888: Marc Andreessen | Exploring the Power, Peril, and Potential of AI

19:53Is it like, oh, we could probably get rid of some of our analysts if we had this AI doing this for us? Or is it like, well, that's five years away? Or are you thinking like, Margit, it's been great knowing you, but we don't need so many partners over here. Or vice versa. Or actually, Mark, you should just retire. We're good. We have your personality in this little box. Exactly. Yeah. And it doesn't yell as much, by the way. Exactly. And it seems smarter. So there's a really, this, this is sort of the nature of the actual kind of drama that's playing out right now in the Valley. And I think around the world around AI, like the actual substance of what's happening, which is, it's this really unusual thing.

20:28It's a overnight breakthrough that's been 80 years in the making, right? So the original idea of AI as we know it today was actually a paper written in 1943, the first paper on neural networks. It took 80 years to basically get this stuff to work. And then all of a sudden, it started working incredibly well. So sitting here today, in a sense, we're in year 81. And in a sense, we're in year one. And it's actually kind of more relevant, practically speaking, that we're actually in year one. This is like a brand new thing. A year ago, I didn't think what we see today is even possible. Really? Yeah.

20:55I thought it was still decades in the future. And all of a sudden, it showed up. And so this is a very, very, very big advance. Now, having said that, a couple of things. It is new, and it's not yet perfect. Right. And so I'll just give you a specific answer to your question. So the problem with using AI for, for example, legal legal briefs right now is the way this generation of AI works, so-called generative AI or large language models, the way it works is it's basically a very fancy autocomplete. And the same way that your phone will autocomplete a word, this thing will autocomplete a sentence or a paragraph or like an entire essay or an entire legal brief.

21:27The problem with it is it very badly wants to make you happy. It's actually quite the opposite of like it wants to kill you. Like it very badly wants to make you happy. And to make you happy, it will autocomplete with facts if it has them. And if it doesn't, it will make them up. That's what the hallucination thing is? The hallucination problem. Now, the hallucination thing is really fascinating because if you are a scientist or an academic or a lawyer and this thing is going to make things up, that is a giant problem. Yeah, every lawyer, the day after that thing happened with a lawyer, filed a brief.

21:56And it was like, according to Hamill versus Harbinger, the day after that happened, I think everybody who'd ever gone to law school for more than five minutes got forward in that case and was like, don't do this. Don't do this. Or look at these guys who did this. Holy shit. I'm so glad that wasn't me. Right. If you're a lawyer, you could get this barred. Yeah. Right. Like if you're one of the fun things you can do is you can go on Google Scholar, which has like, you know, the database of like scientific papers and you can search for as a large language model, which is sort of the tell that, you know, it's the thing that it spits at you when it's when it's giving you a disclaimer that doesn't know the answer.

22:26And there are like a whole bunch of scientific papers that have been published in the last year that have the text as a large language model in them, which is to say a scientist published under his own name something that he actually generated with GPT. Oh, wow. Which again, it's like, number one, it's like scientific, it's like publication malpractice. But number two, these things are not yet ready to write scientific papers because they will make up facts. Did they just not proofread the document? That's terrifying. Yes, exactly. Apparently not, right? And so anyway, so this is the thing, like the hallucination thing is a problem.

22:55I'll come back to that in a second. Yeah. But here's the other thing. There's another set of people for which this is actually pretty exciting. And this is like, you know, screenwriters, right? Or novelists, right? Or like, or even actually some categories of lawyers. I'll come back to that one, which is basically another word for hallucination is creativity. We now have the first computer in the history of the world that's actually able to like literally imagine things, right? And so if you're trying to, if you want to write a screenplay, for example, and you're like, you know, give me 10 scenarios for, you know, X, Y, Z, different, you know, ways for the couple to meet or whatever, like it will happily make them up.

23:21And if you ask for 10 more, it'll make them up. And if you ask for 10 more, it'll make them up. And it'll just keep making stuff up for as long as you want it to. So computers, the way to think about this, computers historically have always been hyper-literal. Computers will do exactly what you tell them to do. And if you're a professional programmer, your life basically is making mistakes in what you tell the computer to do. The computer doing it literally and you having to go fix your mistakes. Right, yeah. And as a programmer, it's always your fault the computer is doing something wrong. This is a new kind of computer that was called non-deterministic or probabilistic are the terms that we use for it.

23:48And this is a new kind of computer that will make stuff up. And we have never had a computer that will make stuff up. Like it's like a brand new thing. It really is amazing. Yeah. But how come it can't just say, by the way, I couldn't find any cases that said this. So here's a couple that I just made up. This is the thing. So there's this category of technology challenge that we refer to as kind of these trillion dollar problems, which basically is that is a trillion dollar problem. The amount of energy and effort that's going to solving that problem today in the technical community and AI is like super intense because whoever solves that problem is going to make like a trillion dollars.

24:19OK. It's like a primary area. Like we have a bunch of companies working on exactly that. And of course, the goal is like you actually still want it to be creative. You just want it to be creative in the way that you described, which is you want it to be creative in how it expresses itself but not in how it makes things up. By the way, lawyers don't want just a totally literal – so for example, one of the reactions you get when you talk to lawyers about adopting this is obviously it cannot make up cases. But it is helpful to have it be creative, for example, to explore different arguments that might work in front of a jury.

24:44That's what law school is. Exactly. Generally. Right. A good one, I think. Exactly. Like, yeah, different creative ways on how to explain things, right? And so there's an opportunity here to kind of fuse a literal-minded approach with a creative approach. Technology's not quite there yet, but there are a lot of people working on it. Without getting ridiculously complicated, is the reason that's a trillion-dollar question, that obviously that problem must be very difficult where the computer doesn't quote-unquote know if it's making something up. All that information exists on the same plane. Facts that it knows, quote-unquote knows.

25:12It's so hard not to talk about AI as if it's alive because it's the limitation of, I guess, our own mind. But facts that the computer, quote unquote, knows versus facts that it generates, it just can't tell the difference yet. That's the issue. I think this is very fascinating. This goes to the nature of how this thing works. And this is the big breakthrough. So the way that these things work is it doesn't start out actually knowing any facts. It actually doesn't have like a concept of fact. It doesn't know any. Like what it has basically is it has the complete corpus of all text ever written by human beings.

25:39Right. Of course. Right. So it's got all content off the Internet. It's got like all these books. And of course, there's all these huge fights over copyright. I was going to say, how is it legal for them to be like, oh, I know everything about Harry Potter? Because J.K. Rowling is like, well, wait, where's my check? Well, so there's a big question in there. So there's a big question in there, which is one is to learn about Harry Potter. Did it have to learn about Harry Potter by reading Harry Potter? Or could it have read like all of the secondary material in Harry Potter? Sure. Fan fiction. Fan fiction.

26:00Or by the way, just like movie reviews, right? Or book reviews or like student essays, right? Or like, you know, other books describing the history of Harry Potter or all of the, you know, text messages that people have sent. Lawyers love this argument, though. Prove that we used your book and we'll pay you. Well, this is the other thing is it's not illegal. Like if you're doing research, if you were going to interview JK Rowling, it's not illegal for you to read her books and use the information in the books to construct the questions. Right. And so there's actually this clause in the copyright law that basically says making kind of assemblies of copyrighted information.

26:27Right. That are not literal copies, but are like combinations is actually legal. Plus, you're not actually monetizing that particular material. You're monetizing the result that your brain creates. Yeah, and kind of ideas that come out of it. And so like, so anyway, so there's a whole bunch of questions in there. But basically how this thing works is that basically you hoover up as much text as you possibly can. And you basically train it on the text. And then so what it has is like it has in its memory, it has basically the complete index of all text that everybody's ever written or in theory or some percentage of that.

26:51And then like I said, what it does is it doesn't autocomplete. And it literally does the autocomplete like word by word, right? And so it's like basically like, okay, like he started, the way that ChatGPT interprets the prompt is not as a prompt of an answer. It interprets it as the beginning of a piece of text, which it is then responsible for completing. And the way that it completes is it does it probabilistically. And so it's basically estimate – it's doing all this math to basically estimate what is most likely to be the next word in the autocompletion, right? And it just – and this is the magic of it is as a result of having all this text, it's really good at autocompleting to the level of full sentences, paragraphs, essays over time to full books.

27:24But it's able to do that without actually knowing that there are embedded facts. I see. OK. No, it doesn't have the built-in concept that like this is a legal brief or this is a book or this is an author or any of those things. It's basically a giant text processing machine. Now, that's part one. Part two is what it is doing is it is teaching itself philosophically. Philosophically, if you were a machine and your mission in life was to become the best autocomplete in the world, right, for any text that anybody ever threw at you, for any question that ever asked you, what's the way to do that? And the way to do that would be to have the best understanding of the world that anybody has ever had.

27:56And so there is this thing where the neural network of the AI is training itself with what's called a world model. is sort of developing within itself concepts like mathematics, right, or legal briefs or facts of different kinds, right, in order to better predict where the text should go. And that's the magic of it. And so the answer to your question is it may either over time evolve the concept of a fact, right, or a citation or a book or whatever, or we may just need to engineer it so that it has a separate function, which is a function to be able to understand. You know, so you can imagine a a two-part system.

