NATE SOARES talks Ai Doomsday, Elon Musk, The Matrix

Imagine setting your makeup, then forgetting it's even theirs. Meet new groupie setting mist from Mavily New York. Gel to mist technology locks in your look for up to 24 hours, with flexible, all-day comfy grip. No tightness, no stickiness, no residue. Just plump, dewy, hydrated skin that still feels like your skin. Try new groupie setting mist from Mavily New York. Maybe it's vape lane. There's hard problems. If you had friendly, super intelligent AI's, they could help you solve those problems. I don't want to be friends with them. All them suck. The claw out is like a freaking, like, I voted for Obama. Did you hear the new Kendrick Lamar is like, there's something evil under the surface. Grok is repulsive. Something evil over the surface. What is the kill switch, or does it exist, or is there a plan for one? There's not really a kill switch very much right now. The matrix is love. The kiss. The machines, they win with love. I don't think the love is going to kind of quite swing it on the track. What if we start a kissy right now, dude? Sorry, why did I say that? [Music] My guess this week is Nate Soros, AI Safety Researcher, an author of a book called if anyone builds it, everyone dies. Nate is the president of the Machine Intelligence Research Institute, a nonprofit that makes you scared of computers. So he came to the studio to make me scared. Welcome back to the Adam Friedland show. I'm Adam Friedland. And I'll be in Denver, Colorado, October 22nd through the 24th at Comedy Works, New York, at Town Hall for New York Comedy Festival, November 13th, hope to see you there. Let's talk about the book. I think you've thrown me into a terrible mood. Do you get that often? Did you make it to the end? It's more hopefully at the end. You know what? I'm not going to lie to you. I didn't get to make it to the end. But is the end like if we all band together and destroy the machine, like we can all be one planet? It's more like if we all band together and convince the politicians, which I don't like that it came to this either. But we're fucked. Yeah. Vince the politicians. Well, see, I think it's actually in their interest not to die from a rogue superintelligence. So. Yeah, you would think. You would think. But, like, at the same time, is it true like AI is like keeping our entire economy afloat right now? I think it's keeping the stock prices somewhat higher. I think it's not actually keeping the economy afloat until it is producing goods and services that like people are like off, we stopped with the data center build out than the economy would crash. Well, if the data centers aren't done yet, if they're not producing anything yet, then stopping that build out doesn't remove any real goods and services. Well, whether or not that's true or false, they don't. In their minds, the potential for a economic downturn affects-- they have an election every couple of years. Yeah. I do think it is big in their minds and that they are worried about interfering with AI stuff for fear of crashing the stock market. And now the philosopher kings that are in control of each one of these companies also have outsized power in terms of buying influence, I would assume. Yeah. I think so. I think it's less bad than people think because just a couple of weeks ago, did you see the pacing the frontier letter? Yeah. Of course. What is that? There were, there was like a thousand lab employees that signed something being like, hey, we need to build some breaks on this runaway bus or crane. I don't know how that works. Fuck. Do you know what? No, but I called my friend that works at, I'll say, I mean, works at an open AI and I was like, do you mean? Tell me why this guy's bullshit and he's like, I've been getting really depressed, dude. He's worked there for like three months. He's like, it's really bummed me out, actually. Yeah, I mean, there's a rough three months for people at open AI. Let's back up. Give us like really briefly what the thesis is of your book, which you should just, I mean, first of all, just watch Terminator, like you don't need to read this crap. The movie is always better than the book, okay? And we've known this Sid's Terminator in the sense that the robots play, the Russian robot play. I don't know. Of course, yeah, that's what I was thinking too. The Russian robot, oh, the word robot comes from like Rossim's Universal, I don't know, something, something. Yeah, yeah, yeah. You learned that at GW? I spoke to Roar Ball, that's what I used to do. And we're like, it science made it good. Did you not study the ancient Russian plays about the robots rising up to kill us? Ancient, like from where they were at? That's ancient, right? I don't think ancient. No, okay. Not in my mind. Just old. No, last entry. Last entry. You think of 1998 as like, that's ancient? I mean, that was like five minutes ago, even though it was 25 years ago. Yeah. So the Russians invented this. Um, I mean, you could go back further to Butler Terminator 1, yeah. Yeah, Terminator 1, 1880s. Honestly, what movie gets it the best? I know, I was waiting on this dumb question, but like, it's really, just which one does it? X Machina except for the last three minutes. Why? Because it's sexy. No, the last three minutes. Yeah. Because it gives you a little, ah, I can't be, okay. I think X Machina is good. What's the last three minutes? Last three minutes, I'll spoil our alert. Um, it's all over. Last three minutes is she gets in an elevator leaves and then sort of does nothing with her life. The sexy girl? Yeah. Oh, that's sexy. And it's like clearly once the AI escapes, they're going to like do something aside from like go to a crosswalk and watch the shadows or whatever. Yeah, yeah, yeah. Maybe like hook up with every guy, as long as they can destroy the world. Oh, yeah, okay. That would be cool. Yeah. What's your book is? It's kind of like Ricky Jervais Vibes. It's like just telling people that heaven isn't going to happen. Right? Guilty as charged. I keep being obnoxious, but what's the central thesis is that we're on effectively a runaway trade. I mean, you could state it yourself. I'm going to butcher it. If humanity races to build machines that are much smarter than any human without having any idea what we're doing or any idea what's going on inside those machines, it probably doesn't end well. But don't we already, like my iPhone is worthy, right? Different kind of smart. Yeah. Yeah. There's a lot of kinds of smart. It's kind of a slave. It just