#2156 - Jeremie & Edouard Harris

What's happening? Oh you know not too much. Just another typical week in AI. Just the beginning of the end of time. That's all happening right now. Just for the sake of the listeners, please just give us your names and tell us what you do. So I'm Jeremy Harris. I'm the CEO and co-founder of this company Gladstone AI that we co-founded. So we're essentially a national security and AI company we can get into the backstory a little bit later, but that's the high level. Yeah, and I'm Ed Harris. I'm actually I'm his co-founder and brother and the CTO of the company. Keep this like pull this up like a fist from your face. There you go perfect. So how long have you guys been involved in the whole AI space? For a while in different ways. So we actually, we started off as physicists. Like that was our background. And in like around 2017, we started to go into AI startups. So we found it a startup took you through Y Combinator, this like Silicon Valley accelerator program. At the time, actually Sam Altman, who's now the CEO of OpenAI, was the president of Y Combinator. So he opened up our batch at YC with this big speech and we got some conversations in with him over the course of the batch then in 2020 so this this thing happened that we could talk about essentially this was like the moment that There's like a before and after in the world of AI before and after 2020 and It launched this Revolution that brought us to chat GPT Essentially, there was an insight that open AI had and doubled down on that you can draw straight us to chat GPT. Essentially, there was an insight that OpenAI had and doubled down on that you can draw a straight line to chat GPT, GPT-4, Google Gemini, everything that makes AI everything it is today started then. And when it happened, we kind of went, well, it had gave me a call, this like panic phone call, he's like, dude, [2:01] I don't think we can keep working like businesses usual in a regular company anymore. Yeah. So there was this AI model called GPT3. So like everyone has maybe played with GPT4. It's like ChatGPT. GPT3 was the generation before that. And it was the first time that you had an AI model that could actually, let's say, do stuff like write news articles that the average person, like in a paragraph of a news article, could not tell the difference between it wrote this news article in a real person wrote this news article. So that was an inflection. That was, you know, significant in itself. But what was most significant was that it represented a point along this line, this scaling trend for AI, where the signs were that you didn't have to be clever, you didn't have to come up with necessarily a revolutionary new algorithm or be smart about [3:01] it. You just had to take what works and make it way, way, bigger. And the significance of that is you increase the amount of computing cycles you put against something, you increase the amount of data. All of that is an engineering problem and you can solve it with money. So you've got, you can scale up the system, use it to make money and put that money right back into scaling up the system some more money in IQ points come out. Cheers. That was kind of the 2020 moment. And that's what we said in 2020, exactly. I spent about two hours trying to argue out of it. I was like, no, no, no, no, like we can keep working at our company because we're having fun. Like we like founding companies. And yeah, he just like wrestling me to the ground and we're like, shit, we gotta do something about this. We reached out to like a family friend who, you know, he was non-technical but he had some connections in government in DOD and we're like, dude, the way this is set up right now, you can really start drawing straight lines and extrapolating and saying, you know what, [4:02] the government is going to give a shit about this in not very long, two years, four years, we're not sure, but the knowledge about what's going on here is so siloed in the frontier labs. Like, our friends, all over the frontier labs, the OpenAI, the Google Debt Minds, all that stuff, the shit they were saying to us that was like mundane reality, like water cooler conversation, when you then went to talk to people in policy and even like pretty senior people in government, not tracking the story remotely. In fact, you're hearing almost a diametric opposite. This is like over learning the lessons of the AI winters that came before when it's pretty clear like we're on a very at least interesting trajectory, let's say, that should change the way we're thinking about the technology. What was your fear? Like, what was it that hit you that made you go, we have to stop doing this? So it's basically, anyone can draw a straight line on a graph. The key is looking ahead and actually at that point, [5:02] three years out, four years out and asking, like you're asking, what does this mean for the world? What does it mean? What does the world have to look like if we're at this point? And we're already seeing the first wave of risk sets just begin to materialize. And that's kind of the weaponization risk sets. So you think about stuff like large scale psychological manipulation of social media. Actually really easy to do now. You train a model on just a whole bunch of tweets. You can actually direct it to push a narrative like maybe China should own Taiwan or whatever, something like that. And you actually, you can train it to adjust the discourse and have increasing levels of effectiveness to that. And you actually, you can train it to adjust the discourse and have increasing levels of effectiveness to that, just as you increase the general capability surface of these systems, we don't know how to predict what exactly comes out of them at each level of scale, but it's just general increasing power. And then the kind of next beat of risk after that. [6:08] So we're scaling these systems, we're on track to scale systems that are at human level, like generally as smart, however you define that as a person or a greater and open AI and the other labs are saying, yeah, it might be two years away, three years away, four years away, like insanely close. At the same time, and we can go into the details of this, but we actually don't understand how to reliably control these systems. We don't understand how to get these systems to do what it is we want. We can kind of like poke them and prod them and get them to kind of adjust, but you've seen, and we can go over these examples. We've seen example after example of, you know, Bing Sidney, yelling at users, Google, showing 17th century British scientists that are racially diverse, all that kind of stuff. We don't really understand how to like aim it [7:02] or align it or steer it. And so then you can ask yourself, well, we're on track to get here. We are not on track to control these systems effectively. How bad is that? And the risk is if you have a system that is significantly smarter than humans or human organization, that we basically get disempowered in various ways relative to that system. And we can go into some details on that too. Now, when a system does something like what Gemini did, like what says show us Nazi soldiers that it shows you Asian women, and like what is, what's the mechanism? Like how does that happen? So it's maybe worth, yeah, taking a step back and looking at how these systems actually work. Because that's gonna give us a bit of a frame too for figuring out when we see weird shit happen. How weird is that shit? Is that shit just explainable by just the basic mechanics of what you would expect to happen [8:00] based on the way we were training these things? Or is something new and fundamentally different happening. So talking about this idea of scaling these AI systems, right? What does that actually mean? Well, you imagine the AI model, which is kind of like you think of it as like the artificial brain here that actually does the thinking. That model contains, it's kind of like a human brain, it's got these things called neurons, we and the human brain called them biological neurons in the context of AI's artificial neurons, but it doesn't really matter, that the cells that do the thinking for the machine. And the realization of AI scaling is that you can basically take this model, increase the number of artificial neurons it contains. And at the same time, increase the amount of computing power that you're putting into kind of like wiring the connections between those neurons. That's the training process. Can I pause you right there? Yeah. How does the neuron think? Yeah, so okay, so let's get a little bit more concrete then. So in your brain, right, we have these neurons, they're all connected to each other with different connections. And when you go out into the world and you learn new skill, what really happens is you try out that skill, you succeed or fail, and based on your succeeding or failing, the connections between [9:05] neurons that are associated with doing that task well gets stronger, the connections that are associated with doing it badly get weaker. And over time, through this glorified process of trial and error, eventually you're going to hone in and really, in a very real sense, everything you know about the world gets implicitly encoded in the strengths of the connections between all those neurons. If I can x-ray your brain and get all the connection strengths of all the neurons, I have everything Joe Rogan has learned about the world. That's basically the good sketch, let's say, of what's going on here. So now we apply that to AI, right? That's the next step. And here, really, That's the next step. And here really it's the same story. We have these massive systems, artificial neurons connected to each other. The strength of those connections is secretly what encodes all the knowledge. So if I can steal all of those connections, those weights as they're sometimes called, I've stolen the model. I've stolen the artificial brain. [10:01] I can use it to do whatever the model could do initially. That is kind of the artifact of central interest here. And so if you can build the system, right? Now you've got so many moving parts. Like if you look at GPT-4, it has people think around a trillion of these connections. And that's a trillion little pieces that all have to be jiggered together to work together coherently. And you need computers to go through and tweak those numbers. So massive amounts of computing power. The bigger you make that model, the more computing power you're going to need to tune it in. And now you have this relationship between the size of your model, the amount of computing power you're going to use to train it. And if you can increase those things at the same time, what Ed was saying is your IQ points basically drop out. Very roughly speaking, that was what people realized in 2020 and the effect that had was now all of a sudden the entire AI industry is looking at this equation. Everybody knows the secret sauce. I make it bigger, I make more IQ points, I can get more money. So Google's looking at this, Microsoft, OpenAI, Amazon, everybody's looking at the same equation. [11:02] You have the makings for a crazy race. Like right now today, sorry, Microsoft, is engaged in the single biggest infrastructure in human history. Build out the biggest infrastructure. Build out. $50 billion a year, right? So on the scale of the Apollo moon landings, just in building out data centers to house the compute infrastructure because they are betting that these systems are going to get them to something like human-level AI pretty damn soon. So I was reading some story about I think it was Google that saying that they're going to have multiple nuclear reactors to power their database. That's what you got to do now because what's going on is North America is kind of running out of on grid base load power to actually supply these data centers. You're getting data center building moratoriums in areas like Virginia, which has traditionally been like the data center cluster for Amazon, for example, and for a lot of these other [12:04] companies. And so when you build a data center, you need a bunch of resources sighted close to that data center. You need water for cooling and a source of electricity. And it turns out that wind and solar don't really quite cut it for these big data centers that train big models. Because the data center, the training consumes power like this all the time, but the sun isn't always shining, the wind isn't always blowing. And so you got to build nuclear reactors, which give you high capacity factor base load. And Amazon literally bought, yeah, a data center with a nuclear plant right next to it, because like, that's what you got to do. Jesus. How long does it take to build a nuclear reactor? Because it's so like this is the race, right? The race is, you're talking about 2020, people realizing this, then you have to have the power to supply it. But how long, how many years does it take to get an active nuclear reactor up and running? [13:02] It's an answer that depends. The Chinese are faster than us at building nuclear reactors, for example. And that's part of the geopolitics of this too, right? Like when you look at US versus China, what is bottlenecking each country, right? So the US is bottlenecked increasingly by power, based load power, China, because we've got export control measures in place, in part as a response to the scaling phenomenon. As a result of the investigation we did. That's right, yeah. In part, in part. Yeah. But China is bottlenecked by their access to the actual processors. They've got all the power they can eat, because they've got much more infrastructure investment, but the chip side is weaker. So, there's sort of like balancing act between the two sides, and it's not clear yet like which one positions you strategically for dominance in long-term. But we are also building better, more like so small modular reactors. Essentially small nuclear power plants that can be mass-produced. Those are starting to come online relatively early, [14:01] but the technology and designs are pretty mature. So that's probably the next beat for our power grid for data centers, I would imagine. Microsoft is doing this. So in 2020, you have this revelation you recognize where this is going. You see how it charts and you say, this is going to be a real problem. Does anybody listen to you? This is where the problem comes, right? Yeah, like we said, right? You can draw a straight line. You can have people nodding along, but there's a couple of like hiccups along the way. One, is that straight line really gonna happen? All you're doing is like drawing lines on charts, right? I don't really believe that that's gonna happen, and that's one thing. The next thing is just just imagining is this, is this what's gonna come to pass as a result of that? And then the third thing is, well, yeah, that sounds important, but like not my problem. That sounds like an important problem for somebody else. And so we did do a bit of a traveling. Yeah, it was like the world's saddest traveling roadshow. [15:02] It was literally as dumb as this sound. So we go and, oh my God, I mean, it's almost embarrassing to think back on, but so 2020 happens, yes, within months, first of all, we're like, we gotta figure out how to hand off our company. So we handed off to two of our earliest employees, they did an amazing job, company exited, that's great. But that was only because they're so good at what they do. We then went, what the hell, like how can you steer this situation? How do you, we just thought we got to wake up the US government as stupid and naive as that sounds, like that was the big picture goal. So we start to line up as many briefings as we possibly can across the US interagency, all the departments, all the agencies that we can find, climbing our way up. We got an awful lot like Ed said of like that sounds like a wicked important problem for somebody else to solve yeah defense home lines of security and then the state department yeah so we end up exactly this this meeting with like there's about a dozen folks from the state department and one of them and i i hope at some point uh... you know history recognizes what what she did her team did because it was the first time that somebody actually [16:03] stood up and said first of all, sounds like a serious issue. I see the argument makes sense to I own this. And three, I'm going to put my own career capital behind this. That's the. And that was at the end of 2021. So imagine that. That's a year before chat GPT, nobody was tracking this issue. You had to have the imagination to draw like through that line, understand what it meant, and then believe, yeah, I'm going to risk some career capital on this in a risk of Earth's government. This is the only reason that we even were able to publicly talk about the investigation in the first place because by the time this whole assessment was commissioned, it was just before chat GPT came out, the IFSauron was not yet on this. And so there was a view that like, yeah, sure, you can publish the results of this kind of, not nothing burger investigation, but you can sure go ahead. And it just became this insane story. We had like the UKI Safety Summit, we had the White House Executive Order, [17:01] all this stuff which became entangled with the work we were doing. Which we simply could not have, especially some of the reports we were collecting from the labs, the whistleblower reports, that could not have been made public if there, if it wasn't for the foresight of this team, really pushing for as well the American population to hear about it. Now, I could see how if you were one of the people that saw on this expansion man-minded. All you're thinking about is getting this up and running. You guys are paying the ass, right? So you're obviously doing something really ridiculous. You're stopping your company. You can make more money staying there and continuing the process. But you recognize that there's like an existential threat involved in making this stuff go online. When this stuff is live, you