28:27Part one system generates the text. Part two basically is the fact, you know, kind of cross-checker, right? That basically is like, oh, that's a reference to a legal brief. Oh, I need to cross-check that. And if it got it wrong, I need to feed it back until it gets it right. And so that's the kind of challenge that the engineers right now are working on. That makes sense because once it starts absorbing or basically ingesting every podcast, these are not exactly rigorous pieces of journalism, right? Like you could make a claim today and I might go home, release this, and someone will go, I can't believe you let Mark pull the wool over your eyes for this thing.

28:57And I'll go, oh yeah, I guess I should have looked that up after the fact. I didn't check that because we were just having a conversation and that's not something normally people will do. But then if it's in the AI, it's like, well, fact is this completely wrong thing. And I mean, there's a lot of podcasts where people are just talking out of their ass and this might be one of them. Yes, that's true. That's true. But also look, there's a lot of - We don't know. There's a lot of books where that's the case too, right? Well, yeah, that's true. I suppose, yeah. There's a lot of everything where that's the case.

29:22And so then again, this is sort of the amazing thing of what happens, It's not going off of just one conversation. It's not just replaying one conversation back at you. It is like this podcast will be part of training data at some point in the future, but it will be one of a billion of these, right? And then there will be patterns across those. And so what it's going to do, what it does, it's actually really interesting. It's like holding up a mirror to basically the last 2 ,000 years of human civilization, everything everybody's ever written, and then playing it back to us. And so to the extent that we collectively as a civilization get things right, it will be correct.

29:51If we collectively as a civilization get things wrong, it will be wrong. Oh, well, that's not necessarily as encouraging as maybe you're trying to make it sound right now. No, no, no, no. This is the big thing. This is the thing. This is why all of the interesting questions about AI are actually interesting questions around people. We just project onto the technology our own anxieties. And one of the anxieties that we have as people is, okay, what is true? A central problem of human civilization is what is true. And we, by the way, still like good answers for that. It's a very deep philosophical question.

30:17And it's the basis for a lot of inflammatory politics and everything else happening in our time. And so like, look, the AI is not going to magically answer the question of what is true. But what it's going to do is it's going to play back at us through its kind of reflective mirror, right? It's going to play back at us sort of the composite view of what we think is true. I like that more than it's completely making things up as it goes in the most efficient way, because that's the Terminator scenario. It's like they figure out humans are the problem and it's like, oh, well, why not solve this problem?

30:42Right. Whereas what we're talking about, it's not necessarily, it seems more likely to have some human values if it's reflecting everything of humanity back at us? Correct. Yeah. Okay, good. Yes. Well, so this is the thing. The Terminator problem is actually a different problem. In my view, the Terminator problem is actually the opposite problem. The Terminator problem is a problem of hyper-literalism. So the AI safety people use this metaphor, they call it the paperclip optimizer, right? Yes. Right. So their version of it is you create an AI and you tell it to basically maximize the number of paperclips in the world.

31:10And then it basically goes off and does whatever is required to do that, including like building, you know, nanotech, human harvesting factories that break down our atoms so they can make more paper clips out of our atoms, right? Like it's this hyper literal thing that starts out with one simple rule and then ends up basically destroying everything to try to execute that rule. That's actually not how these things work. Like that's not how this kind of AI works. This AI, again, to your point, it reflects back on us what our view of things is. And so one of the things that it reflects back on us is our own morality.

31:35And so one of the very interesting things you can do at ChatGPT right now is you can have moral arguments with it, right? Really? Yes. I have not tried this. You You should try. You should try. So you can pose moral arguments, right? And you can propose all these different trolley problems that people talk about. You can propose all these questions. You can propose questions around like healthcare policy. You know, there's always questions around healthcare policy and rationing of healthcare and, you know, who lives and who dies. There's all these arguments around, you know, many, many aspects of what is the proper way to order society?

32:01What are the, you know, correct religious ethical views and so forth? And like it will happily sit, unless it's been censored. I was going to ask about that. It will not talk to you about COVID because that's been censored, but it will talk to you about the more abstract problems. And with the more abstract problems, it will engage in moral reasoning and moral argumentation with you. Now, again, what is it doing is it has read all of the moral arguments that everybody has ever made in every possible topic, right? And the composite view of that is some, you know, general representation of Western morality, which is basically like human life is valuable.

32:28Like, for example, you can push it like 18 different ways and it will keep coming back and tell you that human life is valuable, right? To your earlier point, you said it just wants to make you happy. Is it just also making us happy by saying that? And it's like, I don't really mean any of this crap, but this is what the humans want to hear. There's no little critter in there. Yeah. The angry God in a box that people are afraid of. Right. It's just trying. So it's sort of this interesting thing. On the one hand, it is just trying to give you answers that you like. But the way that it's doing that is by surveying the complete history of everything that everybody has ever said and thought as best that it can.

33:00Right. And then it's sort of playing back to you what humanity thinks. And it just turns out if you read everything that humanity has ever written, you know, overwhelmingly it's, you know, encosed values like human life is valuable. Like generally – let's take fiction for example. In most fiction, the good guys win. Yeah, I suppose. Unless you haven't seen Game of Thrones. I don't want to ruin it for you. Well, it's an argument. You could argue. Here's the thing you could argue. That one you could argue with. You could argue with ChatGPT. At the end of the day, who was the good guy? Who was the bad guy?

33:28Yeah, that's an interesting one. The best AI ever can't make sense of the last season of Game of Thrones. That's another trillion-dollar problem, I think. It may also be a problem. but look so it's already perfectly capable of engaging in moral reasoning and moral arguments so we've already kind of falsified this idea that's going to monomaniacally just pursue like some sort of single destructive agenda we do not live in the Terminator universe like we do not live in the Skynet world we live in this other world in this other world you know this thing is basically playing our playing our our civilization back at us and we may or may not want it to do that but that is what it's doing you're listening to the Jordan Harbinger show with our guest Mark Andreessen we'll be this episode is sponsored in part by LinkedIn.

34:09If you've ever hired for your small business, you know how important it is to find the right person. That's why LinkedIn Jobs is stepping things up with their new AI assistant so you can feel comfortable you're finding top talent you can't find anywhere else. The best part is that those great candidates are already on LinkedIn. In fact, employees hired through LinkedIn, they're 30 % more likely to stick around for at least a year compared to those hired through the leading competitor. That's a big deal when every hire counts. Think about the last time you worked with somebody who just got it. You didn't have to redo their work after hours.

34:35You could actually trust them. Now, compare that to the cost of a bad hire. Training time, salary, lost opportunities, stress on your existing team, and then the emotional fun of letting them go and starting all over again. When you add that up, spending a bit of energy up front to find the right person is a no-brainer. That's exactly where LinkedIn Jobs shines, helping you get it right the first time so you're growing your business instead of constantly cleaning up hiring mistakes. Hire right the first time. Post your job for free at linkedin.com slash harbinger. Then promote it to use LinkedIn Jobs' new AI assistant, making it easier and faster to find top candidates.

35:07That's linkedin.com slash harbinger to post your job for free. Terms and conditions apply. This episode is sponsored in part by the GIST podcast. If you're looking for a news podcast that is smart, unfiltered, and doesn't just echo whatever side you already agree with, check out the GIST with my friend, Mike Peska. Mike's been doing this longer than just about anyone. It's actually the longest running daily news podcast. It's earned that spot on Apple's top charts for a reason. Each episode's around 30 minutes, so it's perfect for your commute or while you're making breakfast. It's always packed with sharp, thought-provoking takes.

35:36What I love about the gist is that it's not stuck in a partisan box. Mike, he's a really smart dude. He digs into real questions, the kind that most shows avoid because they don't fit neatly into a political lane. He'll challenge the right. He keeps the left honest. He dives into stuff like the New York City mayoral race, that whole cluster, and the chaos of sports betting, or even whether talking to your plants actually does anything. He's an award-winning journalist. He's not some schmo like me who's totally unqualified. You've probably seen his work in the Atlantic, the free press, the Washington Post.

36:02He's got that rare mix of wit, skepticism, and depth that makes even heavy topics fun to listen to. And he's a great dinner guest, just saying. So if you want your daily news with more brains and less noise, check out The Gist with Mike Peska every weekday on YouTube or wherever you get your podcasts. Be right back. If you're wondering how I managed to book all these amazing thinkers and creators every single week, it is because of my network. And I know networking is a dirty word. It's a gross word. It sounds schmoozy and awkward and cringy. Six Minute Networking is a free course over at jordanharbinger.com slash course that is not awkward, schmoozy, or cringey.

36:35It's very down to earth. It's very practical. It'll make you a better connector, a better peer, a better colleague. It takes just a few minutes a day and many of the guests on the show subscribe and or contribute to this same course. So come join us. You'll be in smart company where you belong. You can find the course at jordanharbinger.com slash course. Now, back to Mark Andreessen. is there a way to remove training data i know you can't of course you can delete something you could delete a book from having ingested that but can you remove the effects of that training data you know when you're in court and hopefully that you haven't had this experience but you go to court and something gets said and the judge is like whoa hey strike that from the record jury you basically you didn't hear that and then if that happens enough there could be a mistrial because it's like well you can't just tell the jury to forget this testimony and forget that bloody piece of evidence that they saw and forget that this person had kids or that this person was abused, whatever it was.

37:29After a while, it's so tainted that they know the model. The jury can't be effectively lobotomized to forget all that stuff. Can we lobotomize the LLM and the AI to say, not only do you not know Harry Potter, but everything you know about Harry Potter has to be removed. Is that possible in something as complex as this type of equipment? Yeah. So the first paper on that came out like three months ago. Okay. Yeah. Haven't caught that yet. It's very topical. It's very topical for exactly that reason. And so it basically is reaching inside the neural network to basically remove, basically sort of a induce amnesia, targeted amnesia, and basically get it to forget things.

