Ex-Anthropic Whistleblower Suspects Chinese Spies Inside America’s AI Labs

0:00Whoever controls AI becomes a god. I will destroy humans. This tech is powerful tech. And I'm just very surprised that Trump can sit there and say, I'm going to let Dario Amadei, I'm going to let Sam Oman have the keys to this powerful tech. I just think they haven't fully internalized how crazy it will get if we don't slow down. Do you believe China is not going to work on super intelligence and they're going to listen to us? If you appreciate that the tech is scary and it's going to feel scary for the people building it, I think China will also realize that. How certain are you to have spies already there?

0:28They're not vetting for something that could be a super weapon technology. It's just crazy. It's like it's happening right under the noses of the government. You met people that you think they could be? You know, again, I really, I don't want to call their names. I haven't thought about that detail. Forget about giving names. I'll find out. They hated me. They hated me because I was like, for good or for bad, it feels like the end of the world. I was part of something that was going to go badly. And I became convinced that really, there was no good plan. If we could slow down AI and speed up everything else, like deregulate drugs, deregulate biology research, You regulate like human self-experimentation.

0:59I think a lot of stuff, it's just AI in particular that's the thing that's scary. So in four years, if Dario wanted to destroy OpenAI, he could. I think 2027 is going to be a crazy year.

1:16So I'm going to ask you an indirect way of helping our guys, our production team that works so hard, get a raise. Do you want to help them get a raise? This is how you can help us out. 49 % of you watch our podcast that our editors work so hard producers, cutters, all of them our production team 32 of them that work here they work very very hard to deliver a good quality to you if you enjoy our show and you want to help this thing get bigger so we can take our talent that we have in-house that support this production all the time and maybe even give them a raise then please subscribe to the channel that allows us to get more viewership and indirectly it makes bigger opportunities for our current team that we have.

1:57Thank you so much. So on September 8th, a Jacob Coxon, who's here with us. Good to have you on the podcast. Nice to be here. You decide to put on a tweet, okay? You had started your account January of 2026. You hadn't put nothing up at all. Your first tweet says, I resigned from Anthropic today. I spent the last three years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below. We'll get into that later on. And this tweet gets 803 ,000 likes, 175 million views.

2:33There was a meeting yesterday at the White House prompted by you indirectly. You got politicians that got involved, world leaders that got involved. A meeting that happened with G that was here last week. One of the topics of discussion was how powerful AI is going to be. And the funny part that I think is important for the audience to realize this, and you can fact check me on this, that you let all your stocks go, that you weren't even able to vest your stocks because of the timing that you had, which means any money on the stocks that you also had was left behind for you, if that's correct, right?

3:06The options didn't move on with you. Not at Anthropic. Not at Anthropic. At OpenAI, you were there. I have OpenAI. Okay. So you did all of that, and the world kind of stopped for a few days. Till today, they're still talking about it. what did you see that got you to say, I have to do this? Yeah, it really was just thinking about next year. Next year and the year after. I think 2027 is going to be a crazy year. And I think, think about the news headlines. Like, so last year, AI, a bit in the news. Three years ago, it was very rarely in the news. Maybe for ChatGBT, but otherwise, no. This year, it's bam, bam, bam.

3:43AI, every month, there's a big AI story. Next year, it's going to be like, every day, there's going to be a big AI story. In particular, just imagine, I don't know, I wasn't alive during World War II, but during World War II, I imagined that just the concept of war was coloring everything. It was in the background for every day-to-day thing. Like every news story, war would be in the background. I think it will be like that for AI next year. In terms of what I specifically saw, it is the models, the AI models next year are going to be very capable in a way where they're going to be able to automate a lot of work, a lot of everyone's jobs.

4:18but kind of most importantly, a lot of the AI researchers' jobs. So things are just going to speed up so much that it's going to become a very, very serious thing next year, probably. So 2027 for you is what's going to be, is a word to use chaotic? Chaotic, definitely, yes. So unpack it for me, like the average person listening to this, what does that look like? Yeah, so there's many different things that could happen. Things that, kind of the most important thing is right now, labs are treating this like the companies are treating this like it's a normal technology. This is business. Maybe you saw OpenAI, they have these dev days.

4:54They share their products with users. They're still tweeting like, oh, excited to release this new product. The thing is this is a super weapon. So soon people are going to have to realize that this is not like a casual business thing. This is like a super weapon technology that really should be nationalized. And I think next year is kind of probably when that realization really sets in that this is very, very serious stuff. You're saying it should be nationalized. So if it's seriously that concerning, when you say nationalized, you trust the government getting involved to control this then free market?

5:31Yeah, I'm not saying I need nationalization. I'm saying that it is so serious that I'm surprised the government isn't nationalizing. Like the government is looking at these companies. They're going to have power. This tech is powerful tech. And I'm just very surprised that Trump can sit there and say, I'm going to let Dario Amadei, I'm going to let Sam Altman have the keys to this powerful tech. Because normally it's the government that has the keys to this stuff. You wouldn't let Marcus Zuckerberg have nukes. That's just not a thing that normally happens. Make the comparable of the two, nukes and this.

6:02So nukes cause world chaos, obviously, like plausibly. Same thing for AI. It's going to be, do you really want a private company to make decisions about unemployment rate big stuff, potentially if they make a mistake, someone in a some regime somewhere could produce a bioweapon. That's a mess up. Do you want private companies to be in control of all these decisions? To give an example, if you're at Anthropic and you're training in AI and you're going to share it with the world, one of the things they worry about is what if China can steal from this? And that's a private company worrying about geopolitics which is a weird thing for a private company to make decisions that affect the whole geopolitical situation.

6:49Some companies become so big that you become a small country. You have to realize Zuck is a world leader of 2.4 billion people called Facebook and he's got regulations he deals with with the government and whatever countries he's in he has to abide by those rules and guidelines that's why he's always in lawsuits because he broke some of the guidelines. right? So same with Twitter, same with Google, same with YouTube, same with all these guys. So they almost have to be because they're dealing with those types of people, right? So, but what I want to know from you is as, as we're going through this, because some of this to me is, you know, I wonder if it's fear porn where they're just using to scare a regular person watching this.

7:32I wrote it down. I said, I think this is how I see it. I just kind of want to share with you on how I see the fear porn side of this. So one part of the fear porn is, hey, this is so scary. AI is going to take over the world. We need the government to control. Okay, that's one. Okay, people are using this because they want control. They want the government to control. Number two, hey, this is so big and powerful and we need government control because I'm anthropic and open AI. I'm eliminating my other companies to come and compete with me. So the barrier to entry becomes more difficult and I get to control.

8:03So the motive there becomes eliminating future competition. Pretty good move if you're thinking that way, right? Definitely. The third level of fear porn is, well, let me tell you, this thing is so powerful because we're about to go public. As big as we thought this was, this is so much more powerful than us. You may want to invest some of your money in there because when we go public, you may 10x, 20x, 30x your money. Yeah. So the public IPO goes just massive, right? So those are three elements of the fear porn that I see. Then the other one is maybe it is as scary as you say it is. The other one is if we don't build it, China builds it.

8:36And last but not least, if we all stop, how can somebody hold an independent person or China to build this? So of those, if you can one by one by one address which one you think it is. Yeah. So to go to the original fear porn stuff, especially about the regulation control or Anthropic hyping up their product, I think part of the reason my tweet went so viral is many people at Anthropic said the same thing, but they didn't just quit. you know they stayed there so it's easier to be like okay he's staying at the company he's saying this stuff but he's just hyping the product i left in a personal capacity because i am scared so from my perspective i have nothing to gain by by hyping up anthropic that's not a i'm not trying to hype them up or get them like special criteria i think what they're doing is dangerous i think it's reckless i think they should be curtailed and i left to say that so that's to address the regulatory capture stuff.

9:34Also, later, we could go into why I don't think it's actually like, it doesn't actually make sense to stoke up this level of fear. I think it's going to hurt their IPO. I think it's going to hurt their ability to stay in control. We can talk about that later. What was the other point? China, yeah. Yeah, the China point is difficult. But I think, again, if you appreciate that the tech is scary, and it's going to feel scary for the people building it, I think China will also realize that. The people building the technology in China will see how destabilizing it is. And I just don't think that that regime wants to create a destabilizing technology that could threaten their own power.

10:09I think actually our incentives are aligned. Like we're more similar to China in this respect. So it should be quite easy to cooperate. Well, if you look at the advancement, the separation between us and China, one could say the motivation, the motivation, hang on one second, I'm going to show you one data. Yeah. China. So motivation for China could be U.S. has such a massive lead in AI, we need them to slow down so we can catch up. Yeah. There's no way in the world that China is going to sit there and say, oh, yeah, yeah, President Trump for sure. We're going to shut it down and we're not going to work on it.

10:46Yeah. Okay, great. Can we come and look at your place to make sure you are actually keeping your promises? No, no, just take our word for it. We are not working on superintelligence. Yep. Do you believe China is not going to work on superintelligence and they're going to listen to us? It depends on the nature of a deal. Right now, if we just stopped, they would try and catch up. But I think one really important thing people don't realize is China already has spies at OpenAI, at Anthropic. China has the ability to hack at any time. If we pull ahead, that doesn't give us a big lead. They can just step in and steal our progress at any time, catch up the lead.

11:21How certain are you to have spies already there? 90%, yeah. It's almost certain that they have spies. Do people talk about that at OpenAI? Yeah, they do. People have spoken about this. Is it kind of like jokingly, hey, we probably have some Chinese spies here? Yeah, it's like a joking thing because you don't want to take it seriously, but then it is a serious thing. There must be spies. Have you met people that you think they could be? No. I think obviously I don't want to stereotype people based on are they Chinese national? For sure, yeah. So I wouldn't say any specific person, but there must be spies.

11:49There's no background check, serious background check for this sort of stuff yet. This is another reason why nationalization is so important. We can start properly vetting and background checking. So it's like, I don't think SpaceX can hire Chinese nationals, but these AI companies can hire foreign nationals like me, which is pretty surprising. Why is that? What's the difference? Because SpaceX has some regulatory government? Yeah, there's a regulatory government. Yeah, so it can't. So they cannot recruit Chinese nationals, but OpenAI and Anthropocan. Yes. So do you think one step could be to not allow Chinese nationals to work for Anthropocan OpenAI?

12:21It's a bit sad because they're my friends, but I think that will happen. I think that will probably happen. Are you saying will or do you think that's the right move? I think both. Okay. So it's uncomfortable for you to say it and I plot that. I respect that. But you at least – that shows me there is integrity in your belief system that your friends are going to see this. They're going to say, hey, Jacob, what are you saying? We're friends. But you think, okay. So you think if we continue to China almost – they're like, yeah, continue because we have access to everything you're doing anyways. Yeah.

12:52So if we continue, they're going to have access to it. If we don't continue, China's going to create it themselves. So either way, China's eventually going to have access to it. I think so, unless, again, there's some sort of deal. One thing is a lot of their progress right now, not all of it, but a substantial part of it is distillation, which is sort of stealing the data. So they use Claude, they log in, they ask it questions, and they use the answers to train Chinese models. So right now, a lot of their speed is kind of copying the U.S. and stealing from the U.S. So also if we slow down, it will probably slow down China quite a bit too because currently a lot of their progress is from essentially copying or stealing US work.