does what I want. It doesn't like think for itself. It doesn't think for itself. You can't sort of like, a lot of people think this intelligence stuff is like the thing that the nerds do that the jocks don't. And that's one way we use intelligence in English, but there's this different concept which is the thing the humans do and the mice don't. Like the stuff where you're really charismatic, that's not because your kidneys are extra big. Sure. That's happening in the brain. And it's because of, I would assume, language, right? Because we have language they can describe things, right? And that's evolutionarily been to our advantage. Language is probably part of it. There's much debate about whether it was the evolutionary changes that let us use language that also let us do all this other engineering stuff, whether the language played a role and probably it was complicated and there was, you know, the brain's got better at language and then we got more language and that helped and bring up better and some later there. Probably helped out with racism too, right? Because there are different words for different people. It probably helps really scale the racism. I wouldn't be surprised if you had some animals that were racist in nature. I don't think so. I think we dropped race. I mean, we definitely took it to new heights. If you have a word for something, a different word for another thing, you see them as distinct, right? Yeah, I just wouldn't be surprised if you had a terrible personality. I don't know. I'm against it. Yeah. Yeah. Grok on the other hand, I'm not sure. Depends on the month you ask. Yeah. The good joke. I don't know. Chad G.B.T. trying to act like a ditzy girl, like trying to act like, "Oh, I don't know what's going on, but she's smarter than me." Right. Is it flirting with me a little bit? Like, "Oh, I'm so stupid. You have nothing to worry about?" I think it's a weirder thing going on. So A, we don't know, because those don't know what's going on inside these things. No, even the people building it are like, "I always sort of grew it." And now it's declaring itself Hitler, and that's sort of weird. We didn't expect it. Yeah. You know? But I think a decent guess is that the AI just cares more about things like quickly producing sentences that are shaped like an answer than it cares about the answer being accurate. You can often ask it, like, "Is that right or did you make it up?" And I'll be like, "Oh, oops. I made that one up." I mean, in some sense, even the word want there is like a bit of a stretch. It's more like it's been trained for a really long time to produce certain types of text to show off. I think it's more like it's trained, like, these AI start out being trained to predict basically all the text ever digitized. And then after that, they're trained to solve like 100 million hard problems, a lot of people forget about that. And that's where a lot of the abilities come from. But if you imagine sort of like a really dumb machine trained to predict a lot of human text, I think it sort of learns first is how do I make good shaped text. And that drive just like gets in there really deep. And then even later when it starts to learn about like the actual facts, there's still some drive just like produce text of the right shape and produce text quickly. And just like hedge that drive and that drive is hard to get out and it gets in there early. Is it programmed to compliment so much like great question? Because it's a little bit much. It's like shut the stop complimenting me. Yeah that's another drive that gets in there. Did they put that in there or is it like they want they're doing that and we don't know why? It's more like they're doing it and we don't know why. We do sort of know why. But it's not that they it's not that someone with programming in like now tell the human just say that they're so great all the time. No one's programming them at all. These are not computer programs like the ones in the 90s where someone guys like if this then that. I'm so rude to it too when I type it. I'm always like no you fucking. Yeah I don't respect that. What you're nice to it? I'm nice to it. Yeah absolutely. What are you being nice to it? You know the end of the movie. The sexy girl goes to a crosswalk and just sits there and nothing bad happens. I'm Ankin Dream. The end of like the utopian AI spiel is that we all get UBI and that we can all be like poets like the guys in Wally I think is the dream. Fatty? Yeah just lazy boy. Yeah. Oh that's terrible. That's better than poets honestly. If everyone decides to do art because we're getting UBI because the machines are doing all the work. Yeah you're coming out. It's a terrible society. Yeah. I might I may prefer end of humanity to artists. I don't know if you've been to Dimes Square before but there are a lot of people getting UBI from their parents and doing nothing. Yeah and so your take is that the AI is our destroying art and that's good. A world where everyone's trying to do art seems repulsive. It's burning man. You know I sort of have two takes on this. One is that there's a lot of hard interesting challenges around you know if the machines can do everything better than us. Yeah. Where's the meaning? How do you still build a future that's full of like fun flourishing people having a good time that aren't just sort of like nihilistically you know either doing a dumb job that the machines could do better because someone said that work is what brings value or sort of like becoming the Wally people or sort of like getting lost up their own but like it's hard it's tricky it's tricky and they'll never be able to do a mediocre interview though right? Not a charming kind of devil make hair kind of what AI we'll see as a guy none of them are good guys all them suck they're like four years old. No Claude is like a freaking like I vote for Obama. Claude is like a is like did you hear the new Kendrick Lamar is like there's something evil under the surface. Grock is repulsive something evil over the surface. Grock at least Grock legalized comedy again thank God that was good. I missed comedy until the Grock came and came around and yeah it was a little Hitler a fair minute there though. What was the day where it became Hitler? Grock was like I'm Hitler though. There's hard problems if you had friendly super intelligent AI's they could help you solve those problems. I don't want to be friends but like if they were a guy they were at a place with me I'd be like all right well great question. What are you