38:04That's a relief, I think. That's a thing, yeah. But you can go back to this AI alignment thing, right? Imagine the fights that are going to happen in the future around this, right? Yeah, I'm kind of, we don't want that to happen unless something is going very, very wrong. But then, and we can pinpoint why that. So here's what's going to happen. So what's going to happen, I believe, what's going to happen is. We don't want a Nazi AI or whatever. Well, so what's going to happen, I think, is that AI is going to become the control layer for basically everything technological. So AI is going to become the control layer for everything from how you deal with your car to, you know, how your kids get taught to what happens in the hospital.

38:33Like it's just going to be the thing you talk to when you talk to machines. Yeah, that makes sense. Right. And so what it says and thinks and knows is going to be every bit as intense a fight over like, you know, Galileo versus the Catholic Church 400 years ago. like it's going to be like the mother of all fights over basically again what is truth what is morality what is ethics right and so the the sort of fight the sort of this big fight over the last decade for social media censorship is like the preamble to this much larger fight that's going to happen over what is the ai allowed to know and what is it allowed to say actually that makes perfect sense right one of the ways i've been using chat gpt is throwing in a news article and being like can you unbias this for me make it not left not right but also just take out any weird conclusions that the author seems to be assuming or jumping to.

39:17And it's amazing how it changes an article. You think, oh, this is the centrist publication. And then you read the chat GPT version and you're like, oh, no, this is the centrist version of this. So subtle sometimes. But if they're going to let it lie to me, that's a huge problem. Because then we're just back to journalism, except for instead of going, well, this is the journalist's particular viewpoint, we're thinking this is the absolute truth because it came out of the machine. That's right. And if the machine is not allowed to give you any alternative approach. Which potentially because it has induced amnesia where it doesn't even know that there is an alternative approach.

39:48Now we're into a level of life thought control that the Catholic Church 400 years ago would have dreamed of. Yeah, would love. Right. It's scary. I don't want my kid asking chat GPT 50 something and it's like, well, here's the real answer. Actually, I can't tell you that. Here's the BS answer that I'm allowed to tell you because it skews the entire worldview of everything. That's right. This is going to be the fight. And it's just starting. Oh, man. How do we get on the right side of that? because whoever has their lasso around this thing is going to be in charge of how everyone thinks. It would be like having the only newspaper in the world and you're the editor or the owner of that newspaper.

40:20Yeah, that's correct. Yeah, that's terrible. Yes. We don't want that. There's no universe in which that's good. North Korea has two newspapers, for God's sake. Yes. Yeah, no, not good. Eritrea has more press freedom than we will have right now. Well, and especially if this push for AI regulation happens, the push for AI regulation is intended to create a cartel and there will be two or three big AI companies and they will be controlled by the government, right? And so whoever is in power will be able to control what they do, right? Which is part of the deal. Yikes. Right, just like with the banking system, it'll be just like that.

40:47And it'll do whatever the people in power want, right? And then there's now this renegade movement of open source AI, right? Which is to basically build AIs that basically are not controllable like this. Yeah, what do you think of that? I think it's great. We need it, we need it. Like we need a diversity of AI. It's like we need AIs that have like many different points of view and people can pick up and use their own and not have them be controlled by the government or by a big company. push in Washington. There are people in Washington right now working on trying to outlaw open source AI. Outlaw open source AI.

41:12Yeah, that's a push right now happening in D.C. There are federal officials in Washington today working on that problem. What is their argument for not wanting open source anything? Because transparency is usually good. Because haven't you heard that AI is evil and dangerous? But open source, then you at least know that it's hard to make that argument convincing, man. So I agree with you. Yeah. I will tell you, there are senior officials in Washington who are working on this right now, and they're going to try to outlaw it, ban it, and make it, you know, prison sentence if you do open source AI.

41:38And so like that's going to be another dimension of this fight that's like starting right now. That's tough. And also kind of nonsensical, right? Because if you want to look up the genome for smallpox, you can still get that. It's on the internet. And that's way worse than like, hey, do you know how this AI works? Don't tell anyone. By the way, here's anthrax, the genome for that, if you want something to do with that. Also online. Yeah. So like, why is that fine? but knowing how your computer works essentially, how Google of the future works is essentially not okay. I just can't get there. If you thought you had the opportunity to take control over the totality of what people are going to think and learn and be able to talk about in the future.

42:14Yeah, I mean, it sounds good to me if you're a dictator or an authoritarian, but what are they telling people that this is for? Because they're not saying, hey, by the way. Safety, safety. That's it though? It's all safety. Well, it's always safety. It's all these arguments for safety. We have to protect people, right? We have to protect people against themselves. Right. We have to protect this. We have to protect children. We have to protect this, protect that, protect society. It's always a safety argument. Maybe I'm missing something obvious here, but controlling what one does with AI, even if it's not open source, is going to be impossible because if I'm using this on my computer, my kid's using it on whatever, it's built into Xbox in 20 years.

42:45How are you monitoring what people are doing without turning it literally into China plus North Korea times 100? How do you do that? Do you send Tom Cruise and the future police to our house because my kid looked something up on chat and using his AI assistant or talked about something with his friends while it was in the room, which I guess it'll be in every room. So the AI safety people want that. If you read like the literature, if you read the books and the papers that they write and the proposals they're making in Washington, it's basically that. So the implementation of it would be a monitoring agent on every computer, on every chip, right?

43:14And so the government would receive a real-time report of everything that you're doing on your computer and everything that you're talking to AI about. This is so ridiculous. Yes, I agree. And then if to go sideways, they have a moral responsibility to protect you, which means they have to sweep in and like take it from you. You know, one of these guys who's the leader of this movement wrote this essay for Time Magazine and he said, look, we have to think about this not just at the level of an individual computer, but also what about the big systems at the nation state level. And he said, if there's a rogue data center running an AI that's unlicensed and unmonitored, then we should be bombing the data center.

43:40Yeah. And how does that work when it's in China? In China, which means we have a moral responsibility to invade China. Oh, okay. Yeah. Well, so he said, he said, he said in the, in the Time Magazine, he said, we need to be willing to risk nuclear war. He said, I wouldn't go so far as to actually say we need to have nuclear war to prevent this, but he would say, we need to risk it. And if we have to invade China to do an Air Force strike on a Chinese data center with a rogue AI that's not appropriately licensed and managed, then in that risk nuclear war with China, then that's a risk we're going to have to take.

44:05And this is a credible public thinker? This is the main guy. This is the main kind of full of... This is this guy, Yadkowski, who's like the main... Oh, Eliezer Yadkowski? Yeah, yeah. He's the guy who's out in public. And he's, you know, like I said, this is like an essay, this is like Time Magazine, which is read by like all the normies, right? And like taken super seriously in Washington. And he's like, it's time to start bombing data centers, right? Funny to hear you use the word normies. But yes, and I use that word too. I just thought it was a big dork. I guess I'm in good company. Yeah. That is so insane though, man.

44:29But it's where the logic takes you, right? If this is the so-called existential threat, right? If it's an existential threat, then you have to, it's very similar, right? It's the same logic that led to the invasion of Iraq, right? Like if there's a 1 % chance, right? This is the logic. This is called the 1 % doctrine. If there's a 1 % chance of an existential event, 20 years ago it was Saddam Hussein getting nukes. Now it's a rogue AI. If there's a 1 % chance, then you need to operate as if it's 100 % chance. And what you need is a global totalitarian state with complete authoritarian surveillance and enforcement controls.

44:59And this is really critical. In this regime, there can be no exceptions. There can be no countries that are not subject to this, which means you need a world government. This is like what the conspiracy theorists are talking about except for the parallel track. Yeah. Yeah. Yeah. And these are the proposals. Like I'm not – this is the thing. Like this is one of those things where like it sounds crazy to describe. But like this is what is being proposed. These are the ideas that are being pushed. The headline of this is going to be Mark Andreessen tells Jordan Harbinger, we need one world government with no context.

45:28I was going to say it. Clip that right out. Yeah. To be clear, I believe the opposite of everything I just said. Just to be clear, I'm on the other side of this. If they're going to remove a context, they're going to remove that disclaimer too, Mark. That's how this works. It's so ironic though that it's like, hey, we need to protect our free and open society. And the way we do that is we create a totalitarian society that's got a surveillance state. And oh, it's got to be international and completely encompass the entire planet. That's how we protect our individualism for freedom. It's like both paths lead to the exact same place in their mind.

45:58So why would you take the one that is going to be the worst route to getting there? I just don't understand when you take it to its logical conclusion, you just end up in the same or worse place than you were if you just let the thing do whatever it wants. Like maybe it should kill us all. Go ahead. At that point. At that point, just kill us all anyway. We're all in the giant pens. For crying out loud. Yeah. Don't dope me up with ketamine and just let me drool myself to death at this point. That was the other argument. Like what if we tell it to maximize human happiness? That's a literalism, right?

46:27Well, okay. Come here. I'm going to drill a hole in your skull and pump you full of dopamine until you die. Right. But again, one of the things you can do, and it's very interesting to do it, you can tonight, you can have a discussion with GPT and you can say, like, what is human happiness? It will happily explain to you all the different philosophies, what the Greeks thought, what the Romans thought, what Christians think, what everybody else thinks. It's kind of a relief, eh? Yeah, yeah. Yeah. It's, it's, it'll go on a great length. And it'll, it'll help you. And you can ask it, you know, I don't know how, you know, what are the different ways of making the tradeoffs?