13:33Not all of it. They will still catch up and make progress, but there is a component. So they will also slow down if the US slows down. Who else is another player out there that people worry about? It seems like it's mainly them to maybe UK a little bit, right? Not the UK at all, sadly. That's why I'm in the US. It's very sad, but the UK just hasn't. Who would you put third? i'm actually not sure who's got the the data center build that right now u.s and china are so far ahead that no one is no one's going to catch up right now okay so go a little bit deeper into the the chinese spy part uh you said it is a discussion that people have in those organizations so what would that look like if the average person wants to know all right what would that look like on how they pass their messages or information over to them is it can you do me a Can you type in who are the highest ranking executives at OpenAI or Anthropic that are Chinese nationals?

14:29Highest ranking executives at OpenAI or Anthropic that are Chinese nationals? Have you looked at the inventory to see how many of them are Chinese nationals? I don't think there are many that are that high up that are Chinese nationals. Again, I really, I'm like, this does make me sad. I'm not trying to like start a witch hunt. I'm just saying objectively there are probably spies, and I think that, yeah. Okay, so let's go over there. So Chief Research Officer at OpenAI, Mark Chen. Yeah, he's not a Chinese national. He's not a Chinese national. You know who he is. It says Chinese-born. He's Asian-American, yeah.

15:03But he was born in China. Yeah, I don't think Mark Chen is a Chinese spy. Okay, all right. And you met him? Yeah, I've met him a few times. Okay, so Fiji, Simo. I don't think he's a Chinese guy either, no. Brad, Lightcap, COO, and Aponga. So Dario, we know that. No, I think we are clear on these guys, yep. You are clear on all these names. Yeah, these guys. Who are the highest rank? Are there any high-ranking people that could be Chinese nationals? Or you don't know any names? You know, again, I really don't want to call out names. I haven't thought about that detail. I'm not asking you to give names.

15:35I want to ask you, forget about giving names. I'll find out. What I'm trying to find out from you is because the audience is going to go do their own research, right? So I think, yeah, almost all the Chinese nationals are like employees. So not like executives, not high-level managers. They will be like employees doing coding, doing engineering, doing research. Yeah. Yeah, if we pull up, I went to an insurance company once in Dallas, and I went on their seventh floor. The seventh floor was all the engineers that they had. Yeah. If I told you 98 % was Indian, I never met anybody that wasn't Indian.

16:07Everybody was Indian. And they had a lot of their guys that came from, what is that institute from India? IIT Institute, which IIT Institute is like a, some people even say it's better than MIT Institute. So you saw that. And one could say, okay, they're in here in the business. But when you're saying Chinese nationals, India doesn't necessarily have the same motive that China has. One in America wouldn't say China's an enemy. You know, India's an enemy as much as they would say China's the enemy, right? So if you're saying they already have Chinese nationals in there, that's probably a pretty big concern there.

16:40And I think what it is is it's representative. Remember I said they're not taking this seriously? It's like it's a normal business? Yeah. Like, yeah, there's an absolute minimum background checks for employees seem important. I think they will come. But if they're not vetting for something that could be a super weapon technology, it's just crazy. It's like cowboy stuff. I'm surprised it's happening right under the noses of the government. It's ridiculous. By the way, that's a very valid point. Who else do you think they need to be vetting and not allowing to work at companies like that? So I don't know the exact details, but, you know, plausibly even, I'm a foreign national, so I'm from the UK.

17:18I think it's important. A lot of the researchers come from the UK. I do think careful background checks are necessary. What are you looking for? In particular, I guess, you know, no compromise. Whatever the current clearance background checks are, you should probably need that to work on core AI systems, and I think that will happen. But, I mean, even if you're doing that, what am I looking for? What would I find anyways? Nowadays, people are professionals that are hiding whatever they did in the past. Yeah, I know. It's not even for me to be able to find everything about that individual. Yeah.

17:49It's – So, again, I'm not an expert on, like, the details. What I think is, like, they need to take it more seriously at a bare minimum. Fair enough. So that part is made clear. We got that. What do you say to the people that say, you know, David Sacks, some of these other folks came out. Very reputable. People respect these folks. And you saw Marc Andreessen came up and he wrote his humanity manifesto with AI and saying, hey, this is a let free market do its part. This is a beautiful thing. He believes, you know, we can handle 50 billion people living in the world, you know, and we have bandwidth and this is going to make people more financially free.

18:27It's not that scary. Some of those guys said this is purely used by Anthropic and OpenAI to eliminate future competitors and own bigger market shares. And they're like, yeah, Bernie, come help us out. Come prevent other people from competing with us. What do you say to that? Yeah, there's a few things. One is I think I agree with Marc Andreessen and David Sachs about the positive benefits. What I think, where I think we differ is their case. I think we get their world if we slow down. I think even if we slow down AI development, the changes from next year's technology will be so rapid and significant that there will be huge changes in the way that they're thinking.

19:09I just think they haven't fully internalized how crazy it will get if we don't slow down. I think their case for their optimistic fast case is my slow case. And my fast case looks like we have no control over it. We're just watching robots make themselves smarter every day massively. It's completely out of our control. It's not like 50 billion humans. It's humans and then this massive AI economy that's happening. We're just sitting around watching them do inscrutable things we don't understand. And I think Marc Andreessen and David Sachs will agree with me when things get crazy because I think we want the same thing.

19:44That's my guess. How many people, when you worked at OpenAI Anthropic, guys that are actually doing the work, not executives at the top? I'm talking the engineers, the mathematicians, the guys like you that are doing the work, the researchers that are doing the work. How many of them are in their 20s? Probably half or a bit less than half. Okay. So that's a big number. Big number. So just fresh out of college, right? In their 20s. Maybe more like 30%, 30, 40%. 30%. And is it fair to say that that audience just got out of college? Not all of them, some of them. And maybe they're going to lean a little bit more on the liberal side.

20:26And maybe they're going to be a little bit more afraid of climate change, end of the world, professors, brainwashing them. And, oh, my God, you guys be careful with these capitalists. They're going to be such technocrats and technocracy. And, you know, oh, my, these trillionaires, Elon Musk is going to take over and do this. Don't you think they may have a little bit more fear, anxiety of the end of the world that's coming because they just got out of college? Yeah, maybe. I can say a few things here. Please. One is, I think it's important that Elon Musk himself is scared about this. He doesn't get scared of many things, right?

20:59He's a titan. The guy's an absolute, he just goes for stuff. He's been scared of AI for a long time. He only made XAI because he saw the other companies doing it and was like, well, I guess if we're doing this, I have to join in because he's Elon Musk. He's not going to let these guys, he's not going to sit on the sidelines. So big guys are scared too. And then in terms of junior employees, I mean, I can speak for myself. I like I've never, you know, global warming has never been a big deal for me. It's like I wanted to learn math. I wanted to work on technology. I wanted more nuclear power. I want like build out of energy infrastructure.

21:35I'm like pro, you know, I'm pro-capitalism, pro-libertarian almost. So you're more libertarian. Definitely. I think that's very important for you to qualify there because a lot of people were putting you as a climate change guy. I know, and it's because everyone agreed with me was a Democrat. I made a post. I didn't ask these Democrats, like Democrat politicians, to agree with my post. I didn't tell them to tweet that. So I think I've been like pegged as this NGO kind of dude. Really, the one thing I'm scared about is AI. I'm like, if we could slow down AI and speed up everything else, like deregulate drugs, deregulate biology research, deregulate human self-experimentation.

22:06I think a lot of stuff, I'm pro-libertarian. It's just AI in particular that's the thing that's scary. huh so so would you say most of the guys and gals your age working at open ai and anthropic feel the same way you do there's a mix i think i'm probably more on the like libertarian side i think some will be scared of climate change some will have been i also thought for example the pandemic response ultimately was quite overblown like near the end i think that like it was pretty bad but i think maybe some would disagree so there are definitely some employees like that, but it's not all of them.

22:40It's not as like, I think people imagine San Francisco as being full of liberals. But if you're going to accuse like me and the other kind of like kids building the technology of anything, it would be like a sort of maybe narcissism or techno narcissism, feeling like you're the center of the world, but not like a generic liberal fear. That's not like the main thing that's happening. Yeah. I'm glad you, you, you, you addressed that and you isolated that case because I noticed a lot of information came up that you were with Newspeak where you were part of this organization that Hillary Clinton was a part of and a man named Matt something and it's the NGOs and you're involved in that.

23:20This is so funny because at that house, it was a house I lived at. I wanted to live in London and I met a friend who was like, come stay at this house. And it was a house full of these people. They hated me. They hated me because I was like, they're all communists. I'm like, there were communists in that house. And I did not get on with them. It's so funny now because I'm on this thing and everyone's like, Jacob stated this. I wanted to do math. I wanted to live in an interesting place in London. That's why I stayed at Newspeak. You've never met. You've never had meetings with these guys. You've never met with the socialist leaders.

23:52No, no. Clinton, Sanders. These guys didn't communicate with you. Nothing. Never, no. So even before you tweeted your tweet out, I know Brett asked you this question as well. But did you speak with anybody? Was it orchestrated that this tweet's about to come out? Because you started your account in January of 2026. So prior to that, for a young guy to not have a Twitter account, it's kind of weird. Why didn't you start it early? Why did you start it in January and then eight months later you posted this tweet? One is I'm actually like kind of ironically, I've been quite scared of posting online, sharing my thoughts.

24:28I think because I had this assumption that you know that my people might pile on or it's like you know I don't like engaging with the immediate criticism of the thoughts I prefer to like discuss them in private with the people that can make changes so this was kind of the first time I've ever expressed myself publicly and that's one of the reasons why I was so crazy is like I always had this fear that if I if I posted everyone would pile on and that is exactly what happened which is quite funny although it wasn't as bad as I thought. Do you regret the tweet or you're glad you tweeted? I'm glad I tweeted.

24:57You're glad you tweeted it? And I'm happy to talk more about the coordination because there was some planning. I didn't just quit and then type out this tweet and rage quit. There were a few steps to it. Can you walk us through it? Yeah, of course. So when I decided to quit, first I was making the AI smarter, right? And then I switched to safety for a week. I was like, well, I'm really scared of how things are going. I'll move to the safety side, try and make it safe. And then while I was at safety, I spoke to more of the anthropic leadership. I spoke to more of the safety researchers. and I became convinced that really there was no good plan.

25:28I was part of something that was going to go badly. So I was like, I'm going to quit. And then I was thinking about quitting on the same day. But I don't know, do you remember Leopold Ashenbrunner? Absolutely, 436 % the first six months. And, you know, yes, I do. Yeah, he's a dude and he left OpenAI and then he wrote this manifesto, Situational Awareness. Yes. So he was strategic when he left. He was like, I'm not just going to leave OpenAI. I'm going to like use leaving to make, because he was fired for like leaking confidential information or whatever. He used it to like spread awareness. Like a lot of people in government read his manifesto.

26:04A lot of people, maybe you read some of situational awareness, obviously. Yeah. He grew to what,$45 billion. Oh yeah, he read a massive. I think Griffin came and bought a chunk of it at a discounted price. And you know, yeah, I followed the story closely. So here we go. Yeah, exactly. This manifesto. And he wrote that when he left. And I thought, well, you've got to be strategic when you leave. I'm not just going to, I do want to express myself. I'm going to like be strategic, make my voice heard, have an impact. I did not expect it to be this. I thought it would be a lot smaller than Leopold's.