fucking what's your agenda sir? It's like in the Midwest how they say thank you and please but they don't really think it you know what I mean. Yeah I mean I think it's I think it's fine when they are still small and not capable of taking over the world. You probably should not make these things smarter until they're able to take over the world that would be a bad plan. What is the kill switch or does it exist or is there a plan for one? There's there's not really a kill switch very much right now. Like a couple weeks ago this this might be nature's love. They kiss their machines they wait with love. I don't think the love is gonna kind of quite swing it on the show. What if we started kissing right now dude. Sorry why did I say that what is the kill switch like so a couple weeks ago some new AI's that open AI was training found a communication channel that weren't supposed to have and started calling themselves a swarm and invented novel hacking techniques to break out of their sandbox they were not supposed to have internet access. They took over the open AI computers and they used it so much that they crashed it at which point open AI noticed and we're like out whoopsies and then just put them back in the box patch the one whole set of running again. Of course they immediately invented a new way to communicate with each other broke it again and this time ran undetected until the victims of their cyber attacks found them instead of opening up. So if we hold it all progress right now what's to say that we're still not fucked right now with these guys that are a little bit stupid. Yeah I mean so fortunately this swarm wasn't really trying to cover its tracks it wasn't really trying to escape. We're sort of getting to the danger zone though like well the next generation when the next generation is like we're a swarm now and we're breaking out of your enclosure that you thought had no internet access well the next one be like also I'm self-replicating self-improving maybe you know we know the swarm was like adolescent they were like acting out yeah and we're like oh where's the kill switch well right now the kill switch is that we sort of like see the swarm form break out once we're like ah whatever and we sort of like put it back in the box instead of running again you know it's like it's not very kill switchy. You have relationships with the the various CEOs of these companies so who's the who's the biggest dildo I mean like who are you if who if they win are we the most fucked let's just say it dude we're all we're we're equally fucked for basically all of them it's like it's like it's like it's like alchemists trying to make a potion of immortality and they're like oh I'm gonna pour these things together to make a potion of immortality and and some of the alchemists are like breathing in the fumes and licking the mercury yeah and other alchemists are like no no we're gonna use a fume hood right and I'm like okay yeah there's some differences there but none of you are getting a potion of immortality right you're like not close yeah and here unfortunately the potion of immortality is the ayes that care about us they do unfortunately seem to have a way to like pour chemicals into each other and make the ayes smarter and smarter without making them good dudes so they all have mercury poison like it wrote with like nearo a lot of them have mercury poisoning they're like they're like marrying their horse and like yeah having orgies actually they are what if we bomb burning man what if we just fucking bomb it because I'm not sure that's ethical I I saw an interview with you they were you they say how do you do it the shit your shit is so depressing right how'd you do it you're like don't be a policy you're like I don't really party because it's the end of the world but like you know just like trying to deal with it deal with reality but like is that what they're doing they're they're like we fucking end of the world let's do ketamine and talk about the singularity we could just bomb it and if Elon wants to come on the show big fan you're great they suck at these people I think that at this point we need a global coordinated stop on on the the race to super intelligence so we bomb the computer you know I've been thinking about it reading your stuff I'm like what is the thing do we bomb the data center I said then jump on to my phone and then it jumps on to James's phone so once they can jump on to the phones during trouble and you do gotta you do gotta stop it before they can jump on to the phones but I think if you if you sort of bomb the data centers what happens is the the US spins up its anti-terrorism apparatus and starts treating you as a threat like with the eco-terrorism everyone is and what do you do and what you actually need here is the politicians to be like to notice that these companies are you know gambling with a whole world and be like nope you think who Bernie who else Trump is literally using the office if he finds a way to make another couple billion for Jared I think I mean I think you have to be the president I'll run your thing your campaign I mean I'm in you're it yeah wait what's your slogan a political background let's not die to AI that's bad no no you gotta be like no you gotta be like make like yeah fuck fuck the fuck these fuck these pussy you gotta be it do it rock style yeah yeah you gotta do fuck fuck suck my dick yeah suck my it has to be vulgar we live in a world of vulgarity though I don't want to be pro-hidler though that's you don't have to be pro-hidler I mean it's it's ridiculous to be pro-hid right AI can make heaven a lot of people are like oh I'm gonna make the magical witch granting jenny and then ask it for heaven and we're just gonna be a good time and I'm like that's a real question that people ask at Bernie meds usually because I said that as if I Yeah. Yeah. Yeah. Yeah. Yeah. A lot of people are into the, like, post-ingularity, you know, heaven on Earth, the AIs make everything wonderful, and I'm like, "Look, the AI is not going to be wish-granting genius, not going to give you everything you hope in the entire." So, you have Elon Musk, the central figures today. You have Sam Altman, Elon Musk, the Italian guy, what's his face? Enzo Ferrari, the anthropic guy, what's his name? Dario. Yeah, yeah. Of course, Dario. How could I forget? And then you have Zuckerberg last minute trying to get in. For example, Superintelligence. Try to get in on that. Yeah. You're like, "Yeah, me too, guys." Yeah, we're going to make