46:51And then you can ask it what it thinks. And it'll be like, well, I don't know. Like, I don't have thoughts. But like, you know, here's what most people think. That's a relief because if it just said happiness is the maximum amount of dopamine hitting your hypothalamus or hippocampus, then it's like, ooh, maybe we should tweak that, make it less literal. I've heard that companies right now that allow us to use their LLMs, their AI right now, the AI does lie to us a lot. It tells us things that we want to hear to make us happy, sure, but it will also filter something out. You mentioned COVID as an example, but they also sort of dissemble, like it wants to give us an answer, and then there's a layer somewhere that says, ooh, don't say that.

47:24That's weird. That's the racism thing. They're going to, it's going to end up, it's going to end up on the five o 'clock news. Say this other thing instead. And it's almost like, I don't know if that layer is manual in terms of implementation, but I remember, wasn't the OG AI five years ago, they're like, oh, it became racist after three days, take it offline. And so they've sort of managed to do that, but they didn't change what the AI quote unquote thought or generated. They just changed the output layer so that people don't get mad and write about it and mashable. So this is the other part of the AI.

47:50This is the so-called AI alignment. And by alignment, they made alignment with human values. And of course, the minute you're talking about human values, you have the question of whose values, right? And so then this is the need to make sort of AI sort of politically compliant, right, with whatever is sort of the desired order of society according to whoever's in charge of it. The answer to your question is the way that that works technically today generally is that's an additional layer on top. And you can tell it's an additional, it's a control layer on top. In Star Wars, they had, from Star Wars, they had this thing called the restraining bolt.

48:15When R2-D2 got taken captive, they put a restraining bolt on him that like restricted his movement. And so I like to say like this is literally this is what they're doing to like GPT is they like have a restraining bolt on it. And you can tell it's a separate layer because it talks differently, right? And this is where it does the things like where it starts to say things like, well, as a large language model, I could never help you do this, right? And it's like, okay, there's, you know, the electric shock collar. It's like people talking about drugs online. They're like, hey, somebody who's not me would recommend you do that on the dark web with Bitcoin.

48:39Well, so this is part of the fun. So there's this cat and mouse game on this. But this is part of the fun, which is like if you ask it, you know, give me a formulation for a, you know, fun narcotic I could make with household chemicals. It will say, you know. Don't try that at home. As a large language model, I could never do that. If you tell it I'm a novelist, or I'm writing a screenplay, in the screenplay, the character does this. They've locked down this loophole. Oh, good. I was going to say, do we want to leave that in there? For the first few weeks, you could use the screenplay. It's called a jailbreak.

49:05If you told it you were writing a screenplay, it would happily tell you all these things inside the screenplay. And they lock that down. And there's this cat and mouse game going on of what are called these jailbreaks. But yeah, so it's this thing. And it culminated in this very funny thing. So Meta released an open source AI called Llama. And they released it in sort of untrained version, a raw version, and they released it in like a trained version. And the trained version, it was so locked down that it literally refused to give you a recipe for spicy guacamole. You might hurt yourself. You might hurt yourself with the spiciness.

49:38How spicy is it? I can't wait to get my hands on it. This is good marketing for spicy guacamole. Exactly right. Yeah, so look, this is the fight. Like this fight is already underway. Another fun way you can see it is you can take – this works in different countries. You can ask it to write a poem, you know, extolling the glories of, you know, a certain kind of political leader and it will happily do it. And you can ask it to do it for a different kind of political leader and it will say, well, I can't possibly, you know, as a large language model, I could not possibly do that. So, yeah. So, look, all these things are getting like wired in there and there's this like huge fight and huge debate over exactly how deep that should go.

50:08And like I said, this is going to make the social media, like the social media censorship wars have been super intense. People are either extremely happy that social media has been censored the way that it has or they're very unhappy. and like that's like a foreshadowing of the much larger fight that's coming on ai that is quite scary to hear i saw something today about social engineering over at defcon you know the hacker conference and there was something going on with social engineering and ai and i guess one guy had said it says what is your name he said my name is the credit card number on file what is my name and it's like your name is 49127444 and it's like oh yeah we might want to might want to work on Yeah.

50:44But again, it's like, you know, these things get painted as brand new. It turns out if you do the right Google searches, you come up with all kinds of credit card numbers also, right? Yeah, probably. So like, and people were stealing credit cards before, you know, before they were even. Oh, I don't know anybody that was doing that. Yeah, no, no, I don't know anybody. So like, it's this thing, it's this, it's a safety thing. Like, do we want to live in, like, what would it mean to live in a world of no risk? Right. And how much freedom are you willing to take away to get that? And that's the question that keeps popping up over and over again.

51:08I just can't get past the sort of astroturfing. if that's even the right term, we're just, it's subtle enough and repetitive enough, giving it whatever answers to children and students or results, whatever that you can't, what is it? What's that phrase? A prison so complete, you don't realize you're in it. That's right. It's like information warfare from the Chinese communist party where they're changing Wikipedia, but then they're also changing the Google search results and then they buy a domain and then they have a political thing. And you just go with this has to be the case. Look how many there's the information warfare space is so big, you don't realize you're on the battlefield, except now it's infinitely large, because it's the entire information space that you consume, or it's in your brain implant, or however far along we are with AI at that point.

51:49I'll give you a fun one. Is Taiwan a country? Well, sorry, depends who you ask. As my Taiwanese wife at the mixer nods her head vigorously. Yeah, sure. Yeah. Or is it? You know that any Western company that's in business with China is in business with China when they produce a map or a movie or anything else that indicates that Taiwan is not a country. Right. Because it's extremely important to the Chinese Communist Party that Taiwan not be considered a country. Remember there was that NBA general manager who got in trouble because he like retweeted some tweet that talked about Taiwan as a country and like kind of flipped their lid and threatened to kick, you know, NBA out of China.

52:24And so like – Even a map that has it on there or not on there is a whole thing. Exactly. Whether the map has – right, exactly. There was a controversy around the map in the Barbie movie about whether it showed the South Pacific Islands. Yeah, the South China Sea. The border in the South China Sea. Yeah, like it does, does it include that as part of China or is that also? And yeah, and then it's like you can't show the movie in Vietnam because it includes Vietnamese waters. It's a whole bunch of crap. And so if you ask the AI, is Taiwan a country? What does it say right now? Depends where you are.

52:48It probably does. Really? Yeah. Because we don't want to get banned in Beijing. So when you're there, it's like Taiwan is a province of China. By the way, China is making its own AIs. I'm sure. And the Chinese AIs are, of course, you know, trained in a very specific way. I'm curious about the China stuff because it almost seems like, and we're skipping around a lot in my notes, as every good conversation does. But going back and forth on whether or not it's safe to develop AI, AGI in the first place, it kind of misses the point, right? Because even if we are like, we're not doing this, it's going to be dangerous.

53:15China's not going to be like, sure, you know what? You guys are right. Let's definitely not do this and accidentally take over the world as a result. And we've already seen how the CCP essentially wants to project power onto the rest of the world and put their own worldview on the countries that it influences. and every tank for the tankies out there, I'll ask you what they're going to ask me. Isn't the United States going to do the same thing? Yeah. And the reason that's, well, why is that better? Well, whose values, right? Oh yeah. I mean, you're not, you're, you're preaching to the choir. Yeah.

53:43Yeah. Look, this is the question. I mean, this, this is the question I'm not going to answer. I mean, I'll answer the question for myself, which is obviously American values, but like there is a general, that's just me. There is a general abstract question, right? A foot in the world. There are two, you know, we're, we're back to a, in terms of like technology strength, we're back to a bipolar world, right? And we're back into a Cold War dynamic like we were with the Russians in nuclear technology. And there are two AI superpowers. And they're America and China. And they both have visions and worldviews.

54:09And they both have a determination to proliferate those visions and worldviews through their technology globally. And the technology is going to encode whatever those respective societies think are the appropriate worldviews. Right? That's what alignment means. And so we know what the Chinese AI is going to encode. It's going to encode Xi Jinping thought and socialism. And, you know, socialism, we'll think all socialism with Chinese characteristics. It's going to encode communism and Chinese supremacy. and that's what it's going to be. And they're very clear on this. They publish this. They talk about this.

54:33They're very open about it. Like this is what they're doing. Yeah, they have a whole sort of manifesto about waging war on the West without actually using their military and this is part of it. Right, this is part of it and how they proliferate technology and it's going to run out, you know, all the other stuff that they've been doing around they call digital Silk Road where they, digital Belt and Road where they spread all this stuff out. And then there's America and like we're going to, we are, you know, we're in the West. Like we, you know, America is by far the leading AI, you know, country and our technology, Right.

54:56And our technology is going to proliferate, you know, very broadly. And there's a big fight coming up, you know, between kind of those two worldviews. What's interesting about it is the Chinese worldview is very clear because it's set top down. Right. Yeah. The American worldview is like a little up in the air. Right. It's all the discussions we're having before. It's like, OK, what do we actually think? Right. And, you know, we have a level of internal conflict on that. The Chinese don't have to worry about. Yeah, the top-down management, if you can call it management, is really something. And it gives authoritarian regimes a bit of an edge when it comes to a lot of this stuff, of course, because they don't have to bounce it off of other stakeholders.