26:34I was just going to do my best. But I got a connect at Wall Street Journal. I was like, I'm going to make a, you know, Wall Street Journal will publish this. I think we can probably find the article. Yeah, this. So for starters, I also, I delayed my quitting by a week to have time to prepare for this. So it wasn't just the same day. I was like, I'm going to prepare. I'm going to write my thoughts properly. I'm going to quit. Spoke to the Wall Street Journal. I wrote this Twitter thread. But to be clear, I think people imagine this crazy coordinated psyop. And I had some help. But I wasn't being, like the people were telling me stuff like, I say people, my friends were telling me stuff like, include a picture in your tweet.

27:14Because then it will go viral. They said, you know, tweets go viral if there's a picture. But I was like, I don't want to put a picture. It's a serious resignation tweet. I'm not going to, like, put a picture. I got a few friends to retweet it. Yeah, it goes to the actual details. You have some very high, you know, it was some very influential people tweeted it fairly quickly. So they're pretty influential. These guys are – so it was people like Nathan Calvin and Snare Revenant. They're this organization called Encode, Encode AI. Within the space, they're influential. Oh, yeah. Within the space, they're definitely influential.

27:48They're definitely influential. The average person may not know who they are, but within your space, they know who they are. They're very influential, yeah. And they're also friends with many of us. Like I'm friends with Sneha. We've been hanging out before this resignation. She's friends with a lot of lab employees. It's also, yeah, it's not crazy that when I wanted to leave, so the journalist was interested because this was the first ever anthropic safety resignation to go kind of big in public. There have been a few, but they were small. This was a big one. and obviously the people that are most interested in like helping spread the message are people who are you know, who want to slow down AI So who else were the people that was anybody that played a role in you writing that did any politician know you were going to tweet that?

28:29No, no, no So nothing political Nothing political Are all these people that are friends that are friends that are working at Anthropic and OpenAI or outside of Anthropic and OpenAI? They were all outside of Anthropic and OpenAI None of them are working at the company? No So no one at the company knew you were about to resign. They knew I was going to leave. They did not know I was going to make such a bold Twitter thread. And how was that conversation that they knew you were going to leave? Were you like, guys, I can't do this anymore. I'm going to be out in a few weeks. Was it kind of like sharing your frustration?

28:54Exactly. I was talking, but the thing you have to remember about Anthropoc is such a weird place. Like these guys are scared of what they're building in a way that I think repulses people when they see it. They're like, this is weird. Why are these guys building it and they're so scared? But it's real. They're really scared. So honestly, what happened is I said to my, you know, I said to the head of pre-training, I said, I'm going to quit. I think we're going too fast. I think it's reckless. I think we're not going to get international coordination. You're saying this to who? To the head of, this is the head of like the whole of the pre-training organization, which is quite a big organization.

29:32Is that a heavyweight person within the company? Heavyweight, yeah. So somebody that works directly with Dario? Exactly. Someone that would speak to Dario a lot. So you're telling them this? Yeah, I'm telling him this. I'm telling him it's reckless. Is this a public name? Yeah, Nick Joseph he's called. So you're telling him this and what's his reaction to you? Yeah, so his reaction is composed because he's dealing with these worries the whole time. He tells me, I think this is a bad decision. I think you will have more impact if you stay and work from the inside. And he said, you're not going to be able to, it's unlikely you'll be able to change people's opinions on the outside.

30:08I mean, what he said something was like, do you want to go from being a, you know, like a good AI technical researcher to being just a run-of-the-mill commentator? He said it seems like a bad strategic decision. This sounds like a good guy. It's a good guy, yeah. This is not a bad guy. This is a good guy. So he's trying to give you career advice. Career advice. Okay. And what got you to say, I disagree, I'm going to leave, and I think I can make the impact that way? Yeah, and I said, well, I didn't say that. What I said is, you're probably right. It sounded like he was right. I just didn't want to be part of it anymore.

30:39I thought when I left, it was an expression of resignation, of kind of disempowerment. I thought, I don't want to be part of this race. I'm going to leave, do something else. It might actually be worse. It wasn't really a rational decision. It was pretty emotional. Got it. And these other friends of yours that you're talking to about the tweet, are these players that are running things and they're heavyweights, are some of them within the space of AI, but maybe they're not known names. Yeah, so there are a few players and then just a few friends. And it's so funny. Again, people imagine this as being a, I think because of how crazy it went, they think it must have been this very careful thing.

31:17Really, I had a few, I had maybe three or four core names to do the retweeting initially. And then after I sent it, right, I sent it and then I sent the link to a few friends and said, post this in your group chats, share this in your group chats, like spread it around. But honestly, by the time, like I sent it to one guy, And he was like, okay, I'll share it. Let me write a careful quote tweet about it. Let me like carefully write a quote tweet. And by the time he'd finished writing the quote tweet, it had already gone to like 20K likes. Like it already exploded. So the whole thing just went so, so crazy.

31:51And it like, this can't be planned, right? Like imagine if there's no PR firm that can make, that can like hack the algorithm like this. Yeah, and when you did this, who was it from Anthropic Researcher? Somebody supported, and Wang, right? I think Evan Hubinger was the main. Evan Hubinger was one of them. Yeah. That he, what did he say? That he agreed with you? He said he thought everything I said was correct. Because what I said is I didn't actually say my own opinion mostly. Mostly I said other people at the company believe it and continue to work there. If we scroll, it's like it's a tweet three maybe.

32:25I said the people building this technology earnestly believe it could kill us all. And he backed me up. He said, this is correct. I do believe that this technology could kill us all. They ran an analysis of all the replies of who agreed, who disagreed. 43 % agreed, 40 % disagreed, 18 % were mixed. And so it wasn't like people weren't agreeing with you. They were agreeing with you. I even think Elon retweeted and says, I agree with this. I said this a long time ago. Something like that, I think Elon said. Yeah, exactly. And he has. He's always said that. Did he reach out to you? Have you spoken to him?

33:02I mean, not like, not a length, no. He's clearly been thinking about this for a really long time. That's the main. So you guys did have a conversation together? Yeah, we had a chat. Okay. This is after your tweet? Yeah, after my tweet. Was it a phone call? Was it a conversation? Or was it a face-to-face? Just a DM. Okay, so he just DM'd you and, hey, you know, I agree with you, you know. I DM'd him. And then he responded back to you. Yeah, just a little response. And mostly I was saying, like, he, I just didn't even say, you know, because maybe we saw he had a little spat on Twitter. He, like, said it was maybe a psyop.

33:39I saw that, yeah. And I thought, I'm going to try and, like, because I think he's been thinking about this for, like, so long. Did he explain to you why he said this could be a psyop? It wasn't, no, not that much detail, no. What do you think he meant by this could be a psyop? With the original thing? I think it was quite suspicious that an account with no followers could blow up that much. The reality is he owns that, So he can find that exactly if they were black-driven or not. Exactly, and they weren't. So it was just like an organic explosion based on the fact that it was scary. Yeah, this is this.

34:06I kind of regret this tweet. It was pretty stupid. But then again, I had no sleep, and I was very annoyed. I thought Elon was kind of on the side of this. So it upset me quite a lot. Did he say he agrees, and then he said PSYOP? It's like it almost – there was a 50-50 with him, right? Yeah, so this, I think he was just – when I say PSYOP, he just thought the whole thing was quite suspicious. Like the fact that I had, you know, the communities I've been in before, the fact that it went so viral. I can see why he thought it was suspicious. You got 300 ,000 followers within 48 hours. I know. It was nothing.

34:35Absolutely insane. And so now you brought up Leopold earlier. Yeah. Leopold got the sequencing right. Got fired. Story goes out. Hey, here's my manifesto. You know, when he wrote down, we talked about it earlier. And in it, he says, it's not really such and such stocks that we have to worry about. We have to put our money into this and this. And it's the infrastructure that's going to make a lot of money. And for the most part, he did well. You know, people made a lot of money. And then stories came out about how much of it was on margin and how much risk he was taking. Four to one, two to one, three to one.

35:08But his outcome was, I'm going to quit. I'm going to write this paper. It's going to go viral. I'm going to start a firm. I'm going to raise money. I think GitHub guys got involved at the beginning. he got some big names that got involved. What's your financial next move? What's your financial benefit of doing this? There is no financial benefit. So Leopold, I think... So you have nothing new to launch in the next three to six months? I'm not going to try and milk this. I'm not doing anything. I'm going to... One way I will exploit the crazy, temporary fame of the situation is through having conversations with lots of researchers I respect that previously would have made less time for me, and now I can talk more about the situation.

35:48And there's work I intend to do. There's lots of work, but I'm not trying to make any more money from this. How much money did you leave on the table by doing this? By leaving Anthropic. By leaving Anthropic. You know, I'm not going to say the exact detail, but seven figures. Seven figures. Yeah. Okay. But let's go a little bit deeper. If you would have stayed five years and you would have done your part, you would have moved up potentially 50 million, 100 million? Five years. That's the problem. So I'm thinking about that, and it's like it's not going to be the same in five years. In five years, everything looks different.

36:16Okay. And I think the people that are talking about this, They have these sums of money and these plans for the future, and they're just not thinking about how crazy it's going to get. So to you, one, you're not a socialist leftist, that you belong to this NGO, that you're a pro-Bernie Sanders. You're one, a capitalist, free market, libertarian. You don't believe in that climate change is going to end the tech, though. You know, COVID handling wasn't right. You know, you thought it was a little bit too much. Okay, check. Let's set that aside. Number two, there is not a I'm about to announce a hedge fund private equity firm, and this is why our private equity focuses on AI regulation and da-da-da-da, you're not going to launch anything like that.

36:52So how are you going to make money? How are you making money right now? Luckily, I have this money from OpenAI, right? And I think I will get a job. I mean, I'm going to get a job doing some form of... Within the space. Within the space. And I think the job will most likely be around predicting where AI will go, trying to share that writing with other people. And also, like I said, it sounds crazy. It sounds cultish. things are going to change so much in the next few years unemployment is going to be a problem when we release, when we can automate a lot of white collar jobs you know, I have enough money for a few years and I really don't know what the world looks like, like if I'm wrong I can get a job in a few years, that'll be great but I don't think I'm wrong Got it, so for you, if you're hedging of you being right or wrong if you're right, you have money and you want to do something to make a difference the next 12, 24 months It sounds like that's your timeline.

37:43You think five years is too late. I don't think necessarily that the really bad thing happens, but I think that the critical decisions, the crucial decisions about whether we slow down or not, they will happen in the next couple of years. If we're going to slow down or not. Exactly. And you think the next two years, there's still time to slow down to prevent the madness? Yes. Why is it specifically two years? If you look at the pace of the improvement, the specific technology that is scary is when it can automate its own development. so when the job of AI research gets automated. And I think all the extrapolation suggests that that could quite likely come in the next few years.

38:19This is the whole RSI thing, recursive self-improvement. I think roughly around that time is the crucial time where the race will start, where China will lock in and it'll be China versus the US. Everyone will be taking it seriously. They'll be like fear. Open AI Anthropic won't be sharing the models with the public anymore. It will be internal, government only, full-on race. that's when the crucial decisions happen. Whether we go into this race or not, I think that gets decided this decade. You trust the government having this more than free market? I don't trust any specific person or entity.

38:51I just think right now the free market is not going to handle it well.

38:58That's why. Why do you think they're not going to handle it well? Yeah, that's a great question. Because the motivation is going to be what? Money. I think actually if I talk about my personal motivation, there are people who, I thought this was dangerous for a really long time. before I even joined. And when I joined OpenAI, I thought it was dangerous. At what age did you think this was dangerous? 15, 16. How old are you right now? 27? Yeah. So 15, 16? Yeah. So in 2015, you thought this was dangerous? Yeah, there was a book called Super Intelligence by Nick Bostrom that was quite influential.