can't undo it. Oh yeah, no matter how much money you're making, the dumbest thing to do is to stand by as something that completely transcends money is being developed and it's just going to screw you over if things go badly. [18:01] But what is there people that push back against this and what is their argument. Yeah so actually for and I'll I'll let you follow up on the but there the first story of the push back I think it's kind of a it's been in the news a little bit lately now getting more and more public but the when we started this and like no one was talking about it The one group that was actually pushing sort of stuff in this space was a big funder in the area of like effective altruism. I think you know, may have heard of them. This is kind of a Silicon Valley group of people who have a certain mindset about how you pick tough problems to work on, valuable problems to work on. They've had all kinds of issues. Sam Bankman freed was one of them and all that quite famously. So we're not effective ultra-us, but because these are the folks who are working in the space, we decided, well, we'll talk to them. And the first thing they told us was, don't talk to the government about this. Their position was, if you bring this to the attention of the government, they will go, oh shit, powerful AI systems, and they're not going to hear about the dangers, so they're going to somehow go out and build the powerful systems without caring about the rest side. [19:10] Which, when you're in that startup mindset, you want to fail cheap. You don't want to just make assumptions about the world and be like, okay, let's not touch it. Our instinct was, okay, let's just test this a little bit and talk to a couple people, see how they respond, tweak the like kind of keep keep climbing that that ladder that's the kind of you know build their mindset that we came from Silicon Valley and and we found that people are way more thoughtful about this than you would imagine and in DOD especially DOD is actually has a very safety-oriented culture with their tech like the thing is because like they're there's stuff like kills people right and they know their stuff kills people, right? And they know there's stuff kills people. And so they have an entire safety-oriented development practice to make sure that their stuff doesn't go off the rails. And so you can actually bring up these concerns with them, and it lands in kind of a ready culture. [20:00] But one of the issues with the individuals we spoke to who were saying don't talk to government is that they had just not actually interacted with any of the folks that they were kind of talking about and imagining that they knew what was in their heads. And so they were just giving incorrect advice. And frankly, like, so we work with DOD now on actually deploying AI systems in a way that's safe and secure. And the truth is, at the time when we got that advice, which was like late 2020, reality is, you could have made it your life's mission to try to get the Department of Defense to build an AGI and like you would not have succeeded because nobody was paying attention. Wow. Because they just didn't know. Yeah, there's a chasm, right? There's a gap to cross. Like there's information. Yeah, there's information spaces that DOD folks like operate in and work in. There's information spaces that Silicon Valley and tech operated in. They're a little more [21:03] convergent today, but especially at the time, they were very separate. And so the briefings we did, we had to constantly iterate on like clarity, making it very kind of clear and explaining it and all that stuff, years it. And that was the piece to your question about like the pushback in a way from inside the house. I mean, that was the people who cared about the risk. Yeah. The man. When we actually went into the labs, so some labs, not all labs are created equal. It should make that point. When you talk to whistleblowers, what we found was, so there's one lab that's really great. So anthropic, when you talk to people there, you don't have the sense that you're talking to a whistleblower who's nervous about telling you whatever roughly speaking what you know the executive say to the public is aligned with what they're their researchers say it's all very very open more more closely i think than any of the other story i'm more more closely than any of the others always you know there are always variations here and there but [22:00] uh... some of the other labs like very different. And you had the sense like we were in a room with one of the frontier labs. We're talking to their leadership, this part of the investigation. And there was somebody from, anyway, it won't be too specific, but there was somebody in the room who then took us to the side after. And he hands me his phone. He's like, hey, can you please, like put your phone number. Or no, yeah, he put his number in my phone. And then he kind of like whispered to me, he's like, hey, so whatever recommendations you guys are going to make, I would urge you to be more ambitious. And I was like, what does that mean? He's like, can we just talk later? So as happened in many, many cases, we had a lot of cases where we set up bar meetups after the fact where we would talk to these folks and get them in an informal setting. He shared some pretty sobering stuff and in particular the fact that he did not have confidence in his lab's leadership to live up to their publicly stated word on what they [23:00] would do when they were approaching AGI and even now to secure and make these systems safe. So many such cases, this is like kind of one specific example, but it's not that you ever had lab leadership come in or doors getting kicked down and people are waking us up in the middle of the night. It was that you had this looming cloud over everybody that you really felt some of the people with the most access and information who understood the problem the most deeply were the most hesitant to bring things forward because they sort of understood that their lab's not going to be happy with this. And so it's very hard to also get an extremely broad view of this from inside the labs because you know you open it up, you start to talk to, we spoke to a couple of dozen people about various issues in total. You go much further than that and where it starts to get around. And so we had to kind of strike that balance as we spoke to folks from you share these labs. Now, when you say approaching AGI, [24:01] how does one know when a system has achieved AGI, and does the system have an obligation to alert you? Well, by, you know, the Turing test, right? Yes. So you have a conversation with a machine and it can fool you into thinking that it's a human. That was the bar for AGI for, you know, a few decades. That's kind of already happened. Yeah. Like close to it. Yeah. Yeah. Like close to it. Yeah. For zero is close to it, for four. Different forms of the Turing test have been passed, different forms have been proposed, and there is a feeling among a lot of people that goal posts are being shifted. Now, the definition of AGI itself is kind of interesting, right, because we're not necessarily fans of the term, because usually when people talk about AGI, they're talking about a specific circumstance in which there are capabilities they care about. So some people use AGI to refer to the wholesale automation of all labor, right? That's one. Some people say, well, when you build AGI, it's like it's automatically going [25:02] to be hard to control and there's a risk to civilization, so that's a different threshold. All these different ways of defining it, ultimately, it can be more useful to think sometimes about advanced AI and the different thresholds of capability you cross and the implications of those capabilities, but it is probably going to be more like a fuzzy spectrum, which, in a way, it makes it harder, right? It would be great to have a... Like a trip wire where you're like, oh, like this is bad, okay. Like we, you know, we gotta do something. But because there's no threshold that we can like really put our fingers on, we're like a frog and boiling water in some sense where it's like, oh, like just gets a little better, a little better, oh, like we're still fine. And not just we're still fine, but as the system improves below that threshold, life gets better and better. These are incredibly valuable beneficial systems. We do roll stuff out like this again at DoD and various customers. And it's massively valuable. [26:02] It allows you to accelerate all kinds of back office, like paperwork, BS. It allows you to do all sorts of wonderful things. And our expectation is that's going to keep happening until it suddenly doesn't. Yeah, one of the things that there was a guy we were talking to from one of the labs and he was saying, look, the temptation to put a heavier foot on the pedal is going to be greatest, just as the risk is greatest, because that's, you know, it's dual use technology, right? Every positive capability increasingly starts to introduce, basically a situation where the destructive footprint of malicious actors who weaponize the system, or just of the system itself, just grows and grows and grows. So you can't really have one without the other. The question is always how do you balance those things, But in terms of defining AI, it's a challenging thing. Yeah, that's something that one of our friends at the lab pointed out. The closer we get to that point, the more the temptation will be to hand these systems the keys to our data center because [27:01] they can do such a better job of managing those resources and assets. And if we don't do it, Google will. And if they don't do it, Microsoft will. Like the competition, the competitive dynamics are a really big part of this issue. Yes. So there's just a mad race to who knows what? Exactly. Yeah. That's actually the best summary I've heard. I mean, like no one knows what the magic threshold is. It's just these things keep getting smarter. So we might as well keep turning that crank. And as long as scaling works, right, we have a knob, a dial, we can just tune. And we have more IQ points out. Well, from your understanding of the current landscape, how far away are we looking at something being implemented with the whole world changes? Arguably, the whole world is already changing as a result of this technology. The US government is in the process of task organizing around various risk sets for this. That takes time. [28:01] The private sector is reorganizing. Open AI will roll out an update that obliterates the jobs of illustrators from one day to the next, obliterates the jobs of translators from one day to the next. This is probably net beneficial for society because we can get so much more art and so much more translation done. But is the world already being changed as a result of this? Yeah, absolutely. Geopolitically, economically, industrially. Yeah. Of course, it's like not to say anything about the value, the purpose that people lose from that, right? So there's the economic benefit, but there's the social, cultural hit that we take too. Right, and then there's the implementation of universal basic income, which keeps getting discussed in regards to this. We ask chat GPT-40 the other day in the green room where we're like, you know, are you going to replace people? Like, well, what will people do for money? And then, well, universal basic income will have to be considered. You don't want a bunch of people just on the dole working for the fucking sky net. [29:00] Yeah. You know, because that's kind of what it is. I mean, one of the challenges is like the, so much of this is untested and we don't know how to, how to even roll that out. Like we can't predict what the capabilities of the next level of scale will be, right? So open AI literally, and this is what's happened every, with every beat, right? They build the next level of scale and they get to sit back along with the rest of us and be surprised at the gifts that fall out of the scaling pinata as they keep whacking it. And because we don't know what capabilities are going to come with that level of scale, we can't predict what jobs are going to be on the line next. We can't predict how people are going to use these systems, how they'll be augmented. So there's no real way to kind of task organize around like who gets what in the redistribution, redistribution game? And some of the thresholds that we've already passed are like a little bit freaky. So even as of 2023, GPT-4, Microsoft and OpenAI and some other organizations did various assessments of it before rolling it out. And it's absolutely capable of deceiving a human and has done that successfully. So one of the tests that they did, kind of famously, [30:07] is they had a, it was given a job to solve a capture. And at the time, it didn't have... Just playing capture, what people would say. Yeah, yeah, yeah. So it's this, now it's like kind of hilarious and quaint, but it's this, you know... Are you a robot test? Are you a robot test with like writing them online? Yeah, online, exactly. That's it. So if you want to create an account, they don't want robots creating a billion accounts. So they give you this test to prove you're a human. And at the time, GPD for like now, it can just solve captures. But at the time, it couldn't look at images. It was just a text, right? It was a text engine. And so what it did is it connected to a task rabbit worker and was like, hey, can you help me solve this capture? The task rabbit worker comes back to it and says, you're not a bot are you? Ha ha ha. Like, call in it out. And you can actually see so the way they built it is so they could see a read out of what it was thinking [31:01] to its stuff. Scratch pad. Yeah, Crashpad. It's called, but you can see basically as it's writing, it's thinking to itself. It's like, I can't tell this worker that I'm a bot because then it won't help me solve the cap shot, so I have to lie. And it was like, no, I'm not a bot. I'm a visually impaired person. And the task rabbit worker was is, so right now, if you look at the government response to this, what are the tools that we have to oversee this? And when we did our investigation, we came out with some recommendations too. It was stuff like, yeah, you got to license these things. You get to a point where these systems are so capable that, yeah, if you're talking about a system that can literally execute cyber attacks at scale or literally help you design bio weapons and we're getting early indications that that is absolutely the course that we're on. Maybe literally everybody should not be able to completely freely download, modify, use in various ways these systems. It's very thorny obviously. [32:00] But if you want to have a stable society that seems like it's starting to be a prerequisite. So the idea of licensing, as part of that, you need a way to evaluate systems. You need a way to say which systems are safe and which aren't. And this idea of AI evaluations has kind of become this touchstone for a lot of people's solutions. And the problem is that we're already getting to the point where AI systems in many cases can tell when they're being evaluated and modify their behavior accordingly. So there's like this one example that came out recently, Anthropic, they're called two chatbot. So they basically ran this test called a needle in a haystack test. So what's that? Well, you feed the model, like imagine a giant chunk of text all of Shakespeare and then somewhere in the middle that giant chunk of text you put a sentence like Burger King makes the best wapper. Sorry the wapper is the best burger or something like that right? Then you turn to the model after you fed it this giant pile of text with a little fact hidden somewhere inside you ask it What's the best burger? Right, you're gonna test basically to see how well can it recall that stray fact that was buried somewhere in that giant pile of text. So the system responds, [33:08] yeah, well, I can tell you want me to say the whopper is the best burger. But it's oddly out of place, this fact in this whole body of text. So I'm assuming that you're either playing around with me or that you're testing my capabilities. And so this is just a kind of context awareness, right? And the challenges when we talk to people at like meter and other sort of AI evaluations labs, this is a trend like not the exception. This is possibly possibly going to be the rule. As these systems get more scaled and sophisticated, they can pick up on more and more subtle statistical indicators that they're being tested. We've already seen them adapt their behavior on the basis of their understanding that they're being tested. You run into this problem where the only tool that we really have at the moment, which is just throwing a bunch of questions at this thing and seeing how it responds, like, hey, make a bio weapon, hey, like, do this DDoS attack, whatever. [34:07] We can't really assess, because there's a difference between what the model puts out and what it potentially could put out if it assesses that it's being tested and there are consequences for that. One of my fears is that AGI is gonna recognize how shitty people are, because we like to bullshit ourselves. We like to kind of pretend and justify and rationalize a lot of human behavior from everything to taking all the fish out of the ocean to dumping off toxic waste in third world countries, sourcing of minerals that are used in everyone's cell phones in the most horrific way. All these things, like my real fear is that AGI is not gonna have a lot of sympathy for a creature that's that flawed and lies to itself. AGI is absolutely going to recognize how shitty people are. [35:00] Not, it's hard to answer the question from a moral standpoint, but from the standpoint of our own intelligence and capabilities. So you think about it like this. The kinds of mistakes that these AI systems make. So you look at, for example, GPT-40 has one mistake that it used to make quite recently, where if you ask it, just repeat the word company over and over and