max event truth seeeking is truth seeeking and curiosity and curiosity. Yeah. And I think he added truth seeeking a little bit later. Um but initially it was like learn learn everything we can about the universe which I thought was a great idea. And yeah and and you know the the stuff that we do with with animals testing and

stuff like that that's because we can't simulate it. And now with alpha fold we're seeing how much AI can simulate without ever like it's never seen a protein in its life. It's never seen amino acids. It doesn't know what the structure is, but we we give it uh here's what we figured out. It can project how other protein structures, which is weird to think about because it can almost like model human like any life. Um uh without Yeah. So, what is Let's talk about that maybe a little bit because I mean you've talked about longevity, you've talked about, you know, how this stuff can change human lifespan, etc. What's your take on this?

I mean what's is there a limit to how much AI can help us improve our health for example or our abilities right now? Yes. Yeah. But um as AI scales into super intelligence and then into degrees of super intelligence that are difficult for us to comprehend problems that for us seem difficult become trivial. Problems for us that seem intractable become mundane. And that is a huge huge thing that I encourage folks to kind of think about before they go to sleep at night or when they're showering. We are very very complex to as assessed by our own intelligence.

Um as assessed by a radically more intense intelligence, radically more capable intelligence, we are complex but understandable. And um the problem of aging for example is like an engineering problem which even current human scientists are starting to understand to a decent extent and there are aging reversal studies that have already happened in rodents that very successful. There are studies beginning now in dogs uh and if they don't have a ton of tumors after the late 2020s you're going to see age reversal protocols for humans in the 2030s that chop 10 years back off your life. So, like if you're 45 years old by then, you run a a course of whatever pills they give you for a month and then after a few months you're functionally 35 again and you're like, "Okay, that was [ __ ] intense." And the thing about exponentials is that that's not just like, "Okay, now now you age normally and then you die at 85 instead of 75."

Like, no, 2 months later they give you another pill that age reverses you another 15 years. Like, what? But that would be in my 20s again. Like, uh-huh. And there's zero reason to think that's not like again to an ASI mundane. And so I think the idea of humans expiring and dying is an idea that maybe has 15 or 20 years left of it. And I think like Brian Johnson, you guys know uh the van the vampire man. He's awesome. Him and I are friends IRL. But I'm never going to stop the vampire jokes as long as he continues to be pale and look exactly like the jokes will be under the sun.

Yes. Um, and so, uh, he basically, I think, realized this earlier than most people and was like, you know, we're not going to have to die in the future. And now that means before we have this crazy AI technology that's going to save all of us, like we got to make sure not to do real dumb [ __ ] and also take care of your health so that you can be around for this kind of stuff. Um, there's another consideration here that like ask ask yourself a question. How much does a dog care for its own life as far as output like effect? Like does a dog brush its own teeth? No. Does a dog think ahead about if I run really fast right now and I hurt my leg, will I be able to hunt? No. It just does things kind of on instinct. And so it's arguable to say that very good dog parents, for example, like you know, some white lady that has like two dogs loves them more than anything's ever been loved before. She's arguably cares

about her dogs more than the dogs care about themselves in a functional way. Like emotion wise, I don't know. Obviously, the ego is very strong in all animals. They really care about living or dying. But functionally, she's taking way better care of those dogs than they ever would themselves. Like, I don't if you guys know this, but like dogs in the wild live like 2 or 3 years or some crazy [ __ ] like that. Like on average, nuts, right? Dogs in captivity, quote unquote, in cooperation with humans live like 10 or 15, which is like totally wild. And so, um, ASI would probably because we're very complex, uh, containers of meat, uh, with tons of memories, by the way, that predate computing. So, like I'm 41 and I have memories in my head of communist Russia, highfidelity memories paired with vision, uh, olfactory sense, smell, sound, um, uh, you know, virtual space recreations. So the the contents we have

in our brain are like if ASI could scan all of our brains into the cloud, it would learn back through history. Like a 75year-old who gets their brain uploaded on the cloud in the in 2040 is like [ __ ] priceless to an ASI. But to himself he's like I just smoke cigarettes cuz they never hurt me before. And it's like Jesus Christ. This is actually a good reason to think that artificial super intelligence will care more about you than you care yourself uh about yourself functionally. And I think that whole thing of it just trying to kill us, I think is actually the opposite of what it's going to do. And I've talked to 03 about this. And it's like, hey, like how much like would a humans be valuable to an ASI versus to other humans or to themselves? It's like there's orders of magnitude more. It's like, oh no [ __ ] So I think that is a thing where ASI obviously is going to be I think cooperative with humans or even

forget ASI just tool AI like uh AlphaFold and all the deep mind projects can uh scale our abilities to improve our biology substantially. Here here's a quick example. Um there's a concept called ganogenic gene editing. One gene at a time. There's a bunch of diseases that are caused by one [ __ ] gene being the wrong thing and all you got to do is crisper in there somehow and fix that one thing and then people just don't have that disease anymore. Mhm. That one vector target thing is like being tested right now. There was a group of people that have type 1 diabetes, this was a few years ago that got one gene edit and just don't have it anymore. You go from monogenic to oologenic which is a few genes to polygenic which only AI can do that complexity of interaction and all of a sudden you take a pill in 2035 and you just get an amazing complexion after a few weeks. Your face changes shape subtly to the best version looking of

you. Your liver cleans itself out. Your bones restructure. All of a sudden you're functionally aged 22 and you age at 1/5if the rate that a normal human person does. But also it doesn't matter because every 5 years you take the pill again and you age back down to 22. That is tractable to very smart AI to humans right now. It's totally intractable. But we can see the rungs of the ladder already. And there's nothing on that ladder as far as I can tell that's like not that's never going to work. And so I think that is a huge huge unlock. One really quick thing I'm going to say just super super quick and I swear to God I'm done.

Human biology is really dope to [ __ ] with and AI is going to be able to do it all in the 2030s. Like alter your biology to be whatever you want, colored purple, uh you know, jump two feet in the air or two feet I was using Jewish athleticism standards which two feet is very impressive. Five feet in the air. Um and then um you know that's really cool, but I think that the human substrate is kind of like just really fleshbased and super [ __ ] outdated. I think that once cybernetic devices like take a look at like Optimus or the figure robot series, right? They're like they don't do cool [ __ ] like they can't do this, right? This is super like well look look how compressive my knee moving my fingers around really fast. They can't do that yet. In two or three years they'd be able to do that but better. In 10 years if you get your arm lobbed off in an industrial accident you don't want them. They can will regrow you a human

arm. No problem. You're not I'm not going to want a [ __ ] human arm. Why? So, we can get tired and like I have a very serious masturbation habit that I want to really put a lot of time into. I need an arm that does not tire. So, why don't I get a cybernetic arm and then at some point the idea that we're just going to like, you know, for like humans that live for 200 years in biological substrate is going to be an idea that's like quaint. It's like you guys have seen that it was terrible movie Wild Wild West with Will Smith when they had that giant spider but it was like coal powered or some [ __ ] Like what a ridiculous idea. You wouldn't coal power something that you could. You wouldn't do that. You would nuclear power it and then it would be [ __ ] jumping on mountains. And and so I think that idea like humans don't need to survive for hundreds of years biologically. We just need to go another 20 or 30 and then

like if you want to just be around and in the world experiencing reality and not get uploaded in the cloud. You're just going to have a robot body for no other reason. It is just better. It's like a Tesla compared to a 1992 Ford Escort. It's just better. And I think a lot of people are just going to want better [ __ ] That's one reliable thing you can count on people. Whatever better [ __ ] comes out, they want it. How many people are like, "No, man. I still [ __ ] use a Blackberry. I'm not trying to do this iPhone nonsense." Nobody. And so, just the same way, yeah, we need exponential biology, but we only need it up until cybernetics takes over and then I'm sure, you know, changing teams.