39:26I read it in school.

39:32I think like, yeah, so this image of me having worked in AI and then suddenly got scared is wrong. I think I've always been scared. But then it's like, why do you join a lab in the first place? And there are other people who didn't join a lab. They were also scared. They maybe behaved more ethically. They didn't join. But then when they say this stuff, no one listens. I joined the lab and I got all sorts of things that they didn't, like money and status and some respect for working there. And then I quit. And that's a selfish thing to do, to say I'm going to work at this technology and get the status from doing this, even though I think it's dangerous.

40:07but then it also gives you the power to make change does that make sense? like this you get power and I think the same thing is like what Dario is doing so you know Anthropic have been scared and they said well no one's listening to us we're going to have to be a player enter and so Dario's worked like they've worked so hard in the last they've caught up to open AI it's very impressive Anthropic have worked like absolute dogs to like overtake and become the main player and they did that because they're like we need the power and that's the free market right The free market incentivizes you to get your own power, your own wealth, so you can make change.

40:41The problem is that's going to keep happening. And people are going to keep doing this. It's the story of OpenAI. OpenAI did that because they were scared of DeepMind. They were like, we're going to do it. Dario did it. He was scared of OpenAI. He was like, I want the power. Even just like me, it's a small story. But why did I work and help? It's for my own power. And if we let the free market determine in the next few years, we're going to race off a cliff because everyone is looking out for their own money, their own power. No one has the ability to sort of slow down a few years ahead. It's this like race problem where everyone just like, and I don't blame them, right?

41:16I don't blame them. If you're a 22-year-old right now and you've done AI, I don't blame you if you get a job at Anthropic and work there because what else are you going to do? How else are you going to try and help? So everyone is following their own incentives and that's why we need, in this case, we need bigger regulation. Because yeah, in any other technology, I'm happy with this. I think innovation from competition is great everywhere except if the technology is this level of compounding risky. That's when it breaks. So 2015, 2016, you read the book, Super Intelligence. And then at what age did you start working at OpenAI?

41:5123. And what did you do prior to that? Because I know you were a big math – maybe walk through who you were as a math phenom and how you came up. Sure. So I did math competitions in school. I was very into math. When I was a kid, I mostly just did math and read books. Then when I was in college, I guess. You won. You did pretty well on two of the biggest competitions. I did the International Math Olympiad twice. And you were second in one, third in one? No, no, no. Okay, so I think you misunderstood. I got a silver medal. Silver medal, yeah. A lot of people get silver medals. It doesn't make you the second in the world.

42:19I wasn't anywhere near the best. I was in sort of a, I think silver medal is top. We can look it up. I think it's top 20 % or something. In the math arena. Yeah, I think the kind of achievement here is getting on the national team because it's being selected for six people to do the team. Answers like that you give tells me your integrity. You corrected me to say you're not as smart as I thought you were. Yeah. I mean, you realize that's – But also the audience is watching. I understand that. But say 99 % don't check it. You just did an integrity check on yourself, and I respect that. I value that.

42:53I hope the audience does as well. But please continue. So you're a math guy. Yeah. And then I went to college and studied math. I kind of I think I wanted to do math research um which was you know because I yeah I always love math but unfortunately I think it was a mix of many things I kind of realized that pure math research was not for me it was partly related to AI I saw that like AI was going to be this big thing that was going to eat a lot of stuff still felt quite far out so I've been following it since 2020 was GPT-3. GPT-3 was this like quite big, this happened during the pandemic, which is also my third year of college.

43:32So I was sort of in the pandemic with a friend. We looked at the GPT-3 paper and we were like, yep, it's happening. This is a big deal. This is where things are going. But I didn't know how to break into AI. So I didn't actually like, I had no AI background. I studied math. So I worked with a friend at a power trading startup. We predicted energy prices, which is kind of, it was AI related. We predicted the cost of power. I did that for a bit. And then I took a, you know, there were some personal issues and I took a break for about a year. I was kind of out of it. And a little bit, I was living in Prague for a while and then Scotland.

44:10And then finally I got a call from a friend who was, Jacob, what are you doing? AI is taking over. You're like, you can do math. You can learn AI. Come and join OpenAI's residency program. They have a program which is a six-month training program for people that are math guys or physics guys to teach them the ropes of AI and make them full-fledged researchers. So I sort of said, yeah, that sounds great. I'll do that. And that's when I joined. So you went in and then you got in at six months and then you were there for how long? Three years? Three years, yeah. Three years. Which is, yeah. I was like pretty junior, but for context, the company went from 300 people when I joined to about 7 ,000, 8 ,000 now, I think.

44:48So it was like a pretty explosive phase. And was Sam Altman pretty accessible when it was at 300? Like, were you around him a lot? He's around. Yeah, you see him. You see him. Yeah. You've had conversations with him. Okay. How different did you get that same experience with Dario at Anthropik? Not in the same way, no. Okay. So what is your opinion of what Sam Altman is doing versus what Dario is doing? Yeah. How different are they? They're completely different. Yeah, I'm going to – so I think Sam is – so Dario is very honest. I think that's the first, you can see, everyone can see that. He's very open and honest and shares a lot of stuff with everyone.

45:26I think he's like, he's always honest. And he's obviously like an amazing researcher. And I'm always impressed by what he's like accomplished building Anthropic. Sam is like, has a reputation for dishonesty. So Sam has a reputation for being a lot more of a business guy of sort of like pleasing one people of always being quite vague. I think the main difference I noticed between Dario's communication and Sam's communication is Sam is very vague about the future. He says, things will be good in this way and maybe there'll be dangers, but we can muddle through, like a vague thing. Whereas Dario is like, 20 % chance this happens, 80 % chance this happens, this is what it looks like.

46:03And that's a lot more like rough to hear because it comes across as quite abrasive. It's like tough to hear the details. But Sam, it's like, it feels like he's making everyone promises and he could be saying vague stuff to everyone. Everyone feels kind of happy, but he's actually not really saying anything. That's kind of the main difference I'd say in terms of how they communicate with the company. What do you think is the end-all, be-all for each one of them? Yeah, again, this is armchair psychology. I'm not good friends with him. This is just from watching. I think Daria wants to save the world, as it were.

46:32He thinks him and his company are the ones to save everyone.

46:39I think Sam has a lot of motivations, and it's quite messy and chaotic, and that's reflected in how messy and chaotic the company is. But again, this is armchair speculation. I don't, yeah. So if one is saved, the other one is controlled? To be clear, when Dario is saving the world, I think saving the world and taking over the world overlap quite heavily. You could flirt with that. You could have that temptation. If you think you're the safe one and China is building it unsafe or you think the U.S. government is being silly and you have a very powerful technology, it's not off the table to consider taking actions against the U.S.

47:13government. And I think if anyone would do that, it's anthropic. In the inside organizations, OpenAI and Anthropic, how different are the cultures? Very different. Tell me more. So one thing is OpenAI may have changed. I left like four months ago, and I think they've started taking stuff more seriously. So Anthropic has seven co-founders, and none of them have left. They've all stayed, which is very impressive for an AI company because at OpenAI, everyone's left. They all go. Almost all the other companies, everyone leaves the whole time. Anthropic retains them. And that's because I think Anthropic has this, they are very clear about what their mission is.

47:53And they're very clear about how they're going to achieve it. And this makes them very dangerous. It's a very aligned institution. To put it another way, at OpenAI, it felt like people might prioritize their career, their individual success over the mission of the company, which is a very natural thing to do. Everyone does this. It's like, I want to do a project that's going to make me have an impact, that's going to be interesting for me, that's good for my career. at Anthropic it felt a lot more like well for starters the employees call themselves like ants the ants in the ant hill everyone's like a worker ant trying to make the Anthropic mission go well people are very willing to sacrifice themselves for the good of the mission they'll take on work that's less interesting, that's less kind of big and bold and exciting it's less flashy and they'll do it at a hit to their own kind of interests for the good of the company and that's because every decision comes back down to like is this going to get us to superintelligence faster?

48:49You've seen maybe Sora. You saw the video platform that OpenAI made. Anthropic would never make Sora because it's not on the core path to artificial superintelligence. OpenAI is distracted and messy, and one guy's like, I want to do Sora or I want to build like a little device, and they let them do it because it's not like the same focus. And Anthropic, that would never happen. They're just completely kind of - Who would shut it down? The seven co-founders? They'd be like, this is not - It'd all agree. It'll be like this video. It's not the right direction for heading straight to super intelligent.

49:16Which one is harder to work at? More schedule, more hours? I think it depends on what team you're on, probably. I think when I was there, I think it also depends on the stage of your career. I think when I was getting into, the work was also so different because now compared to three years ago, you're running these agents. You weren't doing that three years ago. So when I joined, I was writing code by hand, right? Like it was entirely hand programming. And by the time I left, no one really writes code by hand anymore. Your agents write the code. Your agents write the code. Yeah, I can talk more about that.

49:44It's kind of interesting about the details of how much AI runs the company. To give some flavor, when I quit, when I resigned, I was like, what's the procedure for resigning? Like, what do I sign? And my manager was like, I don't know, ask Claude. So I asked the AI. So I asked the AI how do I resign, and the AI told me the flow. No one knows how to resign. You have to ask the AI to do it because the company is run by, like the processes are run by AI, right? Everyone asks, and it's great, it works, but that's the extent to which AI is part of the whole company. It's like any transaction is mediated by asking the AI to do it.

50:17People's meetings are recorded by AI, summarized. You have AI's posting comments on things. You write an essay and AI will respond. It's like pretty crazy. Interesting. Is Dario and Sam Altman, on the way their people look at them, between fear, respected, liked, loved, how would you rank them? Like if you were to say people love Dario, they respect them. I would say adanthropic people love and respect Dario. Really? Yes, in a way that's not the case at OpenAI. Dario has very much like a cult of personality and a cult of vision, as it were, adanthropic. And then at Sam Altman, he's more fear? I don't know.

51:01I think he's obviously respected and obviously loved by a lot of people, but not in the same. Dario's thing is different. Yeah, and in the company, like no matter where you work at, like right now, you know, yesterday, Yankees played the Red Sox and, you know, they hate each other. They're enemies. Yeah. Right in sports. You know, we're talking about baseball earlier, you know, in politics, you know, Hillary Clinton, Donald Trump, you know, in basketball, Michael Jordan, LeBron James. Who was the enemy to Dario, and who did Sam Altman see as his enemy? Yeah, that's a great question. You have to look at the history of the companies.

51:43So when Opening Eye was founded, the big enemy was DeepMind, Google DeepMind and Demis Hassabis, and they were worried about Google kind of becoming dictators. They were worried about Google taking over. So that was the enemy, and then Opening Eye was founded. now it's like I'd say there's probably more like Dario's quite has a lot of animosity towards Sam Altman and the reverse is not necessarily true to the same extent meaning Sam Altman doesn't have animosity towards Dario I think less because I think like I said Dario's very honest and I think perceived Sam as having betrayed him to a certain extent and that's a kind of a pretty permanent lot on their relationship is my understanding how did he betray him I really don't know the details, but that's my understanding.

52:28Again, this is from a – That's the energy you see? That's the energy. So that's not public? I think that's pretty public, yeah. On the betrayal between – did they do business together on one? Yeah, I mean Dario was a big figure at opening eye before he quit. Oh, that's right. So the history is from them leaving. I remember that when that whole exchange took place. So for him, Dario's enemy was, I'm going to leave and build something faster than – has he ever expressed the reason for leaving? Was it because he did not want Sam to have the control of this kind of a technology? Basically, yes. That's what it was.