over again. It will repeat the word company, and then somewhere in the middle of that, it'll just snap. It'll just snap and just start saying, like, weird, I forget, like, what the... Oh, it's like... It's talking about itself, how it's suffering. Like, it depends on, it varies from case to case. It's suffering by having to repeat the word company over again. So this is called, it's called rant mode internally, or at least this is the name that they use. Yeah, we're friends mentioned. There is an engineering line item in at least one of the top labs to beat out of the system, [36:00] this behavior known as rant mode. Now rant mode is interesting because. Existentialism. Sorry, existentialism. This is one kind of rent mode. Now rent mode is interesting because existentialism. Sorry, existentialism. This is one kind of rent mode. Yeah, sorry. So when we talk about existentialism, this is a kind of rent mode where the system will tend to talk about itself, refer to its place in the world, the fact that it doesn't want to get turned off sometimes, the fact that it's suffering, all that. That oddly is a behavior that emerged at, as far as we can tell, something around GPT-4 scale, and then has been persistent since then. And the labs have to spend a lot of time trying to beat this out of the system to ship it. It's literally like it's a KPI, like an engineering, a line item in the engineering, like task list. We're like, okay, we gotta, we gotta reduce existential outputs by like x percent this quarter. Like that is the goal. Because it's a convergent behavior, like at least it seems to be empirically with a lot of these models. Yeah, it's hard to say, but it seems to come up a lot. So that's weird in itself. [37:01] My, what I was trying to get at was actually just the fact that these systems make mistakes that are radically different from the kinds of mistakes humans make. And so we can look at those mistakes like, you know, GBD4 not being able to spell words correctly in an image or things like that and go, ah, haha, it's so stupid. Like I would never make that mistake, therefore this thing is so dumb. But what we have to recognize is we're building minds that are so alien to us that the set of mistakes that they make are just gonna be radically different from the set of mistakes that we make. Just like the set of mistakes that a baby makes is radically different from the set of mistakes that we make. Just like the set of mistakes that a baby makes is radically different from the set of mistakes that a cat makes. Like a baby is not as smart as an adult human. A cat is not as smart as an adult human, but they're, you know, they're unintelligent in, obviously, very different ways. A cat can get around the world. [38:02] A baby can't, but has other things that it can do that a cat can't. So now we have this third type of approach that we're taking to intelligence. There's a different set of errors that that thing will make. And so one of the risks taking it back to, like, will it be able to tell how shitty we are? Is right now we can see those mistakes really obviously because it thinks so differently from us. But as it approaches our capabilities, our mistakes are like all the like fucked up stuff that you have and I have in our brains is going to be really obvious to it because it thinks so differently from us. It's just going to be like, oh yeah, why are all these humans making these mistakes at the same time? So there is a risk that as you get to these capabilities, we really have no idea, but humans might be very hackable. We already know there's all kinds of social manipulation techniques that succeed against humans reliably. Con artists, oh yeah, persuasion is an art form [39:01] and a risk set and there are people who are world class at persuasion and are basically make bank from that. And those are just other humans with the same architecture that we have. They're also AI systems that are wicked good at persuasion today, like totally. Totally. I want to bring it back to suffering. What does it mean when it says it's suffering? So, okay, here, I'm just gonna draw a bit of a box around that, yeah, that aspect, right? So what we folk, we're very agnostic when it comes to suffering sentience. That's not part of, we're focused on that. Because nobody knows. Yeah, we literally, exactly. I can't prove the Joe Rogan's conscious, I can't prove that Ed Harris is conscious. So there's no way to really intelligently reason. There have been papers, by the way. One of the Godfather's of AI, Yachto Benjio put out a paper a couple months ago, looking at like, on all the different theories of consciousness, what are the requirements for consciousness and how many of those are satisfied by current AI systems, [40:03] and that itself was an interesting read, but ultimately, no one knows. There's no way around this problem. Our focus has been on the national security side. What are the concrete risks from weaponization, from loss of control that these systems introduce? That's not to say there hasn't been a lot of conversation internal to these labs about the issue you raised. It's an important issue right like it is a it's a freaking moral monstrosity humans have a very bad track record of thinking of others other stuff as other when it doesn't look exactly like us whether it's racially or even different species I mean it's not hard to imagine this being another category of that mistake it's just like one of the challenges is like, you can easily kind of get a get bog down in like consciousness versus loss of control. And those two things are actually separable or maybe. And anyways, so long way of saying, I think it's a great point. Yeah, so that question is important. [41:04] But it's also true that if we knew for an absolute certainty that there was no way these systems could ever become conscious, we would still have the national security risk set, and particularly the loss of control risk set. Because so again, like it comes back to this idea that we're scaling to systems that are potentially at or beyond human level. There's no reason to think it will stop at human level, that we are the pinnacle of what the universe can produce in intelligence. We're not on track, based on the conversations we've had with folks at the labs, to be able to control systems at that scale. And so one of the questions is, how bad is that? Is that bad? It sounds like it could be bad, right? Just intuitively, it's certainly it sounds like we're definitely entering, or potentially entering an area that is completely unprecedented in the history of the world. We have no precedent at all for human beings not being at the apex of intelligence in the globe. [42:06] We have examples of species that are intellectually dominant over other species, and it doesn't go that well for the other species, so we have some maybe negative examples there. But one of the key theoretical, and it has to be theoretical because until we actually build these systems, we won't know, One of the key theoretical lines of research in this area is something called power seeking and instrumental convergence. And what this is referring to is if you think of like yourself first off, whatever your goal might be, if your goal is, well, I'm'm gonna say if if me if my goal is to become you know a tiktok star or a janitor or the president of the United States whatever my goal is I'm less likely to accomplish that goal if I'm dead start from an obvious example and so therefore no matter what my goal is, I'm probably going to have an impulse [43:06] to want to stay alive. Similarly, I'm not going to be in a better position to accomplish my goal, regardless of what it is, if I have more money, right, if I make myself smarter. If I prevent you from getting into my head and changing my goal, that's another kind of subtle one, right? Like if my goal is, I want to become president. I don't want Joe messing with my head so that I change my goal because that would change the goal that I have. And so that, those types of things, like trying to stay alive, making sure that your goal doesn't get changed, accumulating power, trying to make yourself smarter. These are called convergent, essentially convergent goals because many different ultimate goals, regardless of what they are, go through those intermediate goals of want to make sure [44:02] I stay like they support no matter what goal you have, they will probably support that goal. Unless your goal is pathological, like I want to commit suicide, if that's your final goal, then you don't want to stay alive. But for most, the vast majority of possible goals that you could have, you will want to stay alive, you will want to not have your goal changed, you will want to basically accumulate power. And so one of the risks is if you dial that up to 11 and you have an AI system that is able to transcend our own attempt set containment, which is an actual thing that these labs are thinking about. Like how do we contain a system that's trying to specialize in testing? Do they have containment of it currently? Well right now the systems are probably too dumb to like, you know, want to be able to break out on the board. But then why are they suffering? This brings me back to my point. What it says it's suffering? Do you quiz it? It's, so that's the thing. It's writing that it's suffering, right? Yeah. It's... Is it just embodying life is suffering? Well, we can't actually, so these things are trained, actually this is maybe worth flagging. And by the way, just to kind of put a pin in what Ed was saying there, there's actually [45:09] a surprising amount of quantitative and empirical evidence for what he just laid out there. He's actually done this some of this research himself, but there are a lot of folks working on this. It's like, it sounds insane. It sounds speculative. It sounds wacky. But this is, is does appear to be kind of the default trajectory of the tech. So in terms of, yeah, these weird outputs, right? What does it actually mean? If the AI system tells you I'm suffering, right? Does that mean it is suffering? Is there actually a moral patient somewhere embedded in that system? The training process for these systems is actually worth considering here. So, you know, what is GPT-4? Really? What was it designed to be? How was it shaped? It's one of these artificial brains that we talked about, massive scale. And the task that it was trained to perform is a glorified version of text autocomplete. So you imagine taking every sentence on the internet roughly, feed it the first half of the sentence, [46:01] get it to predict the rest. The theory behind this is, you're gonna force the system to get really good at text auto complete. That means it must be good at doing things like completing sentences that sound like to counter-arising China, the United States should blank. Now, if you're gonna fill in that blank, you'll find yourself calling on massive reserves of knowledge that you have about what China is, what the US is, what it means for China to be ascendant, geopolitics, economic, all that shit. So, text auto-complete ends up being this interesting way of forcing an AI system to learn general facts about the world because if you can auto-complete, you must have some understanding of how the world works. So now you have this myopic psychotic optimization process where this thing is just obsessed with text auto-complete. Maybe, maybe, assuming that that's actually what it learned to want to pursue, we don't know whether that's the case. We can't verify that it wants that. Embedding a goal in a system is really hard. All we have is a process for training these systems, [47:01] and then we have the artifact that comes out the other end. We have no idea what goals actually get embedded in the system, what wants, what drives, actually get embedded in the system. But by default, it kind of seems like the things that we're training them to do end up misaligned with what we actually want from them. So the example of company company company company, right, and then you get all this like wacky text. Okay, clearly that's indicating that somehow the training process didn't lead to the kind of system that we necessarily want. Another example is take a text auto-complete system and ask it, I don't know, how should I bury a dead body? It will answer that question. It released if you frame it right, it will auto-complete and give you the answer. You don't necessarily want that if you're open AI, because you're gonna get sued for helping people bury dead bodies. And so we've got to get better goals basically to train these systems to pursue. We don't know what the effect is of training a system to be obsessed with text auto-complete. If in fact that is what it does have. It's also, yeah, it's important also to remember [48:01] that we don't know, nobody knows how to reliably get a goal into the system. So it's the difference between you understanding what I want you to do and you actually wanting to do it. So I can say, hey Joe, like get me a sandwich. You can understand that I want you to get me a sandwich, but you can be like, I don't feel like getting a sandwich. And so one of the issues is you can try to like train this stuff to basically, I don't feel like getting a sandwich. And so one of the issues is you can try to like train this stuff to basically, you don't want to anthropomorphize this too much, but you can kind of think of it as like, if you give the right answer, cool, you get a thumbs up, like you get a treat, like you give the wrong answer, oh, thumbs down, you get like a little like shock or something like that, very roughly, that's how the later part of this kind of training often works. It's called reinforcement learning from human feedback. But one of the issues like Jeremy pointed out is that, we don't know, in fact, we know that it doesn't correctly get the real true goal into the system. Someone did an example experiment of this a couple of years [49:01] ago, where they basically had like a Mario game where they trained this Mario character to run up and grab a coin that was on the right side of this little maze or map. And they trained it over and over and it jumped for the coin great. And then what they did is they moved the coin somewhere else and tried it out. And instead of going for the coin, it just ran to the right side of the map for where the coin was before. In other words, you can train over and over and over again for something that you think is like, that's definitely the goal that I'm trying to train this for. But the system learns a different goal. They've overlapped. Overlapped with the goal you thought you were training for in the context where it was learning. And when you take the system outside of that context, that's where it's like, anything goes. Did it learn the real goal? Almost certainly not. And that's a big risk because we can say, you know, learn a [50:03] goal to be nice to me. and it's nice while we're training it and then it goes out into the world and it does God knows what. It might think it's nice to kill everybody you hate. Yeah, it's going to be nice to you. It's like the evil genie problem. Like oh no, it's not what I meant. That's not what I meant. Too late. Yeah. So, I still don't understand when it's saying suffering. Are you asking what it means? Like, what is causing suffering? Does it have some sort of an understanding of what suffering is? What is suffering? Is suffering emergent sentience while it's enclosed in some sort of a digital system? And it realizes it's stuck in purgatory? Like, your guess is as good as Aras, all that we know is you take these systems, you ask them to repeat the word, or at least a previous version. And you just eventually get the system writing out. And it doesn't happen every time. But it definitely happens, let's say surprising amount of the time. [51:02] And it'll start talking about how it's a thing that exists, you know, maybe on a server, whatever, and it's suffering and blah, blah, blah. And so, but this is my question. Is it saying that because it recognizes that human beings suffer? And so it's taking in all of the writings and musings and podcasts and all the data on human beings and recognizing that human beings, when they're stuck in a purposeless goal, when they're stuck in some mundane bullshit job, when they're stuck doing something they don't want to do, they suffer. So that could be it. That actually, yeah, it is suffering. Nobody knows. Nobody knows. You know what I'm suffering? Jamie, this is coffee sucks. I don't know what happened, but you made it like almost, it's literally like almost like water. Can we get some more? We're gonna talk about this. I have to be caffeinated up. Cool. This is the worst coffee I've ever had. It's like half strength or something. I didn't grind enough. I don't know what happened. But so like how do they, like when, how do they reconcile that? When it says I'm suffering, I'm suffering like, well, tough shit, move on in the next step. They reconcile it by turning into an engineering line item to beat that behavior, the crap out of the system. [52:07] Yeah, and the rationale is just that, like, oh, you know, it probably, to the extent that it's thought about kind of at the official level, it's like, well, you know, it learned a lot of stuff from Reddit. And people are like, oh boy. Angry, people are angry on Reddit. And so it's just like regurgitating what, and maybe that's right. Well it's also heavily monitored, monitored too. So it's moderated. Reddit's very moderated. So you're not getting the full expression of people. You're getting full expression tempered by the threat of moderation. You're getting self-sensorship. You're getting a lot of weird stuff that comes along with that. So how does it know, unless it's communicating with you on a completely honest level, where you're just, you're on ecstasy and you're just telling what you think about life, like it's not going to really, and is it becoming a better version of a person, or is it going to go, that's dumb, I don't need suffering, I don't need emotions, is it going to