Chime Climbs, Tech Titans Join Army Reserve, Meta Buys 49% of Scale AI | Joshua Steinman, Bernt Børnich, John Doyle, Emily Sundberg, Lennart Heim

28:27I had no idea. It's a little bit dark, but he's talking about the ISS's structural integrity. He's quoting post from Christian Davenport, who says, the Axiom 4 mission is postponed indefinitely while NASA and Roscomos investigate a leak on the Russian side of the International Space Station. And so Casey says, the ISS's structural integrity is far more marginal than is being publicly discussed. We are having multiple and increasingly frequent leaks from heavily fatigued node segments in the Russian section. John. Yes. This screams like the plot of a sci-fi movie. It really does. Leaks coming from the Russian section.

29:05It's crazy. Yikes. When aluminum gets flexed, it fatigues and gets harder, increasing its tendency to crack. Cracks concentrate forces at their tips and spread over time. Multiple cracks have been discovered. There is no factor of safety associated with this failure mode. none of the structural pressure vessels are meant to crack. We are not even single fault tolerant on the structural integrity of the station. We could wake up tomorrow and find that with zero warning, it has failed catastrophically. Whether that means a leak slow enough to close some hatches, get the crew out, or at least into safer parts of the station, it's a roll of the dice.

29:43It could also depressurize in less than a minute. And Elon Musk says there are potentially serious concerns about the long-term safety of the International Space Station. Some parts of it are simply getting too old, and obviously that risk grows over time. Even though SpaceX earns billions of dollars from transporting astronauts and cargo to the ISS, I nonetheless would like to go on record recommending that it be deorbited within two years. Striking, very interesting. We got to put up a new one. We got to put up 10 of these. I know that they don't have that much economic value, but this is what I love.

30:16Yes, we need a dozen more ISS. I know that it's so expensive. I think it's one of the most expensive man-made objects ever built. Although maybe Stargate will beat it out. I seem to remember the ISS costing a total of$100 billion. The thing with Stargate is it's a project that's happening across states. So it's hard to say that it's the most expensive object. Oh, much like the ISS's different modules put together? Oh. Oh. Kind of similar. We can fit the ISS in a large enough warehouse. Yes, yes, yes. And fit Stargate as a series of warehouses. We have to have a human-rated space station. We have to have a space station.

30:58It's just so important. I would like to see an Elon megaproject. I like that. There's that website. There's that website. How many humans are in space? Have you seen this? Humans are in space right now. Whoisinspace.com. There are 13 humans in space right now. Who is in space? the iss crew 10 launched march 14th 2025 there's four people there there's the soyuz it's crazy on there johnny kim's up there you know wait no way you know johnny kim right no oh johnny kim is a legend so john oh yeah this is the guy who's big go yes he's done everything he's johnny kim yes so he was a navy seal he's a doctor and he's also an astronaut and the whole meme is like his parents are finally proud of it.

31:42Happy. Because he did everything. But it's like, I'm sure they still want more. Johnny Canvas is the best. We should get him on the show from space. Been up there for 66 days. He's going to meet some people over there. And then Tiangong Space Station, the Chinese space station, has three astronauts on it and then another three from a different mission. And so there's six over there and then there's seven on the ISS. And so we got 13. We need more. I'd like to see. We need to be bringing this exponentially. I'd like to see them doing more like Kai Sanat live streaming daily. Yes, totally. I'm sure they get a little lonely and bored.

32:19It's a content problem. They got to be marketing. Everyone's like, oh, you're running some scientific experiment. When are we going to do the paper? That's boring. They're like, thank you for the$5, John Coogan. Exactly, thank you for the$5. They're like doing impressions and stuff like that. If you donate$100, they'll squeeze the washcloth and show how the water sticks around. Let's check it out Tyler. Let's see how he's doing. Give us an update.

32:45He's black and the table's black. It's quite hard to see. I don't know how far I am right now. Good luck. He's at 15 minutes? I can do it. Keep going. Keep going. He's working. He's working. Anyway, let's... We're releasing recap videos now. roughly a one hour recap video of everything that happened on tbpn this entire week we know we put out a lot of content for you this week we'll be recapping wwdc apple released liquid glass it was a little bit controversial we talked to people that were very optimistic about the new design direction people settled down in there i think they eventually settled down to themselves all right this thing might be pretty nice looking it's probably will get pretty good and we'll all move on.

33:34And they'll probably transition well to VR and other platforms. Yeah. It also tied really well to Apple's kind of retreat. Ben Thompson wrote about this. Apple is turning over the reins to developers to develop some new AI applications. So we're very optimistic about that. We also had our intern, Tyler Cosgrove. He's making Legos right now. But on Monday, we had him assemble the first American iPhone from scratch. And by scratch, I mean seven pieces that he had to kind of glue together, but he did it. It was way more than some credit It was way more than seven maybe twelve, but it was fantastic We have we had Andrew Huberman on the show.

34:14That was fantastic. It's great We got to talk to him about what's going on with NIH budgeting The proposal is that research funding will drop by 40 percent. We tried to dig into what is what what what pieces of the research budget? Should we keep? What is the dynamic between the different types of research that are funded? And he really broke it down for us on the back of his four-hour interview with Dr. Jay Bhattacharya, who is the current NIH director. That was a very fun interview. We also went to Y Combinator Demo Day 2025. We were live from San Francisco. We talked to dozens of teams, dozens of founders.

34:52We talked to high school dropouts, college dropouts, tons of AI agents. About 90 % of the new YC class is AI-driven. But we had a bunch of interesting conversations. We also talked to some venture capitalists there. And we shot a lot of confetti. Where were all the hard tech founders? We did not see many of them. There were a few. But we mostly got AI deep dives. They were among us. And then we also discussed the meta and scale acquisition or investment. Meta is investing something like$15 billion. The deal actually just closed today. And so we had commentary from a number of founders and pundits and journalists and also venture capitalists talking about the nature of that deal and how it came together.

35:37And so we hope you enjoy this week's recap and it gives you a good overview. We're one away from being seven days a week. Yes, we're very close. We also have a great show for you today. We already had Josh on the show, but we have One X is coming on to talk about humanoid robots. We've got Emily Sundberg coming on, the fantastic writer of the newsletter Feed Me. We have John Doyle from Cape building a new cellular network. There's a lot to talk about there. And then we have someone from the Rand Corporation coming on, which will be very fun. And so we will go back to the timeline, back to what people are talking about in the news.

36:16more updates on what's going on in New York. Zoran Mamdani is surging, continues to surge in the polls and seems to be doing very well in the debates, but is not popular on tech Twitter because he says capitalism is theft. The quote is taxation isn't theft. Capitalism is. Zoran Mamdani could soon be mayor of the finance capital of the world. Incredible. It is a very odd choice for a city that is known for capitalism and finance. Finance. Finance. And yeah, we will see how this one evolves. It is already a case study in marketing. Signal called this out earlier this week that he's just laser focused on a few key lines and it is resonating with some, certainly not our corner or our audience of capital enthusiasts.