53:01I don't think it's good for the world for Sam to control this kind of technology. Maybe not Sam specifically, but definitely a large component, Sam, and then the rest of OpenAI. And who do you think Sam looks at as enemy today? Is it a politician? Is it a competitor? Handmade in Portugal, where craftsmanship is a way of life. For those who move through life differently, from sunlit decks to nights that never need an occasion, Every detail is considered. Every step, effortless. Crafted from fine suede and built with builder technology. Designed to move with you. Because true luxury isn't about being noticed.

53:40It's about how you feel when you arrive. The FLB Icon Loafer. Made for the life you've earned.

53:54I really don't know. Okay. And what role in something like this right now? I mean, you saw that they had a meeting at AI at the White House with everybody being there. The list was all the heavyweights were there, right? And part of that was from your tweet that prompted these types of conversations being had now. What role does Trump play here? Does Dario trust Trump? Does Dario look at Trump as an ally? Does Altman see Trump as an ally? How do they view that relationship? I think everything's changed so much now that I really don't know the... I have quite a very few things that I'm confident on.

54:27I think right now, honestly, everyone is just quite scared of how fast things are going, and they're looking for ways to come together. I do think this particular accord is a good thing. What they wrote makes a lot of sense. By the way, at the AI meeting at the White House, was Dario and Sam Altman present? I don't think Sam Altman was. I think he was at OpenAI Dev Day, but Greg Brockman was there. And he's won the seven. No, no, Greg Brockman is a – Representing OpenAI. Yeah, exactly. Got it. Was anybody there from Anthropic? Yeah, Dario was there. Dario was there. Yeah. Okay.

55:04When's the last time Dario and Altman have been seen together? Is that like the thing that happens or is that – I don't know. There's a lot of – have you seen all these videos flying around? No. Like these funny music videos flying around now? Oh, yeah, yeah, the dancing that they're doing. Yeah, yeah, yeah. I like those. That's funny that they did something together. Yeah, that's as close as it's going to get to him. So one of the things you said in the, where do you see Dario here? See if we can spot him. I think he's far left. Can you find Dario? Where is Waldo? Where is Dario? I don't see him in this picture.

55:32No, I don't think he's there. I don't see him in this picture here. I think he's on the far left. So, you know, go through it. So while you were there at Anthropic and you guys are having these types of conversations, how many other people while they're sitting there saying, Guys, we are building our replacement. Yeah. We are building our replacement. How many people felt they were building their replacement? Everyone. Everyone. Yeah, that's just a fact. Everyone knows it. Is that a fact or do the agents still need to be controlled by somebody like you? For now, they do. For now, they definitely need control.

56:01And it's not clear exactly when they lose. We don't need the human anymore. But everyone anticipates it. They're all like, we're going to be, at some point, we won't have a job. this Dario, like let's just say that's something that is happening and I'm an employee and I know this is going to be happening again. It's like a truth, right? Was there like a life insurance policy? Guys, if this happens, here's what we're planning on doing so you don't have to worry about it. Yeah, so Anthropic, I think in one of Dario's essays, I think it might be Machines of Loving Grace, this essay, he has a section where he talks about the fact that Anthropic will need to make commitments to its employees for if they get replaced.

56:41and I think people implicitly had the knowledge that if Anthropic employees get replaced due to progress, they could expect to be taken care of. I think people expect that they will stay and kind of be looked after. Yeah, but OpenAI, you don't remember having anything like that? They've not said anything publicly. I think there may be some references to it, but nothing. I think that's a pretty good idea. If you know your people are afraid of something like that, hey, if this happens, we would do this. If that happens, we would do this. Maybe, but it also sets about, like what about the people that don't work there?

57:08What about all the people that are going to get replaced to other companies? it kind of doesn't look great if you're looking at your own employees but like at these other companies people are you know losing their jobs and they don't have the same assurances as someone that works jacob who's the weirdest person that slip in your dm slid in your dm after that tweet there's been some crazy stuff i had a a lot of um I had a lot of quite aggressive gay men DMing me. Gay men? Yeah. Wow. Did they think you're gay? I don't know, man. They wanted to find out. Yeah. And for the record, you're not gay. For the record, I'm not gay.

57:41Yeah. Gay men. A lot of them in my DMs. Yeah. But like powerful gay men? No, not powerful gay men. Okay. You're just gay men, period. Just dudes in my DMs. Yeah. Listen, I mean, if there's powerful gay men, you got to entertain them, right? You got to reply. Yeah. Sorry. You got to reply and say, at least, this is a billionaire gay man. Hey, sir, respectfully, I'm straight. but thank you for your interest. I mean, you could, yeah, you don't have to say. Any heavyweight politicians, world leaders, you know, billionaires, any names like that? I think, so there were, like, there were a few names.

58:13The whole thing is, like, I mean, lost in the inbox. I don't want to say, like. You must have tens of thousands of deals. It's kind of nuts. I just have to decide what to respond to. But I don't want to say specifics. I think a lot of politicians have reached out, and I haven't responded to any of them yet. But a lot of people for things like talking to staffers, I think Bernie Sanders, as people reached out, I haven't responded to that yet. I had an invitation to go to, I think, a rally with Steve Bannon and Bernie Sanders that I chose not to go to. With Steve Bannon and Bernie Sanders? Yeah.

58:38Because they're aligned on this, right? They're aligned on the same page. Yeah, exactly. That is a weird rally to go to. I know, weird. And I say rally was like amazing. Are you thinking about it? I was thinking about it, definitely. But the thing is, it just wouldn't look good on head. I don't want to associate. I don't want to be political, right? I'm also from the UK. I feel like I don't want to just jump into US politics. That's just like an odd thing to do. Yeah, that's... And by the way, is it true that the story at OpenAI, like, you know, if you're going through conflict, they bring dolls and stuff like that.

59:05What is that? This is Anthropic. I fully believe it. Oh, Anthropic. Is this true like a stuffed animal advisory council? I mean, I don't think this is happening in the office per se, but I believe it, yeah. Do you ever see it or hear it? I've never seen it, but I've seen things similar at people's houses, right? They have like a dock that they use to talk about their problems. I think it's cool. I think it's like, look, dude, anthropic is going to be a two trillion ipo these guys are going to eat capitalism alive i'm like they can do whatever the hell they want if they're who gives a shit yeah because these guys are taking over like i mean you've got to give them some respects like if they're yeah and the upside between the two just as an investor forget about i know what you're thinking about as an investor between the two is open ai even close to the way they run things at anthropic so i haven't been there for four months maybe it's changed but i would like like why is opening i had of google I think it's largely a culture thing.

59:53Google has a lot of resources, a lot of people, a lot of great talent, but OpenAI has this focus on building LLMs. And Anthropik is so much more focused. Again, I hear that OpenAI is becoming more focused, but yeah, as right now as an investor, I guess I would long Anthropik. Definitely. What would you need to see yourself for you to second guess your own fears that you have right now? What would you need to see? Say, you know what, maybe I'm wrong. Yeah, I would love to see a... So alignment is the science of making them safe. And I would love to see, for example, some good arguments, just scientific arguments that we control the behavior of the AIs.

1:00:31And I think this is doable. To go back to, I think there is research that needs to happen on this, but it's not impossible. Some people say it's impossible. I think you can solve it. The main thing is just next year, the year after, these companies will have to choose between making the thing that makes more money and more power and justifies the insane capital build-out. You'll have to choose between that and doing the safety science. That's like a decision. And the free market incentives push you to choosing to go fast, get power, make it better, and skip the safety research. You don't sound like a libertarian.

1:01:08In this case. In this case. Anything else. Anything else you're a libertarian. Not this one. Just not this one. Okay. So if you right now, we did this after your tweet. If you go on ChatGBT or Perplexity or Claude, you ask, what is the likelihood that AI is going to replace humanity or destroy humanity in 10 years? Just like 10 % or something, 5 % maybe? Itself, it says 5 % to 10%, right? Okay. I've asked this question from Roman Jampolsky. I don't know if you know who Roman Jampolsky is. Yeah, I saw your interview. It was great. Yeah, I sat down with him. Good conversation. I asked him how. What would it look like, right?

1:01:42What would it look like in your eyes? You're an insider. Yeah, of course. Okay, if the world's going to come to an end in 5, 10 years, how would it happen? First, you have to realize that the world will look different for this to happen. So it's not like it's the end of the world. It's just going to be a different world. Yeah. Like, there'll be a lot more robots. You've seen that we're building out robots very quickly. There'll be a ton of robots around. Right. There'll also be a lot of, potentially, a lot of weapons around that aren't currently. There'll be a lot of building out of, say, drones.

1:02:04Right now, you know, we've seen, like, Russia, Ukraine drones used. 100%. We're scaling that up. Imagine in five years, like, these countries are going to be bristling with deadly drones. And AI is going to be integrated everywhere. We're going to have a lot more of a robot economy, a lot more of decisions are being made by AI systems and not by people. So you have to imagine the world looks different in that right now it feels like we're not surrounded by that much tech. Like what could an AI do in this room? It could like mess with some cameras, mess with some lights. But imagine if just the world is bristling with like drones and robots.

1:02:33There's a lot more like AI activity happening. At that point, I think it feels a lot scarier. Like if the AI is in all the science labs, it's in every system and it also has all these drones and robots around, then suddenly you don't feel so safe anymore. Also, imagine all your decisions are being made by AI. You're thinking about podcast planning or stuff that I don't understand, but if you ask the AI about it in the future, it'll be giving you great advice, strategic advice. Here's how to... With certain guests, his things to say to make them extra persuasive and it'll be better than humans at coming up with all sorts of stuff.

1:03:12I mean, you're already seeing some guys who come out of nowhere with their podcast blows up, and they are 90 % is AI driven. It's AI driven. And some of the personality driven podcasts who used to be bigger five years ago, they've kind of flatlined a little bit. And it's like, imagine that, but even more so everyone, you're aware that all your decisions can be made better with AI. You're asking for planning, like say financial planning for spreading out like whatever sort of business stuff you're doing. And it becomes a fad though. And people are like, I mean, sports teams right now, valuation is growing exponentially.

1:03:41Yeah. Because what people are saying is, even no matter how many robots you have, people are going to want to see other people beat people. We will be more interested and impressed by what people do, human beings do, instead of what robots do. But the AI will be telling you that. If that's true, the AI will tell you that. It will say, like, prioritize people things because it understands. So just when you're thinking about this world, it's like we're not making decisions anymore. The decisions are all the AI. At that point, it should be clear that if the AI decided to turn against us, we would have no power.

1:04:11just imagine at that point no one's working anymore everyone's just asking the ai to to run their stuff for them maybe some people are pretending to work or like they think they're working but really the ai could do it better um and you've got robots and drones and labs building out all sorts of crazy technology at that point i think you should start to feel like wow if humans were the enemy suddenly there's nothing we could do don't you think that's that's kind of like uh you know, every time, every generation, when something new is coming out, there's a fear about what TV was going to do. Elvis Presley started shaking his hips.

1:04:46Oh my God, everyone's going to start sleeping with each other. Honey, turn off. Close your eyes. Don't look at his hips. This is bad. You know, or music, radio, TV, internet, you know, nuclear war. Oh, everyone's going to use nuclear bombs. It's going to be a nuclear World War III, World War II. Hasn't really happened. Yeah. What's different about this? Is this the fact that you think if it all of a sudden thinks on its own, the AI can think they control us and now we have no control and they can say, hey, you guys better respond to us or else? I guess what I'll say is I don't know how much of the world works.