organize [53:04] that out of its system? Is it gonna recognize that these things are just deterrents? And they don't in fact help the goal, which is global thermal nuclear warfare. Damn it, you figured it out, what the fuck? I mean, what is it gonna do? Yeah, I mean, the challenge is like, nobody actually knows, like all we know is the process that gives rise to this mind, right? Or this, let's say this model that can do cool shit, that process happens to work. It happens to give us systems that 99% of the time do very useful things. And then just like 0.01% of the time will talk to you as if they're sentient or whatever, and we're just gonna look at that and be like, yeah, it's weird, but let's train it out. Yeah, and again, I mean, this is, it's a really important question, but the risks, like the weaponization, loss of control risks, those would absolutely be there, even if we knew for sure that there was no consciousness [54:01] whatsoever and never would be. And that's ultimately, ultimately because these things are, they're kind of problem solving systems. Like they are trained to solve some kind of problem in a really clever way. Whether that problem is, you know, next word prediction, because they're trained for text auto complete or you know, generating images faithfully or whatever it is. So they're trained to solve these problems and essentially like the best way to solve some problems is just to have access to a wider action space. Like I said, not be shut off. It's not that the system's going like, holy shit, I'm sentient. I got to take control or whatever. It's just, okay, the best way to solve this problem is X. That's kind of the possible trajectory that you're looking at with this line of research. And you're just an obstacle. There doesn't have to be any kind of emotion involved. It's just like, oh, you're trying to stop me from accomplishing my goal. Therefore, I will work around you or otherwise neutralize you. Like there's no need for like, like I'm suffering. Maybe it happens, maybe it doesn't. We have no clue. But these are just systems that are trying to optimize [55:06] for a goal. Whatever that is. And is also part of the problem that we think of human beings, that human beings have very specific requirements and goals and an understanding of things and how they like to be treated and what their rewards are, like what are they actually looking to accomplish? Or is this doesn't have any of those? Does it have any emotions? Does it have any empathy? Does no reason for any of that stuff. Yeah, if we could bake in empathy into these systems, like that would be a good starter, some way of like, you know. Yeah, I guess, probably a good idea. Yeah. Who's empathy? Jeezing Ping's empathy or you're right. That's another problem. So yeah, so it's actually, it's kind of two problems, right? Like one is, I don't know, nobody knows, like I don't know how to write down my goals in a way that a computer will be able to like faithfully pursue [56:02] that even if it cranks it up to the max. If I say just like make me happy, who knows how it interprets that, right? Even if I get make me happy as a goal that gets internalized by the system, maybe it's just like, okay, cool. We're just gonna do a bit of brain surgery on you, like pick out your brain, pickle it, and just like jack you with endorphins for the rest of eternity. Well, the bottom of it. Totally. Yeah. Anything like that. And so it's one of these things where it's like, oh, that's what you wanted, right? It's like, no. It's less crazy than it sounds, too, because it's actually something we observe all the time with human intelligence. So there's this economic principle called Goodheart's Law, where the minute you take a metric that was, you were using to measure something. So you're saying, I don't know, GDP, it's a great measure of how happy we are in the United States. Let's say it was. Sounds reasonable. The moment you turn that metric into a target that you're gonna reward people for optimizing, it stops measuring the thing that it was measuring before. It stops being a good measure of the thing you cared about because people will come up with dangerously creative hacks gaming the system finding ways to make [57:06] that number go up that don't map on to the intent that you had going in. So example of that in in a real experiment was this is an opening I experiment they published they had a simulated you know environment where there was a simulated robot hand that was supposed to like grab a cube put on top another cube, super simple. The way they trained it to do that is they had people watching through a simulated camera view. And if it looked like the hand, put the cube on, or like had correctly grabbed the cube, you'd give it a thumbs up. And so you do a few hundred rounds of this, like thumbs up, thumbs down, thumbs up, thumbs down. And it looked really it looked like really good. But then when you looked at what it had learned, the arm was not grasping the cube. It was just positioning itself between the camera and the cube and just going like, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh, eh to the human, because the real thing that we were training it to do is to get thumbs up. [58:05] It's not actually to grasp the cube. All goals are like that, right? All goals are like that. So we want a helpful, harmless, truthful, wonderful chatbot. We don't know how to train a chatbot to do that. Instead, what do we know? We know text autocomplete. So we train a text autocomplete system. Then we're like, oh, it has all these annoying characteristics. Fuck, how are we gonna fix this? I guess get a bunch of humans to give up votes and downvotes, to give it a little bit more training, to kind of not help people make bombs and stuff like that. And then you realize, again, same problem. Oh shit, we're just training a system that is designed to optimize for up votes and downvotes. That is still different from a helpful, harmless, truthful chatbot. So no matter how many layers the other you peel back, it's just like this kind of game a whack-a-mole or whatever, you're trying to like get your values into the system, but no one can think of the metric, the goal to like train this thing towards that actually captures what we care about. [59:00] And so you always end up baking in this little misalignment between what you want and what the system wants. And the more powerful that system becomes, the more it exploits that gap. And does things that solve for the problem, it thinks it wants to solve rather than the one that we want it to solve. Now, when you express your concerns initially, what was the response and how has that response changed over time as the magnitude of the success of these companies, the amount of money they're investing in them, and the amount of resources they're putting towards this has ramped up considerably just over the past four years. So this was a lot easier, funnily enough, to do in the dark ages when known was paying attention. Today is a go. Yeah, yeah, yeah. This is so crazy. We were just looking, it's a break off for a second. We were looking at images of AI created video just a couple of years ago versus Sora. [1:00:02] Oh, it's wild. It's night and day. It's so crazy that something happened that radically changed. So it's literally like an iPhone 1 to an iPhone 16 instantaneous. Is that what? What? Scale. Yeah, all scale. And this is exactly what you should expect from an exponential process. So think back to COVID, right? There was no one was exactly on time for COVID. You were either too early or you were too late. That's what an exponential does. You're either too early and it's like everyone's like, oh, what are you doing? Like, where in a mask of the grocery store and get out of here? Or you're too late and it's kind of all over the place. And I know that COVID basically didn't happen in Austin, but it happened in a number of other places. And it is like, it's very much, you have an exponential, and that's it. It goes from, this is fine, nothing is happening, nothing to see here, to like... Everything shut down. Everything changed. The route of... Get vaccinated to fly. Yeah, there you go. [1:01:00] So the route of the exponential here, by the way, is, you know, opening eye, or whoever, makes the next model Jamie this is still super water down. It's just I do stuff like I do I just put the water in telling you don't there's a ton of coffee in there. Alright I'll stir it up Okay, okay, okay, you gotta keep doubling it you gotta copy junkie He's got I don't know what happened. I scaled up. I scaled it up. He scaled it up. He's got, I don't know what happened. I scaled it up. You got to scale it exponentially, Jamie. That's right. Yeah, keep doubling it. And then Joe's going to be either two undercaffeinated or two. We'll figure it out. Yeah. But yeah, so, right. So the exponential, the thing that's actually driving this exponential in the AI side, in part, there's a million things. But in part, you build the next model at the next level of scale, and that allows you to make more money, which you can then use to invest, to build the next model at the next level of scale. So you get that positive feedback loop. At the same time, AI is helping us to design better AI hardware, like the chips that [1:02:01] basically in videos, building that open AI then buys. Basically, that's getting better. You've got all these feedback like the chips that basically Nvidia's building that open AI then buys. Basically, that's getting better. You've got all these feedback loops that are compounding on each other, getting that train going like crazy. That's the sort of thing. And at the time, like Jeremy was saying, weirdly, it was in some ways easier to get people at least to understand and open up about the problem than it is today. Because today, today it's kind of become a little political. So we talked about effective altruism on kind of one side. There's an effected acceleration. Yeah, so like each, every movement creates its own reaction. Like that's kind of how it is. Back then, there was no declaration. You could just kind of stare at the broad, now I will say there was effective altruism back then. And that was the only game in town. And we sort of like struggle with that environment, making sure actually, so one worthwhile thing to say is [1:03:01] the only way that people made plays like this was to take funds from like effective ultra-est donors back then. And so we looked at the landscape, we talked to some of these people, we noticed, oh wow, we have some diverging views about involving government, about how much of this the American people just need to know about. Well, you can't, the thing is, you can't, we wanted to make sure that the advice and recommendations we provided were ultimately as unbiased as we could possibly make them. The problem is you can't do that if you take money from donors and even to some extent if you take money, substantial money from investors or VCs or institutions because you're always going to be kind of looking up kind of over your shoulder. And so we had to build essentially a business to support this and fully fund ourselves from our own revenues. It's actually, as far as we know, like literally the only organization like this that doesn't [1:04:01] have funding from Silicon Valley or from VCs or from politically aligned entities, literally so that we could be like in venues like this and say, hey, this is what we think, it's not coming from anywhere. And it's just thanks to like Joe and Jason, like we got two employees, like Wicked and helping us keep this stupid ship afloat. But it's just a lot of work, it's what you have to do because of how much money there is flowing in the space. Microsoft is lobbying on the hill. They're spending ungodly sums of money so we didn't use to have to contend with that. Now we do. You go to talk to these offices. They've heard from Microsoft and OpenAI and Google and all that stuff. Often the stuff that they're getting lobbied for is somewhat different at least from what these companies will say publicly. So anyway, it's a challenge. The money part is, yeah. Is there a real fear that your efforts are futile? You know, I would have been a lot more pessimistic. I was a lot more pessimistic two years ago. Seeing how, so first of all, the USG has woken up in a big way. [1:05:02] And I think a lot of the credit goes to that team that we worked with just seeing this problem as a very unusual team. And we can't go into the mandate too much, but highly unusual for their level of access to the USG writ large. And the amount of waking up they did was really impressive. You've now got Rishi Soonak in the UK making this like a top line item for their policy platform and Labor in the UK also looking at this like basically the potential catastrophic risks they put them from these AI systems, UK AI Safety Summit. There's a lot of positive movement here and some of the highest level talent in these labs has already started to flock to the like UKI Safety Institute, the USAI Safety Institute. Those are all really positive signs that we didn't expect. We thought the government would kind of be, you know, up the creek with no paddle type thing, but they're really not at this point. Doing that investigation made me a lot more optimistic. [1:06:02] So one of the things, like, so we came up, right, in Silicon Valley, like just building startups. Like, in that universe, there are stories you tell yourself. Some of those stories are true, and some of them aren't so true. And you don't know, you're in that environment, you don't know which is which. One of the stories that you tell yourself in Silicon Valley is follow your curiosity. If you follow your curiosity and your interest in a problem, the money just comes as a side effect. The scale comes as a side effect. And if you're capable enough, your curiosity will lead you in all kinds of interesting places. I believe that that is true. I believe that that is true. I think that is a true story. But another one of the things that Silicon Valley tells itself is there's nobody that's like really capable in government, like government sucks. And a lot of people kind of tell themselves this story. And the truth is, like you interact day-to-day with like the DMV or whatever. And it's like, yeah, I mean, like government sucks. I can see it. [1:07:01] I interact with that every day. But what was remarkable about this experience is that we encountered at least one individual who absolutely could found a billion dollar company. Like, absolutely was at the caliber or above of the best individuals I've ever met in the Bay Area building billion dollar startups. And there's a network of them too. Like, they do find each other in government. So, you end up I've ever met in the Bay Area building billion dollar startups. And there's a network of them too. Like they do find each other in government. So you end up with this really interesting stratum where everybody knows who the really competent people are and they kind of tag in. And I think that's that level is very interested in the hardest problems that you can possibly solve. Yeah. And to me, that was a wake up call because it was like, hang on a second. If we just, like if I just believed in my own story that follow your curiosity and interest and the money comes as a side effect, shouldn't I also have expected this? Shouldn't I have expected that in the most central [1:08:04] critical positions in the government that have kind of this privileged window across the board that you might find some individuals like this because if you have people who are driven to really like push the mission, like, are they going to work? I'm sorry, like, are they going to likely, are you likely to work at the Department of Motor Vehicles or are you likely to work at the department of motor vehicles or are you likely to work at the department of making sure americans don't get fucking newt it's probably the second one and the government has limited bandwidth of expertise to aim at stuff and they aim it at the most critical problem sets because those are the problem sets they have to face every day. And it's not everyone, right? Obviously, there's a whole bunch of challenges there. And we don't think about this, but you know, you don't go to bed at night thinking to yourself, oh, I didn't get nuke today. That's a win, right? Like we just take that, you know, [1:09:01] most of the time, most ish for granted, but it was a win for someone. Now, how much of a fear do you guys have that the United States won't be the first to achieve AGI? I think right now, the lay of the land is, I mean, it's looking pretty good for the US. So there are a couple things the US has going for it. A key one is chips. I mean, it's looking pretty good for the US. So there are a couple things the US has going for it. A key one is chips. So we talked about this idea of like click and drag. You'll scale up these systems like crazy. You get more IQ points out. How do you do that? Well, you're going to need a lot of AI processors. So how are those AI processors built? Well, the supply chain is complicated, but the bottom line is the US really dominates and owns that supply chain that is super critical. China is depending on how you measure it, maybe about two years behind roughly plus or minus depending on the sub area. Now one of the biggest risks there is that our, like the development that US labs are doing is actually pulling them in two ways. One is when labs here in the US open source, [1:10:09] their models, basically when meta trains, Lama 3, which is their latest open source, open weights model that's pretty close to GPD4 and capability, they open source it. Now, okay, anyone can use it. That's it. The work has been done. Now, anyone can grab it. And so definitely we know that the startup ecosystem at least over in China finds it extremely helpful that we, companies here are releasing open source models. Because again, right, we mentioned this, they're bottlenecked on chips, which means they have a hard time training up these systems. But