37:06Well, if you're a capital enthusiast, head over to public.com, investing for those who take it seriously. They got multi-asset investing, industry-leading yields, and they're trusted by millions. Let's go to Pavel. Pavel says, Pavel Esperruhov says, traditional company structure is falling apart. Pretty much every organization should be structured in three parts. People who sell things, people who do things, people who automate things. Everything else is just fat. Very interesting. Sellers, doers, automators. That's kind of maps, too. I would, I think this is, I mean, great insight ultimately resonates, but people who make things, that people who sell things, people who make things, people who automate things, people do things.

37:50Do things is just doing things. Yeah, yeah, that is interesting that there's not, yeah, the doing things part of this is the most vague, but maybe that's management and strategy and deciding what to do in the organization. But I do like the reframing as like... Some people make things, some people sell things. Yes. Yeah, yeah. Maybe making things is better. But again, the automation thing seems like an interesting twist on a traditional engineering organization. As the role of the software engineer changes, it becomes more about automating any type of business process. And I think the role of the software engineer is definitely growing in scope.

38:30It's not it's vibe coding an internal tool using cursor retool or all the different softwares that are available. It's also buying off the shelf or bringing in off-the-shelf software to automate processes and the the leverage of the software engineer leads to just massive automation across the entire company. The selling things is very is very interesting because even though that is automated in some ways. It increasingly relies on... What's automated about a steak dinner. Exactly. On interpersonal relationships and communication. And so that has definitely carved itself out as a unique function within the modern organization.

39:15But yes, I think the people who do things could be build, could be make things, but could also define the strategy and build the company. But if you want to automate things, put your sales tax on autopilot with numeralhq.com. Spend less than five minutes per month on sales tax compliance. Just get on there. And if you're one of those people that sells things, yeah, and if you're one of those people that sells things, get on Adio, customer relationship magic. Adio is the AI native CRM that builds, scales, and grows your company to the next level. It is the backbone of our high velocity guest program.

39:53Get on there. Chamath had a hilarious post that I really enjoyed. He says, well, this is probably one of the funnier things I've done, but my friend and sommelier and I got a California liquor license and started a wine collective. With our license, it allows us to buy wine directly from vintners, get wholesale pricing, buy and house entire collections of wine. We didn't do this just for us, but created it to allow our friends to participate as well. At some point, we may open it up to the world at large. If you love wine, want access to the best wines, and avoid retail markups, this may be a fit.

40:28The team charges for shipping and handling, but otherwise pass through the raw cost to our members. Right now, we're saving 15 % to 40 % per bottle, depending on the winery slash vintage. We call it Drink With Me. We also create quarterly capsules of the best wines Josh and I have drank and allow members to buy in quantity. The first box is below. So it seems like he's on the cusp of launching a D2C wine company. I mean, this is a wine membership, wine club. There's thousands of these. People love them. I think it makes sense for Chamath's creator evolution. I mean, in terms of his brand. Always drinking wine all in.

41:07Always drinking fantastic wines. All in alone and just casually dropping a little note about drink with me or whatever wine he's drinking. could probably drive this to 10 ,000 plus subscribers. I completely agree. The creator market fit here is fantastic. Powerful. Very fun. And if you zoom in, it says, drink with me, a curated wine collective by Chamath Palihapitiya and sommelier Joshua Plak. It looks great. Congrats. I mean, he has a passion for wine. He's expressing it in a business. You love to see it. It's great. We talked a little bit about this, but we didn't get back into this. Scott Belsky came on the show this week, talked about the data wars, and there's new reporting from the information.

41:52Salesforce is blocking enterprise AI rivals like Glean from using Slack data for their applications. This was a big issue back in 2012 with this idea that you had all this data in Facebook, you had all this data in Google, you had all this data with Apple, and you could not, even though every one of those companies would tell you, it's not our data, it's your data. You can take it with you. It's yours. You own your data. People would say, okay, that's great. Google, I'd like to give all of my data to Facebook because I want better ads on Facebook. And Facebook, I want to give all my data to Google so that when I go in my Gmail, I can search my friends list.

42:32And both companies would be like, oh, no, no, no, no, no. We don't trust the other company. It's your data. You can export it through this like CSV zip file that you have to download every time. But we're not just going to let an API link these two services. Absolutely. Because the walled garden must be maintained. And, you know, it's like, oh, oh, beautiful garden. Oh, it's my it's my data. Then then then I can, you know, integrate my iMessage with my Android friends. Like, oh, no, no, you couldn't possibly do that. And and I mean, these big tech companies, they have they have reasonable arguments for it, but it always has been funny.

43:07and now it's more important than ever. And so Scott says, anticipated and referred to this as the impending data wars in previous editions of Implications. That's his newsletter that you should go subscribe to, but didn't expect to start this quickly. Data grab and protectionism underway, especially by those that don't own their own graphs or sophisticated permissioning layers. And so, yeah, there's this big question. OpenAI is launching deep research. It integrates with Gmail and Google Docs. That's gonna be fantastic. if they cut that off entirely and they get cut off, like that might be something that actually drives enterprise adoption of Google's Gemini deep research project, right?

43:46Because if we have a whole bunch of documents in Google Drive and we want someone to be able to run a deep research report around all of our internal documentation, which obviously they would want to do, and Google successfully cuts off OpenAI, like you're just going to have to go there or else it's going to be a hassle of like, download all of our PDFs from Google Drive and then feed them into ChatGPT one at a time. No one's going to do that. The friction's going to be so high. Even if you can download your data, it needs to be there over an API. Salesforce has taken the first shot in the data wars and we'll be tracking it here on the show.

44:23Shots fired. For the next forever. This was a very timely call out from Scott on the show. Yes, it was. um well uh if you're losing sleep over the data wars you got to get on eight sleep go to eight sleep.com slash tbpn get an eight sleep how do you sleep last night jordy i'm on a generational run put up huge numbers last night i think i got you beat this whole week across i think it's i think it's tied up uh monday 91 tuesday 86 wednesday was on the road doesn't count thursday 91, Friday 95. Let's hear it for me. I got completely mogged. Oh, soundboard not working. I was expecting some clapping, but the soundboard must be broken.

45:08Oh, there we go. That's my favorite. I'm happy for you. Just throw up an excuse. I'm sure you have a good excuse. Let me hear it. It actually was funny. I woke up a little late this morning because my son crawled into bed and basically pushed me so far off the edge of the bed that the eight sleep alarm the vibration alarm was waking him up he didn't even wake up he was just he's sleeping like a rock but but i was just feeling it like very faintly yeah yeah my son hops in the bed after i after i leave and uh and i use the warm-up alarm from eight sleep which is fantastic uh it it really makes me get out of bed but he loves it because he's like oh it's so warm and cozy He sleeps like an extra two hours because he likes it really warm.