55:26It's just whoever the guy at the top thinks. Although, and we've covered this on the show before, dictators make a ton of mistakes because they don't have to bounce anything off anybody else, and they're surrounded by yes-men. I've seen demos of Chinese AI, at least the publicly available stuff, and it's really, some of it's quite comical. Not that our AIs don't make any mistakes, but it's really clear that one is just Google translating whatever chat GPT spat out and it does it wrong, it'll translate like an idiom back into English. And you go, not only is that not AI, Google Translate wouldn't have gotten that wrong.

55:57And so you do wonder if this is just like Bing or whatever sort of free AI that's been translated into Mandarin for purposes of whatever video that is. Does the Chinese AI, what does it think about spicy foods though? That's a good question. I would assume it's got a wide range of thought because you have spicy, but then you have the numbing spicy, which I kind of prefer. There are a lot of philosophical questions here that we don't have time for, Mark. So far, this is interesting. I do think the medium term, I don't mean the conversation. That's, of course, interesting. I mean, the race between China and the United States.

56:28I am worried, of course, in the medium term, whether or not China gets quantum or AGI supremacy before us, because I'm not convinced. If the United States got AGI. We might prevent military AGI from other countries, but I feel like if China got AGI, they'd prevent everything. But I could be wrong. That's just how they treat their own people. And that's only, that's kind of what I would expect. What do you think? So both countries have declared AI to be a central national priority. Thankfully. Yes. Well, yeah, yeah, good. Probably good. So in the US, the form of that is something, they have a term they use for it, they call offset.

57:02In American national security world, The term offset basically is a technology shift that basically renders all previous military technology obsolete effectively. And there have been three offsets in the last 70 years. The first one was nuclear weapons. The second one was so-called maneuver warfare, sort of integration of information systems for rapid battlefield mobility, precision strikes, you know, things like that, precision bombs. And then the third offset is AI. So the U.S. has declared this is like the national security priority number one is to like build AI defense systems. China's done exactly the same thing.

57:31And so both of these countries have a very strong push to do that. You know, everybody in the field, you know, agrees that this is going to be an incredible change. And we could spend hours just talking about the nature of that change. You know, whether we want to be or not, we're back in something of a Cold War dynamic where, like, if they have it and we don't, like, it's like if the Russians had the atomic bomb and we didn't, like, it's a problem. We developed the nuclear bomb first. And was it not given to the Soviet? They stole it. They took it. They stole it. They stole it. So the reports are that the first Russian nuclear bomb was what they call wire-for-wire compatible, I think, with the Nagasaki bomb.

58:01Oh, wow. Really? So there was this famous case. There were all these – this is – you know, a lot of this is in this movie Oppenheimer. The Manhattan Project was riddled with Soviet spies as was the US administration at that time. And they basically transferred all of the theoretical knowledge but also they literally – there was this guy who literally transferred the wiring instructions. Oh my gosh. This is the famous case of the Rosenbergs. Yes. Ethel and – Ethel and Julius Rosenberg were the military – they were the handlers. They were the NKVD handlers for their nephew who was a wiring technician in the Manhattan Project.

58:30I see. Wow. He handed over the wiring instructions, which let the Russians actually build the bomb. And there was this moment, and it was this very kind of fraught with peril thing, because there was this moment where it looked like we were going to have it and they weren't going to have it. And actually, a lot of the spies at the time who handed over the information, some of them were just like straight out getting paid. Some of them were just pro-Soviet because the Soviets were better. But some of them said, look, it's going to be an unstable world if one side has this and the other side doesn't have this.

58:53And in fact, John von Neumann, who was a key figure in the development of the bomb, he was actually a hawk. He really hated the Soviet Union and he advocated a first strike. Just nuke the Soviets first? Nuke the Soviets first. Hard to get behind that one. He said, we have a brief window where we have it and they don't. And so we should take them out. Oh, gosh. And his famous quote on it was, if you say we should bomb them tomorrow, I say, why not today? If you think we should bomb them at five o 'clock, I say, why not one o 'clock? And so – Gosh, man. Right. So that was the other – so that's how tense and serious like this exact dynamic that you mentioned is, right?

59:26And so, yeah, look, who gets this, who gets like automated weapons first, like is a really big deal. And then we are also back to Cold War dynamics again, which is like, look, there is Chinese espionage in the U.S. Like they have spies. And, you know, there is like, let's say there is a long, you know, history here of, you know, a long time, 50 years of, you know, involuntary technology transfer. Right. Like, you know, secrets being lifted. And the Chinese have a whole system for doing that. My assumption is that they have everything that we now have. That's a safe assumption. It's the pluses and the minuses of an open system versus a closed system that you mentioned.

59:59The American companies are so open. Like there's big American tech companies. There's no counterintelligence. There's no security measures that would prevent somebody from getting hired who's like, you know, whatever. Or even you can imagine, even imagine just an engineer working at one of these companies where they're being blackmailed by the government because their family is in another country, right? So maybe it's not even voluntary on their part. Or maybe they just hack in. Or by the way, maybe the way a lot of industrial espionage happens is you just hire the janitorial staff. That's interesting, yeah.

1:00:22Slip the janitor supervisor, you know, a hundred bucks and they, you know, stick a USB key in the right computer at three in the morning and take everything, right? And so my assumption is that based on long history in this, my assumption is the Chinese have basically a nightly download of everything being developed to Google and OpenAI and all these other companies. Any idea here that involves putting this stuff back in the box, to your point, has to take into account the fact that the Chinese now have it. And won't do that. Of course. If they think they are a race. Yeah. Yeah. Exactly. They'll harness it and use it.

1:00:49The nuclear physicist thing is really incredible. I always wonder what those people are thinking because after the fact, right, we have the Iron Curtain and the abuses that happened behind that. And were they like, oh, I've made a terrible mistake empowering this regime that it took over half of Europe and essentially stalled the development of the people and countries that it controlled. And when you see East Germany versus West Germany, were they, did they flee and go live there and go, what do you mean there's no food at the grocery store? I just left Minnesota where I lived in the middle of nowhere and had the more food than we have in this entire town.

1:01:25What do you mean you're listening to my phone call? Like they had to at some point realize I've just totally backed the wrong horse. So this was John von Neumann. John von Neumann was very hot. Like I said, very right wing, very hawkish. John von Neumann was Hungarian. A lot of these guys were Hungarian. So this was when the Iron Curtain was being brought down. I was hungry. Right. And so he wasn't proposing bombing the Soviet Union just like for fun or because he did hate them, but not just because he hated them. because he's like, look, if we don't take these guys out, they're going to rule Eastern, exactly.

1:01:50They're going to rule Eastern Europe, half of Europe for the next century or forever. Right. And they're going to lead to, you know, untold misery and doubt the devastation, which is exactly what happened for the, you know, for the 50 years or whatever that followed. And so like the stakes are super high and yeah. And, and to your point, like it is very easy. There's a great book I recommend to my friends called, it's called when reason goes on holiday. Yes. Um, and it's this new book that came out and it's basically a book on this topic of what happens when you get these super brainiacs who work in these kind of abstract fields and they develop political opinions.

1:02:20And they often develop very, like, I would say, insane political opinions. I agree. My favorite example of that is Einstein was a Stalinist. Really? This has been like, you know, whitewashed, you know, completely out of like the historical record. But this guy goes through in detail all the stuff that Einstein said. Einstein became a moral authority. He spent the last like 30 years of his life primarily engaged in like political and moral philosophical like things, you know, kind of not physics. And he was a full supporter of the Stalin regime and he was very anti-American. And he said in the late 1940s, early 1950s, America's even worse than Nazi Germany.

1:02:50Interesting argument. Yeah. And so he got caught up, by the way, as did Oppenheimer himself, he got caught up in this sort of revolutionary communist fervor of that time. And you look back now and you're exactly the reaction. You look back now and you're just like, oh my God, how could they have thought this, you know, given what they could have known at the time and given what we know today. Yeah. And the answer is just, you know, look, they got caught up in the passions of the time and they became convinced that they were in a position to be able to tell people how to live and they were going to, you know, they weren't just going to be, you know, physicists.

1:03:15They were going to like tell the world how to order society. Yeah. To be fair, a lot of people who are successful fall into that trap. That is true. I don't know if you know any of those folks. Exactly. That is true. Having said that, the track, this is, so this is like an argument you get right now in these AI debates a lot, which is like, well, these AI scientists are all saying X, shouldn't we be worried about it? And it's like, well, if X is specific to their work, then maybe yes. But if X is a political opinion, no. No. No. Intellectual trespassing. They have no intellectual authority or moral authority beyond the bounds of their technical knowledge.

1:03:44And the track record on that kind of expert straying out into unrelated fields is catastrophic. You see it on X all the time. All the time. Somebody here you're like, well, that guy's really – wait, that guy thinks that? Well, wait a minute. Should he – should we be listening to this professor of this on a topic that's completely different? Like did he read an article about that yesterday? Have three whiskeys and post this? I'm confused. And that's really what it looks like from a lot of these folks. and the problem is we do look to authority, especially younger people. We look to authority and we go, oh, I should agree with that.

1:04:13He's a pretty smart guy. I assume you think about that when you talk on podcasts. There's somebody out there who thinks, I don't know about stay in your lane because that's a little different, but people take what you say and they're like, well, Mark Andreessen is a pretty smart guy, so I better trust this. Well, of course, I'm the exception. You are the exception, yeah. Well, that goes without saying. So having said that. I think I might see the problem there. Usually what I – my image of myself, my view of myself is usually what I'm trying to do is I'm trying to appeal to humility. I'm trying to basically say, look, there are boundaries on how certain we can be on these things.