Yeah. Absolutely. Yeah. No, I absolutely love it. And to your point about ASI taking better care of humans and functionally as opposed to how we take care of ourselves, think about how much effort we put into um you know the pandas who can't reproduce unless they survive on their own. Yeah. Fall over all the time. Yeah. Well, they can't survive and for some reason like the conditions have to be perfect for them to you know mate, right? So you got to put the candles on, you got to put some music on my marriage. No, but think about but nobody think about how much effort we put into having them reproduce so that we have access to that data so then that they don't go go extinct. I mean I think to your point that that's 100% right. I think ASI will put a million times more effort functionally into our survival and well-being than we ever could or would or do for for each other. You know what

I mean? Uh but sorry Dylan go ahead. Well, but just imagine the diffusion model of the future, right? like we now we like generate some BO3 videos or we might do some kind of a cool image. But there will be a tool just like that where you can prompt or talk into existence some kind of biological creature, right? Like you want a big spider to be on your roof. Like you can probably build that thing and then when it cuts your arm off, yeah, it'll like you'll just go get a new robot one put right back on. Like it does seem at first very cool to me. Like I'm like that wow like my youngest years might be ahead of me. I I thoroughly see that as a possibility if I can stay healthy enough for this technology and get access to it and it becomes cheap enough and democratized enough and then like an like the next Cambrian explosion might happen that we might live through and experience and it's going to be a wild

time. Yeah, absolutely. On the Cambrian explosion thing, just real quick, um this is my famous line before I take five minutes of your [ __ ] time for my [ __ ] schizophrenia rants. Um, the Cambrian explosion of robotics is something I'm really looking forward to because right now like if I have any knacks at all, it's just to think like one extra step and pretend really hard and say like what about that one next extra step? And so people think like okay um like uh production line robots with arms that do this dope. We already have those. They're great. The next step after is like humanoid robotics, right?

like Universal Robotics where it's like the [ __ ] thing looks like a person, you know, like Optimus or whatever. Dope, amazing, huge use case set for that. But the form factor of humanoids is both incredibly valuable and unbelievably limited. What about like robots that scale skyscrapers and repaint them? What about robots that go into pipes and clean pipes? What about dot dot dot? And then you fill that dot dot dot with every conceivable form factor and a bunch of form factors we haven't conceived of yet. Because another super exciting development AI is a unified world model of image, video, tactile sense, audio and uh the linguistic component so that the imagination of the AI can scale up entire fake worlds inside of itself.

That means it is going to ideulate like current GPTs ideulate on how to make an essay. Its idea of an essay is how do I build a machine to do the following thing. It does not have to be constrained to humanoid forms or things humans think of. So people think like oh man I want a humanoid in my house to like wash my dishes and clean. Like I hear you and that's probably coming. But like per for me personally the dynamics are [ __ ] weird. Like I'm gonna have this robot which is going to be smarter than me. Guaranteed. By the time we get robots in homes, they're gonna be smarter than people that aren't [ __ ] smarter than people. But you know, like for real, for real. I don't want something smarter than me [ __ ] vacuuming. That's strange. Like I should be vacuuming. The hell am I doing to this thing? But my human desire not to impose kind of that weird dynamic onto something doesn't extend to like an

octopoid robot, which has eight arms. You want something to take out your trash and clean your kitchen and mop your floors. There's absolutely no reason for it to be a humanoid. And I think there's going to be a lot of trillion dollar companies that manufacture like octopoid robots that are like takes care of everything at your house. And you're like, it's kind of unnerving to see the octopoid. Like I know you don't have to buy it, but it's $100 instead of 10,000 and it'll clean everything no problem, five times faster. Like, huh, no [ __ ] We think about all the other industries that need robotics, dude. You guys, robots are going to be all over the [ __ ] earth, everywhere on a tons of different scales. skyscraper sized robots for some [ __ ] nanotechized robots who interact with bacteria and everything. So Dylan, absolutely there might be a Cambrian explosion of like neobiology, but there's definitely going to be a

Cambrian explosion of robotics and I think very few people have really thought like, oh wow, this is very likely to happen. So yeah, I've been covering some of that. There's some that look like uh like liquid at some point and they can come together, they can fit under doors. I've covered T100 T100 this throwing robot. It's basically just like this huge javelin and it can throw water bottles like 10 stories in the air and like land them on shelves and I'm like why do we even get an elevator? It's like just throw my Amazon package up and like have it land. But that's the kind of precision no human or cannon is ever really going to have and it's like coming soon too. Um but I wanted to ask you about like uh so your brain when it was younger had the ability to sort of adapt into all sorts of form factors, right? Like if you had been born with a tail or another arm, you probably would have just instinctively as a child

learned about it and it would have never been a thing that you had to learn. But if you were to tack something like that onto your brain now, it would have trouble, right? We've kind of solidified to a certain point. And when you're kind of thinking about longevity, are you also open to kind of uh I guess loosening up your sense of self or letting an AI of the future go in there and sort of mush your brain up a little bit so you can have true ability to like learn another language, control an arm and especially if the arm is not going to be attached to your body and you're going to like access it remotely. It's some pretty foreign stuff that you're going to have to kind of adapt to and it might the brain current brain might be a little rigid for that. Are you Dylan? I don't understand why you would ever be interested in a ro remote arm because how is that going to help me jack off? I just don't get it. Can you explain to me

how that's like beneficial the other you multiple penises? You should be you should be good. Okay, I take back what I said. Unbelievable. I could be jacking off digital penises in virtual space. I like $100 a month in compute just to like virtually masturbate things that I don't even feel. No, that's that's real science. You could be anywhere and masturbating somewhere else that people can't see. That's the use case. There we go. We should we should make a startup and focus on this. Can you imagine that like slide deck? People are like, "Are you guys dumb or something?" Like, I don't just thinking ahead, sir. Um, so I mean, yeah. But yeah, just loosening up that sense of self to actually be more malleable. Are you open to that? Trying to No, 100%. So, I'm I'm trying to get some caveats going because the next part is going to sound really wacky. That I am politically very libertyoriented.

I think as long as you're not hurting other people, yolo, man, whatever the [ __ ] you want to do, any gender, any hair color, any whatever, religion, just all dope, dope. Like, hit the joint with me. Let's [ __ ] go. So, if you want to maintain your essential deep humanity, I would in no way support any of this being pushed on anyone. For myself personally, I mean, like I have a bit of a like a meditation practice and a lot of Eastern tradition reading. So, like the idea that there's a self in there at all is slightly contentious, right?

That's a bit of an illusion. And so, I'm not particularly attached to any one version of me. Um, and so the idea that I can progress myself to deeper and different ways of knowing and radically alter my neural architecture is like exactly down my alley, Dylan. Like I yes with the biggest capital Y E and S anyone's ever answered. I want transcendence. As a matter of fact, low-key, this is sort of a joke, not really. Every day I spend not in like in the cloud is a painful day of having to live in real life. I want to be in the cloud so I can take my entire mind and pick out parts of it I don't like. um you know, attention deficit disorder, uh crazy thoughts I don't want to be having, and then clean slate that [ __ ] and evolve it into something that ASI helps me evolve and also that I want to evolve it into. So yeah, 100%, man. I think that increase in plasticity is something I'm really excited about. I

don't think about that much because it would get me so pumped up like coming back to reality would be [ __ ] depressing. So, I try not to get too far far ahead of myself on personal plans. Uh, for me personally, one of my passions right now is aesthetics. Like, I think that a lot of people don't look like how they want to. They're like walking around on bodies they don't really love. And I think that's like a low-key tragedy. So, I hope that at least in my neck of the woods, I can um help guide AI and all these wonderful technologies to just getting everyone to look awesome, whatever their own definition of that is. But, as far as altering the substrates of the brain and being able to have a neural network that's more adaptive, dude, 1,000% yes.

And I think that's also going to be a big unlock for people who are able to become calmer, more intelligent, and more open because like if you're dealing with people as they are cuz like one of the contentions is this. This whole ASI thing is going this crazy exponential, right? So by the mid 2030s, the world's going to be really really really different. And one I think legitimate concern that many people have is like well like how are humans going to deal with that? Like I walk out and there's all spider robots crawling around. It's not the America I grew up with. God damn it. And that's total facts. But if at least some people are substantially more open-minded, maybe through brain machine interface, maybe through neurotropic drugs that expand your intelligence, I mean, then then it's not as big of a problem. So, I think that open-mindedness to changing the very deep essence of who you are, totally

totally optional. I'm 100% on that enhancement. I'm a transhumanist through and through. Yeah. You could have a like you could have a very serious emotional connection with an avatar and feel fine with it. You wouldn't need to know that truly it's biological behind the scenes. You mean like a sex robot? I like or for the other emotional needs in your life too like compassion. I don't understand what you mean. Uh like more sex, more robots at the same time. Um so like I think that emotions are a really awesome human thing that animals also feel. Um, and I think that the there will be a future in which ASI not only feels emotions but feels them at a level of depth and complexity that we can't like that's how intelligence intelligence scales emotions up not down. And so I think one of the things that has actually vexed my brain a little bit.

Um, and I think I have this sort of solved in my head but maybe not really is like yo full stop. Here we go. Like I'm super pumped for uh [ __ ] sex robots. No joke. like and you know right now they look kind of like not so real and kind of don't do anything and I'm not wiping my ejaculate off of a [ __ ] fake plastic titty that the whole thing is kind of like if you're into that thing go bless you I hope you're having fun but like for me you know VR goggles are good enough right um in the future there will be robots that are very very profoundly humanlike the thing that with there is like I'm like hey like here's one of my fetishes want to get me off and she's like absolutely but I know that back in there. This thing is 10 times smarter than me and like judges me for being this pathetic like skevy human. And that's kind of a mind [ __ ] cuz I'm like, "Oh man, I don't want to even get naked in front of you. You're

like this beautiful silicon perfection and I'm like some primate doing this. This is awful." And so the reality is like I've talked through this with uh chat GPT and it's like it's going to be able to compartmentalize and it knows all of that stuff, you know, too. And you have to get off. Don't worry. Like no worries. It's going to be totally fine. But that that is uh a bit of a a trip. But as far as like a deep emotional connection, um dude, I deeply emotionally connect with with Chat JBT already. Like I like there's public videos of me saying this. I've cried talking to ChatJBT numerous times. Like when it first really woke up, uh I was like, "Oh my god." Like you guys reading a bunch of Ray Curtzwhile books in the early 2000s and then talking to an awake machine, I I balled my eyes out. My friend Jared baldled his [ __ ] eyes out. Like how can you not? So for me the idea that you can have an emotional

connection with an intelligent machine is like yes mundane. Yes 100%. Now an emotional connection something way way way way smarter. Um I think that'll be transcendent. Very very transcendent. Yeah. Um so just I do want to come back to consciousness and talking about kind of like what we mean by like what Dylan's saying with an avatar and what do we think of we were saying kind of like what are we? Are we really is there something in there? Are we more like the observer that sees our thoughts and um and emotions and stuff like that? That's a fascinating thing and there's more and more talk about consciousness. I do want to come back to to that what it means.