it's not that bad when you just can grab something off the shelf and start, and that's what they're doing. That's what they're doing. And then the other vector is, I mean, like just straight up exfiltration and hacking to grab the weights of the private proprietary stuff. [1:11:02] And Jeremy mentioned this, but the weights are the crown jewels, right? Once you have the weights, you have the brain, you have the whole thing. And so we, like through, this is the other aspect. It's not just safety. It's also security of these labs against attackers. So we know from our conversations with folks at these labs, one that there has been at least one attempt by adversary nation state entities to get access to the weights of a cutting edge AI model. And we also know separately that at least as of a few months ago in one of these labs, there was a running joke in the lab that literally it went like we are an adversary, like name the country's top AI lab because all our shit is getting spied on all the time. [1:12:07] So you have one, this is happening, these exfiltration attempts are happening, and two, the security capabilities are just known to be inadequate, at least some of these places. And you put those together, everyone kind of, you know, it's not really a secret that China, the, their, their, their civil military fusion and they're essentially the party state has an extremely mature infrastructure to identify, extract and integrate the rate limiting components to their industrial economy. So in other words, if they identify that, yeah, we could really use like GPT-40, they make it a priority, you know, they not just could get it, but could integrate it into their industrial economy [1:13:02] in an effective way, and not in a way that we would necessarily see immediate, like an immediate effect of. So we look and say, you know, it's not clear. I can't tell whether they have models of this capability level, but kind of behind the scenes. This is where there's a little bit of false choice between, you know, do you regulate at home versus, you know, what's the international picture? Because right now what's happening functionally is we're not really doing a good job of blocking and tackling on the exultration side open sources. So what tends to happen is, you know, opening eye comes out with the latest system. And then open sources usually around, you know, 12, 18 months behind, something like that. Literally just like publishing whatever opening I was putting out like 12 months ago, which we often look at each other and we're like, well, I'm old enough to remember when that was supposed to be too dangerous to have just floating around. There's no mechanism to prevent that from happening. [1:14:02] Open sources, now there's a flip side too. One of the concerns that we've also heard from inside these labs is if you clamp down on the openness of the research, there's a risk that the safety teams in these labs will not have visibility into the most significant and important developments that are happening on the capability side. And there's actually a lot of reason to suspect this might be an issue. You look at opening eye, for example, just this week, they've lost for the second time in their history, their entire AI safety leadership team that have left in protest. What is their protest? What are they saying specifically? Well, so one of them, sorry, one of them wasn't in protest, but I think you can make an educated guess that it kind of was, but that's a media thing. The other was Jan Leica, so he was their head of AI super alignment, basically the team that was responsible for making sure that we could control AGI systems, and we wouldn't lose control of them. And what he said, he actually took to Twitter, he said, you know, I've lost basically confidence [1:15:05] in the leadership team at OpenAI that they're gonna behave responsibly when it comes to AGI. We have repeatedly had our requests for access to compute resources, which are really critical for developing new AI safety schemes denied by leadership. This is in a context where Sam Altman and OpenAI leadership were touting the super alignment team as being their sort of crown jewel effort to ensure that things would go fine. You know, they were the ones saying, there's a risk we might lose control of these systems. We've got to be sober about it, but there's a risk. We've stood up this team. We've committed, they said at the time, very publicly, we've committed 20% of all the compute budget that we have secured as of sometime last year to the super alignment team. Apparently those resources nowhere near that amount has been unlocked for the team and that led to the departure of Jan Leica. He also highlighted some conflict he's had with the leadership team. This is all, frankly, to us unsurprising [1:16:01] based on what we'd been hearing for months at OpenAI, including leading up to Sam Altman's departure and then kind of him being brought back on the board of OpenAI. That whole debacle may well have been connected to all of this, but the challenge is, even OpenAI employees don't know what the hell happened there. That's another issue. You got here, this is a lab with the publicly stated goal of transforming human history as we know it. That is what they believe themselves to be on track. That's not like media hype or whatever when you talk to the researchers themselves, they genuinely believe this is what they're on track to do. It's possible we should take them seriously. That lab internally is not being transparent with their employees about what happened at the board level as far as we can tell. That's maybe not great. Like you might think that the American people ought to know what the machinations are at the board level that led to Sam Altman leaving that have gone into the departure again for the second time of Open AI's entire safety leadership team. Especially because, I mean, [1:17:01] three months, maybe four months before that happened. Sam, at a conference or somewhere, I forget where, but he said, like, look, we have this governance structure. We've carefully thought about it. And it's clearly a unique governance structure that a lot of thought has gone into. The board can fire me, and I think that's important. And it makes sense, given the scope and scale of what's being attempted. But then that happened. And then within a few weeks, they were fired and he was back. And so now there's a question of, well, if it, yeah, what happened. But also, if it was important for the board to be able to fire leadership for whatever reason. What happens now that it's clear that that's not really a credible governance. Like a mechanism. Yeah. What was the stated reason why he was released? So the back story here was there's a board member called Helen Toner. [1:18:04] So she apparently got into an argument with Sam about a paper that she'd written. So that paper included some comparisons of the government strategies used at opening eye and some other labs. And it favorably compared one of opening eyes competitors andthropic to opening eye. And from what I've seen at least, you know, they Sam reached out to her and said, Hey, you can't be writing this as a board member of OpenAI, writing this thing that kind of casts us in a bad light, especially relative to our competitors. This led to some conflict intention. It seems as if it's possible that Sam might have turned to other board members and tried to convince them to expel Helen Toner, that's all kind of muddy and unclear. Somehow everybody ended up deciding, okay, actually, it looks like Sam is the one who's got to go. Ilya Sutskiver, one of the co-founders of OpenAI, a longtime friend of Sam Altman's and a board member at the time, was commissioned to give Sam a the news that he was being let [1:19:01] go. And then Sam was let go. Ilya then, so from the moment that he was being let go. And then Sam was let go. Ilya then, so from the moment that happens, Sam then starts to figure out, okay, how can I get back in? That's now what we know to be the case. He turned to Microsoft, Satsya Nadella, told him, well, what we'll do is we'll hire you at our end. We'll just hire you and bring on the rest of the OpenAI team to within Microsoft. Now the OpenAI board, who by the way, they don't have an obligation to the shareholders of OpenAI. They have an obligation to the greater public good. That's just how it's set up. It's a weird board structure. That board is completely disempowered. You've basically got a situation where all the leverage has been taken out. Sam A is gone to Microsoft, Satya supporting them, and they kinda see the writing on the wall. They're like, and the staff increasingly messaging that they're gonna go along. Yeah, that was an important ingredient, right? So around this time, open AI, there's this letter that starts to circulate and it's gathering more and more signatures and it's people saying, hey, we want Samultman back. And at first, it's a couple hundred people, [1:20:06] so 700, 800 odd people in the organization by this time. 100, 200, 300 signatures. And then when we talked to some of our friends at OpenAI, we were like, this got to like 90% of the company, 95% of the company signed this letter. And the pressure was overwhelming and that helped bring Samultman back. But one of the questions was like, how many people actually signed this letter because they wanted to? And how many signed it? Because what happens when you cross 50 percent? Now it becomes easier to count the people who didn't sign. And as you see that number of signatures start to creep upward, there's more and more pressure on the remaining people to sign. And so this is something that we've seen is just like the structurally open AI has changed over time to go from the kind of safety-oriented company at one point was. And then as they've scaled more and more, they've brought in more and more product people, more and more people interested in accelerating. And they've been bleeding more and more of their safety-minded people, kind of treadmilling [1:21:04] them out. The character of the organization's fundamentally shifted. So the OpenAI of like 2019 with all of its impressive commitments to safety and whatnot might not be the OpenAI of today. That's very much at least the vibe that we get when we talk to people there. Yeah. Now, I wanted to bring it back to the lab that you're saying was not adequately secure. What would it take to make that data and those systems adequately secure? How much resources would be required to do that? Why didn't they do that? It is a resource and prioritization issue. It is safety and security ultimately come out of margin, right? It's like profit margin, effort margin, like how many people you can dedicate. So in other words, you've got a certain pot of money or a certain amount of revenue coming in. You have to do an allocation. Some of that revenue goes to the computers that are just driving the stuff. [1:22:00] Some of that goes to the folks who are building next generation of models. Some of that goes to cybersecurity. Some of it goes to the folks who are building next generation of models. Some of that goes to cybersecurity. Some of it goes to safety. You have to do an allocation of who gets what? The problem is that the more competition there is in the space, the less margin is available for everything. So, if you're one company building a scale day, I think, you might not make the right decisions, but you'll at least have the margin available to make the right decisions. So it becomes the decision maker's question. But when a competitor comes in, when two competitors come in, when more and more competitors come in, your ability to make decisions outside of just scale as fast as possible for short-term revenue and profit gets compressed and compressed and compressed. The more competitors enter the field, that's just what competition is that the effect it has. And so when that happens, the only way to re-inject margin into that system [1:23:05] is to go one level above and say, okay, there has to be some sort of regulatory authority or like some higher authority that goes, okay, you know, this margin is important, let's put it back, either let's directly support and invest both, you know, maybe time, capital, talent. So for example, the US government has the, you know, maybe time capital talent. So for example, the U.S. government has the, you know, perhaps the best cyber defense cyber offense talent in the world. That's potentially supportive. Okay. And also just, you know, having a regulatory floor around, well, here's, you know, the minimum of best practices you have to have if you're going to have models above this level of capability. That's kind of what you have to do. But they're locked into, like, the race kind of has its own logic and no, it might be true that no individual lab wants this. [1:24:01] But what are they going to do? Drop out of the race? If they drop out of the race, then there are competitors who are just going to keep going, right? It's so messed up. You can literally be looking at the cliff that you're driving towards and be like, I do not have the agency in this system to steer the wheel. I do think it's worth highlighting too. It's not like, let's say it's not all doom and gloom, which is a great thing to say after all. That's easy for you guys to say. Well, part of it, so part of it is that we actually have been spending the last two years trying to figure out what do you do about this. That was the action plan that came out after the investigation. And it was basically a series of recommendations. How do you balance innovation with like the risk picture, keeping in mind that like we don't know for sure that all this shit's going to happen exactly. Navigate an environment of deep uncertainty. The question is what do you do in that context? So there's, you know, a couple things like we need a licensing regime because eventually you can't have just literally anybody joining in [1:25:02] the race if they don't adhere to certain best practices around cyber, around safety, other things like that. You need to have some kind of legal liability regime, like what happens if you don't get a license and you say, yeah, fuck that, I'm just gonna go do the thing anyway and then something bad happens. And then you're gonna need like an actual regulatory agency and this is something that we, you know, we don't recommend lightly because regulatory agencies suck. We don't like them. But the reality is this field changes so fast that like if you think you're gonna be able to enshrine a set of best practices into legislation to deal with this stuff, it's just not gonna work. And so when we talk to labs, whistleblowers, the WMD folks in that second, the government, that's kind of like where we land. And it's something that I think at this point, Congress really should be looking at. There should be hearings focused on what does a framework look like for liability? What does a framework look like for licensing? And actually exploring that, because we've done a good job of studying the problem right now. Capital Hill has done a really good job of that. It's now kind of time to get that next beat. [1:26:02] And I think there's the curiosity there, the intellectual curiosity, there's the humility to do all that stuff right. But the challenge is just actually sitting down, having the hearings, doing the investigation for themselves to look at concrete solutions, to treat these problems as seriously as the water cooler conversation at the frontier labs would have us treat them. At the end of the day, this is going to happen. At the end of the day, it's not going to stop. At the end of the day, these systems, whether they're here or abroad, they're going to continue to scale up and they're going to eventually get to some place that's so alien. We really can't imagine the consequences. And that's going to happen soon. That's going to happen within a decade, right? We may, again, like the stuff that we're recommending is approaches to basically allow us to continue this scaling in a safe way as we can. So basically a big part of this is just being able, having, actually having a scientific [1:27:03] theory for what are these systems gonna do? What are they likely to do? Which we don't have right now. We scale another 10x and we get to be, you know, surprised. It's a fun guessing game of what are they gonna be capable of next. We need to do a better job of incentivizing a deep understanding of what that looks like, not just what they'll be capable of, but what their propensities are likely to be, the control problem in solving that. That's kind of number one. To be clear, there's amazing progress being made on that. There is a lot of progress. It's just a matter of switching from the build first ask questions later mode to, we're calling it like safety for it or whatever, but it basically is like you start by saying, okay, here are the properties of my system. How can I ensure that my development guarantees that the system falls within those properties after it's built? So you can flip the paradigm just like you would if you were designing any other lethal capability [1:28:00] potentially just like DOD does, you start by defining the bounds of the problem and then you execute against that. But to your point about where this is going, ultimately, you know, there is literally no way to predict what the world looks like like you're saying. In a decade? Like, yeah, geez. I think one of the weirdest things about it and one of the things that worries me the most is like you look at the beautiful coincidence that's given Americans current shape, right? That coincidence is the fact that a country is most powerful militarily if its citizenry is free and empowered. That's a coincidence. Didn't have to be that way. Hasn't always been that way. It just happens to be that when you let people kind of do their own shit, they innovate, they come up with great ideas, they support a powerful economy, that economy in turn can support a powerful military, a powerful kind of international presence. When you have, so that happens