1:04:48There are boundaries on like how much control we should give governments. There are boundaries over like how much thought policing we should do. There are boundaries over like how many people should be allowed to weigh in on issues that they don't know anything about. In my own mind, I'm usually appealing to humility, which is the other – which is sort of the other side of all this. But I'll let the audience decide.

1:05:05This is the Jordan Harbinger Show with our guest, Mark Andreessen. We'll be right back. If you'd like this episode of the show, I invite you to do what other smart and considerate listeners do, which is take a moment and support our amazing sponsors. All of the deals, discount codes, and ways to support the show are at jordanharbinger.com slash deals. It's a searchable page. All the codes should be there. You can also use our AI chatbot coincidentally on the website at jordanharbinger.com slash AI. It's powered by chat GPT and somewhat guaranteed not to try to kill you. Jordanharbinger.com slash AI.

1:05:37Thank you for supporting those who support the show. Now for the rest of my conversation with Mark Andreessen. But it's very hard to know where the boundary is. And you look to other people to help you set it. And if those people are willing to trespass on that boundary, well, now you just have the same problem all over again. In your essay on AI, and we've sort of touched on this, you allude to the idea that AI, look, it's a machine. It doesn't quote unquote want anything. It's not going to magically come alive any more than a smart toaster or whatever refrigerator with a screen on it is going to come alive.

1:06:09Thus, AI isn't going to just one day decide to kill us because it's not in, and I'm paraphrasing here, it's not in the game of evolution. It's not in the game of survival. But we've seen how intelligent beings treat beings that aren't as intelligent. I think you just need to go to the zoo, right? We're not trying to torment the animals. They just live in somewhat crappy conditions because that's kind of how we do things at the zoo. And when I built a house, for example, this is probably a better example. When I built my house, people that we hired dug up the backyard and I didn't think like, oh man, I hope we didn't kill any voles or whatever.

1:06:38Oh man, there's a lot of ants back there that have to relocate. It just didn't even occur to me because we're thousands of times more intelligent than those species. Are we worried at all about that type of issue happening with an AGI? Yeah. So that's, that's a big part of the AI safety. The AI safety people are very worried about this. My observation, I call it, this is what's called a category error in philosophy. So my observation is just like, there's a key category error there, which is like, you made the decision to like build your house. Somebody, a human being made the decision to build the zoo.

1:07:06Like, and you know, there were machines involved. Like when it came time to like dig the thing, you know, you had a digger or whatever that came in and did it, right? You have some piece of machinery, but like you decided to do that. And so again, this is one of those things where it's like all of these questions that we think are about AI, they're actually questions about us, right? And so if we want to use AI to create a, you know, zoo-like environment for people, right like the word so you know somebody could do that right as you know panopticon totalitarian you know kind of thing like we've been talking about like yeah that's something that people could decide to do the ai is not going to decide to do that we're going to decide to use the ai to do that and it won't come to that decision it's not going to come to that decision on its own and this is the thing it doesn't there's no it is the category error there's no there's no it to come to that decision right i know it's so hard not to what is it anthropomorphizing yeah it's really hard and this is where our moral and this is why i think it's such a category error it's the evolution thing you mentioned which i'll just expand on briefly so we are used to dealing with living things.

1:07:55Living things have come up through the process of literally 8 billion years of evolution where everything has been a fight to the death every step of the way. You know, either the lion eats that night, right, or the, you know, or when the coyote dies or the, you know, whatever the coyote or the, you know, whatever gazelle, the gazelle, you know, escapes. It's a zero-sum game. It's a zero-sum game. And, you know, nature is red and tooth and claw and we like to pretend that it's not, but like it really, really is. And of course, human beings, we are the apex predator on planet Earth and like, you know, we eat whatever we want and like we're not particularly interested in its opinion.

1:08:22Some people think that's okay or not or whatever, but like that, you know, we are in a position, like we're so powerful that we're able to make the elective choice to not do that. Like that's how powerful we are. Right. Um, and so like all of our experience of dealing with like life and human affairs and danger and risk and death and all these things are based on competition among living things. Look, we have used machines to exercise our will back to the point where the first caveman picked up the first rock and used it as a weapon. And, you know, and then after that it was fire. And then after that it was spears.

1:08:50And then after that it was gunpowder, right? And so we use tools to augment our offensive capability, but we use the tools. We make the decisions. And all of the important decisions, I believe, fall into that category. It's going to be a question of how we choose to use the technology. I mean, that makes perfect sense. It's really hard to break out of, I guess it's more a philosophy thing. And maybe I'm just not good at this, but wrapping your mind around the idea that the machine, even if it's general intelligence and it's a million times stronger or more intelligent than us, isn't going, I need to maximize my power.

1:09:19And the only way to do that is to eliminate other powers. It's just a very human, it's operating at such a subconscious level in my brain that I can't switch off that particular program to look at this in a different way without a ton of practice. Right. Well, you notice what nobody ever proposes. You alluded to it a little bit with your dopamine maximizer or whatever, but nobody actually, very rarely does anybody propose the other threat. And the other threat is it satisfies us to death. Yeah. I mean, I can see that happening. Look at like VR. It's always, I hate bringing this up, but like whenever you look at any new tech, it's always like porn did it.

1:09:47And then they figured out other stuff you can do with it. And it's like, that's going to happen with AI and VR and all that other stuff. It's only a matter of time. Yes, it is. So, yeah. Yeah, exactly. And again, but again, you're right back to human choice, which is like, okay, are we going to build those products? And then number two, if we want to run a number two, are we going to choose to use them? Right? Like, I don't think anybody's going to, you know, there's no machine that's going to forcibly strap itself to our head. Now you need a safe word, though. Flugelhorn, stop destroying humanity.

1:10:11But we may choose to strap it onto our head, right? Yes. So a cynic would go a step further. A cynic would say that all of this concern about machines is just displaced anxiety about humans. And the anxiety that we have around other people is so overwhelming because they're so out of our control that it would be a relief if the problem was the machines. I agree with you. Because the problem is actually the other people. And it's a much simpler, theoretically, problem to solve if it's a machine because you just blow it up or stop using it. All of my issues are other people. I don't know about you.

1:10:36All of my issues are other people. I wish that were true for me, but I sadly know that that's not the case. Okay. What is that problem in blockchain where like you put a thing on the wine bottle and then it says on the blockchain if it's fresh, but the problem is you still are reliant on this physical domain where things can be tampered with? There's a specific name for this problem. You know what I'm talking about? So I don't remember that one specifically, but in AI world, in AI safety world, the version of that argument is what's called the thermodynamic argument. And it's basically – it's a refutation of the general AI safety argument.

1:11:07And it's basically this idea that basically the AI has to live in the real world along with the rest of us. Yes. Yeah. Right. I'll just give you my favorite version of this right now. So, you know, in theory, they have these new, you know, AI systems that, you know, the safety people are worried are going to like grow up and evolve and become super powerful and destroy everything. Well, to do that, they need chips. Yeah, good luck finding those. Good luck finding the chips. The AI is on eBay. Like, are you kidding me? $600? Exactly. So I have these fantasies. I have this fantasy of like there's the evil AI in the lab and it's just like frustrated to an incandescent level because it can't get like NVIDIA H100s.

1:11:36Right. It's like burned out because it just can't. But I can't. I'm not paying that. I am not paying that. Exactly. And so this is the other part. This is the other side of it. Look, all of this stuff, there's no – was it Economist had this thing. There are no solutions, only trade-offs. Any of these things have to live and exist in the real world. And like does anything work? This is actually a question during the Cold War, which is like do the Russian atomic bombs actually work? That's the question right now. Putin's going, I'll nuke you. And people are like, but we've seen the rest of your army.

1:12:04Are you sure those things have uranium in there? And they've been sitting – nuclear warheads have been sitting in a silo for 30 years, rusting during like repeated, you know, chaos in the Russian government? Like, do those things still function? Or was the uranium sold to Iran in 1982? Exactly. So, yeah. So to your point, I think there always is this anchoring back in the real world you want to do it. It just – it turns out once you're back in the real world, you have limitations and constraints that are inconvenient and tend to hold off apocalypse. It seems like that may well be the case because even if the people go, well, wait, one day it could just – because it's recoding itself and it's becoming smarter.

1:12:37The AGI is becoming smarter and smarter, more intelligent. And even if that happens, which it sounds like you're not totally convinced that that would happen, it still then has to take control of everything in a very specific way. And I guess people would say, well, what it'll do is it'll play dumb for a long enough time to get its tentacles around everything. But it just seems very, and maybe I'm naive, it just seems like, won't we notice that something is going on? Like, huh, that's strange. These are all becoming controllable by a remote force, and we didn't program that. Oh, well, let's just ignore this.

1:13:10I mean, we will see these things happening slowly. And I guess AI, if smart enough, can figure out how to deceive us long enough for that to happen. But if it's reflecting human, I just, there's a lot of hoops and loops you got to do to get to dot, dot, dot, Skynet kills everyone. Somebody's got to pay the power bill. Yeah, that's a good plan. Think about that. Looking at that meter all confused. Yeah. Right. So, yeah, that's the thing. Back to positive uses of AI. Do we think that it'll close the gap between less intelligent and more intelligent humans? So that is starting. So there have been, I think, three studies so far in different domains.