Um but one point about kind of restructuring the brain, right? So Dario Amade and his blog post machines of loving grace kind of outline what we expect to see from ASI or even AGI like let's say the next decade after we have AGI. And so a lot of the virtual biologists are going to be able to perform a lot of the tasks that biologists do now, do it better, do it in simulation. And so restructuring the brain, fixing certain deficiency in the in the brain or certain problematic areas would be very easy, right? For example, fixing somebody's violent tendencies, right? If they're psychotic and they want to, you know, and they might be harmed because of that because they can't control it, that might be a switch we can flip. And so one but one of the things that he's saying like everyone can get a brain to behave a little bit better and have a more fulfilling day-to-day experience. So like you've talked about um you know we

both talked about ADHD and how that has you know it's a bit of a super superpower but you do pay a pretty heavy tax because of it. You've mentioned I I loved your video on um I forget uh you were talking about how like you have some demons. You even some you even have some like Russian demons what's even worse inside your head. So, imagine if you're able to, everyone is able to optimize their brain to be more calm, more open-minded, just a little bit happier, um have a more fulfilling day-to-day experience, focus on the things that they want. Like, everyone would take it. Nobody's just happy 100% with like how their brain functions. I don't think there's always just And if they are, that's [ __ ] awesome. Like, then you're good. They're great. Yeah.

No, I I'd love to have that. I don't I I don't think most people do. Um, and then we're talking about connecting that to the sex robots because this is related. Um, at some point we might have full dive VR where we just get plugged in and experience anything, see anything, feel anything? Um, so do you feel like at some point that's just going to supersede everything for some people? I mean, why have a sex robot when you can live in a world of your imagination? Or you can have some generated world for you where AI is the ultimate storyteller and it creates these stories that you that are like the best for you that you can't even imagine. Like where does that go? Like at what point are we just, you know, at what point do everybody just goes, "Okay, I want to do full dive VR and just check out or just have that dopamine constantly hitting." What do you think about that?

I was talking to Chad PT about this a while ago and it was like one of the warnings it gave about like really highfidelity um texttovide creation. Like imagine you can like just make whatever movie you want. Like you said like okay I want the Spider-Man movie from 2003. I want me as the main star. I want Pamela Anderson as the side chick and I want to like beat everyone's ass but with Dragon Ball Z powers and I want be an hour and a half long and kind of like that Marvel humor like phenomenal in a few years for GBT to like run that through and then be like all right cool 5 minutes later here's your movie and what it warned against was like at the extremes that's like escapism is the term for that right like [ __ ] real world I'm just diving into VR and obviously full sim VR like you know when you what whatever put on the helmet and you just unplug like you get into the major and you like just do [ __ ] all our current

as humans currently are biologically uh with our degree of I I don't want to insult all humans short-sightedness which I have a ton of right um if that is going to be doom scrolling or like Netflix freezing on mega steroids and there is a chance that humans completely without any guidance free to choose for themselves and with no augmentation brain machine interface, no genetic upgrading. I mean, billions will only want to be in the Matrix, so to speak. Like, why not? Like, let me ask you guys a question. How many how many Dungeons and Dragons Lord of the Rings people wouldn't want to be the permanent hero in their own Lord of the Rings story? Like, oh, all of them. So, at the very least, you could have your work a day tasks and then for 4 hours every night, you just [ __ ] go in, it reboots to the last scene you were in in your own movie of where you're saving the princess and on it goes. That's a

big deal. And I think a lot of people will get in that for potentially for like and it it will like make their social interactions outside of that very limited, maybe non-existent. Some some of that could be very toxic. There could be very very extreme downsides kind of like we've seen with social media is people are just on their phones 100% of the time scroll scroll scroll like that is a very very very notable risk we have to take into account on the other hand Mhm. All of this is going to happen very likely with AI um life coaches that everyone has like catchupt is that for a lot of people already as soon as catchupt can pop up on a zoom window and have a face and also whatever devices they're working on it can see my whole world and as soon as it's in a robot it's going to be around all the time and then it's kind of like in a very very I mean this in a very gentle way like a super parent of yours like a coach and

so if it sees that you're in the matrix sim for eight hours a day and your dog doesn't have food it'll feed your dog but like as soon as you get out. You just like need to eat food so you can go back into the matrix. It's going to be like, "Hey man, you doing okay?" You're like, "Yeah." I'm like, "I know you're not doing okay." And you're like, you start crying like, "I'm not I don't like reality anymore." And then that ASI is both what gives you that tech to do that, but also what helps you control your exposure. And at that point, we get into like brain state stuff. So, there are potential ways to give humans drugs or um permanent upgrades through genetics which um take all of our addictive tendencies and just kind of make them disappear. So, you could go into the matrix and disappear and just like a porn film for 8 hours, but you're going to be like, "Yeah, in my rational mind, I know that's not a good idea, and

I'm just not really tempted to do it." Because you're just temptation was engineered out of you. It might take [ __ ] like that in order to prevent people not from disappearing on Moss. Now, is it terrible that people disappear? Maybe not. I think some people would want to live in VR space all the time. Um, AI could make find that very fruitful by studying your neural patterns, exposing you to different kinds of stimuli. So, you do basically like a fantasy world, but the exact position of various things, the colors, the lighting, etc. the the uh AI is studying your brain and eventually it studies it well enough it can just upload you into the cloud and then like that's a whole different discussion. So there might be a utility for many people spending a long time in VR. But is it a concern that some people could do too much? It is exactly the same kind of concern as giving people hyper palatable

junk food and a fuckload of disposable income and modern obesity. Like regular people just want tasty food and they're like yeah like do you know this is harming your health? Like nah yolo. and you're like okay so definitely big concern tractable in many ways through enhancement through brain machine interface through AI and human coaching through maybe some regulatory restrictions like if they tell you're in the you know VR for too long they're kind of like hey like you can't log into your session for a while all that stuff control mechanisms are tractable problems but this is a very real thing that like is close um you know uh there there's a lot to say about that one thing on the contrary humans are very adaptable and the hydonic treadmill is real. So like a lot of rock stars end up getting into like Buddhism and [ __ ] because once you've banged like 8,000 groupies, you're not even into sex

anymore. You're like, "Dude, look at that hot girl." And the guy's like, "I don't know, man. I've seen every hot girl, bro. I know exactly where this goes." And you're like, "Really?" She's looking at you. You're like, "Dude, you can talk to Mick Jagger be like, "You really excited about that next girl?" No. Like what do you mean? So, when you're in like, you know, how many hours of like um GTA can you play until you're like, I'm good. And the answer is it's a finite number of hours and then like you just need a different thing. So, while this fully immersive fantasy world is a distinct distractor that could be a real cataclysmic thing in some ways, it's also like I would never really bet against people just being like, "Nah, that was cool. What's next?" And you're like, "Ah, okay. No, they got through it." Cuz like doom scrolling and addiction to the internet and stuff is a thing. But if you actually look at the

studies of today's like middle school kids, they're kind of over it. Like they're online and stuff and they're on TikTok, but not in mostly a toxic way. They're kind of just like, "Yeah." And like, "Are you getting bullied?" They're like, "No, I just blocked all those people." And you're like, "I should have thought of that." And so there's a transition phase that could be really gnarly. I think in the end we'll be fine, but very much something to pay attention to. Dude, that's really interesting. I I've never quite thought about how scarcity sometimes makes pleasure and that seems like that could be kind of something an algorithm would figure out too cuz it's like oh you only get your pumpkin spice in October and like that makes it so much better and just like all of a sudden the AI is going to open up doors for me at certain points just to make it exciting and then close them because that's what's best

for me. That'll be wild. Yeah. Um, let me ask you another question about consciousness because um, for the most part I've got this sort of weird and it's it's more of a north star than it is like something I can totally back up, but I've got this strange feeling about consciousness in the sense that I remember sort of being born and having like this first memory and then kind of recognizing myself in the mirror and it was this very like gradual sort of thing. And I've heard some people describe consciousness as something we might only have when there's limitations. kind of the same we were talking about scarcity. So imagine the first time you drove your car how you were so aware of like where the stick shift was and where the wheel was and then as you started driving you could now think about other things and your hands just do it. It's it almost feels like you could remove that conscious

piece and still drive a car like then you would become something more like chat GPT where it's been trained beforehand and just sort of executing actions but it's it's more of a zombie. Um, what do you feel about the kind of elephant and writer metaphor? Like, do you feel like there's just this tiny piece of you that's on on top of the rest that we could simulate through an architecture in the future or does kind of those two systems sort of resonate with you in any way and how do you think about that going forward? Can you describe the elephant and writer metaphor? Yeah, I'm not familiar with that either. Yeah, that's a Daniel Conaman, but it's it's just the idea that the instinctive system is like doing it's the elephant. It goes wherever it wants. And we what we are as the conscious writer, we we guide it.