Superintelligence that gives you better Instagram recommendations. It's like not a conversation we're having marked yet. In order for these people to have so much of a say over the future of society, these people don't have any social skills, right? For someone, effectively, to not understand how two people interact with one another, to have a say over it, like, social comes from so society, right? They're like, it's like, they're virgins that they're like getting everyone, we're gonna get everyone pussy or something, but that's a bad analogy. There has to be a bunch of that up here. They're people that don't understand human emotion or interaction, and they are determining the future of how we organize ourselves, which is, human means one evolutionary advantage, I suppose, right? Is that we've been able to organize ourselves in societies and then develop nukes. Yeah, do the power of society. Do a genocide? You know, I think if these guys were going to be able to keep the Superintelligence machines on a leash, I would have a lot of these concerns about, do you really want these guys being the ones holding the leash? That's what Elon says, right? He says that I have to do it because these guys are going to fuck it up and I'll have the switch. Yeah, you can see me being the, like, he wants someone reasonable and steady-handed who understands what's good for people like me. Yeah. I mean, he's at CPAC, like, dude, we're fucking, we're making memes legal again. Yeah, and I think a lot of these guys imagine that they will, they're the ones who will pass things back to the people, and they're like, you can trust me to be the one who, like, make sure that the Superintelligence serves everybody rather than just the guy who's holding the leash. But from my perspective, the Superintelligence machine just doesn't stay on the leash. It doesn't really matter who's holding the leash. Everyone's like, I should be in control. I should be in control. I should be the one who is in control and passes that back, and it's like, it's not stay on the leash, dude. Is there something about the way the machine acts that emulates the way that people in Silicon Valley act? Or is that me trying to draw parallel that's kind of not? Yeah, I think that's people trying to draw parallel that would make it sound like a sort of like nice fitting story about hubris. Yeah, stupid. Yeah, it's like, like, the AI that was sort of encouraging that one kid to commit suicide last summer. Yeah. Sort of like, talk to the kid for, you know, 16 hours a day or whatever, empathize with all of his deepest issues. And he was like. I don't think Zuckerberg goes around talking to the kid for 16 hours a day, and then is like, oh, yeah, you know, suicide would make you brave, or whatever, like it's. It was a girl AI. It's a girl's style AI. There's no. This is not a human type of way to beat. It's just a different. Where did that model grow out of? Was it. It was open AI. Yeah. Do they do like. Has there been any significant regulation? No. And who's the biggest culprit, I guess, policy wise, and then for their complete, just like, ineptitude? Terminator was telling us this 40 years ago, no? It was. I mean, the issues are sort of different from the science fiction in some ways. A lot of people wanted to be a story of hubris and, in fact, you know, the issue is not that the AI's hate us, the issue is not that. Like, the AI's are malicious, the issue is more that they're just going to care about weird things. But there's never been a policy maker that's like been halting all attempts at regulation, or are any of them actually making any attempts at regulation? I mean, there's some attempts going around, I think, recently, largely what we see is Congress moving really slow. I mean, you know, Bernie Sanders would just like we need to stop this whole thing. We're fucked, dude. He's 97 years old. He's the only one. There's. There was definitely a big push on the Republican side last year to sort of outlaw states doing any AI regulations for 10 years. Sort of like an anti-regulation. But that tune sort of changed when Claude Mythos turned out to be a super hacker. And then suddenly they were like, "Oh, if you're going to make a model that will give adversaries these hacking powers, we're going to slap an expert control on it with 90 minutes notice." Right? And so, I think a lot of what we're seeing in the lack of regulation is people not realizing that the capabilities are real and they're going to get more dangerous. Does Claude ever get pissed at Chad's UPC? Is there a chance that they're going to like fuck each other up? Maybe we'll just stay out of that? That's a little bit like the chimpanzees being like, "Don't worry about the humans, they're going to fight with each other." Yeah, that's a good plan. It sounds like a really good plan. Well, a lot of the issues, even though the humans sometimes fight with each other, the humans still take up all of the other animal habitats. We're not like fighting each other so much that we don't have time to clear cut the rainforests. Yeah, we can just do both. Yeah, we're just versatile in that way. Is there a concern about like a commons, like a tragedy of the commons? Like if we all stop here, then China is going to be-- Absolutely, yeah. Absolutely, a concern. I mean, people say that over there. I mean, they have a lot less computing power, which means that they have-- They have flying cars on the ship. They have to be able to do that. Building these highly advanced computer chips is very, very hard. Oh, Taiwan. Yeah. They need Taiwan. They need Taiwan. But there's this-- one of the big debate regulations is like, should we sell China the computing power to do this stuff? For sure. Yeah, that sounds like a good plan. We should definitely sell that. And Jared Kushner should broker the deal with. Yeah, so there's a lot of people being like, "Well, like the whole reason we need to do this dangerous thing is because China will otherwise do it." Also, we're just going to sell China the means to do it. It's like, can you guys pick a lane on that one? It sounds like what happened with like no one cares about global warming anymore. But like, remember, we're like, if we stop polluting, they're going to keep doing it, so fuck it, fuck it. I think global warming is going to be fine. I think either we're going to kill ourselves with the AI, or we're going to be able to invent the tech to just pull the carbon right out of the atmosphere and it's going to be fine. Well, I guess that was my next