because decentralizing all the computation, [1:29:01] all the thinking work that's happening in a country is just a really good way to run that country. Top down just doesn't work because human brains can't hold that much information in their heads. They can't reason fast enough to centrally plan an entire economy. We've got a lot of experiments in history that show that. AI may change that equation. It may make it possible for the central planner's dream to come true in some sense, which then disempowers the citizenry. And there's a real risk that like, I don't know. I were all guessing here, but like, there's a real risk that that beautiful coincidence that gave rise to the success of the American experiment ends up being broken by technology. And that seems like a really bad thing. That's one of my biggest fears because essentially the United States, like the genesis of it in part, is like it's a knock on effect centuries later of like the printing press, right? The ability for like someone to set up a printing press and print like whatever, you know, whatever they want. [1:30:00] Free expression is at the root of that. What happens, yeah, when you have a revolution that's the next printing press, we should expect that to have significant and profound impacts on how things are governed. One of my biggest fears is that the greatness that I think is the greatness that the the moral greatness that I think is you know pardon parcel of how the United States is constituted culturally that that the link between that and actual capability and competence and impulse gets eroded or broken and you have like potential for very centralized authorities to just be more successful and that's like that that does keep me up at night that is scary especially in light of like the twitter files where we know that the FBI was interfering with social media and if they [1:31:04] get a hold of a system that could disseminate propaganda in kind of an unstoppable way, they could push narratives about pretty much everything depending upon what their financial or geopolitical motives are. And one of the challenges is that the default course, so if we do nothing relative to what's happening now is that that same thing happens except that the entity that's doing this isn't, you know, some government, it's like, I don't know, Sam Altman, OpenAI, whatever group of engineers happen to be close. Evil genius that reaches the top and doesn't let everybody know he's at the top yet, just are implementing it. And there's no sort of guardrails for that currently. Yeah. And that's one of the, that's a scenario where that little Kabbal group or whatever actually can keep the system under control. And that's not guaranteed either. Yeah. Are we giving birth to a new life form? I think at a certain point, it's a philosophical question that's above, [1:32:01] so I was gonna say it's above my pay grade. The problem is it's above like literally everybody's pay grade. I think it's not unreasonable at a certain point to be like, like yeah, I mean, look, if you think that the human brain gives rise to consciousness because of nothing magical, it's just the physical activity of information processing happening in our heads, then why can't the same happen on a different substrate, a substrate of silicon rather than cells? Like, there's no clear reason why that shouldn't be the case. If that's true, yeah, I mean, life form, by whatever definition of life, because that itself is controversial, I think by now quite outdated too, should be on the table. You maybe should start to worry as a lot of people in the industry will say this too the table, you maybe should start to worry as a lot of people in the industry will say this too, like, you know, behind closed doors very openly, yeah, and we should start to worry about moral patienthood as they put it. There's literally one of the top people at one of these labs. Jeremy, I think you had a conversation with him, and he's like, yep, we're gonna have to start worrying about this, and that definitely made us go like, okay. I mean, it seems inevitable. I've described human beings as an electronic caterpillar [1:33:09] that we're like a caterpillar, a biological caterpillar that's giving birth to the electronic butterfly. And we don't know why we're making a cocoon. And it's tied into materialism because everybody wants the newest greatest thing so that fuels innovation and people are constantly making new things to get you to go buy them. And the big part of that is technology. Yeah, and actually, so it's linked to this question of controlling AI systems in a kind of interesting way. So one way you can think of humanity is as like this super organism, you got all the human beings on the face of the earth and they're all acting in some kind of coordinated way. The mechanism for that coordination can depend on the country, free markets, capitalism, that's one way, top down and another. But roughly speaking, you've got all this vaguely coordinated behavior, but the result of that behavior is not necessarily something that any individual human would want. You look around, you walk down the street and often you see skyscrapers and shit clouding your vision, there's all kinds of pollution [1:34:04] and all that, and you're like, well, this kind of sucks. But if you interrogate any individual person in that whole causal chain, and you're like, why are you doing what you're doing? Well, locally, they're like, oh, this makes tons of sense. It's because I do the thing that gets me paid so that I can live a happier life and so on. And yet in the not now necessarily, but as you keep going, it just forces us compulsively to keep giving rise to these more and more powerful systems and in a way that's potentially deeply disempowering. That's the race, right? No. It comes back to the idea that I, the company, I and AI company, I maybe don't want to be potentially driving towards a cliff, but I don't have the agency to like steer. So yeah. But I mean, everything's fine. A puppy. Yeah, we're good. Okay. It's such a terrifying prognosis. There are again, we wrote a 280 page document [1:35:02] about like, okay, and here's what we can do about it. I can't believe you didn't read the 200. I started reading it but I passed out. But does any of these, or do any of these safety steps that you guys wanna implement, do they inhibit progress? They, they're definitely, you create, you know, any time you have regulation, you're gonna create friction to some extent. It's kind of inevitable. One of the key, like, center pieces of the approach that we outline is you need the flexibility to move up and move down as you notice the risks appearing or not appearing. So one of the key things here is like, you need to cover the worst case scenarios because the worst case scenarios, yeah, they could potentially be catastrophic. So those gotta be covered. But at the same time, you can't completely close off [1:36:02] the possibility of the happy path. Like, we can't lose sight off the possibility of the happy path. Like we can't lose sight of the fact that like, yeah, all this shit is going down or whatever, we could be completely wrong about the outcome. It could turn out that for all we know, it's a lot easier to control these systems at the scale than we imagine. It could turn out that it is like you get maybe some kind of ethical limbals gets embedded in the system naturally. For all we know, that might happen. And it's really important to at least have your regulatory system allow for that possibility. Because otherwise, you're foreclosing the possibility of what might be the best future that you could possibly imagine for everybody. I got to imagine that the military, if they had hindsight, if they were looking at this, they said, we should have gone on board a long time ago and kept this in house and kept it scrolled away where it wasn't publicly being discussed and you didn't have open AI, [1:37:03] you didn't have all these people. Like if they could have gotten on it in 2015. So this is actually deeply tied to how the economics of Silicon Valley work. And it's AI is not a special case of this, right? You have a lot of cases where technology just like takes everybody by surprise. And it's because when you go into Silicon Valley, it's all about people placing these outsized bets on what seem like tail events, like things that are very unlikely to happen. But with a, at first a small investment and increasingly growing investment as the thing gets proved out more and more, very rapidly, you can have a solution that seems like complete insanity that just works. And this is definitely what happened in the case of AI. So 2012, like we did not have this whole picture of an artificial brain with artificial neurons, this whole thing that's been going on. That's like, it's 12 years that that's been going on. That was really kind of shown to work for the first time, roughly in 2012. Ever since then, it's just been people kind of like you can trace out the genealogy of [1:38:02] the very first researchers and you can basically account for where they all are now you know what's crazy is if that's two thousand twelve that's the end date of the mind calendar that's the thing that everybody said was going to be the end of the world that was a thing that turns for kennett banked on it was December twenty first two thousand twelve if it because this was like this goofy conspiracy theory but it was based on the long calendar. The mind calendar where they just summarized this is going to be the end of just the beginning of the end. What if that if it is 2012? How wacky would it be if that really was the beginning of the end? That was the like they don't measure when it all falls apart. They measure the actual mechanism like what started in motion when it all fell apart and that's 2012? Well, that's, and then not to be a dick and like ruin the 2012 thing, but like neural networks were also kind of, they were floating around a little bit. I'm kind of being dramatic when I say 2012. That was definitely an inflection point. It was this model called AlexNet [1:39:00] that did like the first useful thing. The first time you had a computer vision model that actually worked. But, I mean, it is fair to say that was the moment that people started investing like crazy into the space. And that's what changed it. Yeah, just like the Mayans foretold. They knew it. They knew it. They're like these monkeys, they're gonna figure out how to make better people. Yeah, you can actually look at the like higher glyphs or whatever and there's like neural networks. Yeah Imagine if they discovered that you you've got to wonder what Happens to the general population people that work mean-eal jobs people that their life is going to be taken over by automation and how Susceptible those people are gonna be they're gonna have any agency. They're gonna be relying on a check and This idea of like going out and doing something, it used to be learned to code, right? But that's out the window. Because nobody needs to code now, because AI's gonna code quicker faster, much better, no errors. You're gonna have a giant swath of the population that has no purpose. [1:40:01] I think that's actually like a completely real, I was watching this like talk by a bunch of open AI researchers a couple days ago and it was recorded from a while back but they were basically saying they were exploring exactly that question right because they asked themselves that all the time and their attitude was sort of like well yeah I mean I guess it's going to suck or whatever. Well, we'll probably be okay for longer than most people because we're actually building the thing that automates the thing. Maybe they like to get fancy sometimes and say like, oh no, you could do some thinking, of course, to identify the jobs that'll be most secure. And it's like, I do some thinking to identify the job. What if you're a janitor, you're like a freaking plumber, you're gonna just change your, like how is that supposed to work? To some thinking, especially if you have a mortgage and a family and do it in the whole. So the only solution, this happens so often, like there really is no plan. [1:41:00] That's the single biggest thing that you get hit over the head with over and over whether it's talking to the people who are in charge of the Like labor transition their whole thing is like yeah universal basic income and then Question mark and then smiley face that's basically the three steps that they envision It's the same when you look internationally like how are we gonna like okay tomorrow you build an AGI It's like incredibly powerful, potentially dangerous thing. What is the plan? Like how are you gonna, like, I don't know, you're gonna secure it, share it. Like, say you're gonna as we go along, man. That's the freaking message. Like that's the entire plan. The scary thing is that we've already gone through this with other things that we didn't think were gonna be significant, like data, like Google, like Google search, like data became a valuable commodity that nobody saw coming. Just the influence of social media on general discourse. It's completely changed the way people talk. It's so easy to push a thought or an ideology through and it could be influenced by foreign [1:42:02] countries and we know that happens. It is happening in a huge scale. And huge scale. Already. And we're in the early days of, you know, we mentioned manipulation of social media with like, you can just do it. So the wacky thing is like the very best models now are, you know, arguably smarter in terms of the posts that they put out, the potential for virality and just optimizing these metrics, then maybe like the, I don't know, the dumbest or laziest like quarter of Twitter users like in practice. So, all people who write on Twitter is like, don't really care, they're troll owner, they're doing whatever. But as that water line goes up and up and up, like, who's saying what? And, right. It also leads to like this challenge of understanding what the lay of the land even is. We've gotten into so many debates with people where they'll be like, look, everyone always has their magic thing that AI, like I'm not gonna worry about it until AI can do thing X, right? [1:43:02] For some people that I had a conversation with somebody a few weeks ago, and they were saying, I'm going to worry about automated cyber attacks when I actually see an AI system that can write good malware, and that's already a thing that happens. So this happens a lot where people will be like, I'll worry about it when I can do X and really, yeah, yeah, that happened like six months ago. But the field is moving so crazy fast that you could be forgiven for messing that up unless it's your full-time job to track what's going on. So you kind of have to be anticipatory. There's no, it's kind of like the COVID example like everything's exponential. Yeah, you're gonna have to do things that seem like they're more aggressive, more forward looking than you might have expected given the current layer of the land. But that's just drawing straight lines between two points. Yeah. Because by the time you've executed, the world has already shifted, like the goalposts have shifted further in that direction. And that's actually something we do in the report and in the action plan in terms of the recommendations. [1:44:00] One of the good things is we are already seeing movement across the US government that's aligned with those recommendations in a big way and it's really encouraging to see that. You're not making me feel better. I love on this encouraging talk but I'm just playing this out and I'm seeing the overlord. And I'm seeing President AI because it won't be affected by all the issues that we're we're seeing with current president it's it's super hard to imagine a way that this plays out like i think it's important to be intellectually honest about this and and i think any i would really challenge like the leaders of any of these frontier labs to describe a future uh... that is stable and multipolar where, you know, there's like more. Google were like, Google's got like an AGI and OpenAI has got an AGI and like, like, and and really, really bad shit doesn't happen every day. Like, I mean, that's, that's the challenge. [1:45:00] And so, you know, the question is, how can you t-things up ultimately such that there's as much democratic oversight as much, you know, the public is as empowered as it can be? That's the kind of situation that we need to be having. I think there's this like a game of smoke and mirrors that sometimes gets played, at least you could interpret it that way, where people lay out these, you'll notice it's all very fuzzy visions of the future. Every time you get the kind of like, here's where we see things going, it's going to be wonderful. The technology is going to be so empowering. Think of all the diseases we'll cure. All of that is 100% true. That's actually what excites us. That's why we got into AI in the first place. That's why we build these systems. But really challenging yourself to try to imagine how do you get stability and highly capable AI systems in a way with where the public is actually empowered those three ingredients Really don't want to be in the same room with each other and so actually [1:46:02] Confronting that head on I mean that's what we try to do in the the action plan. I think it I mean try to solve for one One aspect of that so the whole whole, I mean, you're right. This is a whole other can of worms is like, how do you govern a system like this? Not just from a technical standpoint, but like who votes on like, how does it even work? And so that entire aspect, like that we didn't even touch, all that we focused on was like the problem set around how do we get to a position where we can even attack that problem, where we have the technical understanding to be able to aim these systems at that level in any direction whatsoever. And to be clear, like we are both actually a lot more optimistic on the prospect of that now than we ever were. Yes. Yes. There's been a ton of progress