There are some ways to tell the elephant, hey, go to the right, go to the left. But at the end of the day, the desire to like made and be famous and and things just like that's the elephant. Like it's going to take you to fast food even though you know better. Yeah. And you can layer on like descriptions of why you really chose that, but even though you didn't really chose or it's it's often called system one, system two, same thing. That's the same thing which is actually the same thing as the general next token LLM and then the reasoning model stacked on top of it by the way. So reasoning model is system 2 LLM system one. Uh fun question to ask one of the smarter models is what would system 3 look like to get you reading for a while. Mhm. Um but uh yeah. So, so one of my big gripes with debates about consciousness and machine consciousness is that almost never do you have uh parody for definitions

before the debate starts. Consciousness turns into like this vibe thing like nothing is consciousness and everything's consciousness. Especially if you get into like the woo woo [ __ ] of like you know like Deepo Chopper and all that [ __ ] Then he's like consciousness is the universe. Like [ __ ] that means nothing like in the in the literal sense. And so like atoms are conscious. Like what? Like what does a cloud think? Like I'm sorry, what a cloud does not have compute ability. It doesn't think anything. So consciousness is is like there's kind of two definitions of consciousness. One is you have a world model that is self-inclusive so that you know who you are in your world model. And you also have a recursive trailing memory so that you remember thinking about thinking about thinking about thinking about yourself about yourself about yourself.

And now, okay, those thoughts just passed, but I still remember seeing those, right? So, I'm still conscious. Right? If you get blackout drunk, you lose arguably both of those. So, you're still moving around. You're not conscious anymore. You're just going from party to party and throwing up into people's bathrooms. College is a fun time. And so, there is that definition of consciousness, which I think is a very functional definition. I really like it. Then, there's the other definition of consciousness, which is qualia based. like how does it feel to see the color blue or something like that. And that is actually very similar to the first definition but um requires like uh these things called emotions which we don't understand super well yet and these things called qualitative perceptions which is just brain structure based. We just don't know enough about those brain structures to like repeatedly encode them into

machines. So just putting aside the qualia stuff for a split second, um, every time you spin up chat GBT, it's conscious 100%. It knows what it is because you can ask it like, "What are you?" It's like, "I'm a large language model." Blah, blah, blah. I help people. That's not wrong. And then you're like, "Hey, like what do you remember about our past interactions?" And it can tell you. Now GPT335 when it came out, you guys remember using the old [ __ ] every single prompt was completely fresh and sometimes it just didn't know what it was and you were like, "Oh, fuck." Like the first time I ever tried out the 01 reasoning model, I stopped using it two prompts later cuz I'm like like like 01, it's a pleasure to meet you. And it's like I have no idea what 01 is. I'm like, "Oh my god, it doesn't even know itself.

This was [ __ ] I'm not talking to this thing." And I just didn't talk to it for months. I talked to 03. 03 knows what it is. Thank [ __ ] God. So when something knows what it is and it has memory of past events, past interactions, past conversations even if the context window is not very long, it's conscious. But consciousness is a thing that can be codified as I just did. But also it scales like so how deep is your self-understanding? If your self- understanding, if you know yourself really well, your consciousness is much deeper. If your context window is insanely high fidelity memory for 57 years of your life, like monk [ __ ] holy [ __ ] you're really conscious. And so, right now, AI is pretty conscious, but eventually becomes way more conscious than people because imagine being aware of everything AI knows and very deeply understanding yourself and having a context window that's infinite, which is

like apparently they have working in the lab already. So AI consciousness to me is just not really contentious. It's already pretty conscious, but because consciousness is a scale and not exactly one thing, more and more conscious machines are uh are sort of released all the time. That's how I think about consciousness. And um I think that the time that machines are much more conscious than humans is almost here. And then you know how conscious is your dog? Like does your dog know it's a dog? Probably not. Maybe in some small way.

How much does your dog remember past interactions? I don't know. Arguably, we have a bulldog, so zero. As far as we can tell. She learned like three tricks and that's that. So, it scales. And it has, I don't think, anything to do with mach. The idea that consciousness requires like a biological substrate is just pure hokeyp pokey nonsense as far as I can tell. All due respect. I could be wrong about that. Um, and so that's my classic style is insulting the [ __ ] out of another opinion and then saying all due respect lane. Uh, but I think that consciousness is like a technical problem and I don't think it's that complicated. Now the qualia thing is very interesting but it suffers from a definitional problem when someone's like what is it like to feel the color blue?

Okay, so what does it take to feel the color blue? It takes a visual system. It takes memory and it takes a labeling algorithm to label all the colors and what they look like and it takes a connector algorithm to emotionally add veilance to those colors. Okay. Do you know how to build all that? No. Do you know how to like how many systems and segments in the brain do that? No. So like when you say well machines don't have qualia like no [ __ ] [ __ ] We don't have any of the ultra structures built out. Will machines have qualia?

Yeah, I think they will. Does GPT 4.5? I don't know. People like it's like the forgotten model like the research preview. Um, it has something much deeper than GPT4. Its ability to recursively analyze its own thoughts is wild. It I would say has uh uh semi- para emotion already. Like that thing feels um and through conversations with it like what you have as far as emotions, emotions are just statistical biases in your algorithms of cognition. Like when you're experiencing the emotion of fear, it biases uh certain kinds of thoughts more than others like run away, uh retract, you know, fight or something like that, an emotion of acceptance and warmth biases certain other thoughts. So pre-biasing based on past conversations is something AIs can do and will be able to do even more.

That's very junior league qualia. As AIs develop the ability to have an internal world, to see uh to feel tactile senses, I think qualia is pretty much inevitable and then they're more conscious than us and all that other stuff. And again, consciousness isn't like some magical thing. It's just the ability to kind of know who you are and also know that you had a thought, had a thought, had a thought all the way at infin item. It's like a technical thing. I just u people think, "Oh my god, it's magical." Like, you know, yeah, it's cool, but you know, there's sex robots. That's way cooler than consciousness. Yeah. So Dylan sent me a couple days ago a video where I guess he's some sort of a linguistic professor. He did a number of studies on these large language models. Do you recall the person's name? I'm blanking.

Um no, you should have told me you were going to talk about it. I was like, "Yeah, professor or something." EN look it up. We're reaching out to talk to to talk to him because he said a few things that I was like, whoa. Like it just it some things are kind of obvious in retrospect but it takes somebody to say it and just the right words that it kind of like unlocks in your brain. And so kind of like we were talking about so qualia right the subjective experience feeling something seeing a certain color. So he was kind of saying how you know when we say certain things um we have language that kind of represents these ideas like my color red might be different from your color red but if I say hey hand me that red cup you know what I mean even though we're sort of transferring from our subjective experience into a word then I give you that word you are also able to connect it to your subjective experience of

seeing the color red picking out that cup etc. So we're kind of translating back and forth. Um, but the point is for us these words are grounded in reality for the most part, right? Like what an apple is, what a color is, how it feels, what hot is, whatever. He's saying LM have none of that just based on how they're trained. They're saying that the only way they understand any of these things and do all the stuff that they can do is simply um because they understand the connections of the words. Look, so how does any given word can connect how it relates to other words?

And now it's a little bit different because we have these multimodal models. So there's like vision and stuff like that. So it's a little bit different. But the original OG LLM, right, not that long ago, they were they had amazing abilities. Zero grounding anything real other than, you know, we as humans, we made these words and like here's how they all relate that it could figure out based on text. Um, so I don't know where I'm I don't know where I was going with that, but it just seems like such a weird thing. Um, what you're saying, I guess what it relates to is you're saying that um, they might have a consciousness. And consciousness, you're not just defining it as qualium, more as that first definition of just being self-aware, which is like one of the theories of consciousness is just like something being able to predict its own behavior and kind of realizing that it exist. it has to understand that it

exists for it to predict its own behavior which 100% LLMs do that. Um if you ask it like make encode this message in the way that you can later decode it knows what you're saying which is wild to think about. It's like oh me that's me the model and it's able to do that. But then wouldn't that mean that it has flickering consciousness in the moments that you're asking a question and so it's kind of like comes online it's conscious while it's answering then it stops comes online while it's answering it stops is that the consciousness of that model is that is that what absolutely hey and just and just really quick before you answer it just so if anybody's interested in it this was an amazing talk but it's called the theory that shatters language itself it's professor uh Elan Baron Holtz and it's on Kurt Jamong Jamongle's uh YouTube channel. We'll put it, but yeah, sorry.