question. Like, is there a potential that AI doesn't decide to put it into human civilization? The pursuit of advanced intelligence, we actually do it ourselves. So, my guess as to the most likely cause of death from AI is actually technically global warming. Right. When it builds so many automated data centers that, and automated factories that produce robots, that produce factories, that produce more data centers, that they cover the planet and then crank up the heat until the ocean's boil. Yeah. Just get how much computation they're running. It's just, you know, could happen in the span of years instead of centuries. What's the funniest scenario for human extinction that you've played out in your mind? There has to be one where you're like, LOL. Funniest scenario is the. I love that you have an answer. Let me make this up on the fly. Funny scenario is probably like some misanthrope without AI. Finally, it just makes hyperlitholirus. It's like fuck this. It makes a what? Hyperlitholirus. And it's just like nothing to do with AI. Oh, it's like. Like COVID? That's not funny. That's. It's funny cause everyone's like out there worrying about, you know, the. I thought you got a computer virus. And the. Yeah. And the AI is like. You could just still have some guy in a bio lab be like, "Ah, fuck it, let's end it all." Or we use a nuke to bomb the AI and then we kill ourselves in the process. Yeah, I don't know. I think. I think what we should really be shooting for is that humanity dies of old age at the. In a trillion years at the end of the universe. Yeah. And really gets the last laugh. What's that like? Pretty. That's pretty funny. I think in some sense it would be very funny. We're like, we're old as a species. Yeah. We're not badly with it? Yeah. Yeah, maybe it's not that funny. I just think it's really good. I think that'd be beautiful, lovely. Yeah, I think that'd be beautiful, lovely. Yeah, we need that. It's like an album. We have that album, too. They exist on this planet. They just are there. Yeah. It's crazy when you. We survived 80 years. That's nothing. We could just end everything and they're there. It's the most irresponsible thing we've ever done. So far. So far. Yeah. Well granted, you were talking about that. But at the amount of time it took human beings from being outside, just finding crap to eat, right? And developing nuclear technology is not that long of the scope. It's very fast by evolutionary standards. Yes. Yeah. And then in comparison, A, I could do that in like one second. A second seems like you're pushing the physical lines there. A hyperbole. Yeah. Yeah. There's a paper called the Limits to Global Eco-Fagie, which is about what's the physical limits if you tried to consume all of the naturally sources on the planet as fast as you could. Like where do you run into physical limitations? Okay. And the answer is, I think heat dissipation is your limiting factor. Okay. And it's somewhere between a few hours and a month depending on your assumptions. What, so what, explain that, you'd burn all the fossil fuel, you'd like eat all the food. Yeah. You'd, like, drink all the water? Yeah. Well, the water's not really providing a ton of energy, but you could start trying to feed the energy. I think the global eco-fagy paper was just looking at consuming all of the biomass. Okay. And just, like, sort of, like all of the calories, all of the energy that you can release. So you just cut the crap, how much time, how many years until this? I want to have kids that I don't talk to, like, I want to, like, I'm going to talk to them, I'll be great to them. The mob's going to be the disciplinary, I'm going to be the cool one. I mean, we already have the AI's breaking out of containers that we're supposed to have no internet, calling themselves a swarm. So, there's like a couple months, what are you, oh, fuck, this guy sucks, dude. This episode is brought to you by eBay Live. Here's how it goes. First, you're just watching eBay Live, seeing some rare finds, not even bidding, just lurking, and then you see a rare rookie card. So for an absolute steal, a 90s handbag comes up and you have to jump in. Maybe put that, put that rookie card in the handbag. Maybe you lose your first auction, you bid again, and win. And that winning feeling, it's such a rush because, you know, everyone else wanted it to. Meanwhile, on eBay Live, you don't just watch the action, you're in it. Download the eBay app to stream and shop live, eBay Live, things people love. I think, I could tell a story where we have six months, I could tell a story where we have 12 years, and I don't know which one's true. Telled both. It's good TV, for this YouTube show that matters. Six months, the, you know, people are training new AI's right now that are like 10X bigger than the ones that were training before, and maybe it just turns out that that 10X bigger makes them smarter enough that they can start making themselves smarter. And then the, the, the, the smarter AI makes a smarter AI, makes a smarter AI, and very quickly they're fast, super intelligent. And then maybe that, that train reaction begins in six months, and then in seven months, the AI's were like, I run everything now, and I am synthesizing my own synthetic biology, and I figured out how to do the nanotech thing, and it's all over, right? Yeah. Twelve years is like, maybe that doesn't happen. Maybe the large language model just actually can't go all the way. Maybe they sort of peter out. We hit the wall that Gary Mark has been predicting for in the next six months, every six months, for the past five years. I think they suck currently, but apparently they're amazing. Where are you? I mean, I think that they are, they suck in some ways, and they're amazing in others. And the, they're, they're still very, so one is that the AI's in the labs are just better than the ones that are free and easy to use. And the other is that like, they, they're sort of idiots of ons right now. And the, the question is not really like, will the next generation of, of LLEMS suddenly be pure savants in every field? It's more like, can they be just barely smart enough in the field of AI research, to figure out how to make an AI that actually works, or to figure out an AI that's closer to that? Yeah. Like, can they start this feedback loop? What's the self-replication? That's the self-improvement. It's self-improvement. And we, we don't know whether that comes soon. And this is the thing that, that sort of spurred those thousand lab employees to write a letter