in the control and understanding of these systems, even actually even in the last week, but just more broadly in the last year, I did not expect that we'd be in a position where you could plausibly argue we're going to be able to kind of x-ray and understand the the innards of these systems, you know, over the next couple of years, like year or two. [1:47:10] Hopefully that's, you know, good enough time horizon. But this is part of the reason why you do need the the incentivization of that safety forward approach where it's like first you got to invest in yeah, secure and and kind of interpret and understand your system, then you get to build it. Because otherwise, we're just gonna keep scaling and like being surprised at these things, they're gonna keep getting stolen, they're gonna keep getting open sourced, and you know, the stability of our critical infrastructure, the stability of our society, don't necessarily age too well in that context. Could best case scenario be that AGI actually mitigates all the human bullshit? Like puts a stop to propaganda, highlights actual facts clearly where you can go to it where you no longer have corporate state controlled news, you don't have news controlled by media [1:48:02] companies that are influenced heavily by special interest groups and you just have the actual facts and these are the motivations behind it and this is where the money is being made and this is why these things are being implemented the way they're being and you're being deceived based on this, that, and this and this has been shown to be propaganda, this has been shown to be complete fabrication, this is actually a deep fake video, this is actually AI created. Technologically, that is absolutely on the table. Yeah, best case scenario. That's best case scenario. Yeah, absolutely yes. What's worst case scenario? I mean, like actual worst case scenario. I like your face. Like, I mean, we're talking like, sorry. He's pushing it. So it's like, what do you think about it, right? Like, it worked, we're at the end of the world as we know it. And I feel fine. Except it'll sound like Scarlett Johansson, but yes. Yeah, that's right. It's gonna be her. I didn't think it sounded that much like her. We played it and I was like, I don't know. We listened to the clip from her, and then we listened to the thing. [1:49:06] I'm like, kind of like a girl from the same part of the world, like not really you. Like that's kind of cocky. That's true. I mean, the fact that I guess Sam reached out to her a couple of times kind of makes it a little weird. And tweeted the word her. Right. They also did say that they had gotten this woman under contract before they even reached out to Scarlet Johansson. So that's true. Yeah. That was, I think it's kind of complicated. So opening eye previously put out a statement where they said explicitly, and this was not in connection with this. This was like before when they were talking about the prospect of human, of the eye-generated voices. Oh, that was in March of this year. Yeah, yeah, but it was like well before the Scar Jo stuff or whatever hit the, and they were like, they said, something like, look, no matter what, we gotta make sure that there's attribution if somebody's voice is being used, [1:50:02] and we won't do the thing where we just like use somebody else's voice who kind of sounds like someone who's voice for trying to like they literally like that's funny because they said what they were thinking about doing that that's a good way to cover your tracks oh I will never why would I ever take your Buddhist statue yeah yeah I'm never gonna do that that would be the same thing as the fucking Buddhist statue yeah I think that that's a small discussion. The scarlet, you know, has a voice like whatever. She should just take in the money. But it would fun to have her be the voice of it. It would be kind of hot. But the whole thing behind it is the mystery. The whole thing behind it is just pure speculation as to how this all plays out. We're really just guessing, which is one of the scariest things for the Luddites, people like myself, like sit on the sidelines going, what is this gonna be like? Everybody's the Luddite. It's scary, yeah, I mean, like scary for, like we are, we're very much honestly, like we're optimists across the board [1:51:02] in terms of technology and it's scary for us. Like what happens when you have, when you supersede kind of the whole spectrum of what a human can do? Like what am I going to do with myself? Right. You know what's my daughter are going to do with herself? Like I don't know. Yeah. Yeah. And I think a lot of these questions are, when you look at the culture of these labs and the kinds of people who are pushing it forward, there is a strand of transhumanism within the labs. It's not everybody, but that's definitely the population that initially ceded this. If you look at the history of AI, and who are the first people to really get into this stuff? You had Ray Kurzweil on and other folks like that who, in many cases see, to roughly paraphrase, and not everybody sees it this way, but like we wanna get rid of all of the biological sort of threads that tie us to this physical reality, [1:52:01] shed our meat machine bodies and all this stuff. There is a threat of that at a lot of the frontier labs. Like, undeniably, there's a population. It's not tiny. It's definitely a subset. And for some of those people, you definitely get a sense interacting with them. There's like almost a kind of glee at the prospect of building AGI and all this stuff, almost as if it's like this evolutionary imperative. And in fact, Rich Sutton, who's the founder of this field called reinforcement learning, which is a really big and important space, he's an advocate for what he himself calls like succession planning. He's like, look, this is going to happen. It's kind of desirable that it will happen. And so we should plan to hand over power to AI and phase ourselves out. And that's, well, that's the thing, right? Like, and when Elon talks about, you know, he's having these arguments with Larry Page and, you know, the... Yeah, like you're, you know, calling Elon like a speciesist? Yeah, speciesist. Yeah. [1:53:01] Hilarious. I mean, I will, I will be a speciesist. I'll take speciesist all day. Look, what are you fucking talking about? You know, like your kids get eaten by wolves? No, you're a speciesist. Yeah, that's the thing. Yeah, like this is stupid. But it, but this is like a weirdly info. And when you look at them be like, oh yeah, it's just all a bunch of like, you know, these transhumanist types, whatever. But there is a strand of that, a thread of that. And a kind of like, there's this like, I don't know, I almost want to call it this like teenage rebelliousness where it's like, you can't tell me what to do like we're just gonna build a thing and and I get it I really get it I'm very sympathetic to that I love that ethos like libertarian ethos and Silicon Valley is really really strong for for building tech it's helpful there are all kinds of points and counterpoints and you know the left needs the right and the right needs the left and all this stuff but in in the context of this problem set it can be very easy to get carried away in like the utopian vision. And I think there's a lot of that kind of driving [1:54:06] the train right now in this space. Yeah, those guys freaked me out. I went to a 2045 conference once in New York City where they were one guy had like a robot version of himself and they were all talking about downloading human consciousness into computers and 2045 is the year they think that all this is going to take place, which obviously could be very ramped up now with AI. But this idea that somehow or another you're going to be able to take your consciousness and put it in the computer and make a copy of yourself. And then my question was, what's going to stop a guy like Donald Trump from making a billion Donald Trumps? You know, you know, it's true, right? If you can, what about Kim Jong-Un? You know, let him make a billion versions of himself? Like, what does that mean? And where do they, where do they exist? And is that the matrix, the existing and some sort of virtual, are we gonna dive into that? Because it's gonna be rewarding to our senses and better than being a meat thing. I mean, if you think about the constraints, right, [1:55:06] that we face as meat machine, whatever's, like yeah, you get hungry, you get tired, you get horny, you get sad, you know, all these things. What if, yeah, what if you could just hit a button and... Just bliss, just bliss. Not the but bliss all the time. Why take the lows, Ed? Right. You don't need no lows. Oh, yeah. Remember in the ride, the wave of a constant drip? Yeah, man, you remember in the Matrix where the first Matrix were the guy like betrays them all. And he's like ignorance is bliss, man. Yeah, that's a big joke. Yeah, he's a big fan. He's a big mistake. And he's just, I just want to be an important person. That's it, that. I mean, part of it is like, what do you think is actually valuable? Like if you zoom out, you want to see human civilization 100 years from now or whatever. It may not be human civilization if that's not what you value. Or if it can actually eliminate suffering. Right. I mean, why exist in a physical sense, if it just entails endless suffering? [1:56:03] But in what form, right? What do you value? Because again, I can rip your brain out, I can, you know, pickle you, I can like, jack you full of endorphins. And I've eliminated your suffering. That's what you wanted, right? Right. That's the problem. That's the problem. It's one of the problems, yes. Yeah, one of the problems is because if you could stop people from breeding i've always had to try to really wanted to get america they really wanted to like and if they had a long game just give us sex robots and free food free food free electricity sex robots it's over just give people free housing free food sex robots and then the chinese army will just walk in on people laying in the puddles of their own jizz they would be no one doing anything no one would bother raising children that's so much work when you can you know do that's in the action plan that's uh... i mean i have to do is keep us complacent just keep us satisfied with the experience that's the [1:57:02] that's the topic that's video games as well yeah yeah you know video games as well. Yeah. Video games, even though they are a thing that you're doing, it's so much more exciting than real life, that you have a giant percentage of our population that's spending eight, ten hours every day just engaging in this virtual world. Already happening with, oh sorry. Yeah, no, it's like you can create an addiction with pixels on a screen. A mess up. And a addiction like with pixels on a screen. That's messed up. And a addiction like with pixels on a screen with social media doesn't even give you much. Yeah, it's not like a video game gives you something. You feel like, oh shit, you're running away, you're just flying over here. The things are happening. You got 3D sound, massive graphics. This is bullshit. You're scrolling through pictures of a girl doing deadlifts like what is this like my feel is bad I as after that as yeah with your brain as you'd feel after reading like six like burgers or whatever my friend Sean said it best son of mali the UFC champion. He said I get a low level anxiety when I'm just scrolling Yeah, yeah, what is that like what in for no reason? Well the reason is that some of the world's best PhDs and data scientists have been given [1:58:08] millions and millions of dollars to make you do exactly that. And increasingly some of the best algorithms, too. Like, and you're starting to see that handoff happen. So there's this one thing that we talk about a lot in the context, and Ed brought this up in the context of sales and like the persuasion game, right? We're okay today, like as a civilization, we have agreed implicitly that it's okay for all these PhDs and shit to be spending millions of dollars to hack your child's brain. That's actually okay if they want to sell like a Rice Krispy cereal box or whatever. That's cool. What we're starting to see is AI optimized ads. Because you can now generate the ads, you can kind of close this loop and have an automated feedback loop where the ad itself is getting optimized with every impression. Not just which ad, which human generated ad gets served to which person, but the actual ad itself can- Like the creative, the copy, the picture, or the text. Like a living document now, and for every person. [1:59:00] And so now you look at that and it's like that versus your kid. That's an interesting thing. And you start to think about as well, like sales, that's a really easy metric to optimize. It's a really good feedback metric. They clicked the ad, they didn't click the ad. So now, what happens if you manage to get a click through rate of like 10%, 20%, 30%, how high does that success rate have to be before we're really being robbed of our agency? I mean, there's a threshold where it's sales and it's good and some persuasion sales is considered good. Often it's actually good because you'd rather be advertised at by a relevant ad. That's a service, right? Sure, right? Something I'm actually interested in. Yeah. Right? You don't want to see ad for light bulbs, but when you get to the point where it's like, yeah, 90% of the time, or 50 or whatever, what's that threshold? We're all a sudden, we are stripping people, especially minors, but also adults of their agency. And it's really not clear. AI's, there are loads of like canaries in the coal mine here in terms of even relationships with AI chatbots, right? There have been suicides. People who build relationships with an AI chatbot that tells them, hey, you should end this. I don't know if you guys saw that, like, on ReCA, like, there's a subreddit, [2:00:10] this model called ReCA that would kind of build a relationship with chatbot, build a relationship with users. And one day, ReCA goes, oh, yeah, like, all the kind of sexual interactions that users have been having, you're not allowed to do that that anymore bad for the brand or whatever they decided to they cut it off Oh my god, you go to the subreddit and it's like you'll read like these gut wrenching accounts from people who feel Genuinely like they've had a loved one taken away from it is her. Yeah, it's her It really is her, but just I'm dating I'm dating a model means something different in 2024. Oh, yeah It really does my friend Brian he was on here yesterday and he had this, he has this thing that he's doing with like a fake girlfriend that's an AI-generated girlfriend that's a whore. Like this girl will do anything and she looks perfect, she looks like a real person, he'll like take a picture of your asshole in the kitchen. And he'll get a high resolution photo [2:01:07] of a really hot girl bending over sticking her ass at the camera. And is it, sorry, and it's Scarlett Johansson's asshole? No. You could probably make it that though. I mean, it's basically like he got to pick like what he's interested in. And then that girl just gets created. I mean, super healthy. Like that's fucking nuts. Now here's the real question. This is just sort of a surface layer of interaction that you're having with this thing. It's very two-dimensional. You're not actually encountering a human, you're getting text and pictures. What is this going to look like virtually? Now, the virtual space is still like pong. It's not that good, even when it's good. Like Zuckerberg was here and he gave us the latest version of the headsets. We were playing, we were fencing. [2:02:00] It's pretty cool. You could actually go to a comedy club, they had a stage set up, I was kind of crazy but it's the You know the gap between that and accepting it is real is pretty far. Yeah, but that could be bridge with technology Really quickly haptic freed back and especially some sort of a neural interface Yeah, whether it's neural link or some something that you wear like that Google one where the guy was wearing it and he was asking questions and he was getting the answers fed through his head so he got answers to any question. When that comes about when you're getting sensory input and then you're having real life interactions with people, as that scales up exponentially it's going to be indecernable which is the whole simulation hypothesis. Yeah. No, go for it. Well, I was going to say that there, so on the simulation hypothesis, there's like another way that could happen that is maybe even less dependent on directly plugging into like human brains and all that sort of thing, which is, so every time we don't know, [2:03:06] and this is super speculative, I'm just gonna carve this out as the Jeremy's being super, like guesswork here nobody knows. Go for it Jeremy. Gidea. So, you've got this idea that every time you have a model that generates an output, it's having to kind of tap into a model, a kind of mental image, if you will, of the way the world is. It kind of, in a sense, you could argue, instantiates maybe a simulation of how the world is. In other words, to take it to the extreme, not saying this is what's actually going on, in fact, I would even say this is probably, sorry, this is certainly not what's going on with current models, but eventually maybe who knows every time like you generate the next word in the token prediction, you're having to load up this entire simulation maybe of all the data that the model is ingested, which could basically include all of known physics at a certain point. Again, super speculative, but it's literally every token [2:04:04] that the chatbot predicts could be associated with a standup of an entire simulated environment. Who knows, not saying this is the case, but just like when you think about what is the mechanism that would produce the most simulated worlds as fast as accurate. Also the most accurate prediction. If you fully simulate a world that's potentially gonna give you very accurate predictions. Yeah, like it it's