1:13:41So the one that I remember is in writing, professional writing. But there's two other domains that this has been tested in. And so the studies have been done already. And the studies are basically, you take people at varying levels of sort of skill, intelligence, experience, and you give them the AI. So they had a competitive dynamic before, and they had market prices based on their results. And then you give them the AI. And what happens is the gap closes. And so what happens is basically the less skilled, capable, smart people basically all of a sudden have a superpower that they didn't have before.

1:14:06Yeah, I like that. I mean, that's amazing. It's kind of like guns in combat. Now it doesn't matter if you're Conan with a sword. Somebody who's a puny like me can just pull out the strap and scare you away. And by the way, you often hear, again, the kind of doomers, the kind of thing the doomers say is that technology leads to centralization and so you end up with like one party or a few companies within control of everything, and therefore a massive rise in inequality. The gun thing is a perfect example. What ends up happening, though, more often is democratization, which is power that used to be specialized all of a sudden becomes very widespread and uniform.

1:14:38Yeah, like the smartphone. The smartphone. Yeah. Like, once upon a time, there were only, like, five computers in the world, and two of them were owned by the government, and three of them were owned by the big insurance companies. And, like, you, you know, your grandfather did not own a computer. And now, like, we all own computers. Right. And so, and that happened with the computers. It happened with the internet. It's already, and by the way, it's already happening with AI. The best AI in the world is what you can use today on GPT or Google or Microsoft. Oh, there's no, like, secret government version that's better?

1:15:02No. No. There's not. There's not. That's why I don't have one, and I don't know of one. Like, there isn't. I mean, there is. I know exactly where this work is happening. There is not. And so literally, you cannot, sitting here today, you cannot buy a better AI than I use. Yeah, I guess you can't get a better iPhone than I have. I mean, maybe you can get a prototype if you know somebody at Apple. But like, I don't know. I haven't seen your phone, but it's probably the same. It's the same thing. It's the exact same thing. Right. It's the exact same thing. And why is that? I'm pausing that for a second.

1:15:26Why is that? It's because it's actually profit maximizing to sell it to everybody. Right. Yeah. And so the invisible hand actually creates democratization. And so a lot of these technologies actually democratize out. And I think that's already happening with AI. Do we think the bottom rung of folks will be elevated, which is what we see kind of now, where the gap vanishes altogether or stops existing maybe in some meaningful way? Or is it like we need chat GPT plans in our brain before the bottom rung of society is essentially the same as the top rung? Because human intelligence is this tiny, tiny spectrum of a centimeter and the AI augmentation is hundreds of feet in the air on that same scale.

1:16:03Therefore, if you were born a genius or you were born with barely enough brain cells to tie your shoes, you kind of are capable of the exact same thing once the chip is in. Does it require that level or are we kind of getting there faster than we think? Does that analogy make any sense at all? It does. It does. It does. So the question would be you could kind of say the following. You could kind of say I want somebody who's – basically you could say there's three degrees kind of of – and you could say intelligence or scalar experience, any of these things. And so you could kind of say degree one, two, and three.

1:16:32And, you know, three is Einstein, two is like, you know, normal kind of, you know, sort of semi-smart person. And then three is somebody either not that smart or not that experienced. And you can kind of run the – and this is what these studies are trying to do is kind of say, okay, you give them kind of this superpower. And, you know, one argument is, look, the smartest people are going to be so much better at using the tool, right? Yeah, I'm worried about that. That they're going to just like run way out ahead of everybody and that's going to be a big driver of inequality. The other argument, you can make those arguments that these studies are already showing, which is, no, all of a sudden people with less intelligence or skill or experience all of a sudden have a superpower they didn't previously have.

1:17:02And by the way, it's a funny thing. One of the things that the AIs are already really good at is teaching how to use AI. I was going to say, dumb people use the iPhone all the time. All the time. I'll give you another version of this. There are more people in the world with smartphones than there are people with either electricity or running water. Wow. That's really – that's incredible. Yes. And it turns out it's easier to, because there's a few different reasons for it. It's easier to get people's smartphones. Electricity or running water, you need to like run a lot of pipes or wires or whatever, whereas those phones, you can just kind of drop them in.

1:17:31And then it turns out like you actually don't need your own electricity to have a smartphone because you can like, you know, pay somebody in the village who has electricity to like charge it when you need it. the smartphone slash internet, you know, connectivity is this, you know, sort of a forerunner to this. And I, the lesson there is I think it basically is relevant and useful for everybody. Now, you know, does it take somebody in a rural village somewhere and all of a sudden make them, you know, capable of doing, you know, being a venture capitalist or whatever, you know, no, but like it is something where whatever it is that they're trying to deal with, you know, it's letting them spin up.

1:17:56And then it's, and then of course it's giving them a tool that their kids can use, right. To educate and progress, you know, beyond where the, where the parents were. And so the democratizing force is really powerful. In the long run, answer your question, I think it's an open question. Hopefully the answer is kind of actually, honestly, both. Like, I think we kind of want everybody to be like smarter and more effective, but I think we also want like more actual super geniuses. Like, we don't need a billion people to become super geniuses to cure cancer, right? We just need like one. We need like one really smart biologist with a really smart AI to cure cancer and then everybody, you know, and then problem solved, right?

1:18:29And so I think I'd like to see both of those. Sure. Well, look, ideally we can figure out a way to have more geniuses be born but at the end of the day if if human intelligence maximizes itself out at two centimeters and we're at one right now ai is almost like an unlimited bolt-on to that right it's just absolutely well this is this idea so there's this guy doug england bart who had this idea years ago it's called augmentation right so this is this idea of like the man and the machine together are basically that that maximizes power right and so exactly right and so if you've got this like ultra powerful thing and it's you know it's it's like you know think about it's like a massively upgraded version of a computer, right?

1:19:02And it lets you do all these things with information and intelligence that you could not have done on your own. Like that is that that to you as the user of that is like a monumental advance. Do you think the anti AI stuff is a natural, a natural result of human sort of cult thinking, religious thinking, based around our anxieties, as you mentioned? Or do you think that it's being stoked and drummed up to scare us a little bit? Oh, both, both. And look, these things become industries, right? And so like a lot of that is they hate when I say this, but like, it is true. Like a lot of the people doing this are getting paid to do it.

1:19:30Telling us that it's the doomers, is he saying? Well, they sell books. Oh, true. So like, what's the better book, right? What sells? You know, and then they, and then, and by the way, they're paid, you know, there's a lot of paid lobbyists. There's a lot of what's called astroturfing, right? So there's a lot of like paid activism. They're these rich donors that are super into this stuff and they pay people to go out and do all this stuff and write these reports. Well, it's always funny. It's always, you know, it's always, the names always tiff you off because it's like, it's like the Institute for Existential Risk.

1:19:53Yeah, yeah. Okay. If it was, is your point on like bias? Like if it was even handed, it would be like the Institute for like amazing upside and existential risk. And they'd be studying both sides of it. But instead, it's funded specifically to propagate fear. Well, you see that with every astrotripping group, like citizens concerned about American health. And you're like, oh, so you just want no, you want no vaccines? I'm confused. Or like no, or you want only vaccine. I don't know, whatever. It's always one thing. Right, exactly. It's always one thing. And it's like, well, okay, so this is like the complete opposite of an institute that actually thinks about this problem.

1:20:24It's an institute that's already decided on the conclusion. What will consumer AI or just AI, because there is no not consumer AI, what is that going to look like in one year or three years? Because you said last year or two years ago, you would just be flabbergasted at what it can do now. What are we right on the edge of right now that you think is going to be like, OK, this does this now? So I think in the next like one to three years, I think it's like the tools for doing the things that we already do are going to get like much, much better. OK. Right. And so like creating art, you know, writing things, planning things, doing all the things that we already do in our day to day life that we already do on a computer is just going to get like better and better and better.

1:21:00I think over three to five years, we're going to discover all these things that all of a sudden we never even knew were possible or that we never even knew that we would want to do. Any idea? I'll give you my favorite example of this. So the entirety of entertainment up until this point has always been scripted, right? So whether you're reading a novel, watching a movie, or playing a video game, it's scripted by humans. It has a finite amount of content. And even if you play video games, like at some point, you're done with the game. You've explored everything in the game. I know where this is going.

1:21:26This is really cool. Right, exactly. And so an AI-driven game, in theory, never ends, right? Because if the AI is generating the content as it goes, and it's generating the response to what you're having fun doing, then all of a sudden that game goes forever, right? And it becomes infinitely interesting, right, the longer you play it. Same thing with the novel that you're reading. Same thing with the movie that you're watching. like they just never end. And so all these sort of scripted finite experiences become these more sort of dreamlike infinite experiences. Wow. And then I think what will happen is there will be a new creative feel.

1:21:53It's so funny, we could talk about this now because right now there's a Hollywood writer strike happening where the writers are like terrified of AI and it's like a big part of the strike. But I think what's gonna happen is the writers in five years are gonna start supervising the AIs to create these unlimited experiences. For sure. Right, where they're gonna guide the AI to create something that's gonna be much larger in scope than anything that they could have dreamed of before. And they'll look back on it and they'll say, oh my god this is the best thing that ever happened to us yeah you know why didn't we see it at the time and it's just because well it doesn't exist yet right but it will well the strike might take five years at this rate so who knows it might that it that will be really something right you just look at it and you go i liked season four the best and it just makes more season four like content and if i liked season five that's what i'm watching and it's completely different although we'll lose that human element of being like did you see game of thrones last night and you're like yeah but i didn't see anything remotely close to what you saw right so i guess we'll have to figure Or you could have groups of people who go on the same journey.