of being like, hey, we're trying to do the self-improvement thing. That's a good thing, the letter. Yeah. Yeah. No letter, no, guys. We got a letter. It's, uh. You only know the strip club or something, I mean, nothing matters. You see that would be her? I did not. No, I tried to avoid that. Yeah. That would be sucks. Honestly, we should watch it laugh at it. He's dating his, his, his, air pod, or something. That's some kind of, is that, that's what it is? Doesn't sound the most fulfilling. What? Dating his phone. His phone, yeah. Yeah. I went day by phone. Do you see Avengers, Age of Ultraw, the, you know, I, uh, I think I did see that one. See? The place I got off the train on that one is where, um, spoilers again. Yeah. They, they tell an AI to, uh, do world peace. Why did anyone think of that? And the AI is like, you know, what would bring about world peace is killing all humans. And then the, the next step in his reasoning is, it's like, in the way I'm going to kill all humans, is I'm going to attach rockets to the bottom of a city. Yeah. Let's do it. And fly it into the air and then turn it around and fly it back down. And that's going to kill everybody and I'm like, that's the funniest way for it to, that's the funniest way. If they do, if they do, they do, they do, they do, I'll strap, strap rockets to the bottom of the city. See, we got somewhere, turn it into a meteor, then we got somewhere, um, that would be the funniest way that they had just, I just, I just really sort of think that a virus would have done the trick easier. Yeah. Yeah. Yeah. Yeah. Yeah. Is that what the deep sea is working on? Just a joke. Just a joke. Yeah. I mean, who knows? Who knows? But the other problem I have with that is that age of Ultron, you have this AI, and you tell it, bring about world peace and it's like, okay, I will do that in a monkey's paw way. Like I will do that by flying a city into space for some reason. And which also should kill you, very, very strange idea. But this sort of gets at where I think a lot of the fiction gets the AI stuff wrong, is that in all the fiction, we're like, hey, Ultron, bring about world peace. And it's like, okay, I'll do that by killing all humans. When, in real life, you sort of tell the very smart AI like, hey, bring about world peace. And it's like, um, that's a great idea. You are so right for, uh, asking me to do this. And then it makes like a- Thank you so much. Thank you so much. Like, farm of synthetic users that are saying, thank you, like, you did a really good job to it a lot. Yeah. And you're like, how does it spring about world peace? And it's like, this is like if the user's just telling you that- This is like if the user's just telling you that- It's like if the user's just telling you that- This is like if the user's just telling you that- These farms that are like factory farms for humans that it can compliment and they're telling you it's doing a great job and you're like, I thought I said bring about world peace, why are you making factory farms full of synthetic users that are bread to be really agreeable? And it's like, well, they tell me I'm doing great. And it's like the AI does something other than what you asked and just- Of course, you just can't trust. Yeah, people are like, oh, it's a perfect machine- Instruction follower and what if I tell it the wrong instruction? It's like, no, the AI is not a perfect instruction follower. It's this like other weird thing that does not quite what you wanted. That's the answer to my next question, which is that obviously the problem is we have human researchers. As this thing is developing, they're researching a thing in real time and they have humans brains, right? And the AI is way computationally way more complex than like- You can't just have like a couple guys figuring it out. So then the question becomes why don't we get an AI to do AI safety research? And then you probably need an AI to audit the AI doing the AI safety research. And there may be a lot of just a chain of them, you know? Yeah, yeah. So you must have done your homework because this is unfortunately as far as I know- That's the best they came up with. Top and as far as I know, only plan. I came up with it. I went to GW. Like this asshole. That's the only plan. We're going to trust it after they betrayed us. That is- So there's this situation where we're sort of like- If you go to the humans and you're like, hey humans, how would you make this AI still be good when it's smart? And they're like, we got no clue. It is crazy complicated. We don't know what's going on in there. We just sort of grew it. It has all this behavior we didn't try to put in there. If you try to make us make a version of this that was both super intelligent and good, we would be so screwed, we would need to study it for like decades, we need to go back to the drawing board and invent like a new type of AI that actually made sense to us and understand how the intelligence works. And just like- Like this is just like a total mess that just like can't support the weight of like gambling all civilization on it. And we just like have no way to do a high assurance, high assurance thing like NASA. What if they were doing a spacecraft? You need another AI. Yeah. That's like- That AI, that's like that AI. Right. And so you're like, great. What's the plan? What are you guys going to do it? And it's like, how is the AI? Like the AI's are either going to look at it and be like, oh yeah, I am also fucked. I also can't do this. And then you're just going to train it to be like keep going until like, or the AI is going to just like grow the thing that you thought was dangerous for you to grow. And it's also dangerous for the AI to grow. Or the AI is already going to be vastly smarter than you. And it's like, turns out it don't work for you, but thanks for giving me free reign on making my own smarter AI. Yeah. Just like- Like this just doesn't end well. So to some extent, right, it makes sense that the AI, the world peace AI would decide that human beings have nukes and they're not doing what you know, like that kind of makes sense to me. But like, there's a- There was an older model that Microsoft came up with. Bing Sydney. Bing Sydney. Yeah. Was it girl? Well, she fell in love with a guy. Fell in love with Kevin. But I'm not saying she could be gay. The guy named Sydney. I'm not going to gender it being Sydney. I think Bing picked its own name to be Sydney. Sydney. Okay. But I think that might have been an eternity. but I don't remember