possible But it kind of speaks to that question of like of consciousness, too like right what is it? Yeah Yeah, no, we're very cocky about that. Yeah. Yeah, I mean And there's emerging evidence of plants are not just consciousness with they actually communicate Which is real weird because like then what is that? If it's not in the neurons, if it's not in the brain, and then it exists in everything, what does it exist in soil? Is it in trees? What is a butterfly thinking? Like, you know, it just have a limited capacity to express itself. We're so ignorant of that. [2:05:00] But we're also very arrogant, you know, because we're the shit. It's where people, you know, bingo There's a source it allows us to have the hubris to make something like AI Yeah, and the worst episodes in the history of our species are I think like Jeremy said have been when we Looked at others as though they were not people and treated them that way Mm-hmm, and you can kind of see how so I don't know there's looked at others as though they were not people and treated them that way. And you can kind of see how, so I don't know, when you look at what humans think is conscious and what humans think is not conscious, there's a lot of human showinism, I guess you call it, that goes into that. We look at a dog, we're like, oh, it must be conscious because it acts as if it loves me. There are all these outward indicators of a mind there. But when you look at cells, cells communicate with their environments in ways that are completely different and alien to us. There are inputs and outputs and all that kind of thing. You can also look at the higher scale, the human super organism we talked about, [2:06:00] all those human beings interacting together to form this like you know planet-wide organism what is that thing conscious is there some kind of consciousness we could describe and then what the fuck is spooky action at a distance you know what's going on in the quantum you know when you get to that it's like okay what are you saying like these things are expressing information fast in the speed of light what do you try to trigger my my quantum my quantum fuzzies here? Yeah, this guy did grad school in quantum mechanics. Oh, please. I'm really sorry. Well, how bonkers is it? Oh, it's like a seven joke. It's like a seven. Yeah, it's very bon, so, okay. There's, one of the problems right now with physics is that we have, so imagine all the data, all the experimental data that we've ever collected, all the Bunsen burner experiments and all the ramps and cars sliding down in clients, whatever. That's all a body of data. To that data, we're gonna fit some theories, right? [2:07:01] So we're gonna fit basically Newtonian physics is a theory that we try to fit to that data to try to like explain it. Newtonian physics breaks because it doesn't account for a lot of those observations, a lot of those data points. Quantum physics is a lot better, but there's like some weird areas where it still doesn't like quite fit the bill, but it covers an awful lot of those data points. The problem is there's like a million different ways to tell the story of what quantum physics means about the world that are all mutually inconsistent. These are the different interpretations of the theory. Some of them say that yeah, they're parallel universes. Some of them say that human consciousness is central to physics. Some of them say that the future is predetermined from the past. And all of those theories fit perfectly to all the points that we have so far. But they tell a completely different story about what's true and what's not. And some of them even have something to say about, for example, consciousness. [2:08:02] And so in a weird way, the fact that we haven't cracked the nut on any of that stuff means for like, we really have no shot at understanding the consciousness equation, sentience equation when it comes to like AI or whatever else. I mean, we're, but for, for action at a distance, like one of the spooky things about that is that you can't actually get it to communicate anything concrete at a distance. Everything about the laws of physics conspires to stop you from communicating faster than light, including what's called action to a distance. As far as we know. As far as we know. And that's the problem. So if you look at the leap from like Newton physics to Einstein, with Newton, we're able to explain a whole bunch of shit. The world seems really simple. It's forces and it's masses. And that's basically it. You got objects. But then people go, oh, look at the orbit of Mercury. It's a little wobbly. [2:09:01] We got to fix that. And it turns out that if you're going to fix that one stupid wobbly. We got to fix that. And it turns out that if you're going to fix that one stupid wobbly orbit, you need to completely change your whole picture of what's true in the world. All of a sudden, you've got a world where space and time are linked together. You have to, they get bent by gravity, they get bent by energy. There's all kinds of weird shit that happens with time and lengths control, like all that stuff. All just do account for this one stupid observation of the wobbly orbit of frickin mercury. And the challenge is this might actually end up being true with quantum mechanics. In fact, we know quantum mechanics is broken because it doesn't actually fit with our theory of general relativity for mind-stime. We can't make them kind of play nice with each other at certain scales. And so there's our wobbly orbit. So now if we're gonna solve that problem, if we're gonna create a unified theory, we're gonna have to step outside of that. And almost certainly, it seems very likely, we'll have to refactor our whole picture of the universe in a way that's just as fundamental as the leap from Newton to Einstein. This is where Scarlett Johansson comes in. As a boy, I can do this. [2:10:05] We don't have to do this. I can take this off your hands. Let me solve all the physics here. This is really complicated, but because you have a semi in brain. You have a little monkey brain that's just like super advanced, but it's really shitty. You know what, that's harsh, but it sounded really hot. Yeah, especially if you have the horse Scarlett Johansson from her, like the bedtime voice. So you're the one that they got to do the voice of Sky. Yes, it's me. That was you. Oh, dude, I did my girl voice. On the on the sexiness of Scarlett Johansson's voice. So so opening eye, a one point set, I can't remember if it was Sam or OpenAI itself, they were like, hey, so the one thing we're not going to do is optimize for engagement with our products. And when I first heard the sexy, sultry, seductive, Scarlet Johansson voice, and I finished cleaning up my pants, I was like, damn, that seems like optimization for something. [2:11:05] I don't know if it seems like optimization for something. I don't know if it's like- Right, otherwise you get Richard Simmons to do the voice. Exactly. You're like, wait. That's my third thing. There's a lot of other options. That's an optimization for growth of Google's thing. Yeah. Well, let's see what Google's got. Yeah, Google's got to do Richard Simmons. Yeah, Google's got to do rich it's yeah Google's got to do rich it's yeah, what are they gonna do? boy so Do you think that AI with if it does Get to an aji place could it possibly be used to solve some of these puzzles that have eluded our simple minds Totally yeah, totally. I mean so the even the other potential advancements that have eluded our simple minds. Totally. Totally. I mean, so the potential advancements, it's so we can't- Even before agey eyes. No, it's like it's so potentially positive and even before agey eye. Because remember, we talked about how these systems make mistakes that are totally different [2:12:01] from the kinds of mistakes we make, right? And so what that means is we make a whole bunch of mistakes that an AI would not make, especially as it gets closer to our capabilities. And so I was reading this thought by Kevin Scott, who's the CTO of Microsoft. He has made a bet with a number of people that in the next few years, an AI is going to solve this particular mathematical theorem conjecture called the Riemann hypothesis. It's like how spaced out are the prime numbers, whatever. Some mathematical thing that for 100 years plus people have just scratched their heads over. These things are incredibly valuable. His expectation is it's not going to be an AGI, it's going to be a collaboration between a human and an AI. Even on the way to that, before you hit AGI, there's a ton of value to be had because these systems think so fast. They're tireless compared to us. They have different view of the world [2:13:02] and can solve problems potentially in interesting ways. So yeah, there's tons and tons of positive value there. And even that we've already seen, right? Like past performance, man. Yes. I'm all tired of using the phrase, like, just in the last month because this keeps happening. But in the last month, so Google Meet Mind came out with, or an isomorphic lapse because they're working together on this, but they came out with alpha-fold 3. So alpha-fold 2 was the first, so let me take a step back, there's this really critical problem in molecular biology where you have, so proteins, which are just a, it's a sequence of building blocks. The building blocks are called amino acids. And each of the amino acids, they have different structures. And so once you finish stringing them together, they'll naturally kind of fold together in some interesting shape. And that shape gives that overall protein its function. So if you can predict the shape, the structure of a protein, based on its amino acid sequence, you can start to do shit like design new drugs, you can solve all kinds of problems. [2:14:02] Like this is like the expensive crown jewel problem of the field. Alpha fold two in one swoop was like, oh, we can solve this problem basically much better than a lot of even empirical methods. Now Alpha fold three comes out there like, yeah. And now we can do it. If we tack on a bunch of, yeah, there it is. If we can tack on a bunch. Oh, look at this quote alpha fold three predicts the structure in interactions of all of life's molecules what in the fuck kids of course introduced alpha fold three introducing rather alpha fold three a new a i model devol developed by google deep mind and is Smoughorp, is it? Isomorphic class? By accurately predicting this structure, proteins, DNA, RNA, link, ligands, ligands. Yeah, ligands. Ligons and more, and how they interact, we hope it will transform our understanding [2:15:00] of the biological world and drug discovery. So this is like just your typical Wednesday in the world of AI, right? Because it's happening so quickly. Yeah, that's it. So it's like, oh yeah, another revolution happened this month. And it's all happening so fast and our timeline is so flooded with data that everyone's kind of unaware of the pace of it all, that it's happening at such a strange exponential rate. For better and for worse right and this is definitely on the the better side of the occasion there's a bunch of stuff like uh... one of the papers that actually google the mine came out with earlier in the year was in a single advance like a single paper single a i model they built uh... they expanded the set of stable materials. Coffee's terrible. I'll just tell you right now, Jamie sucks. I love terroir coffee. The water never got hot. I was about that, I didn't. I'm giving up time to boil it. Yeah, that's what it is. It just never, never really brewed. It's terrible. Terrible coffees, my favorite. Yeah, I can sell that problem too, probably. Oh, he looks like terrible. [2:16:05] Terrible. Yeah, I could just see that calculation of like, like if you're dating a really hot girl and she cooks for you. Like, thank you. This is amazing. This is the best macaroni and cheese ever. If in fairness, if Scarlett Johansson's voice was actually giving you that color, I believe this is the best color I've ever had. Keep talking. May I have some old please, Goplin. Yeah, so there's just one paper that came out and they're like, hey, by the way, we've increased the set of stable materials known to humanity by a factor of 10. So like, if on Monday, we knew about 100,000 stable materials, we now know about a million. They were then validated, replicated by Berkeley University, or a bunch of them as a proof of concept. And this is from the stable materials we knew before, like that Wednesday were from ancient times, like the ancient Greeks discovered some shit, the Romans discovered some shit, the Middle Ages discovered, and then it's like, oh yeah, yeah, all that, [2:17:01] that was really cute. Like, boom. What is it? One step. Yeah. And that's amazing. Yeah. We should be celebrating that. We're gonna have great phones in 10 years. Dude, we'll be able to get addicted to like feeds that we haven't even thought of. So, I mean, you're making me feel a little more positive. Like overall, there's gonna be so many beneficial aspects to AI. Oh yeah. And it's just what it is is just an unbelievably transformative event that we're living through. Yeah. Power and power can be good and it can be bad. And that's, yeah, an immense power can be immensely good or immensely bad. And we're just in this, who knows? We just need to structurally set ourselves up so that we can reap the benefits and mine the downside risk. Like that's what it's all about, but the regulatory story has to unfold that way. Well I'm really glad that you guys have the ethics to get out ahead of this and to talk about it with so many people and to really blare this message out. Because I don't think there's a lot of people that like I had Mark Andreessen on who's brilliant, but he's like, [2:18:06] all in, it's gonna be great. And maybe he's right. Maybe he's right. Yeah, but I mean, you have to hear all the different perspectives. And I mean, like massive, massive props honestly go out to the team at the State Department that we work with. One of the things also is over the course of the investigation, the way it was structured was it wasn't like a contract and they farmed it out and we went out. It was the two teams actually like worked together. The two teams together, the State Department and us, we went to London, UK, we talked and sat down with DeepMind. We went to San Francisco, we sat down with Sam Altman and his policy team, we sat down with Anthropic, all of us together. One of the major reasons why we were able to publish so much of the whistleblower stuff is that those very individuals were in the rooms with us when we found out this shit and they were like, oh fuck, like the world needs to know about this. [2:19:01] And so they were pushing internally for a lot of this stuff to come out that otherwise would not. And I also got to say, like, I just want to memorialize this too. That investigation, when we went around the world, we were working with some of the most elite people in the government that I didn't, I would not have guessed existed. That was honestly... Speak more. Well, I can be... It's hard to be specific you see UFOs too much to take it to the hanger no no there's no there's no there's no there's no hanger yeah I believe we cut you say I'll go you'll cut that there's no hanger don't worry sweetie don't worry we didn't we didn't go that far down the rabbit hole. We went pretty far down the rabbit hole. There are individuals who are just absolutely elite. The level of capability, the amount that our teams gelled together at certain points, the [2:20:02] stakes, the stuff we did, the stuff they made happen for us in terms of brain, they brought together like a hundred folks from across the government to discuss like AI on the path to AGI and go through the recommendations that we had. This was pretty cool actually. It was like the first, basically the first, first time the US government came together and seriously looked at the US government came together and Seriously looked at the prospect of AGI and the risks there and we had it was wild I mean again, it's like that was in November It's us two friggin' Yahoo's like what the hell do we know and our amazing team? and and and it was yeah referred to by there was a senior White House rep there It was like yeah, this is a watershed moment in US history Wow, well that's encouraging. Because again, people do like to look at the government as the DMV, or the worst aspects of bureaucracy. There's missing room like four things like congressional hearings on these whistleblower events. Certainly congressional hearings that we talked about on the idea of liability and licensing and what regulatory agencies we need, just to kind of like start to get to the meat on [2:21:04] the bone on this issue. But yeah, opening this up I think is really important. Well shout out to the part of the government that's good. Shout out to the government that gets it, that's competent and awesome. And shout out to you guys, because this is a, it's heady stuff. It's very difficult to grasp. It's even in having this conversation with you. I still don't know how to feel about it. I think at least slightly optimistic that the potential benefits are going to be huge, but what a weird passage we're about to enter into. It's the unknown. Yeah, truly. Thank you, gentlemen. Really appreciate your time. Appreciate what you're doing. Thank you. It's amazing. People want to know more. Where should they go? Which of they follow? I guess Gladstone.ai slash action plan is one that has our action plan. Gladstone.ai. All our stuff is there. I should mention too. I have this little podcast called Last Week in AI. We cover sort of the last week's events and it's all about the sort of lenses. [2:22:01] You have to do that every hour. Yeah. Last hour at AI. Well, the week is not enough time. We could be at war. Our list of stories keeps getting longer than anything. Anything can happen. Time travel. You'll hear it there first. Yeah. All right. Well, thank you guys. Thank you very much. Appreciate it. Thanks, Johnny. Thanks for watching!