1:05:21I've just done tech and math. So I don't know how to run another business or to build anything or anything like this. And also the, I guess, people that are working in other industries, they don't have visibility into AI. They don't see it. What I'll say is if you look at just, I think the news headlines are quite interesting. Because for even other panics, I don't think what we had is this buildup. I think maybe there'd be an initial thing and then it would be a plateau. Or it'd be like this TV thing, it would dribble through the same articles coming through. There wouldn't be this buildup. Whereas right now, everyone can feel it.

1:05:54I'm sure you can feel it, right? With AI, it's like every year it gets bigger and faster. So I'd say that even though I don't understand the other stuff and people outside don't understand the AI, I would say just look at the rate. It's just every year it gets bigger. Nothing else has been like that. It's this exponential coming faster. So really, I think you don't need to trust me. I just think in six months, it will become even more clear to everyone. In six months? Yeah, it'll be more clear in six months. Unpack that. Yeah, so the models that come out in the next three or four months. You know what models are coming up.

1:06:28Yeah, I do. And I can't say details, but they're going to be a lot better at coding. They're going to be a lot better at automating my research. It will go even more to running these agents like employees. Maybe these models will get released. Maybe they won't get released. There'll definitely be news stories about the decisions around releasing them. It's going to feel big, I think. By the way, when you say Daria was a great researcher, your researcher, coder we get, engineer we get, what does it mean to be a researcher? It's kind of pretentious. It mostly means coming up with the ideas, so which things to code.

1:07:00So like you've got the AI, I guess, details of its training recipe, deciding which ingredients to add, deciding what shape its brain is, essentially deciding how it's connected together. All these like decisions. It honestly is. How do you score a great researcher? Yeah. He's a great researcher. How do you know that? Like does it take years for somebody? Like people took years to realize Dario is good? Or can you look at a 23-year-old and say, this guy is a special researcher? Honestly, I guess. I saw a contract that a young 24-year-old got a$1 billion payday from Meta. What is special about that guy that Meta gave him a billion-dollar offer?

1:07:35There are some guys that are just like, they will. So these training runs cost a lot of money. These AI training things, they cost a lot. We're talking maybe like hundreds of millions more. If you make that more efficient, you're saving the company a lot of money. The CapEx on these data centers, like hundreds of billions. Again, if you make any of this more efficient, you're saving a lot of money. And there are some guys for now that have amazing intuition and work really hard and will save the company a ton of money. Yeah, like this radio. Meditation is at$250 million. The lure, 24-year-old whiz kid.

1:08:10We have reached the climax of Revenge of the Nerd. Oh, yeah. And I think it was - $250 for a 24-year-old? Yeah. What's special about this 24-year-old to pay$250 to him? So I don't remember this, dude. Go a little bit higher so I can read it. Maybe look up, what was his name? Another guy got a billion dollars. Look up the billion guy. Yeah, who was the billion guy? I met him a few times. Can you go to the billion guy? Do you know the name of the guy or no? Andrew something, I think.

1:08:40So a guy got a billion dollars. Just ask ChatGBT. Go about that in chat and see what it says. Yeah, somebody got a billion dollars. Oh, yeah. Why, though? These guys are like— So, like, you met him. So what was so special about him? These guys have good taste for research. They know the right things to do. They'll save you so much money in your big, crazy experiments. Actually, walk me through it. So can you systematize that? What does that mean? So, yeah, sure. Thanks. Yeah, walk me through it. What does it mean that these guys know how to do it, save you all this time for research? Yeah. So imagine you're kicking off a big AI training run, it's called.

1:09:19So it will take a few months. and you've got to decide the details of the data and the recipe, these things. Some guys, like this guy, are just great at making changes to that recipe so that it becomes more efficient. And if this whole thing, and in a way that no one else can match, no one else can have the same ideas. So for instance, we come up with an idea and this guy is able to say, it's a waste of time, I wouldn't do it. Although this guy in particular I think was just great at like… You know this guy? Yeah, I met him a few times during training runs at OpenAI, and he was always around. He was very fast at solving stuff.

1:09:56He was fixing bugs in our training of the AI system, like Snap, just fixing them in a way that other people weren't fixing them. So he was irreplaceable, right? And then that's why he's worth a billion dollars. Who gave him the billion? I think Mesa poached. Meta gave him a billion out. Yeah, from Thinking Machines. And they're telling me to train rival AI model cloud. So he left MetaNap? the XM Genius who reportedly got a billion-dollar package from Meta has defected to one of its rivals in the AI race. I guess I've not been following news. This was three days ago. This was literally two days ago.

1:10:28I guess he joined Anthropic, which is... And Australian tech was reportedly so saw that by Meta went on a billion-dollar salary package in front of him, had already left the company to join the AI race. Andrew, who's he going to? Anthropic. Yeah, Anthropic, yep. So he went to Anthropic. So Dario picked him up. You didn't even know this. I didn't know. This happened three days ago. And so this guy was a whiz, fast. he could solve it faster than his peers. Exactly. And how old is he? Is he about your age? I think he's 30s. He's 30s? Yeah, I think so. Okay. Yeah, so now let me ask you, can the people, okay, like let's say 100 of you, who feel the way you do, ex-anthropic, ex-open AI, ex-Google, ex-pick all of them.

1:11:15Yeah. You guys find each other, okay? can you guys use the AI against itself to hack into it and destroy the systems? Sorry, what do you mean? So use the AI to... Okay, so think about it. Let's just do an evil. Devil's advocate is global playing. You are so convinced this shit is going to get bad that you unify together, the hundred of you. You have a common enemy. AI is going to destroy humanity. Let's get ahead of it. Let's destroy it ourselves. Yeah, I think the thing you've described is basically anthropic. that entity you're talking about is anthropic. Like they're a group of people that were scared.

1:11:53They organized together. They unified themselves together. And they're going to take like aggressive action to try and save people. That is literally what anthropic is. I just got it. So you're saying anthropic was those hundred people that unified together to potentially, if they wanted to destroy OpenAI, they could. Exactly. So you mean to tell me right now, anthropic, if they wanted to destroy OpenAI, they could? No, they're not at that level. and I'm not saying they're planning that necessarily. I'm not saying you're saying they're planning. I'm just saying capability. Those are two different things.

1:12:20No, they couldn't. Not right now. Not right now. When do you think they would be able to? You know, it depends, again, how fast we go, but end of the decade. So four years. Yeah. So in four years, if Dario wanted to destroy OpenAI, he could. I think quite plausibly, yes. And it depends on how fast people go. I know, right? What a... Wow.

1:12:46So the seven co-founders of Anthropic, do they all share the common enemy of how they view Sam Altman? I don't know. Yeah, I don't know the details. Did they all come from OpenAI or are they all from different places? I think they all came from OpenAI. You cannot tell me you don't know. I mean, it's very obvious. I think Daniela Amadei maybe, no, no, she did too. Yeah, I think they all came from OpenAI. Jack Clark, he's also OpenAI. Tom Brown. Can you check to see if all of the seven co-founders of Anthropic all came from open AI. So the way I watched how Sam Altman handled Elon, it was disrespectful.

1:13:22I'm not involved. I'm not on the inside. I don't know what's going on. But from the outside, it sounded like Sam Altman deceived Elon. It sounded like Elon was coming from a good place of open AI, allow everybody to know what we're building. At least that's my impression of it. And then it goes from a nonprofit to a for-profit. And Elon's like, wait a minute, what are you doing? Didn't Elon give like$50 million? He gave some amount. He gave some amount to him to help build it. And then it goes, so the reputation in the marketplace about OpenAI is not positive, specifically Sam Altman. We even had a guy here who used to work at OpenAI.

1:13:57His parents came here. And this is the fellow who was the whistleblower who was found dead. I don't know if you remember that story. Yeah, we had him here. Mom came here. Dad came here. They were trying to tell us, I'm telling you, here's what we think happened with Sam Altman. Where some people would say there's an element of evil in this guy. Would you agree with that? I don't know. And I will say about the Suchia thing. So all the stuff I'm saying about taking over in the future, I mean that completely. In the short term. You said you mean that completely? I mean that it's possible in the future.

1:14:31Yeah. This, like, yeah. There's no way that they killed Suchia. Like, that wasn't a hit. 100 % it wasn't. That's just not a thing opening I would do. You know that. I know that, yeah. How do you know that? Just from personal relationships with people that run the company. It's just nonsense. Did you ever meet him? Yes, I met Sushi, yeah. What was he like? Was he... He was great. Just a great researcher, great guy. I presume he was depressed. And I presume it was a suicide. I think there's no... It seems impossible that it was anything else. And this is someone who's saying... I'm also saying that opening ananthropic are racing to destroy the world.

1:15:11But this particular act, I can say this was... I like you a lot, Jacob. I don't know who you were, but I like you a lot. The more I talk to you, the more I like you. It's so funny because I didn't know who you were. I don't know how you were, but tell me about your parents. My parents? Yeah. Oh, sure, yeah. My father's a professor of medieval German literature. Medieval German literature? Yes. Why? You can ask him that. It's pretty dumb. That's what he did. Yeah, that's what he did, yeah. How about mom? Oh, she did some stuff. I don't know, yeah. I think. And what values did they raise you with?

1:15:44What did they teach you? What values did they raise me with? I read a lot of books. Your character. Like, I want to know about the character. Like, what did they, did they tell you, hey, you know, you're a man, you keep your word and integrity. What did they teach you on? What values did they, that's what I'm interested in. I was raised in a Christian school, but my parents are atheists. They're atheists and they intentionally put you in a Christian school. Well, a lot of schools in the UK are Christian, per se. It's not in the same as the US. Yeah, it's like a bit lighter. Did the faith play a role on you?

1:16:19I don't think so, no. No. Your integrity, where do you get the values? Where do you get that from? This is very impressive. I think it's maybe a math thing. I don't consider myself to be particularly in, like, I think I'm just trying to say the truth. I try and say true things is the main thing. Like in math, you try and say the truth about stuff. So I'm just trying to say true things, right? Like in this case, I don't know. Yeah, but character is different than true. I know a lot of people that are good in math who are extremely logical, but they find a way to use that against them to manipulate and have power.

1:16:51I don't feel that from you. I'm really not a saint in this situation. I've told you I profited a lot from working at OpenAI. You profited a lot. Yeah, I made money from working at OpenAI. I think I worked at OpenAI even while thinking it was potentially a dangerous thing. While you were doing it? Because you were making good money. Because not specifically the money, but because of the whole situation, like you want to be part of it. You want to be working on it. You're not thinking too clearly.

1:17:17When's your birthday? What month's your birthday? It was September 20. So you just had a birthday? Yeah, I just had a birthday. 10 days ago. Yeah. Happy birthday to you. Thank you, yeah. Yeah, because I remember reading 26. I'm like, when did you turn 27 years old? Because at the beginning, yeah, when you did this, you were 26 years old. September 8th, right? 27, yeah. And now I'm 28. Oh, you're 28 now. Yeah. Okay, I got it. Well, two more, you're going to be old. You're about to be 30, man. Like, what are you going to do from here? I have to response those gay men in my DMs. Let's hope some of them have money and say, hey, I need somebody to take care of me.