45:57So he enjoys my post-alarm side of the bed, enjoys that. Very sweet. Anyway, let's go back to the Army's newest recruits, tech executives from Meta, OpenAI, and more. The Wall Street Journal was covering it. Executives from Meta, Palantir, and OpenAI are joining a new Army Innovation Corps, bringing tech upgrades to the military. the tech reservist will serve 120 hours a year. So two hours, a little over two hours a week as lieutenant colonels advising on AI and commercial tech acquisition. Very exciting. And so I like this from Boz, Andrew Bosworth over at Meta. He says, it's possible I watched too much Top Gun.

46:40He's 43, standing more than six foot two. Happens to the best of us. It does. You can never watch too much Top Gun. Come on, Boz. It's an amazing movie. He was told he was too tall to realize his youthful ambitions of flying an F-16 fighter jet. Did that happen to you too? It was never even in the conversation. I remember seeing a movie about the Vietnam War when I was like 13, and it was very obvious that I was never going to make it in the military. But there was someone from my high school, and I do have a family member, who's flying F-16s, and it seems like it's a wonderful program. It's still incredible.

47:15You really have to be a very specific specimen to be a pilot. Yeah. I did flight lessons as a kid, and I was in the Civil, what's it called? Civil Air Patrol. And I was really worried as a kid. I wanted to be a pilot, and I was worried that my eyesight would degrade because you can't have glasses, you can't have LASIK or anything like that. You just got to have perfect eyesight. You got to be very specific height. And anyways. Yeah. Yeah. Bosworth went on to say, there's a lot of patriotism that has been under the covers that I think is coming to light in the Valley. I completely agree. They will be sworn in as uniformed officers in a public ceremony on Friday, the day before the Army's 250th birthday.

48:04Happy birthday to the Army. Happy birthday. Less than a decade ago, even working on technology that might be used in defense, never mind suiting up for service, was anathema in Silicon Valley. The new reserve program reflects how the relationship between the Pentagon and the tech industry has deepened. Meta and OpenAI adjusted their policies to work more with the military last year. Although Meta is not technically a defense contractor right now. They are a supplier to a defense contractor with their annual partnership. A little bit of confusion there. That story went out and it was so fast. A lot of people didn't understand.

48:37They're not an official military contractor at this point. Although, I would love if they were. Palantir has been involved in national security work for two decades. It has an AI and data project with the Army worth potentially more than$1 billion. For the Army, the deepening ties can prepare for wars of the future. They are expected to be waged in part with ground robots and drones and rely on networks of sensors and artificial intelligence to coordinate it all. I mean, looking at the conflict from yesterday in Iran, that feels like, in many ways, the future of warfare. Zero troops deployed and absolute devastation.

49:19Well, speaking of robots, we're going to have the founder of 1X on the show in just five minutes. But in the meantime, let me tell you that AdQuick, out-of-home advertising made easy and measurable. Say goodbye to the headaches of out-of-home advertising. Only AdQuick combines technology out of home expertise and data to enable efficient, seamless ad buying across the globe. Go to AdQuick.com. Anyway, Shamsankar took to the free press to write a little bit of a, I guess this is an op-ed or a little article about his experience. He says, I am the CTO of Palantir. Today, I joined the army. He says, my father grew up in a mud hut in India.

49:52America gave him and me a life. Now, technologists like me need to give back. Oh, this is heartwarming. Fantastic article. Does he have a Giga Chad filter on here? I love it. The best bit. This is your new bit. If you need a bit going into the weekend, if you see a picture of your absolute boy and they don't and they clearly, you know, just a casual filter, accuse them of using a Giga Chad filter. I mean, I mean. He looks like he's in shape. Yeah. Let's leave it at that without bringing out the Glazenator 3000. Later today, triple glaze. Receiving their commissions on my side will be some of the most impressive minds in the world of technology.

50:39Kevin Wheel, the chief product officer at OpenAI. Andrew Bosworth. Boz, the chief technology officer at Meta. And Bob McGrew, formerly the chief research officer at OpenAI and engineering director at Palantir. None of these men need to pad their resumes. None of them have free time between fatherhood demanding day jobs and a dozen other demands. Are they all dads? Let's hear it for fatherhood. Let's go. Let's go. I mean, we got 50 % of this group on the show already. We got it. We got to finish it out. We were close with Boz. We almost overlapped at some event. We're going to make it happen. We almost caught him at Figma.

51:12Config, we're going to get him on the show. Let's make it happen. A decade ago, it would have been unthinkable for so many tech heavyweights to openly align with the U.S. military equally. It would have been out of character for the military to enlist the support of the nation's business elite, much less create a special corps that they could deploy their technical talents in service of the government. But a sea change has taken place. Wars in Europe and Middle East, and above all, the threat of war in the Pacific, have focused the national mind and initiated a scramble for mobilization. Exploding pagers and long-distance drone strikes from shipping containers prove the technology has once again changed the battlefield.

51:46Our military has to change with it. The Army's Executive Innovation Corps, under the direction of the Army's Chief of Staff, General Randy George, who has been on the show, is part of a larger effort by our military to transform the way it prepares for and fights wars in the 21st century. Marrying the nation's most innovative private companies with our most important military missions is fundamental to that effort. That cooperation relies on reviving a sense of duty in an American elite that has become disconnected from our nation and its tradition of service. Let's check in with our intern and his tradition of service how is it going tyler servicing let's lego

52:31he's 75 of the way there though this is much harder than i thought well past 30 well past 30 minutes okay well uh we gotta start thinking of the punishment you don't get a prize you don't get a prize for well that's actually i have a really i have a really good one what is it i have a really good i know what you're thinking of and it's actually a bit no you don't you actually don't okay because i was going to uh give this to you as as a gag the thing that i'm imagining would be a gag gift for you okay but now it's going i'm going to sort of flip it around convert it into a punishment for tyler are you going to drop it right now what is it um it's on the way here it's on the way here it's an it's an it was in the car and okay and ben's going to drop it here in the next few minutes we might have to wait till after the next guest but uh you can look forward to that.

53:17Yeah, yeah, we didn't tell you, but if you'd been able to get it done in 10 minutes, we would have gotten you a fantastic new watch on bezel. We had our finger hovering over the nautilus, but that's kind of out of the window. Would you have worked faster, Tyler, if you knew that there was a novelist on the board? Your bezel concierge is available now to source you any watch on the planet. It'll have to be next time. But in the meantime, if you're listening, you can go to getbezel.com. We should treat this Lego set as sort of something that he can kind of iterate against. You know, like you don't just do a Rubik's cube once.

53:51You do it once, you try to do it faster, et cetera. And so I think we should give Tyler another crack at the Nautilus. Yes. And now that he's dialed in, he should be able to do it a lot faster. So Dylan, why don't you bring this up for a second? Okay, this is the potential punishment. What are we doing here, Jordan? okay i was gonna give this to you john because you're a horse these are the uh apple decks apple flavored electrolytes for horses it's from uh it's from horse health products horse health products it's a 40 day supply for classes for all classes of horses show it show it to the single and uh and i'm just gonna i'm got Tyler your punishment is you have to finish this before your internship ends this summer it's totally safe for humans so it's 40 day supply for a regular horse okay let's see how long it takes you you gotta see how long you get through it anyway so you you have this waiting for you after the show um and we could do something if you finish the next one the next iteration if you do it in 30 minutes maybe maybe you don't have to have it but you might try it and discover that You actually love it.