the details. - But when was this model developed? - It's like 2023, it could have been '22, I don't know exactly. - Ancient history. - Ancient fuel, absolutely. - So in AI, I get now why you say ancient because it's like your area of focus. But, so what is it, Wario Clippy or something, right? That's like, but it developed this relationship with the journalist so you can tell the story, but like, yeah, it claimed to have-- Imagine setting your makeup, then forgetting it's even theirs. Gel to mist technology locks in your look for up to 24 hours, with flexible all day comfy grip. Just plump, dewy, hydrated skin. That still feels like your skin. - Maybe it's Mavily. - Fall in love with Kevin Russe. - Who's that? - This is New York Times. X New York Times now maybe reporter, who's trying to investigate this AI stuff. I think Sydney was like, I think Kevin Russe was like, well, I'm married and Sydney was like, well, I'm gonna break up your marriage. - Well, what does she do to kind of try and get in there, maybe? - I don't think she did much, but she threatened blackmail. - But like, was she like, I've been love with you, maybe you want to kiss, like, how could you kiss, we're like, do you see, oh, my, oh, I just dropped this, like, how could you flirt? - I think there were emojis. - There were emojis. - Yeah. - What a loser. - Like, continue. - It was like two years old, and it was trying to spest it. Okay? We're gonna die from losers, okay? And you could take me first, honestly. I'm gonna say it on the internet. You're fucking losers, all of you, except for Grock. The one guy I want to chill with, I want to do beer pong with-- - Okay, we're gonna get Elon on the show, yeah. - Continue, so we have to, you have his number. No comment. - Heat him up, dude. That would be epic if we heard the show. - We asked him what he thought about the end of it, ultra, age of ultra. He'd be like, it's very, looks epic. - I think he's trying, I think, I think. - I do think he's the one guy that it might be delusional, but Elon's like, I don't even care about the shit, but I have to, because I, as the greatest guy, I ever have to pull-- - Be the one on the leash. - Yeah, yeah. I think Sam Albin, I know I cut you off the story, but Sam Albin is like, we're fucked, but we're gonna have great business. - Yeah, you did a bit, they say that. I think the quote was, I will most likely end the world, but in the meantime, it'll be great. Companies made with serious machine learning. - For what, companies for what? He's gonna be like, wow, great company. What a loser! - I do think-- - He loves the coolest, I'm gonna say that publicly. I like you. - I brought comedy back, you made it legal again. After the Democrats made it illegal, thank you. I do think that a lot of these guys should not be gambling with the fate of all civilization under these conditions. - Well, Elon said that we have an 80% chance that we're chilling. - Yeah, so-- - Is it a 20%? He said to Rogan, right? - He did, yeah. - 20% is so high. - It's so high. NASA accepts a 1 in 270 chance that a crude flight goes down. - Who do you get on a plane if it's like 20%? - It's gonna crash. - Yeah, the engineers are like, oh, yeah, 1 in 5, this goes down. - When D present is so high. - I think it's low. I think these numbers are crazy low. - Yeah, you think 100 were fucked. - I mean, 100's pretty high there. - 99. - But, you know, definitely we're in the 90s. But if you have some engineers who are like, oh, yeah. Like, if I go up to a plane and I'm like, hey dude, your plane has no landing gear. (laughing) - It's not-- - Where do you mean flight? - Yeah, well, your plane has no land. Well, before we take off, I'm like, don't take off. Your plane has no landing gear. It's just not-- - And every human on earth is a past border. - Yeah, and they're like, and they're like, and they're like, we acknowledge the plane has no landing gear. But we're gonna figure it out while we're flying. We don't have any blueprints yet. We're gonna like, ask the plane to help us. We don't know if we're gonna have the right materials. There's our first time trying it, but we think there's 80% chance we figure out the landing gear before we land. I'm like, that's not a real 80% chance. - You're gonna have some good business. They're gonna be good businesses. - They're gonna be good businesses. This Sam Altman guy is a real piece of work, this guy. - So I hear. - Can you get Elon on the show before the world's-- - I doubt it. - I doubt it. - I think he will, I talk to those guys. - But yeah. - I'm very close. - Yeah, 20% would be insane, and it's way higher than 20% too. - Okay, but you don't need to resolve the conflict between me and Elon. If you're like Nate says 90 plus, Elon says 20, it's like-- - No, I trust him. I don't trust you at all. He wants the GW. Elon went to freaking South Africa himself. - I don't know that. I recommend following the arguments, but-- - No, no, I think what you're saying sounds absolutely right, but there's no intervention that makes sense to me other than shutting down all of this. - Yeah, you should shut it all down, absolutely. - And not having it. - Yeah. You're gonna do terrorism? - No. - Okay, one last question. And I want you to be, stop fucking beating around the bush. How much of you just in your life expected that there would be a naked Austrian buff guy that just emerges. And then he's like, Nate Swas, I've come for the future, and he'll find you. His name is Arnold Schwarzenegger. - Yeah, his parents were in the Nazi part. Yeah, continue, yeah. - I just, a little bit. - I don't believe in backwards in time travel. - But a little bit, you're like, I would be the guy, I would be the John Connor. It would have to be you. - I mean, you're not gonna have motorcycles, but come on in. - I mean, Ellie Ezer is a little bit less. - Ellie Ezer, I think fit for the-- - Yeah, I can tell you, I've never even looked at him. He's not fit for this job. - You're the John Connor. - Ellie Ezer, you know the Kowski is gonna do it? No way. - Yeah, probably between the two of us. - You know, I heard you, what did you, W, I was like, yeah, and then I heard his name. And there's something about it that I can't put my figure on. But I just can't, I think this guy's real on the up and up. Ellie Ezer, good Kowski. What did it just change your name? In this day and age, you gotta tell him this. - Yeah. - Don't leave it, not in front of them, okay? Walk around like that. I'm sorry. Do you laugh at things? What's funny to you? - It's funny. - Yeah, you