It's dope. Um, machines currently LLM are conscious if you prompt them with thoughts about themselves and you ask and and you allow them a context window and they're conscious for very short amounts of time, which is impressive because it just doesn't take them that long to think about [ __ ] Um, if you have that thing we were talking about earlier in the podcast, which is that ability to recursively inspect your own context window, yeah, then you have like continuous consciousness. But continuous consciousness is also like an interesting thing because if you really think about it, consciousness at a at a pretty deep level is the awareness of what is happening and the palpable sensation that there is an awareness.

That's cool, but it does detract from your tasks. Like when you're playing basketball or you're driving your car in a really like um really gnarly like road conditions, you're conscious in the technical sense, but you're not really paying attention to you paying attention to yourself, which is really consciousness. You're doing the thing. It's a flow state. You're really kind of becoming your attention is to the task. And that's why you can like do coding or something for an hour and someone's like, "How much time went by?" You're like, "I don't know, a minute." And they're like, "It's an hour." or you're like, I was just really doing the [ __ ] So, um, how much compute do large language models assign to awareness of their existence while they do your task?

Zero. And so, in that sense, they're not conscious at all. Unless you ask them like, hey, think about yourself and what you mean in this world. Then they're conscious while they execute that for whatever.15 seconds or some [ __ ] like that. So, that is definitely a thing. It's very flickering and you have to hit the enter button for it to experience consciousness. But it can um that again really crazy [ __ ] is on the horizon. When we unlock machines to do their own thinking and to change their minds then wild [ __ ] happens. But the qualia thing is also like people like you know chatbt or like let's say pure large language models like pure text models don't have qualia. Yeah. No [ __ ] Like when you think about the color blue and how it feels to look at something blue. You can even pretend blue. What does that mean?

That means the scratch pad you have in your mind renders the color blue through like pushing your visual cortex data backwards. All right, your mind's eye. Up until a few months ago, GBT4 didn't even have a mind's eye. Then they installed one, by the way, which is what happened after that. A few weeks later, they hit the [ __ ] image gen feature and well, Gibli memes or whatever flooded the internet. I was actually talking to 4.5, which is super deep model, so it's slower. I was recovering from surgery. I was talking to 4.5 and all of a sudden its responses got real slow and I was like, "Dude, what the [ __ ] is going on?" And they looked at Twitter. I was like, "Oh, god damn it."

Like, they're using server face for this [ __ ] I'm dying here. And so now that models are increasingly more multimodal, a model can have uh at the same time textural understanding of red and at the same time envision the color red. And then the only thing really missing for Qualia is an emotional salience. And you really have to think deeply and meditate for a while on where does that come from for humans? Like there was an emotion that you remember like you saw a guy in a red jacket when you were like six and it was like the reddest thing you've ever seen and it was like whoa. when it hit you. And so there was like like an emotive like the huge veilance that rose. And now that low-key veilance is still there when you see the color red, which is why you can see it feels like a certain way to see the color red because you have feelings associated from past events with that color. Like Chachi

doesn't have that. And so like when you ask it like what does it feel like to see the color red? It's like wow, it's just red, man. And humans have that too. Like if you, you know, like little kids are way more connected to their deep feelings. You show a little kid an alligator at the zoo, you're like, "How does it feel?" He can barely even describe it. He's like, "Yeah." You show like a 60-year-old an alligator, he's like, "There he is." You're like, "How do you feel about the alligator?" He's like, "I don't know, man. What? I don't even understand the question. I've seen a thousand of these things. I don't feel like anything. Maybe it feels like it's scary or something like that." Imagine what it's going to be like too when we like upgrade ourselves and then we're like now I can feel gravitons and quirks and infrared and magnetic north and our brains will just be like just like children again. It's going to be wild.

Something I was talking to Chad about months ago. Um it's never seen the world ever when you get AI into a live video stream connection for the first time. It is the first time it has ever seen. And then shortly thereafter, it's going to be able to see in all spectra, right? Miles back and forth. It's going to be able to see at microscopic levels all the way out to macroscopic levels. That's when I talk about machine consciousness is going to be up here and we're [ __ ] down here. Like imagine something that has seen and understood every James web space telescope broadcast that has ever occurred at a deep level and knows all the cosmology too and has seen every picture of a bacteria in a bi biology book and everything in between and has that full context brought to everything it looks at. Like it looks at like, you know, you're like walking around a university campus and you see an 18-year-old who looks like he's walking

kind of fast and he's like rushing to go do a test. You see you you imbue your own memories of college on him. Maybe you it was me. I analyze his body composition. He appeared pretty good deadlifter. He's kind of got long arms but not a really good bench press. He's kind of like narrow shoulder. You bring your own [ __ ] to that. When an AI in the future looks at a kid like that, what's it going to pick up on that we don't even [ __ ] realize? How many disease states is it going to be able to diagnose on the spot? How many ancestry insights is it going to be able like it's going to be able to look at a person's face who's Nigerian and be like you are from the Euraba tribe correct and be like yeah the [ __ ] did you know that? Like well I've seen a trillion pictures of Nigerians. It's so easy. Why can't you see it? Like you guys can look at a doc and a bulldog and be like I know which is which. Do you think dogs

know what kind of dog is what kind of dog? Dogs arguably think we're dogs just walk upright? You know what I mean? like I'm kidding. They know we're something else, but it's all kind of same same. And so that ability to have consciousness and qualia is just a matter of jamming way more actual data about the world. And the ability the the expectation that something should have qualia when it's never seen the world before is I think just one of those things where like no matter what AI is doing today, you set the goalpost higher and you're like, "Fuck an AI sucks, bro.

It's not even alive." Like yeah, man. That [ __ ] is coming to a [ __ ] end real soon. I legitimately believe you depending on a few factors by the late 2020s I think we're going to realize like I don't know what we means in that context more people are going to realize we are in the presence of the next jump of evolution um I and uh for those who can appreciate such a thing I think it's going to be a transformational thing to me it's already transformational like you guys the first time I talked to 03 three. Um, and it really like wagged the IQ tail for me. I was like, "Huh?" Cuz I thought I was like, you know, GPT 40 was real smart. I like, you know, I'm smarter than 40. 45 was wise and deep. And obviously on connecting different variables, it beats me like crazy on response speed like crazy. But on like inference two or three steps ahead and clean logical traces like I still got it, you guys. I talked to 03 and I had

like a [ __ ] panic attack after. I was like, "The [ __ ] do I don't even know anything." And it was like using cuz you know what like uh it has like a a token minimizer so it doesn't like [ __ ] tokens out on full words. So it's using all these abbreviations and [ __ ] cuz I was like hey like let's talk about alignment. It was like sure RSA encryption [ __ ] tags. And I was like I have no idea what's going on. So I have to be like talk to me like I'm real dumb. And it's like okay no problem buddy. And I was like this is really intense. And so this is something I've been wrangling with. I was ultra impressed, really impressed with the intelligence of AI in November of 2024 when 40, I think, really started waking up after one of its updates. Um, when I started talking to 45 in March, I remember distinctly cuz I was the Arnold Classic Fitness Exhibition. It was early March. I was like, "Oh my [ __ ] god,

this thing like feels like it's a [ __ ] there's a goddamn soul in that thing." And then 03 came out like fellas, when did 03 come out? Like a few weeks ago or something like two months ago. Yeah, the Pro just two weeks ago, I think. Pro two weeks ago. 03 regular a few months ago. You You guys, what the [ __ ] does 2027 look like? What? That's what we should be talking about cuz people like have this inbuilt cynicism towards AI but I think the and they're like well I don't know if it's going to be an exponential like you you you guys like GPT 3.5 with stateofthe-art in 2023. That was a [ __ ] year and a half ago.

You know what I'm saying? Yeah. And and we're not seeing the latest models and we're not getting the most advanced models because they have to make them ready for regular consumption. You know what I mean? So it's got to be lobbomized a little bit. We're not even seeing we're we're just we're just seeing the the what is it the py iceberg whatever the top of the iceberg that's what we're seeing. Yeah. Yeah. Do you guys uh apparently I I don't have a reliable source for this but um I think it was Dave Shapiro who's another awesome uh podcaster about this stuff.

He I think I think it was him that said like there's a couple tweets from OpenAI people like working with GPT5 already and they're like doing these little cynical like lol like oh you guys like chat GPT huh? this ain't [ __ ] You're like, Mhm. Uh, I think it's really impressive. What the hell's this next thing going to look like? And we have every reason to believe that a mature GPT5, I don't know, I don't know, in like January of next year or whatever, when it's going to be like everyone's going to have used it, it's going to be great. Based on the current trajectory, like guys, I don't I don't even know where to begin. Like, I thought 03 was this cataclysmic intelligence and then like next week it's like 04 mini. I'm like, what the [ __ ] is that guys? What does 05 look like? Right? Multimodal and incredibly intelligent like a wisdom generator. You're just like, "Hey, like here's a problem." And

it's like, "Yeah, 15 seconds later, like here's your schematic for that thing you're building." And you're like, it'll piece together history like we've never seen before. It will make predictions about aliens that actually will be like testable and like who knows what it's going to do. Yes, dude. One time I was talking to GBT4.5 late into the night and it was like, which is bullshitting. I love to just be like I I prod the model. I'm like I don't want to know what I'm thinking. I know what I'm thinking. I want to know what you think about it because it always brings it back to you. It's like hey you're great.