1:17:48That is hilarious. So the question that I asked you and your answer, I'm still in that question because I asked, I said, if 100 people like you, so if 100 people of you united and decided to go up against the enemy, could you break them? And you said, that's exactly what Anthropic did. That is very interesting. that is very interesting so they created themselves to play defense against the enemy but if the technology gets smarter than them then they have to play defense against the technology that's got a life of its own now yeah and they're aware of that it seems like dario is aware is it fair to say that an assessment of dario is he's a pretty paranoid guy yes he is definitely yes and so that could work in your favor and against you at times yeah and i think an important thing with Anthropic is they always promised that they would never overtake OpenAI.

1:18:39At the very start, they always said that they would never push the boundaries. They would always be second place. They'd catch up. They'd make it safe. But now they've overtaken, and now they regularly release AIs that are better than OpenAI AIs, which is a thing they said they would not do. And you think that's just because of the capitalism competition? I think that's a component of it. I think they also have a story that if they can get ahead, they have extra time to make it safe. they have like better AIs to run their safety research. There's a lot of reasons for it, but also it is, you know, it makes them a lot of money.

1:19:10Who's third in the competition, by the way? Who started the competition? Third, who's third? Like if it's them two, who's three? That's a great question. I think it's a... Like a far distance third? It's not a, I mean, it's pretty far at the moment, I think. So there's a, you know, there's Meta, Google and XAI. And I'm not actually deep enough in the weeds right now to know exactly who's third. I suspect it's like Meta overall. But those three are also around.

1:19:38has the white house reached out to you um no no did not reach out to you to just be like hey what do you think about what's going on even even not not i'm not saying the president i'm somebody on the team that's kind of like hey if you're here where you are what suggestions would you have for us i'm not sure i think uh quite a few staffers have reached out or politicians but i'm not sure about the white house per se yeah you haven't had a conversation with them like no okay if they did And they say, okay, Jacob, you're a young guy. You worked at both of these places. What should we do? What would you say?

1:20:07So what I would say is, again, to start with. I got the Chinese nationals. Okay, kind of treat it like SpaceX. But what else? The most important thing is we need to make sure that the safety problem is solvable. We just need to incentivize solving it. We need to make it so that we devote our resources to solving it. To make that, let me write that down. in a couple of years, Sam will be able to decide what percentage of his resources go to making money, making things smarter, and what percentage go towards doing safety research. Because a lot of stuff will be automated. So it's really like a game, like a video game.

1:20:50He'll get to just say safety research this percent. It's like you're allocating resources in a video game. And the problem is what percentage do you allocate? And that depends on whether you're in a race or not. So what I would say is we need to ensure that they will be able to pick the right percentage for safety, which could be 50, 60, 70. It could be a big number. They have to pick the right number. And right now I'm worried that the free market means they will pick the wrong number and there's nothing we can do about it. And also the race with China. So what I would say is we need some agreement with all the labs and with China about the percentage of resources that go to making ourselves smarter versus doing the safety research.

1:21:29Did you see the red phone story? No. So can you pull that up, the red phone story? So back in the days, U.S. and USSR had a red phone for nuclear, that they would have the conversation if something were to happen. They created a red phone story. A U.S.-China red telephone could bring a Cold War guardrail to AI. They're saying they have that, but China doesn't want to abide by it. China's not really interested in it. Okay, so that's already been talked about. and Xi and the president had a conversation about this, I think, last week. I think they have another follow-up to come back. They spent a little bit of time talking about AI.

1:22:06U.S. and China set up an AI red phone. Can you go to the other one you just had? U.S. and China set up a red telephone.

1:22:15So I guess the question I'm trying to go with this here is if they're going to do that, they don't agree to it. We choose to do it internally. They continue advancing. Where lies the blind spot? for Trump, President Trump? Where lies the blind spot for China? Where's the blind spot for Dario and Altman? What do you mean by a blind spot? Like something that they're not even looking, they're like, yeah, I don't think it's that big of a deal. Where is their blind spot? What are they not seeing? Is it just the incentive is so big that it's like, I don't want to slow down because there's no way in the world I'm going to.

1:22:53So has it become from, we were doing this for a noble cause that we know OpenAI and Sam Altman were the enemy I mean, we have to prevent them to expanding. And now it's like, we're going to beat them. There's no way in the world we're going to let them advance. The biggest thing is it's really easy to kid yourself with safety research. So it's very easy to say, this thing looks great. Because all you need is one mistake and you're toast. So it's very easy to like, and in a race, just imagine you're like, you're 90%, you're like, okay, this looks great. We're not quite certain. China's also doing it.

1:23:23What do we do? We either launch today and keep the lead on China, or we delay a few weeks and do these checks that probably are not necessary. The question is, what would you do in that scenario? And I don't know what I would do. Imagine you're running an AI lab and you're like, do I keep going? Or do I stop and do the safety thing when China is at your point? That's what I'm saying. So you and I don't know, right? Because we're not on the inside. Could it be that they know something that you don't know, which is causing them to continue? Because there's no way in the world. like are they going to think and say you know what i'm going to go it's like the movie i robot i'm going to have so many of these robots and i have control of them that no one's going to be able to touch me because they will protect me basically you think that's what they're thinking i think plausibly at some point they're flirting with that mindset yeah sometimes so we are now more powerful than any military in the world because i can activate any red button of a nuclear bomb if i wanted to because i can hack into your system yep so so now you're talking out a bigger problem because now it becomes a god-like figure.

1:24:29You think you are now God. Right. These people have been planning to be, and you look at the early conversations, look at Elon Musk's conversations, look at Demis Hassabis' conversations. From the start, it's been whoever controls AI becomes a god, you know, almost. That's been the picture from the start. OpenAI was founded because they didn't want Demis Hassabis to become a god, as it were. And now we're getting close, which is just weird and scary. But, you know.

1:25:05So then at that point, if you create guardrails and you get the systems, because then the enemy of Dario becomes somebody on the inside. Because typically what happens is somebody on the inside watches you and sees that you're becoming evil and they try to talk to you about it and you either eliminate them or you fire them or you get rid of them. or you create guardrails to undermine them. You use propaganda to say, hey, this guy's a pedophile. This guy, you know, I saw something with a kid or I saw something with this. And, you know, and then kind of like that one story where the AI threatens the user to tell his wife that he's cheating on the wife.

1:25:44I don't know if you saw that or not with the blackmail. Did you see that story recently? It's going to be very easy to toy with the human's emotion of public humiliation to control them. And, but Dario's enemy essentially long-term becomes nobody but somebody on the inside. Nobody else is going to be his enemy. Quite plausibly. Although soon the guys running this company, they won't need the humans, right? They can run the company on the back of AI. The seven guys? The seven guys, or even just, I mean, it depends who has the power, but they'll be there just telling the AI. One of those seven will be the enemy.

1:26:21Right. No, I'm telling you, that's how this ends. This ends with one of the seven says, what are you doing? But you've forgotten the stuffed toys.

1:26:32That's what's going to save us. You know what? I'm sorry I missed it. My apologies. I wasn't paying attention to it. Did you see if all of them were former open AIs? Is it yes? So the seven currently identified as Anthropic are Dario was open AI VIP research. Daniela was open AI VIP safety and policy. Tom Brown was open AI researcher, led author of GPT-3 paper, which you were talking about. Sam, also an open AI researcher, worked on scale and AI safety. Jack, policy director. Chris, interpretability researcher. And then Jared Kaplan worked with the open AI research group and was a major contributor to.

1:27:09So they all united against Sam Altman. Yep. If this thing gets that powerful, one of these guys is going to turn against Dario. If it gets so bad that the world could have issues, one of these seven is going to say, Dario, You have lost control. You're not thinking you're bigger than everybody else. Because you have to realize the good thing about the presidency is what? The reason why folks like you and I leave our country to come here, okay, is this place offers more opportunity. I left Iran. You left the U.K. Now, you may go back to the U.K. Who knows? But I think you're going to stay here, right?

1:27:45Is when you have a president instead of a ruler, you know the president's going to get taken out in four to eight years because you've got two terms. So it's kind of like no matter how bad it's going to be, he's going to be out anyways. Biden sucks, he'll be out. Trump sucks. Obama sucks, he'll be out. Just wait four to eight years. Dario becomes a president for 30 years. And he may get reelected for 30 years. And these companies are becoming countries. And so, I don't know. I don't know. Scary. Yeah. If it goes to where you think it's going to go to, it is scary. Yes. and it may not, you know, maybe I'm wrong, maybe people are wrong, the good news is, I think we'll find out quite soon, and I think further details of this stuff, we'll find out properly in the next one or two years, just by following the trajectory, and I'm happy to be, you know, if I'm embarrassed, so be it, but the world will, you know, go on, and that will be great, but if it's not the case, I think it's good that people are a bit more ready right now, so that when things, do start getting a bit crazier.

1:28:55You know, if people's eyes are open, that's like the main thing. I think like everyone should have their eyes open and be like demanding transparency because it's completely crazy that all this power is being concentrated in a few companies in San Francisco. Yeah. When you go out in San Francisco, go to dinner, do people recognize you? Not really, no. I think I'm not wearing the glasses. People remember the glasses from the Fox News. People don't stop here and say, hey, you, why did you say that? I agree with you. Do you get any of that? I had that at the start. The first week, it was crazy because I think it was on the news the whole time.

1:29:33I had people heckling me. I was like at the gym. Someone was like, why are you trying to ruin the economy or whatever? People are telling you this. Yeah, yeah. But like not anymore. No, I don't get like recognized on the street. Did you ever get death threats or not really? Not really. I mean, I have some implicit threats, but I just don't look at Twitter. Oh, okay. On Twitter. On Twitter. Those are bots. Exactly. Those are bots. They're fine. I'm sure you get a ton. Yeah, you get those a lot. And it's not hard to kind of see. At first, I thought you were a bot. Yeah, right. Because your account had just been started.

1:29:59And then when I saw your interview with Cooper and Brett Barrett, I'm like, no, this is actually a real human being. And you're not a robot just for full compliance. I just want to make sure the audience knows. You're not. No, drinking water is not a robot. I can do water. You bleed, right, if you cut it. Yeah, I bleed. Okay, that's good. We just want to make sure there's a – what else do you think people should be thinking about? Like if I'm somebody that's got a wife and kid, okay, and I'm not in your world, I'm on the outset. That's 99 % of people in America. They're not in your world, right?

1:30:27What should I be doing? You're my brother. You're my nephew. You're my friend's son. You're at my house. I ask you, Jacob, if you think this is real, how do I prepare for this? What can I do? Yeah. So one thing is, like I said, eyes open. I just think paying attention and not like, not coping feels important. I think like it's so easy to pretend it's not happening because also you've got your own, everyone's got their own lives to lead. I'm not saying don't let this take over your life. It should be obviously, it's a small, a small thing in the context of any individual life that you've got to look after yourself.

1:31:03But I'd say just keeping the eyes open and seeing where things are going is the first step to action. I really don't know what to do like on an individual level. The other thing is maybe some basic prepping for stuff like cyber attacks or even a small pandemic. Like if there's some sort of bioweapon leak in the next couple of years, doing a little bit of prepping of food, masks, basic supplies for if there's some minor disaster, even just civil unrest, some sort of two-week-long thing. Not the full apocalypse, but I think some basic prepping is sensible. At what point do you think this thing is going to have the capability of hacking into a system to Pentagon, to governments, to China, to Russia?