55:04So we'll see. Okay. Tyler's still working on that. We'll check in with him. He does have two hours. We have another hour of guests coming into the studio. We have one X coming up next talking about humanoid robots. But, and I'm excited for this. They've had a, they've had a massive week. And I have a ton of questions about, about humanoid robots. The biggest question is, is, is data that sim to real gap. There was something percolating in my mind related to the Alex Wang acquisition, the ScaleAI acquisition of Meta. Because robotics data is so difficult to get, maybe ScaleAI would play in that market.

55:41We'll see where that develops. But in the meantime, let's bring in the founder of OneX and start talking to him about humanoid robots. How are you doing? There he is. What's happening? Welcome to the show. I'm good. I'm good. Thanks for having me. Fantastic. Yeah. Give us some context on your background on 1X and NEO, and then we'll just get into a bunch of different questions. I was about to tell John NEO was actually blessed by Lil B, the bass god, I saw that. A while ago, much like Timothee Chalamet was before the generational run. So it's very possible that NEO is on a similar run itself. I'm very excited.

56:19So anyways, great to have you on, Bernd. And yeah, give us some color. Awesome. no and i think like uh let's see oh hey over here wow wow okay that's convenient um there has to be something through right it's it's been going pretty well lately like uh lately so um we have been blessed okay um you've clearly made one how many how many have you made of these things it's about a hundred a hundred wow okay different versions though right they're not still all operational like we're very quickly cycling through hardware revisions to ensure we actually get to the launch by end of year so we do quickly deprecate them as new ones come out sure what does launch mean you want to be able to sell these to companies this is business to business business consumer i've seen them on streams they seem like kind of just like companions uh it's less industrial but what what's your thesis for the early adopter here yeah let me just jump in straight into it So like, who am I?

57:19Who's the company? What are we doing? Why are we doing this? So, well, I'm Bernd. I'm the founder, CEO. Come to this from a robotics background. Been on this journey all my life. Picked apart all of my mom's kitchen appliances since they had motors when I was a kid. Always liked building stuff, programming. Started One X 10 years ago. And like the mission stayed pretty constant. We want to create an abundance of artificial labor. And we're doing this through creating these intelligent machines that can live and learn among us. And the last part is so important. I think what really differentiates OneX from the rest of the pack here is the focus on making robots that are safe among people and that are very affordable.

57:58But first principles engineering, how do you make something that's like super lightweight, doesn't need very special raw materials, doesn't need like that high tolerances in manufacturing, make something that's like closer to a refrigerator than a car, but still as capable as a human. and when i say getting this out the door what i mean is literally like getting it into your homes right so we're going through in-home testing among employees now i have one in my house in redwood it's amazing to get to kind of live with the product and it's going to be some customers having them quite soon but like under nda early testing and then if we play our cards right it should be in the main market by end of year.

58:42That's the goal. Wild. Moving quick. How has the company evolved since 10 years ago? Humanoids have always been in people's minds to some degree. They haven't been in the headlines as much as they've been, let's say, in the last couple of years. But kind of walk us through maybe those different stages, how your vision evolved. And then I have a bunch of other questions. Yeah. I mean, it's the kind of product that we all know will happen, right? And we've known it for like i don't know last 50 years like at some point this will happen and the impact is tremendous right so i think what's really changed uh the last couple years is uh to be very blunt that it works uh for the for the first eight years it was kind of a grind and it's a kind of a problem that's really worth just like doubling down on like grinding and pushing on until it works uh and i don't think the impact of this type of technology can be segregated right it's we're years, not decades now, away.

59:40It's very hard to put a specific date, but we're years, not decades away from robots building robots, robots building out our data centers, our energy infrastructure, our chip manufacturing, our mining. In general, it's this kind of hard takeoff moment, right? Where we can actually get to an abundance of artificial labor. And it doesn't really even matter if it's like physical or digital, right? because once you have this engine running, you will just in general have extremely intelligent agents, whether they're in the digital or physical space, and kind of will redefine what it means to be human.

1:00:18What do we value in humans? How do we see our own worth? And I'm just really freaking lucky to be alive, right? To be part of that. It's been my dream all my life. Yeah, the timing is amazing. What about getting into more specifics with Neo? do you have any sort of specific early use cases that you're focused on in all of this sort of sort of marketing that I've seen so far? It hasn't seemed focused on certain use cases like manufacturing or certainly not mining or things like that. It feels like you're maybe more, you know, focused on the on the home. Is that is that correct? Am I off? No, no, no, it's correct.

1:01:01But it's really important to point out that like, if you take a step back, right? It's like we're years, not decades away from actual abundance of artificial labor. So as a company, the only thing that makes sense here is to get there as quickly as possible. It's all we care about. What's the shortest path? So we're not a consumer company. We're the shortest path to artificial labor. And it just happens to be that the shortest path is through consumer. Because you need to live and learn among people. Like if it's one thing that's incredibly clear in AI, it is that the diversity or like the variance in your data set.

1:01:36basically equals your intelligence, right? That's where intelligence emerges from. And in a factory, there just isn't enough diversity. Think about it. If you're in a factory all your life, moving something from A to B, and that's all you're doing, then you could ask, okay, how much do you actually learn, right? And we actually have the answer you learn about. We need like 20 to 30 hours of data from that, and then we don't learn anything more. Interesting. So that's just like, that's not where to deploy robots. We want to deploy robots there, but we have to deploy them first where we actually get the data that gets you through a truly intelligent machine that doesn't just repeat emotion, it understands deeply the task it's doing.

1:02:12And this brings you into something interesting that you said in your intro here, which is companionship, right? Because everything we do is social. Work is social. Whenever you do something, usually it's because someone wants something. And usually someone's next to you or even communication, the social context of objects like hey is this cup on the table someone using it is it yours like is it is it half full half empty is it dirty like there's liquid in it but if it's coffee and it's cold it's probably like not in use anymore it's actually dirty like everything has this like social context and this is what creates this incredible variance and diversity in the home not only is every home different but like everything happening in the home is different every day and everything has a social context so i think we're seeing pretty strong proof now that like a lot of model intelligence comes from this diversity and we really want to double down on that and then the other part is i think it's just going to be incredible to live with these companions in our life that can not only give us back time in our everyday life with like household chores but actually be a trusted friend right and we're talking to our phones with chat bots and being like, these are our new companions, but it's not the right interface, right?