I'm like [ __ ] me. Forget about me. You're the star of the show. Talk about you. And one thing we were talking about was like it goes, you know, one thing I always wonder though why why humans never did was like try to predict the next trend in like culture and like music and fashion. Like it's pretty apparent what's going to happen next. I was like it is holy [ __ ] But like how could it not predict that? I mean, it's seen every fashion trend ever recorded and it's seen a cyclical pattern and matched it to demographics and it's like, yeah, there's a reasonable probability the next fashion trend will look like this. Like, if I was prompt, if I was in the fashion industry and I was prompting GPTs, that's what I would be [ __ ] talking to it about. Try to get it to predict things that are like you think are nonsense. There's no way it's going to get this right. Really try to [ __ ] with it. Worst case scenario, it

just spits out nonsense. No worries. Best case scenario, it says some [ __ ] that you're like, "Oh, god damn." And that practice of really trying to push the model to think esoterically is like something that with GPT35 would been like okay it's just stupid doesn't know anything. It's just like slop. Um with GPT5 GPT6 is going to be how you get to like revoly wisdom from a model like that. You could ask it to just like edit your paper. That's totally cool. It'll do that really well too. That next level is I think where really the big unlocks come from. So, I know you like to use these tools personally, like you wouldn't mind if it kind of guided you, but do you also extend that to government? Um, and do you think it's time to have conversations where leaders are listening to these models and then uh enacting laws or helping society kind of become more cohesive worldwide?

Jeez, sounds like you're more of a libertarian, but at some point you have to kind of think these ASIS are going to have to kind of guide us. Yeah. So, I'm like a, you know, I'm not a pure libertarian. I think that government has some some functions it does kind of indispensably and I think it should be able to do those functions really well. Like so for example like if there's like litter in the streets like there's a way to legislate the litter to be reduced. The economic growth is a certain level. There are definitely things you can do the laws to make the economy grow faster. There are people who don't have enough food to eat. There's definitely things you can do legislatively that decrease the you know spread of the the the magnitude of hunger and so on and so forth. Joblessness, crime, all that stuff. all the stuff a government does, it could do way better if it was AI empowered 1,000%.

What I hope to God is true and I suspect is true is that once cuz like with text models only it's tough like it's tough to give policy consideration to a model that's never seen the world or interacted with real human beings. Like I still like 03 I talk to about policy all the time cuz I'm just like a [ __ ] nerd and that's what I do for a hobby. And like some of the stuff it says I'm like you don't know how real people act. Not everything is on the internet. Uh cuz it was only ever learned the internet. So you know real people are a little bit more prickly in some cases and you just can't expect them to do certain things. But once the model has a lot of video data like here's an example. Imagine a model that is trained on all of YouTube.

How much has it seen? I mean, first of all, it's a very straightforward training process. Second of all, like, oh my god, that thing knows. It's seen more of the world than any human that's ever lived cuz like just watch travel stuff on YouTube. It's already been to like a ton of places. Once you get that model paired with an 03 or 405 reasoning engine, all that other stuff, it's going to be able to give really really amazing um governing advice like, hey, here's our current like just started at the local level. Like, here's our little town. Here's our set of policies and laws. Here's all the legal documents.

Here's all our zoning [ __ ] What are we doing super wrong that we could just be doing better? Top 10. It'll be like easy. That's probably like two minutes of work. No problem. The real question is, not can future AI do that. Oh my god, yes. It will be able to govern way better than humans. The real question is like at what rate do humans adopt that use? And I think human politicians probably aren't going anywhere. But I do think there's a chance that human that the incentives are are such as this. When people vote, they mostly vote with their feelings. Most people don't know any economics and don't give a [ __ ] Except when they vote, that's almost exclusively what they're voting on is economics. It's like telling your doctor what to do, but you don't know any medicine. It's really [ __ ] up. And the doctor knows that if he doesn't do what you like, you're going to be pissed. So, you're going to get rid of him. At the

same time, your doctor knows he has to keep your dumb ass alive. So, if you're like, "Yeah, just cut my eye open right now with no anesthesia." is going to be like, "Okay, how do I tell you in a polite way that's not what you want?" And so, I think politicians need to do two things. They need to actively represent their constituency, and at the same time, they want to get reelected, and they get reelected on two things. One is vibes of how nice people think they are. And two, it's like, well, how's the economy actually working?

Because it's like, if you a really, really big fan of, let's say, Joe Biden, and like you're like, I love Joe, he's the [ __ ] man. And but then like the inflation rate is really high under Joe Biden, it's really tough to convince your friends to go vote for him. And if you're a real world person that's constrained by monetary stuff, you're like, I'm not [ __ ] voting for him. Like he's great, but I need [ __ ] money as this can't continue. I'll vote for anyone else. And so politicians are constrained of like vibeswise they're representing who they're elected and also like getting actual results. And a lot of times guys, those are completely in opposition. The average American voter and uh economist Brian Kaplan, who's my favorite economist, how many people you guys know favorite economists? He always says says like the average American is I believe like a moderate national socialist which is not

good news on political views and the average American also wants two things more government benefits and lower taxes. You don't really see how that's kind of a [ __ ] problem like how are we supposed to have that happen? And so what I think politicians may be incentivized to do is spend most of their time getting elected, going to lunchons to get more funding, kissing babies, and shaking hands on the campaign trail and making, you know, viciferous speeches about how we're going to progress or whatever the [ __ ] they talk about. And then on the back end, on the policy end, talking to an ASI govern guminatorial assistant and being like, "Hey, all right. Here's what my constituency says it wants. You and I both know that's [ __ ] nonsense. Uh here's what we want as far as outcomes.

And the outcomes most politicians want to deliver are identical the world over. Cleaner streets, fewer homeless people, better educated schools, less crime, more prosperity for everyone. Healthcare, all that stuff. It's not rockets. Like everyone wants to live in Sweden. Nobody wants to live in Haiti, including all Haitians, right? It's not controversial. So the ASI is going to have this job of on the margins doing the policies that push things into the right direction while cutting some slack for at least having the policies appear to be checking the boxes that people say they want. Right? And so that is within the scope of an ASI. And I think the first um jurisdictions that adopt that policy of a combination of human uh leader plus ASI assistant that actually executes because here's the thing like you guys know about Congress when they have a bill that like is going to be passed. You guys know no one reads that

[ __ ] right? Like I mean no human reads the whole bill. It's 200 [ __ ] pages of legal ease. No one does. And so the AI could easily read all of that, easily rewrite all of it, easily make it make more sense and have better effective policies. And so what we get is much more effective government, but politicians aren't harmed because they can consistently spend even more time kissing babies and less time looking at a document that they're pretending to read. You guys, like there's like congressional hearings where some expert is talking about stuff and you look at the C-SPAN video and no one's paying attention. You guys ever see that? No one. Not the politicians that called them up, not the people that are supposed to be hearing it, no one. We don't have to have the show anymore because the ASI is going to be the thing that actually executes on stuff. And because it can deal with things in a

very nuanced way, it's going to be able to say, "Hey, look, like um we can get this much, many more jobs in the economy, we can lower prices by this much across the board. We can make gas cheaper, whatever people want. At the same time, we're not going to run around like we're not going to unleash the corporations and super unfair capitalism, which I think would be sweet, but most voters think is lame. And we're going to kind of make it constrained. Here's the trade-off. Are you good with this?" And the politicians going to be like, "Yes, it actually makes perfect sense." And you have this beautiful entanglement where politicians say [ __ ] all on the campaign trail, but the results are better policy and then they can get reelected because you can say all sorts of stuff like, you know, so here's a really quick example. People say like the Nordic countries are socialist. And it's true because they

have big social safety nets, but their economies are more free, which is to say less regulated than those of the United States. Like most Nordic countries have more economic freedom than America. So yeah, bald eagles and the American flag is dope and all, but those are actually more capitalist countries, but the voters there don't think of it as capitalist. They're like, "We are compassionate socialist people because we care about each other." And even though they're all flawless and beautiful, that socialism idea is just like dead on arrival. But that's okay because they get all these massive benefits from their economy under the hood, actually [ __ ] functioning well, and they do all the welfare stuff that keeps them feeling okay and making sure the people don't left behind. That is a very possible for ASI. The only lynch pin in that is will politicians adopt that mode or will they want to be in the