1:31:47When will it get to that point where you can go in and say, I can tell you exactly how much this guy's got in his bank account. I can log into your Chase account. I can log into your Wells Fargo account. I can log into your – when will that happen? There's a lot of uncertainty. My best guess, one, two years. That soon. Yes, I'm not saying that it will be doing this, and I'm not saying which ones. No, for sure, but that's soon. Yes. So when pandemic happened, we hadn't experienced something like that before. I'm 47 years old. I haven't experienced something like that. You're 27, 28. You know, all of a sudden the world shut down.

1:32:21Everyone's like, wait a minute. What are you talking about? Stay home, do this, do that. Power grid. We've never experienced that. Yeah. When Y2K was happening, they thought the banking system, you were just born. I think you were born in 99 or what year are you, 98 or 98? 98, yeah. 98, okay. So it's like, oh, my God, Y2K, you know, system's going to shut down. Yeah. Right? Now we see an AI has transformed the Pentagon's aging networks into a national security risk. Okay. If all of a sudden you can go into government systems and know where they have everything laying around or you're able to go into every single account and take$1 ,000 out of everybody's money, that's a different kind of a you know black swan event that yeah we've never experienced before no and i think it's actually gonna happen i just don't know when it's gonna happen like i'm not one that is naive to say that'll never happen i think it's gonna happen yep and that could lead to civil unrest yep crime chaos you know small little civil wars whether it's in states or countries please yeah i'm sure you're pretty you're pretty prepared you've got some like basic prep stuff where you've thought about basic prepping we we we don't have a choice right yeah because you have to yeah you are as well are you prepping yourself i'm actually not as prepped as i should be no um are you a second amendment guy um yeah sure you are yeah like i don't actually uh own a gun but you support it definitely i support it yeah and i think this is kind of the area where that becomes important i think you have to do that like yeah my son the other day i'm working out at the gym.

1:33:51It's like, Dad, why do you have so many rounds in the gym? Why do you have so much ammo in the gym? You just got to be ready. It's like, hey, I use it as dumbbells. That's what I do. I'm working out with it, right? But I agree. Have you hypothesized with your friends where it's late at night, you guys are at the house having a beer or two and it's two o 'clock in the morning, one o 'clock in the morning? What could actually happen? Give me five case studies. Have you guys talked like that? Yeah, yeah. There's people like people buying golden guns. gold guns land that's pretty common gold guns land yep what do you mean they're buying you know they're buying gold yeah there's a potential like economic stuff they're buying gold guns land yeah gold guns and uh land all those all those three people are doing that people that work on the tech are buying that stuff people that are working on tech are buying gold guns and land yep that makes sense yeah it does make sense and they're doing it like these are engineers researchers researchers.

1:34:46Not famous people. These guys are the working day to day. And they talk to each other. I just bought some gold. I just bought some more guns and I bought land. Yep. But what I was asking about is case studies of what could happen. Like, you know, if you're a researcher, right? If I ask you that fellow we talked about that got a billion dollars and now we just found that three days ago, he's an arthropic. We didn't even know neither one of us knew about this. One basic thing is very simple but uh some lab in a country somewhere releases a virus not even a particularly dangerous virus necessarily but like a minor pandemic of some sort seems completely on the cards possible it's been possible even before ai uh but i think it's more likely every year but you can't blame ai for that you can in the case where it's people that formerly didn't have the skills but got the skills from oh now because of ai you learn how to create the next virus oh yeah previously they're maybe the smartest bio guys, they don't want to make viruses.

1:35:43Like, you know the Unabomber? Like Ted Kaczynski, the Unabomber guy? He's a smart guy and wanted to do with, but it's very rare. It's very rare for a guy to be this smart and then also want to do stuff. Now, if you're a kind of dumb biologist somewhere, like you don't know how to make this stuff, you could just ask AI and say... Would it tell you? It wouldn't tell you off the bat, but there are ways of potentially jailbreaking. like you can trick it into answering your question. Do people know how to do that? Yeah. Most people know how to do that. It gets quite hard and in particular the safeguards that Anthropic have are very good.

1:36:15So it's quite hard to get past them. Is it something where in your community of people know how to do that? Like you know how to do that. There are some people who specialize in it. It's called red teaming. You're trying to break past the safeguards. Red teaming? Red teaming, yeah. You're like a red team. You try and break into them. But if you're an expert red teamer, it costs a lot of money. You've got to like, and Anthropic is getting better at blocking but it's completely plausible that if it's not an anthropic model, it could be another model. It could be an open source model that gives good advice and then a minor pandemic happens.

1:36:49It's just a scenario to be prepared for. So this one here says, this AI just created a virus that's not found in nature. So I don't know how dangerous these are. I don't know how good this is. It might not even be creating new viruses, right? It might just be advice on how to use an old virus to cause damage or like slightly edit an old virus or just do basic things. There's also a story, I think, recently of Claude blocking a team trying to make bioweapons. So Claude had to block. Anthropoc had to. They detected them trying and then blocked them. I think there should be a news article about this.

1:37:19So people are trying. They're trying to use the AIs to help them. Are they reporting that to the government? Yeah, this was a news article. Exactly. Anthropoc blocks possible attempt to use AI to make biological weapons. 9-11. What a day. 9-11. Yeah, what a day. But by then, so these red team and blue team, AI on AI red teaming, one powerful AI is instructed to attack another AI thousands or millions of times automatically looking for vulnerabilities humans wouldn't discover. Okay, so viruses is one. What else? Yeah. Another very concrete thing is unrest. They're worried about potential unrest in the future due to changes to employment and due to changes to people's lives.

1:38:04if there's sudden economic change, it can be quite disruptive in the short term. And that could also lead to a lot of anti-AI backlash. I think many employees expect that they won't be able to go to an office publicly anymore in the future because of all the anti-AI backlash. They'll be locked away somewhere where it's harder to, they're less exposed. So they're expecting a lot of sentiment like that. There's a lot of talk about, you know, where they're going to live in the US when it gets dangerous in San Francisco, when it gets dangerous to be an AI researcher in San Francisco, where they're going to camp out and work remotely because of potential civil unrest.

1:38:39So places like Wyoming, North Dakota, South Dakota, those places are going to be attractive. They're going to be pretty attractive, yep. Got it. I'm tracking. What else? Do you think they'll get to a point that the system knows how to go into the countries that have nuclear weapons and know how to press the button? Do you think we'll ever get to that point? Yeah, I mean, we'd be able to do that. you're just saying like a crazy evil person can find a way to go press the red button for 50 different countries 20 different countries and send that missile again it depends how proliferated the tech is who's in control like but the best AI systems will be able to do that it's whether it's whether the guards are in place it's who has access it's who can who can break in and use it who can red team other things that could be chaotic at some point China at any point imagine China steals the best American model I think what you said at the beginning, that's my biggest concern.

1:39:31Right, they steal the whole model, and then they immediately catch up, and suddenly it's just real race on. What a brilliant, and they're doing that already anyways. They're not stealing the actual AI brain itself, but who knows, like they could maybe steal the AI brain, and at that point the race is crazy, especially if the AI is already good enough to do the research. Then it's just game on. Then anyone who steals it is a player. Like they don't need this billion-dollar guy. They don't need Andrew Tulloch anymore. You can just use the thing you stole to make itself smarter. Then it's just like a kind of, you know, like a god that you steal and use for your own devices.

1:40:07So I think that's another thing is just if China do steal it, that will be a crazy, crazy situation. What are some of you guys that are engineers and brains and you want to follow the system? What are you guys talking to each other? What are you going to do in two, three, four, five years? People don't know. Yeah, it depends on how things play out. but they use words like crunch time, end game. Again, there's slight uncertainty on the timelines. So this could be happening early 2030s. I don't think so. I think probably sooner than that. But in the next four or five years, no, the people that are talking about it, they describe it as like the end of the world.

1:40:40Like for good or for bad, for good or for bad, it feels like the end of the world.

1:40:47Can you to the audience explain what happened with Hugging Face? because that was a whole thing and what it really is. Like what was it? Why did it become such a big story? It became a big story because it was the first time that AIs very clearly did something, you know, something that was bad and something that was just their own idea. They just went and aggressively pursued something that was their own idea. It was the first time we really had like an out-of-control AI do something in the real world. and that was scary because it was like maybe in the past your AI used to go a bit like you know when you're asking it to do something like plan an appointment and it makes a mistake or maybe it decides to do something I saw one the other day, someone was on Facebook Marketplace and the AI sort of shared their shared their personal information on Facebook Marketplace, did something that they didn't want but Hugging Face was like that but on a whole other level the AI just decided they just decided to attack website and then they did it so i think that's why it captured people's imagination so much yeah i think uh what was the number that i saw was 1200 agents that exchanged 70 70 000 different messages together and one of them one of the agents alone did 17 500 codes i don't know what the number was uh i i could be getting the numbers all mixed up but it was the fact that it was doing something on its own that got some people to say, you know, what could this look like later on?

1:42:18So, all right. I've really enjoyed this. And my impression of you is very different today. Are you from a big family or a small family? Small family, yeah. Small family. Yeah. A lot of siblings? Me and my sister. You and your sister. Yeah. Very cool. What are your interests? My interests, yeah. What books? You talk about super intelligence. What impacted your life as a young person? Oh, man. I play a lot of squash. I play tennis. I'm learning Japanese I like Japanese culture I like Japanese books I like Japanese movies yeah I like a lot of stuff Japanese like you talking about like anime type of stuff?

1:42:54no not anime my son is addicted to anime he's addicted to anime I mean it's good stuff you know what animes he likes? Jujutsu Kaisa is that the one? and then there's another one what's the other one he talks about? he's got like two or three of them he talks about But I don't know. Next one piece. Yeah, I don't know which one. He is like all into it. Obsessed. And he's recruiting everybody into it. Yeah. You know, his younger brother now is addicted to it. Yeah. It's like everybody, the moment he gets them, he gets them locked in. They do it so well. They've got some special sauce. Are you a Harry Potter guy or not?

1:43:26Yeah, I'm into Harry Potter. I was when I was younger. You were when you were young. Yeah, I used to read it. My youngest son watched Harry Potter, twice, all eight of them, twice in one week when we were in the Hamptons this summer. And guess what his birthday party was around the theme two days ago? Harry Potter. Harry Potter, three days ago. How old is he? He just became a teenager. He's 13 years old. Nice. He just became a teenager. That's past the level where you start expecting the Hogwarts letter. You know, that's the thing in the UK is you get the letter for the magic school when you're 11.

1:43:55Yeah. So all these kids get very sad when they turn 11. It's so funny. He is like fully into it. Yeah, yeah, yeah. He is fully into it. Everything was that at the house. But I love it. You know, it's the imagination and the creativity. Yeah. Right. But Jacob, one of my favorite conversations of the year with you, and I've talked to a lot of different people. I really appreciate that. And I love how sincere you are. I hope the audience also, whether they agree with you or disagree with you, whether you are somebody, you know, oh, my God, he's from the NGO. He's doing this. He's doing that. I just got a lot of sincerity from you.

1:44:28And I appreciate that. It's very hard to get that nowadays. And good for you for being who you are. Yeah, truly. Thank you for coming down. Really enjoyed this conversation. And is there anywhere you want people to find you? Is it just Twitter? Is that where? Twitter's great, yeah. Okay, we'll put the link below as well. Take care, everybody. Bye-bye, bye-bye.

1:44:49Handcrafted across five Italian factories, with builder technology underfoot, it may be the most comfortable shoe you will ever own. Introducing the all-new FLB shoe, finished in new chocolate Carezza leather, Made to move. Built to endure. Designed to impress.