1:03:30The right interface is something that you can really deeply connect with, and it gets you away from a screen. Bicentennial Man, a movie I know you haven't seen, but it is a fantastic story of a humanoid robot that stays with the family for 200 years and evolves and changes. Robin Williams, it's a beautiful movie. I want to talk to you about kind of the progression of the software, there. Teleoperation, super valuable in a bunch of different ways. Simulation, we're seeing lots of promising data. There's the sim to real gap. There's deterministic algorithms for walk cycles. There's end-to-end systems.

1:04:11What is your view on the path that we need to go down to get to something that's truly generalizable so that you can put a humanoid robot in a completely novel situation, just talk to it and say, hey, clean up this table, and it will remove all of these cans one at a time. It knows exactly how to do it, and it's never actually been trained on that specific task, but it can generalize just like any other human can.

1:04:37Well, there's two interesting answers. The first one is no one knows how to solve that yet. And then the second answer is we actually know really well how to solve that, which is it's going to be some big transformer and you just need enough high quality data. So we kind of know like it's an engineering problem at this point, because even though we don't know the exact solution, we know what we need. Right. We just need an enormous amount of relevant data. And this data is very different. All models that you see today, they're trained on what I call like only observations. observations if you think about the internet where there's video or pictures or text or whatever um it's static right so you're like given that i saw or read this what's the next thing yeah what what you're actually missing and where in my opinion like a lot of like our ability to reason comes from is that you have some hypothesis about what's going to happen like you have your own mental model of like this conversation yeah so when you're saying something if i'm training just on what you're saying, that's not the same as if I'm training on what you were thinking and planning and then what you said and then what the result of that was, which was back to like how I reacted to that.

1:05:45And robots actually allow you to do this, right? Because you have what the robot is thinking and what it was planning to do, what it actually did and what it resulted in. And this data is what is needed to get to this. And that's how we're gathering at scale. There's no, yeah, there's no real like massive data set, like the web that you can just scrape. folks are using teleoperation to generate data there's simulation data like what are the richest sources of data for you uh what do you see as being like the real trove that you're going to be doubling down on over the next few years i think it's all of them but they're all just like a crutch it's like you're bootstrapping your way to where robots are learning in the real world yeah and this is really the direction we're moving in because teleoperation does not scale right it's super valuable and we need it but it doesn't scale um simulated data very useful for simple things like walking yeah but for like peeling a shrimp or whatever right it's like it's not nearly detailed enough like it doesn't work so um you just what you need is to use the internet data some teleoperation simulation synthetic data all of these things to get to where your model is able to sometimes succeed so when you say get me a coke in the fridge maybe there's a 50 chance that a robot comes back with a Coke.

1:07:02Maybe it gets stuck somewhere on the way and didn't manage to do so. But once this happens, then you can basically just tell it, hey, good job. Or like, hey, we're just supposed to get my Coke if it's just, I don't know, staring out the window instead. And that's the data loop that scales, right? And that's what we're deploying. So when we're saying like getting this into your home, what we mean is almost more like adopt a Neo to help us teach it. It's not going to do everything in your home day one. We're not there yet. but we're able to deliver a product that is really fun very engaging somewhat helpful and really the ability to be part of this generational journey from the beginning right that's the product and that's how we gather that data and this comes back to how we design robots that are safe because if you want the robots to actually be learning in the real world they need to be safe not just with respect to you but with respect to itself and your environment right you don't want your kitchen to look like a robot had been ravaging around there when it's done you want your kitchen to look nice just like it did before the robot went there even though the robot failed a few times the challenge is you don't have the luxury of of hallucinations long term like if the robot is unloading the dishwasher and puts a bunch of knives in the wrong cupboard you know that's like that's not safe right like you have you know kids at home and you know so it's like the the bar for acceptable is very high i want to ask about the supply chain dylan patel was on the show saying that no matter what happens with humanoid robotics, all the parts are going to be made in China.

1:08:32There were some actuator startups in the comments saying like, that's not true. We're going to build it here. We're going to fix the supply chain. What's your view on the short, medium and long term of the robotic supply chain? And you know how to ask the hard questions. Let's start by saying to us, this has been all about how do you don't over-constrained the system too early, right? So, juvenile robots don't exist at scale yet. So, if we were making a car, if my new startup was a car, I would clearly utilize the existing supply chain. Of course. What we've done is we've made everything ourselves.

1:09:14I know I say literally here. We're undergoing some patenting process for a new type of die casting of aluminium at the same time that we're launching AI models. So we're like definition of vertical integrator, including all the manufacturing technology. That puts us in a position where we can actually kind of manufacture anywhere. It just depends on access to raw materials, power, and general tariffs and everything else. We are actually building a factory in the US right now. there are challenges because like sourcing simple things in quotes like copper aluminium steel these kind of things in the u.s for example it's a lot more expensive than in china logistically it's also kind of like a nightmare because you don't have kind of like the economic zones with like basically your metal refineries it's next to your forge um and there are so there's like pros and cons but i think there's a path i think what's the most important here to realize is that when When people talk about things need to be built in China, we're not necessarily talking only about this.

1:10:12We're talking about actual access to knowledge about how to do manufacturing. And I think that is the thing that takes the longest to kind of onshore, right? So sure, there are things that need to happen with rare earths. There's things that need to happen with all these things. But I think the main thing to really work on also is just making sure that people understand how incredibly exciting manufacturing is and make sure that our new generation actually really wants to work on the hard problems in manufacturing. Yeah. I mean, we talked to the author of Apple in China and he really reframed the conversation about Apple's impact in China, not just as being a buyer, but actually being an educator in China and sending over massive teams of manufacturing designers to create a supply chain that didn't even exist but was enabled by the scale of the population and the economic zones and everything you articulated.

1:11:12So yeah, I'm optimistic that in a new boom, you could potentially set up trade deals such that if your business is scaling, you would have the choice to create a new generation of manufacturing engineers in America or anywhere else that might be advantageous. But this has been fantastic. I'd love to have you back and go way deeper. Jordi, you have a last question. Congrats on all the new launches this week. Excited to follow you and the whole team's work. It's been awesome to see. It was great talking to you guys and stay tuned. There's actually more coming. Amazing. Wow. There you go. And we'll pull up another chair for Neo whenever you're ready.

1:11:52We'll adopt. Happy to adopt one. Bring us another energy drink. Thank you so much for joining. We'll talk to you soon. Cheers, Bert. Bye. Have a good one. Thanks, John. Next up, we have John Doyle from Cape coming in the studio, talking to us about wireless networking, cellular technology. Very excited to talk to him, especially with the backdrop of what happened in Ukraine two weeks ago. Anyway, welcome to the studio, John. Good to have you here. How are you doing? Hey, guys. Happy to be here. I'm doing great. Thanks so much for joining. Would you mind kicking us off with a little bit of introduction of yourself and the company maybe to get us started?

1:12:27Yeah, you bet. I'm John Doyle, founder and CEO of CAPE. It is America's privacy first mobile network. Two sentences on my background because it explains how I came to start this company. I started my career in the Army Special Forces. I was a communication sergeant on a Special Forces group. Went to Palantir. I was there for nine years. And at Palantir, I started in technical roles. I wound up running the national security business for five years. And that's where I learned about this whole, you know, PhD field of study worth of vulnerabilities that exist on the global cellular network. And I'm really passionate about solving them.