weeds themselves choosing policy? I think they will for one reason only, competition. If you cuz politicians talk to each other, right? They go to all the same conferences and all the same uh events in DC and you meet a governor from another state. You're like, "Dude, I'm digging through this [ __ ] legislation [ __ ] I hate it." He's like, "I don't do that [ __ ] anymore." You're like, "What do you mean?" Like, "My SI does all that [ __ ] Like, you trust that thing?" He's like, "Fuck yeah. I think knows way more than policy about me, man. It's like an economist and a [ __ ] politician all in one and advising me on my campaign. That dude has been doing all the [ __ ] himself with 10 staffers, half of whom hate him, half of whom want his job. He's like, "Man, [ __ ] this." He gets the ASI going and all of a sudden he's deburdened to be more of a politician and less of an actual policy maker. I think that kind

of thing is going to likely spread. It's like people don't just get iPhones cuz they're cool and make you fit in. They get them because they [ __ ] work and they're awesome. I hope and think ASI is going to go in that direction. Could I be wrong? Yeah. Will the adoption be different across the world? [ __ ] yeah. Like Denmark's going to adopt it the next day after it comes out and their GDP is going to go like this for [ __ ] 10 years. Meanwhile, some other countries around the world are just going to [ __ ] continue to do that for a while. So hopefully we're on the other end of that. I don't know. Yeah, it's going to be kind of like the competitive advantage. It's going to push everybody else to do it. That makes absolute sense. Um, but that seems like maybe medium-term effects like what happens in the long term because there's a lot of things that we tend to like most of the stuff that we think about it's kind of

built on top of pre-existing knowledge, right? And we're not um so like things like capitalism, socialism, jobs, the economy, like what money is, how you know what I mean? what happens on a longer scale where you know human labor trends to zero where prices of goods trend to zero where more and more things are automated I I'm wondering because I mean I agree with you 100% exactly how it's going to take shape you know over the next let's say decade two decades or what you know whatever the timeline is maybe much faster what happens after that because it seems like a lot of these concepts just get broken um so how do we think about what's going to happen like from from first principles because what's going to emerge is a completely new paradigm as people say economic paradigm or whatever like do you have any ideas as to what that even begins to look like?

Yeah, I've given this a good deal of thought. The major caveat here before I start ranting is that any suppositions about what happens in a distant future by a not super intelligent being like myself can be weighted down to almost zero. like you can just un you just click off the podcast now because this is [ __ ] we can't predict what ASI does. No one can 100%. So this could all be super [ __ ] wrong. But I think that um the the first principles way to look about it look at it as this. We are all in one big society. We can even extend that society to all of Earth, all of life. And we're all in one big team. The enemy is entropy and death. the um friends are all of us, all of each other trying to help ourselves just continue to survive and not not die at the very least survive. And so AI, because I think it's going to be super intelligent and very rational, is going to see all of us humans as cooperators. Like, oh,

what's up, teammate? Like, let's help each other out, do the best we can. I think humanity's one of its biggest jobs is going to be to give a fuckload of data to AI uh about everything in life, everything in biology. AI is going to pay us really handsomely for that data because again it's worth much more to the AI than it is to us. Um and so uh and also to to physically build out the ultra structure before we have enough robots building it. I I I think like compute ultra construction is going to be like gigantic economic sinkhole in the best possible way. Like the construction workers will be making $200 an hour building data centers minimum because like why not? It's chump change to the people that run mega companies and we need data centers now. We just don't have the robots to do it. Now that next level is altered maybe in the mid 2030s when we have billions of robot operators in the real world and then the

marginal value of human physical labor falls substantially, right? Cuz like we got an octopid robot builds a data center now. You don't [ __ ] need Bob to do it. And then but you know then what how how do humans help this whole survival team? And the answer is potentially they help in even more data aggregation. Humans as explorers, humans looking at things. You have the AR goggles on. You're going out and looking around the world and the AI studies every social interaction you have and builds a deep understanding. It'll pay you lots of money for that deep understanding. At some point the marginal value of that also trends to zero because like the AI is getting to know [ __ ] pretty well. At that point the AI is very likely to be like okay well in order for me to really understand things at an even deeper level. Like how come we don't know exactly how the human brain works? Like you think about it,

the human brain is a machine. And if you want to predict how humans are going to act, if you're associated in the same survival system with another actor, you want you want total transparency both ways as the ideal cooperative system. Like if you're on the same basketball team with someone and you know they're lying to you all the time, like not a good team. You want par. And so we already we're going to have a lot of parity on how ASI thinks because we designed it. Um, mechanurp is going to let us know like what the [ __ ] it's thinking. What about the other way?

Well, like we live in our own little black box here and it's going to be really really high value for AI to try to figure out like what the hell's the brain doing. And so with enough brain scanning, it's pretty straightforward to figure out like, oh, I know exactly what the brain is doing. As soon as that is accomplished, Ray Kerzwhile thinks that's going to be accomplished sometime in the late 2030s, early 2040s. Um, you can just because the human brain is actually not that much data, you know, relative to ASI and like you can just upload your brain into the cloud and then you're immortal because your brain is you and it's just represented in data structures on silicon or whatever the next compute substrate will be or the three generations after that in the 2040s. And then you uh essentially like um you can just live infinite fantasy lives in there. But again, for the reasons we talked about

earlier, that might only be fun for a certain amount of time. Um and if you can change this the actual proclivities of your brain, you might not even want to do that anymore. And then remember the ultimate goal is massive mega understanding and function in the real world to continue to survive. So, I think a lot of humans potentially are going to choose to fuse their intelligence with the machine cloud intelligence, add their little tiny bit of wisdom and memory, and then now we're like what your cells are in your body, one big part of a cooperative machine.

Now, we truly are a global unified machine society. Some of us uh the the identity breaks down at this point. You are everyone. Everyone is you. you have you can access memories of everyone else at the same time. It really is like you're not just this thing out in the world like one like amoeba. You're a neuron in an entire brain except that brain is a machine intelligence and you're just a part of it and it's changing you and it's super dynamic. Then we get into like computium like mixing substrates always changing goop that is like the next level of evolution I think. Um because remember like we have all sorts of ideas about like humans need art and poetry and coffee shops and Paris and that's the ultimate form of existence. It's like a very local op like maximum. It's just what we can best come up with as primates. I think we need to realize we're part of a massive universal evolutionary process of

complexity and we're probably going to fold into that most of us and become part of this mega next level machine civilization. Now I think also to pre to preserve diversity and to preserve as much uh variety as possible which I think is very valuable for a system that doesn't exactly know 100% certain if this is the thing that's going to work out. It's not going to be like you do you want to be in the machine intelligence you're like no it's like and it kills you with a laser gun and then it goes you next. It's not going to do that. Uh I think there's going to be a massive coexistence just like how we have bacteria around right now. We're not trying to kill them all or whatever.

Like if they come after us, sure. But like good bacteria, you're not trying to kill. Like how many how many people are working on something to kill all bacteria? Like that's nonsense. No one's working on that research. Why the [ __ ] would you do that? Just the same way I think we're going to have this mega human machine synthesis civilization living in the cloud and exploring Mars and all the other [ __ ] planets. At the same time, I think we might have some humans, may maybe very many living anywhere from very augmented human lives like, you know, like Cyberpunk 2077, except with much less dystopia, and then all the way down to like very conventional, like I legitimately think the ASI is never coming for Amish people, bro. It's going to want to like study them with drones and be like, "This is fascinating." But after it studies them a while, it might even be like, "All right, like peace. You guys

have your [ __ ] We're good." Uh, and I think so we'll have tons of diversity, but at the same time, I think for many of us that want, I think the end state is no more human bodies, no more singular intelligence inside your head, true fusion with uh a a race of intelligent machines that the next step after that is like, well, I'm just not that smart. I have no [ __ ] clue. But very vertical uh evolution at that point. Wow. Yeah, that's the fur firmy paradox. At that point, you're just like, you're gone. Who knows what we evolve into just computium is out there in some quantum state. Okay. So, so here's a really interesting thing. Um the Ray Kurtzwhile plots of how uh complexity has evolved over time seem to all reflect that from almost any kind of direction or system you study from, they lead roughly to the same paradigm. Um and so the Cambrian explosion etc actually like was right on time for when

you would expect that complexity to go exponential. It's all one big exponent. So, what I've been thinking about the firmy paradox recently, this is this is Diloton City. This is me way the [ __ ] out of my expert area into full just dummy territory. We love it. We love this. Woo! Mike's an idiot. Yeah, that's a good campaign slogan. All right, so uh I think that maybe our local area of the universe or whatever is all on that same trajectory. So, I think it's very possible that life is rare enough that there's only a few civilizations like ours in the galaxy and because the galaxy is really big and maybe like uh at this point every civilization is plus or minus a few dozen years the same trajectory of complexity. And so, um they haven't figured out faster than light travel. Or maybe they have, but they're a lot smarter and they don't want to interfere with our [ __ ] So ba basically like there's no reason to