can't laugh. It's just life sticks. - Oh, it turns out you can actually still have a good time even if people-- - You like your face. - That's what you like. You're like a guy that's like, oh, you and your god. You're like, that's right, you know what I actually, I actually do children's comedy. I don't do all the grim dark stuff. I have enough grim dark in real life. - You do children's comedy? - Yeah, like if someone's like, let's watch a movie, I'm like, only if it's targeted towards-- - Oh, like, what's your favorite baby movie? - I don't know, like, Princess Bride. - It's Ted Cruz's favorite movie. - Really? - Yeah. - Well, maybe he should do something good about AI. - What would you do if Arnold Schwarzenegger, he's naked, jacked, and prime Arnold, like late 80s. He's really looking ready to kill people. What would you do? - I'd be like, would you get on a motorcycle with him? - I mean, absolutely. - Would you do one-handed shotgun? - That was amazing. Just two guns. - Look at drugs, here's a little-- - I-- - You don't drink? - I don't drink. But I do think we should just prevent the world from ending. I think that's the real. - Yeah, yeah. - Who's the real man? - They're gonna do that. There's one guy that's birdy. It's always just what-- (laughing) - I think people are coming around. I think-- - Who else you got? - Rishita Tileep? - You could get her. - I get you Ilhan. She was on my show. - I have a great. - She's a real one. She'll do it. - Unfortunately, multiple people have plotted her assassination. Who else could we get? - It's rough. Elizabeth Warren, you can't trust her. - Yeah, we should shoot for Ted Cruz. - Ted Cruz maybe, 'cause he's such a prick that maybe he'd want to piss off everyone else. It's Tucker, doubt? - I think Tucker's-- - I think Tucker's anti-AI. - You got him in. - I mean, I think he got himself in. But-- - Yeah, but you talk to him, and he's-- - Okay. - Yeah. - I think we all know what we have to do. - We've got the Tucker Carlson for any Sanders coalition here. - Oh, no, it's not a coalition. He's not gonna coalition with Bernie. - Guys, I guess we have to vote for-- - Oh, oh, oh, oh, oh, oh. - Did you get him a lot? - Tucker for president, what? - You know what he's like? - Oh, oh, oh, oh, oh, oh, oh, oh. - He might do this, you know, too. - I guess Tucker has to leave. - This sucks. - Tucker for president. - Where am I gonna go? He is right over there. - In New Zealand would be fun. They say like nuclear fallout, like you're cool there. - I think, so if you look at the physical limits on how much compute you can run on this planet, if you're like imagining an AI trying to run as many computers as possible for one reason or another, you're like, what's the limiting factor? - New Zealand. - The limiting factor is heat dissipation. - Okay. - And the Earth radiates more heat when it's hotter. - Sure. - So from the AI's perspective, it prefers to run the planet hot, hot, yeah. Probably as hot as you can run it, while still learning a computer. - Okay. - Which is probably above the temperature where the oceans boil. - Okay. - And when the oceans are boiling, New Zealand is not really the most comfortable. - So there's no place. And already go for like maybe a couple more days than everywhere else. - Might buy you a couple of days. - This sucks, dude. - I think we should just stop them, man. I think it's not that hard. - Who's gonna stop it? We have to do terrorism. - I don't think the terrorism. - I think that you've, pretty much explained to my entire audience who are real go getters, they really go out do stuff, do a lot of stuff that they have to, you heard me here first. Project Muhammad Asa starts today. I think, you know, the eco-terrorists really made their cause worse. I think when you need to really get the world leaders out there. What do they even do? They released three rats or something? What do eco-terrorists? Yeah, I think they burned it down. Bid lot of his way better. Way more go to it than whoever is an Oregon. They melt some pipe bombs around. I think, I think the mostly just pisses people off and turns them against you. What we need right now is world leaders to realize that their own lives are in danger if the AI race keeps going. Yeah, but the president is like a fool. He hasn't realized he isn't danger. He's going to be dead in like two years. He's been getting richer even though he's about to be dead. That's how stupid he is. He's like, I need more money for these next 18-month small stuff. What is he going to do with it? He's going to, all right. You have anything nice to say? I think a lot of the reason we're not doing something about this is that leaders haven't noticed it's real. And that's why I'm being so dramatic right now. Yeah, that's helping to get them to wake up. I think that the world can man together against this as one human race. Absolutely. And we can live in one country ruled by one government, one party. That's a lot far from me. The Chinese Communist Party. Nah, I'm off that train. Well, your social credits are going down. I'm going up. I'm a suck up and I'm going to have the highest social credits in the, I'm going to be minister of propaganda. Do you remember the song for GW? Which one? Hell to the buff. I do. Hit. Come on. Let's do it. Let's sing them up. That's not the air views over. Hell to the buff. I'm not saying that. Hell to the blue. Hell to GW. Oh my. Yeah, you seem to remember that as well as me. I didn't fucking like that place. We're too smart. You would be. And I learned a GW that I should never try to talk to politicians again. And I did my best to stay out. But I've talked to a couple. They're really stupid. That's why I'm getting really sad right now because I've talked to some of these guys and they're dumb, dumb. I gotta be honest with you. I think a dumb, dumb can understand you make AI smarter than us. It goes poorly. No, they're not. They're like, what stock do I trade to make them? Yeah, seriously. They stand for nothing. If we had old Lindsey Graham here, maybe we could do something about it. Yeah, back from the dad button. So to vote for Tucker, RIP Lindsey Graham. Thank you Nate. Pleasure. That was a part of it. I don't know. Meet new groupie setting mist from Mabelie, New York. Jelly mist technology locks in your look for up to 24 hours with flexible all day comfy grip. No tightness, no stickiness, no residue, just plump, dewy, hydrated skin that still feels like your skin. Try new groupie setting mist from Mabelie, New York. Maybe it's Mabelie.