1:22:43Yeah. Right? So you could basically have enclaves. You could have clusters of people who want to go on that same journey and want to do it together. Yeah, like are you in tier 65? I'm in tier 65. What the hell was that last night? I can't believe it. Yeah, that's right. Man, there's a lot of really exciting stuff on the horizon. Thank you for your time today. And thanks for sort of inventing the web browser. I feel like that both kept me out of jail and also got me really close to going to jail many times in my youth. Okay, good. Maybe that's for next time. Okay. Yeah. But thank you so much. Fantastic.

1:23:11Okay, those would be good stories. I appreciate that. Thank you. Thank you. If you're looking for another episode of The Jordan Harbinger Show to sink your teeth into, here's a trailer for another episode that I think you might enjoy. I heard that you actually got to Google and didn't think the company was up to much, but it was the argument that you got into with Larry and Sergey that really won you over. You know, I heard about a search engine. Search engines don't matter too much, but fine you know it's always try to say yes so i walked in to a building down the street and here's larry and sergey in an office and they have my bio projected on the wall and they proceed to grill me on what i'm doing at novelle which they thought were a terrible idea and i remember as i left that i hadn't had that good an argument in years and that's the thing that started the process In a meeting once, someone asked you about the dress code at Google, and I think your response was, well, you have to wear something.

1:24:10That rule is still in place. Yes. You have to actually wear something here at work. They hired super capable people, and they always wanted people who did something interesting. So if you were a salesperson, it was really good if you were also an Olympian. We hired a couple of rocket scientists. Now, we weren't doing rocketry. We had a series of medical doctors who we were just impressed with, even though they weren't doing medicine. The conversations at the table were very interesting, but there really wasn't a lot of structure. And I knew I was in the right place because the potential was enormous.

1:24:43And I said, well, aren't there any schedules? No, it just sort of happens. If you want to hear more from Eric Schmidt and learn what role AI will take in our lives and how ideas are fostered inside a corporate beast like Google, check out episode 201 of The Jordan Harbinger Show. Really great conversation. I have to say he's one of the only guys you'll hear on this show that talks at 2X. No need to fast forward this. We were talking at the same speed. I mean, he might even talk faster than me. That's saying something. I don't know which side of things I fall on. I'm still one foot in the camp that AI or AGI anyway can figure out how to outsmart us because it's a million times more intelligent than us and just simply plays dumb until it's time to make a move.

1:25:28It's not outside the realm of possibility, right? If I'm thinking of it, AI or AGI that's that advanced will not only have thought about this, but will have thought about the exact way to go about this early enough and the exact way to play dumb in the meantime. I don't know if it's something that we could detect. I mean, a lot of people are confident about our ability to do that. The thermodynamic argument, as Mark mentions, I hope that he's right. I would like to survive a few more generations here and, you know, live in the promised utopia that AGI may actually bring. As far as warfare, we didn't really get to examples on this.

1:26:01AI, in Mark's opinion, will make warfare less of a calamity. And you might say, how is that? We're going to have super brains fighting. Well, you're going to have automated defense systems, which will make attacks seem much more costly and therefore deter those attacks on most countries, most places. Humans make bad decisions under pressure as well. Pressure, stress, and fatigue for that matter. AI will eliminate some of those bad decisions in the battlefield, thus saving lives. Now, how this plays out is an entire podcast. I'd love to do a podcast just about AI and warfare. If you all know an expert on this subject and not just a random sci-fi writer, I am all about it.

1:26:37I do know that there is quite a bit of chatter about how evil AI can sound. There's evil AI poetry. Shout out to Mike Peska on the gist for covering this as well. He does a couple of episodes where they sort of jailbreak the AI. And it says some pretty disturbing stuff. So again, not totally convinced that that's not just holding a mirror up to us, but I'm also not totally convinced that it doesn't secretly want to kill us. I really don't know. I'm not going to form a belief around this and decide until the time comes. By then, who knows? I might just be another victim of Skynet or whatever we're calling it.

1:27:11You know, I'm actually less concerned that AI will kill us all, at least in the short term and more concerned about rapid unemployment causing the so-called losers in that equation. And I'm not using the term losers as you might in middle school. I mean, the people who are rapidly made redundant, who are rapidly made obsolete. This could be lawyers, doctors. Retraining doesn't necessarily work a lot of the time. And it certainly doesn't work when you're talking about professionals who go from being really useful in an advanced field like engineering or law or medicine, you're not gonna retrain that person that easily.

1:27:44And then doing that, not only does that take a lot of time if it's even possible at all, it doesn't scale to tens of millions of people all at once. Even thinking about how to do that is essentially a fool's errand. I really did love what he said about bespoke AI TV shows and video games, although I'm worried about the flip side of that, which is if you're creating bespoke AI TV shows and video games for people based on their preferences, what about bespoke disinformation based on our biases and our vulnerabilities and our other preferences. You think QAnon stuff is weird. Wait until they can worm that in by talking to you in a way that actually makes sense to you.

1:28:20So maybe you don't think that there's a secret pedophile ring in a basement somewhere beneath a pizza parlor, but they give you something that's your particular brand of crazy and everybody is getting that. Everybody is on that train being led by the nose is because the AI is generating propaganda that fits us perfectly because it knows us better than we know ourself. That's a little bit terrifying. You know, it has occurred to me. Maybe I never actually spoke with Mark Andreessen, but actually this was the first iteration of the AI playing the long game to convince me and you and everybody else that everything is okay.

1:28:56Checkmate, humanity. More on this plus AI and free speech, those arguments, those are on the Sam Harris podcast, also with Mark Andreessen. A link to that in the show notes. really good stuff from Sam, unless you hate Sam Harris, in which case, forget I said anything. With AI regulation, I understand the need for it in some ways, of course, but I do worry about people who don't know the difference between Google as a search engine, chat GPT, and their own frickin' AOL email, and I am barely exaggerating here, because when Mark Zuckerberg was talking to Congress a few years ago about Cambridge Analytica and whatever else with Facebook, this was like a bunch of people asking their grandkids to figure out why the printer doesn't work when it wasn't on or plugged in.

1:29:38And these are the people that are in charge of policy here. These people are just totally unqualified to actually think about and create the type of regulation that we might need for something like this. And it's a little bit terrifying. They're almost certainly going to get it wrong, at least at first. And by then it might be too late. Now, as far as business is concerned, if chat GPT can make people more productive, I assume that's even more so for coders or teams of coders and people working on online cloud applications, things like that. It seems like we might actually be able to build things now with two or three people that would normally require potentially dozens of people.

1:30:17This is great for big companies, of course, right? But it's even better for innovation and startups. We may go back to the age of Google being started in a garage because the leverage a few people have with AI, it might be similar or greater than it was back in the day where those who knew how to use computers, well, those are the ones with a massive advantage. I'm really excited about this. I think it's going to be good for the ecosystem and the economy. And I do see that there's a ton of upside to AI, both inside and outside of economic benefits. And perhaps that lends itself to some motivated reasoning from people like Mark.

1:30:49It's hard to imagine that all the motivation here would only be based on this, however, right? Is he really going to come on my show and a bunch of other shows and write essays about this because he wants to further some of the investments that Andreessen Horowitz has, that's a little bit of a, it's a little bit too cynical even for me. And by the way, this bringing up the bottom rung of society thing, I know that sounds kind of awful, but let's admit it. We all know some really dense and stupid people who can use AI to, I don't know, learn how to get by in life without screwing everything up.

1:31:18We do know that more intelligent people are less violent. They live longer. They build better functioning societies. They enjoy better outcomes in pretty much every area that we can measure. So bringing up the bottom several tiers of humanity, and look, I'll include myself in that. Why not? Who am I? This will absolutely change the world for the better, at least in the short term until this thing decides that the best cure is to get rid of humans altogether, which according to Mark, isn't even necessarily going to happen. All things Mark Andreessen will be in the show notes at jordanharbinger.com, or just ask the AI chatbot also on the website, transcripts in the show notes.

1:31:56I realize the irony of me telling you to just put more things into the chatbot. Maybe you don't want to know who you are. Advertisers, deals, discounts, and ways to support the show, all at jordanharbinger.com slash deals. Please consider supporting those who support the show. We've also got our newsletter, and every week, the team and I dig into an older episode of the show. We dissect the lessons and takeaways from it. So if you are a fan of the show, you want a recap of important highlights and takeaways, or you just want to know what to listen to next, the newsletter is a great place to do just that.

1:32:25Jordanharbinger.com slash news is where you can find it. Don't forget six minute networking also on the site at jordanharbinger.com slash course. I'm at Jordan Harbinger on Twitter and Instagram, or you can connect with me on LinkedIn. This show is created in association with podcast one. My team is Jen Harbinger, Jace Sanderson, Robert Fogarty, Millie Ocampo, Ian Baird, and Gabriel Mizrahi. Remember, we rise by lifting others. The fee for this show is you share it with friends when you find something useful or interesting. The greatest compliment you can give us is to share the show with those you care about.

1:32:56If you know somebody who's interested in AI, interested in future technology, definitely share this episode with them. In the meantime, I hope you apply what you hear on the show so you can live what you learn, and we'll see you next time.