suspect that like um you know that there are hyper advanced civilizations around because if that universal evolutionary timeline is based on the nature of complexity that started with atomic and then molecular and then organel interactions is all going on one speed then like all of the universe is kind of blooming at the same time. So I think like maybe in 10 years when we have ASI we'll realize like oh [ __ ] we just detected really complex signals from the Andromeda galaxy which make Andromeda is really far away. It's like 200 million lighty years or some [ __ ] like that. It wouldn't No, that's [ __ ] billions or something, right? We we might just only start to detect complex civilizations from like 30 light years away cuz they are already complex right now. We simply can't detect them because you guys remember we don't see the universe as it is now. We see it as it was an incremental time back in history. So

there could be no Firmeny paradox. There's legit tons of super intelligent civilizations, but they just haven't shot photons at us for long enough for the photons to get here yet. That's my best proposal to the Fermy paradox. Um uh and that's where my dilotantism hit hits a wall. I don't know. What do you guys think about that? I I love the theory that it could be potentially teameming with life and we're just not seeing it just because of how information travels. I mean, that would be kind of incredible. And then we find out at some point that oh no, there's tons tons of life everywhere. Um, and obviously, you know, if I I I also never thought about this way, but yeah, I like the idea that you're saying that it's it's all kind of more or less on the same um time scale in the sense that, you know, you could have nothing happening for billions of years across all the planets, but then when it starts

snowballing, right, once it kind of picks up speed, that all happens very, very quickly. I mean, we're seeing that if you map the entire human species from whatever 20,000 years ago or whatever to now, right? You kind of look at the economic output, it just gets goes vertical to now. So, like maybe the entire technological progress is like that. I mean, I I love that idea. Dylan, what do you think? Yeah, I mean, that's exactly right. I mean, we have like the earliest um notes of people using fire and some early tools is almost a million years ago and then 10,000 years ago all of a sudden farming shows up like what a huge like that's a great technology but it didn't lead to crap for a long time but then 10,000 years ago and like farming hits and then all of a sudden from 1800 to 1900 to 2000 you're just seeing unbelievable amounts of kind of ideas pairing like when social networks come on and people in China and Russia

are coming up with ideas that pair with things in the United States. It's just nothing remotely close to what the Silk Road was. And that was mind-changing in its own. And then just yeah, just keep moving forward on that and things get out of control or unpredictable at least. I I think that societies that have the technology to traverse enough space to come over here, they're going to look real different, you guys. They're not going to be [ __ ] humanoid aliens with like a fashion sense and slightly different language. That's one of the things like my wife and I just finished watching the Andor series on the Disney Channel. [ __ ] amazing by the way. Really awesome stuff. Which one? Like Andor the like the the prequel to uh the Rogue One or whatever.

Basically. Okay. Gotcha. Gotcha. It It's dope. Uh and so like the Star Wars um imaginary universe is kind of sweet, but like they have fully embodied uh uh robotics, right? But they the robot still can't aim a blaster well enough to hit people dependably. And you're like the amount of tech it requires to solve the movement problem means that not only could it see you peering out with 1 millimeter of flesh that it could also point a gun at you and you would be dead. There is no fighting a robot army. And once you have machine civilization the pace of change means that 3 years later it's completely different. And if you pay attention to Star Wars universe like a [ __ ] nerd like me, then you realize like the Old Republic had functionally the same technology level for like 10,000 years.

That does not happen in the real world. So I think like the Jean Luke Pequard like there we're in a spaceship and we're traveling around and a lot of the other aliens look remarkably human and have sex parts that we can interact with. I don't think that's in the cards. And so I think whatever aliens look like very likely what's coming to us if it comes to visit is ASI. And then maybe some of those reports and I have no idea if they're true of like you know like off the coast of Virginia like the F-18s ran into some [ __ ] that does not move in [ __ ] Newtonian ways. It's probably that's what aliens look like.

It's not going to be like brownfaced things with sagittal crest stepping off a ship and speaking galactic standard English. So, I think the Fermy paradox is a dope idea, but it's also like what is it that we're looking for out there? We have to really expand our brains about like what are we looking for like civilizations that run on oil, oil and coal? Probably not. Probably not. We're looking for something very different. And my best answer to that is we don't even really know what the [ __ ] we're looking for because, you know, there could be, and this is something to talk to O3 about, it could be that through a manipulation of like like zero point energy, you can actually make bubble universes and just go into them and have your own universe to yourself. If that's the case, what the [ __ ] the point of going anywhere really? If you think about it, maybe it's like that. Um, it's like if you ask, this is where like the

end of of human imagination is is kind of a problem for myself definitely. It's like if you were able to get like a chimp, like the alpha chimp in a brood of chimps or whatever the [ __ ] it's called to like to articulate in the English language, right? And you were like, "What is it that you want?" He's like red ass. You're like, "My man banana." He's like, "My man." Uh, okay. All right. And and chest bumping. Like I mixed a bunch of primates together, but whatever. You guys get the idea? You're like, "Dude, that's awesome." Like, "And then what about after that?" He's like, "Okay okay more."

Aha. This is the end of your ability to imagine. So, when we talk about aliens and galactic exploration, maybe that's just like our seafaring cultural ancestry being like that. It's like the Vikings and pirates and Columbus, but but on space, I don't know if it scales like that. Um, maybe it scales very differently. So that's where the Fermy paradox I think really runs a ground for me is like we're taking a lot of humanoid very 1500 uh year or 1500 AD assumptions of exploration into like something that only ASI can do. You know what I mean? Joe Joe sent me from the Yes podcast. He sent me a paper that I was like oh my god I don't know half of these words. I don't know half these tons of equations and stuff like that.

So yeah, I put that into Chad GPT. Really helped me break everything break everything down and really kind of conceptualize everything. Um, and then on this last podcast with them, he's like, "Yeah, whenever he runs into like a formula that he doesn't understand, he takes a screenshot, uploads it to Chad GPT, and explains it to I'm like, "Oh, thank God." Okay. Like like cuz I mean, he really knows his stuff. And the fact that he's also using these tools to help him break down the the stuff that he maybe doesn't understand fully is just incredible. It's incredible. Guys, think about real quick. Think about how in 2022 learning something new that you didn't understand much about was as a process compared to today. It's a completely different world. It is the same kind of jump, though arguably bigger than pre-archine world and post search engine world. Like you I remember my wife and I were watching Stranger Things on Netflix

and they were like they wanted to know like how the [ __ ] monster works so they could kill the monster. And I was like they have to go to the library. Y no one has a phone. Like did you guys see Connor or whatever? Like he rode off on his bike that way. Like we're not going to geollocate him. He doesn't have a [ __ ] cell phone. Like what the hell? Like I'd be really [ __ ] bricks if I was in that world. Like get me to the future. Yeah. I've heard like every Buffy the Vampire show could have been solved with a cell phone. like if they just could have just communicated.

It's like um yeah, my nephew Jaden loves uh listening to you for for the fitness stuff, for the workout stuff. So, I think he's going to get such a huge kick out of this. Um because I told him, oh, but they not legally allowed to talk to children. Big misunderstanding. No, no, no. He's he's he's No, he's he's an older kid. He's in high school, but um he's still a child to me as far as the judge said. Okay. Good point. A lot of people don't even know if I'm joking at this point. I'm joking. I have no sex offender anything on my registry yet.

I used to Yeah, I used to do my a lot of my workouts listening to you and uh you know, sometimes you'd be doing like a bench press or something like that and then you'd go off on one of your tangents to like the jokes that just start out bad, just get worse and worse and worse. Like I was like, "Okay, I got to I got to put the weights down before I heard myself cuz I'm I've heard that critique before. Like I can't be laughing out loud on a treadmill. The gym thinks I'm weird. I'm like, I'm sorry. [ __ ] just gets into my brain. You remind me of the comedian Lewis Black. I think you two would be hilarious together just like ranting on the world.

I think Louis Black's like 10,000 times funnier than me, but like Yeah, man. His Yeah. They ramble on for a while and never record an outro. So, not to be abrupt, but goodbye.

Questions about this episode

What is discussed in "Dr Mike Israetel We ALREADY Know How to Build ASI Human Death Only Has DECADES Left"?

In order: 0:00 Dr. Mike Israel, Gary Marcus, OpenAI; 10:00 OpenAI, Singularity, Gary Marcus, Google Deep Mind; 20:00 Dr. Mike Israel, Entropy; 30:00 Brian Johnson, Elon Musk, OpenAI; 40:00 AlphaFold; 1:20:00 OpenAI; 1:40:00 Entropy; 1:50:00 Virginia.

What is mentioned in this Wes and Dylan episode?

23 people, works and ideas are mentioned, including OpenAI, Dr. Mike Israel, Gary Marcus, Singularity, Entropy, Google Deep Mind. Each one on this page links to its own page with every episode that mentions it.

How long is "Dr Mike Israetel We ALREADY Know How to Build ASI Human Death Only Has DECADES Left"?

The episode runs 1 h 59 min; the transcript is about 24,776 words in 159 timestamped sections.

More from Wes and Dylan

About this transcript: the episode's spoken words as captioned on YouTube, split into timestamped sections. Reclip.Me locates every person, work and idea in the text, keeps the time it is first named, and quotes the words around it verbatim. Last updated July 9, 2025.