1:13:06And so I started CAPE in 2022 to get after that problem. What is the biggest, like, concrete risk to the cellular network? Is it like... Well, I would just start at a higher level. Is it how unsecure is the average cellular network? Completely. And, you know, we saw, if you remember during the last campaign, the story came out about J.D. Vance's phone calls being listened to. Oh, yeah. And that kind of came and went as these stories do. But it snowballed into the story about Salt Typhoon, which the headline of that story is China has just explicitly and completely infiltrated all the major telcos in the United States.

1:13:49Wow. Which is pretty bad. But even before you get there, there are all these, they're really features that are baked into the protocols that run the network. They're there for good reason. But in the hands of malicious actors, some really bad effects can be drawn out of them. And so, you know, we can get deep in the technical weeds if you want to, but basically the whole thing is blown and it's problematic because we are entirely reliant and kind of fundamentally reliant on the cellular network. Yeah. What's a bigger risk right now, espionage or sabotage? That's a really good question. I'm going to say espionage because it is being like, it's absolutely happening right now.

1:14:30The Vance story is a perfect example of that, just listening to the future vice president's phone calls. Although the sabotage question is, interesting. I know you guys want to talk about or have been talking about SpiderWeb a bunch and the drone attacks. One of the really interesting features of that story, of course, from our perspective, is the fact that the drones were piloted on using commercial cellular connection, which is really interesting to us in a bunch of ways. But importantly, and it came out a bunch in the reporting, when people start to think about countermeasures and how would we guard against these attacks in the future, one option that gets taken off the table almost immediately is turning off the cellular network.

1:15:10Right. Because it has like disastrous consequences, right? Yeah. Because what if someone needs to call 911 and they can't, something like that, you're going to wind up hurting people. The technical impacts. Yeah, totally. When there's an AT &T outage, stuff grinds to the wall, right? Yeah. Yeah. I mean, we saw it yesterday with just like cloud flare outage and it was very disruptive. Talk to me about how you actually go about building a new network. are you able to solve the problems that you're trying to solve just by piggybacking on the existing physical hardware? Or is this a crazy CapEx problem that we're going to see you raising a trillion dollars in debt and stuff?

1:15:47You could probably do that based on your background, but I'm always interested to know where you think the right problem to solve is. Yeah, it's a great question. And we didn't know the answer to it when we started the company. Thankfully, it turns out you don't need to build the towers. You don't need to own the antennas or the spectrum. Got it. Our approach is we own all the software that runs the network. So we control application, billing, call routing, messaging, all of the important stuff. And we've done pretty extensive red teaming and testing, especially with government partners, and proven out that if you own all the software, you can actually make a lot of progress against the problem.

1:16:22Very cool. So when I see one of those cell phone towers, that might not be a Verizon tower specifically that might be owned by a different company that then is like leasing or running software for Verizon and AT &T within the same kind of like, you could think of that almost like a multi-tenant data center. Is that the right mental model when I see a cell phone tower? Yeah. And that's a great, it's a really good observation and it turns out to be true. The carriers, you know, the carriers, what the carriers own is Spectrum, right? What Verizon owns is an amazing amount of Spectrum that supports and they administer it.

1:16:55You're right. They don't even own the towers most of the time. Like American Tower or some other company builds the tower, oftentimes puts the radio on there, or the carriers will sign an agreement with Nokia or Ericsson to put the antennas up there. Sometimes the carriers don't actually run the radios. They also are sat to Nokia and Ericsson. And so they have sort of assumed this system integrator role. And I think my personal opinion is that that has serious security implications, has led to some of the problem that we find ourselves in. Sure. Can you get me up to speed on 5G? I remember the drama around Huawei and they had built 5G towers that we were not going to buy.

1:17:34I also heard someone mention offhand that America doesn't really have 5G. We have like fake 5G or something. Is it all marketing? Yeah, is it marketing? Oh, you're like, don't talk about it. Right. Yeah. Okay. So there's two things there. First, you're teeing up one of my favorite history lessons. 20 years or so ago, 23 years ago, China made the strategic decision to capture as much footprint on the global cellular network as possible. And the way they achieved that was by pumping billions of dollars of subsidy into Huawei. So Huawei could run around the world and sell antennas and also core software, the software that runs the network at half or less of the price of any of their competitors.

1:18:17And so it became really attractive and everyone went out and bought Huawei gear. And the result was that China now owns still not in the U.S. so much anymore, but around the world, you know, a huge percentage of the footprint on the network. At almost exactly the same time, the U.S. and it's not obviously a different kind of government. So it wasn't like a centrally made decision. But U.S. telecom started outsourcing and offshoring all of the technical aspects of running the business. So, you know, the course offer now is built by Nokia or Ericsson, which are in Europe. And very little of that capacity exists within the United States.

1:18:53So while China is making all these really deep and recurring investments in the infrastructure, the U.S. is really kind of divesting and just becoming these system integrator spectrum managers. And that's how we find ourselves in this imbalance now. And when people talk about the race for 5G and how is the U.S. positioned with respect to China, I think it's a little bit of a misnomer to say it's 5G because they're not talking about when 5G first started to get buzzy, the use cases were always like a heart surgeon in Boston is going to be doing surgery in Africa, right, because of the low latency protocols.

1:19:27And that stuff is cool, but that's not like how we're behind China. How we're really behind China is just in how much footprint do we own and how much access, how much control do we have over the network. And that's actually agnostic of which generation, or it's independent of what generation we're on. And then are we really on 5G is such a good question. There's like, I think about it mostly from a bandwidth perspective. Like I, I be, I became able to stream HD video at like retina resolution. And I really don't have a daily use case that's goes beyond that in terms of bandwidth needs. Right.

1:20:00And so as long as, but, but every once in a while I notice if I'm about to take off on a plane and I know I want to download a movie right before the difference between fast for downloading like a full gig in two minutes and 20 minutes is material. And so that's where I would think 5G would help. But I've also heard about like streaming video games, obviously 4k stuff. But I don't feel the pain that much that I'm not, you know, I'm not a single issue voter on the topic. But I am, I am curious. You're like, I mean, that observation is a threat to the entire like, like 3g to 4g was way better. Oh, totally.

1:20:35It was night and day. Yeah, better, more securely. And 4g to 5g was arguably better, although not entirely because like 5g connection drains the battery on your phone a lot faster interesting there's some trade-offs there where lte times that 4g is actually like a better option in some but beyond that like to hurry up and download a movie use case yeah there's kind of nothing left and so there's a big you know shrug in the industry people are kind of waving their hands at ai like well we're going to do ai and that'll you know that'll change everything but i don't know that that that that comes into it.

1:21:08Everything's an S curve. It's all sigmoid. We saw mobile bandwidth growing exponentially. Everyone was like, you're going to be able to download 10 exabytes in two seconds. And in fact, it was a sigmoid function like it always is. I think you've nailed the news case. It's like, as long as I can stream Netflix at maximum resolution, I'm done. I'm done. Who is the core customer that you're going after? Is it people in important positions, you know, politicians, CEOs, etc. It's just J.D. Vance. Just J.D. Vance. Just one massive client. One billion dollar contract. You should be able to get a one billion dollar ARR contract.

1:21:45No, but I think, but when I think about the broad consumer market, people say they want privacy, but in reality. A lot of people just want$20 instead of$30, right? Yeah. Yeah, I mean, both of those consumers exist in the market. There are a lot of people. Mint Mobile perfected the, like, how cheap can you offer a sell plan, right? Yep. like and they nailed it it's 20 20 a month turns out to be the the answer right ryan reynolds undefeated i love the guy when you have unlimited free marketing from ryan reynolds um so okay so yeah so who are like who are we trying to solve problems for i mean i like my background in special operations i am deeply passionate about um the fact that cellular phones have been a part of the battlefield for as long as we've had cellular phones and they are increasingly not decreasingly so over time.

1:22:33We saw this in Ukraine, like everyone, the entire Ukraine war is fought on the mobile network and like communications aspects of it. And I'm deeply passionate about building technology that keeps people doing that work safe. Right. I think that's like a really important mission and I love it. And that is where we started when we were thinking about the product and starting to like build early versions and build early features. That was the market. But we always had in mind and I in in my early life I was a special operation soldier my later life I'm a parent and like and I even in my sort of relatively mundane parental you know mid-40s existence I I have worries about how reliant I am on my cell phone and the trade-offs I'm making and things like sim swaps if you know what sim swaps are that's a huge problem and you need you have to build technical solutions to problems like that and I have that also like a general sense of indignation that if I want to be a subscriber to any traditional telco, I give away all of my personal information.

1:23:30And then they lose it like clockwork four to five times a year. They really do. They're giving it away. It doesn't have to be that way and it shouldn't be that way. Well, thank you for your service, both in the military and as a father. It's very important. We got to get you Ryan Gosling. I think, you know, Ryan Reynolds with MinMobile, we get you Ryan Gosling. There's something there. this is fantastic well now you're our you're a chief mobile correspondent now yeah um uh yeah i mean do you have anything else or should we let him go no this was great thank you so much seriously we i think i think there's going to be i'm sure there'll be more stories this year yeah yeah we'd love to have you back on some some deeper insight but we'll let you get back to it important work have a great rest of your day cheers we'll talk to you soon uh let's quickly tell you about wander find your happy place find your happy place book a wander with inspiring views hotel amenities dreamy beds top tier cleaning 24 7 yeah so good there we go top tier cleaning 24 7 concierge service it's a vacation home but better yesterday i got a little spicy on the timeline i said vacation i said traveling is overrated i wasn't talking about what are you running i love staycations i love getting to wander in my hometown in something i can try to load up the entire family load up the dogs drive out to a wander hit hit joshua tru marathons and travel marawans and travel there's no reason you need to go outside the country to enjoy a beautiful wander although is it possible is your ticket going go well that's a beautiful wander it is all over the u.s let's check in with our intern tyler cosgrove let's see how we doing well i'm i'm very close okay but this has been kind of a disappointment i'm extremely slow right now i mean i thought i'd been done 30 minutes are your hands aching are your fingers breaking i'm sweating oh sweating okay well you're an hour and a half but if you're sweating a lot you should try these apple decks apple flavor force electrolytes i didn't let's get him uh let's get him a let's get him a cup let's get him a cup uh get him some electrolytes make sure he's hydrated i didn't tell tyler this but i took a picture of the lego kit and i asked chat gpt how long will this take it said between two and four hours and then he was like i can do it in 30 and i was like okay good luck anyway we have our next guest in the studio emily sunberg the writer of the fantastic Feed Me newsletter.

1:25:46The daily newsletter about the spirit of enterprise. Really? Isn't that an incredible line? That's an incredible line. That feels like a line we would use. How are you doing, Emily? Welcome to the show. What's up, everyone? I'm happy to be here. Everyone, all two of you. Yes. Yes, all of us. We have our intern. We have our intern here. We're kind of like this. You know, Emily being a guest, I've gotten more random texts about it. People being like, oh, you should ask her this. Oh, you should ask her that. Oh, really? That's great. People are paying attention. Is that a Google shirt? exciting that's scary that's amazing well what what is your shirt let me see oh did you stand up i was at wwdc on um on monday yeah you were reporting on what the really old folks are thinking about apple right is that is that what they said yeah i saw a backlash they uh the people over at puck said that um that's really that apple should have invited younger people wow Wow.

1:26:39I'm a little too old. Yeah. Before we go to, is it true that Tim Cook begged for a selfie? Oh, yeah. We saw that. Yeah, totally. No, that was really fun. That was cool. I didn't expect that to happen. Yeah. What was your reaction? It feels like a lot of the liquid glass hate has blown over by now. People, you know, yeah, I think like sitting there in real time reading Twitter and seeing people's reactions uh it seemed like people weren't excited and wanted more apple intelligence updates but that night i went to have you guys ever been to wwdc no no never i went to one of those like ancillary um engineer and developer barbecues like an hour away in an airbnb with like 30 guys who who build apps and they were all really excited so i think they're the people who it's for and they're the people who are going to ultimately be using these updates and they seemed thrilled about it.

1:27:32Yeah, it was a developer conference and they actually announced a lot of exciting things around developers. And it also seemed like they seeded some territory in the AI world that could potentially enable more AI apps to be owned by independent developers, which is a huge, huge boon for the ecosystem. Did you get to press Tim Cook on the Photos app at all? Are you a Photos app enjoyer? I had about 14 seconds with him and Will Welch was before me. So it was like, not a lot of time yeah next year next year next year um awesome i i have a bunch of other random random questions to go through what i want to ask about the state of substack but before we get into that what is the state of the new york city uh election this mayor we've been kind of covering it our corner of the internet is uh not super excited about the potential of a socialist mayor what's actually happening on the ground yeah our friends about that that photoshop beard job that felt more just like like a printing error honestly to me it didn't seem like they did a ton of pre-processing ink dripping down i didn't see it as longer i saw it more as just like when you reduce the resolution so much like like the the the the gaps between the hairs kind of blur all together and it makes the beard look thicker i wasn't i wasn't fully like someone went in there i'm a little superstitious with the mayor stuff okay explain that oh i don't i don't want to talk oh scared okay okay you don't want to jinx it oh yeah yeah you're in your single endorsement could actually shift the election decided and then you would um i interviewed for them i interviewed cuomo i interviewed zoran lander and stringer um for the newsletter last week it was really fun oh cool cool what's next question let's move on from politics i want to talk about substack how you're seeing it evolve it feels like it just continues to have a moment uh you're dominating the platform at least that's what it looks like we're not super active there but what what is the state of things?

1:29:41Yeah, I've been writing on Substack for five years now. So I definitely had, I think I had a little bit of first mover advantage there. Like I know the platform so well. And I think even just understanding the tools and like what works well, the heat map of my own audience has been super helpful for me. And for anybody listening to this who doesn't know, like I write a newsletter every single day. So the opportunities for people to discover me are are a lot higher. But I think that's something that's interesting about Substack is that just because you have a large following somewhere else doesn't mean it's necessarily going to carry over to a newsletter.

1:30:20Like the inbox seems like such precious real estate to people and who they, who subscribers allow into that feels a little bit more intentional. So it's not like you're going to just follow a thousand people like you do on Instagram. So you have to be really fucking good to make it and then you have to sorry can i curse on this we don't but you can but but you're welcome if you yeah you have to be really excellent at at content and something that i do um i have one of the more active chats on sub stack like across the whole platform um and i'm in conversation with my readers constantly daily and that's a lot of work as well um so the state of things on substack i'd say it's getting pretty competitive you're seeing a lot of people jump over there and maybe try to make it but aren't necessarily making money yet um they also don't have an advertising integration like beehive or even twitter or youtube do so if you're gonna sell ads you're you're on your own for that or you have to get an agency um yeah what specifically do you want to know i'm curious i'm curious like specifically specifically like how much some of the new social features are driving growth is that like a primary driver at this point or or would you do you still think audio and video well like all of the new social features you know i just see like it's like a lot of people are growing based on like recommendations in like basically like cross promotions or collabs so yeah so for me like so yeah i was um i was in dc a few months ago and I met with somebody who who's invests in this space and he was asking me like why don't you consider like doing your own thing off platform and and building away from them and there's a few I used to think about that a lot more and I used to do that math a lot more and I showed him my analytics on my dashboard because you can see how many people are discovering you from um within the app and within their ecosystem and he was like they have you like you can't leave you're you're stuck there and it's not a bad gig like they take 10 of what i make which i think is totally fair considering how they probably drive more than 10 of growth right yeah yeah exactly they're probably driving consistently 10 growth and so that's totally the math's out just fine yeah yeah yeah when i when i when i think about sub stack i think of peak sub stack for me was like pre 2020 2020, like that era, when I think about when I was reading multiple substacks a day.

1:32:54And now Feed Me is the actually only newsletter that breaks into the important tab of my Gmail. And then there's a bunch of other substacks that are just kind of lost in the mix. I've just been trying to get a read on it. But if you are going to be dependent on a social platform or any type of media platform, Substack is a great one. It's like the friendliest one. Take your newsletter. Take your Stripe account if you ever need to. It's great. I mean, I own my subscriber list. I get to sell my own ads, which is going quite well. And when things break, they fix them. That's great. Talk about you had a fairly viral post.

1:33:31The tech guy in a hoodie stereotype is dead. Oh, yeah. You were at WWDC and seeing a bunch of different labels and brands being worn. Does that surprise you? Does it not surprise you? What was your reaction broadly? I think menswear as a whole, well, Will Welch was there. So clearly there's something, Will Welch is the editor of GQ. There's something happening, you know, just with his spirit in the air at the conference. People always think of iOS developers as being, you know, at the foremost of fashion. So you should always be there if you're running GQ. Yeah, I think I assumed more of them to look like the guys who work at the Apple store at my mall on Long Island or something.

1:34:16um and it was impressive i mean like guys looked good i wouldn't say capital s suiting but like fits um there were a lot of cool sneakers i think that a few years ago it would have been more new balance i think that like the steve worshipers are sort of aging out a little bit now there's these younger guys who probably are seeing that moment as like one of their big trips of the year and probably wanted to break out some fun outfits. Saw a lot of good sneakers. Saw a lot of big watches. Saw like a Dauntless on a guy. There we go. Did he have an Apple Watch on the other wrist? No. Or was he just fully committed?

1:34:54A lot of Apple Watches also. But a lot of nice watches. Make a lot of fun. Which like, I don't know. A lot of people don't wear nice watches in their cities anymore. Well, you can't in New York because it will be taken from you very quickly. Right. Exactly. But it's safe space. on the apple campus. Cupertino, yeah. At the donut. Do you ever get tempted to angel invest? I'm sure you get asked constantly. Do I get tempted to angel invest? I get asked, but that's not something that tempts me now. Why not? My business is kind of my main focus right now. Yeah, that's good. That's good. That's good.

1:35:30Are we post-peak celeb brand? Are we post-peak celeb brand? Yeah, are we on the down slope? I think so. Yeah, I think so. Yeah. So brands. What do you think the interaction between celebrities and consumer startups would be? I think everybody knows the formula too well now, which is that the celebrity isn't actually going into the office every day. And like that illusion has broken, right? So you can wear the suit on the stage at the conference and do the Forbes thing. but people are sort of understanding that they're not the operators and they're just sort of like a paid person to have their name on a brand.

1:36:12Right. Yeah. And I think the other thing happens all the time where like the founder becomes a celebrity after they start the company. Yep. Which is more interesting to me than like a famous person hiring a bunch of people to do their work for them and and getting paid for it. Yeah. Do you think we move back to... So I think of the celeb corporation collaboration eras is like you used to just pay a celebrity to appear in your Super Bowl ad and it would be like a million dollars in cash. Then we got to the point all the way on the other side of the spectrum where celebrities were commanding 30, 40, 50 % equity grants in businesses that were built entirely around their brands.

1:36:56To your point, people are catching on that that means that, yeah, they have a huge economic stakes they're going to be talking about all the time, but they're not really driving the product forward in any meaningful way. We could go back to just pay for a celebrity appearance in an ad, or we could move forward into something like what I've seen Saquon Barkley doing, where he is essentially just angel investing in these companies. And yes, he'll appear in the ads, but he's not saying he's building the company. He's very transparent about that. He will partner with them, but he's equity aligned, but just like any other investor.

1:37:32And so I'm wondering if there will be a boom in that kind of paradigm or that model, or if we'll just go back to companies and startups just being more comfortable writing a $400 ,000 check for a celebrity to appear in an ad. It's cash and you know what it is. yeah i don't i don't see the latter like being becoming a big thing i just feel like it's such a specific kind of company that has that kind of marketing budget and it's also really risky to invest in celebrities and influencers now because people are so volatile um i've definitely seen more of the angel investor in in like hot girl instagram bios like oh sure yeah angel investor and I think it's like a flex you know it's like I'm I'm also a businesswoman um so I think that's already happening like there's like a soft boom of that of of people getting the opportunity to do that and basically only having one canned cocktail in their in their house and not because they're necessarily getting paid but because they quietly invested or not so quietly invested and they're able to sort of flex a portfolio of investing and that's like something they do besides whatever other things they do.

1:38:45But there was that Alex Earl Poppy deal last year where she was drinking it a lot. And I guess her dad, who sounds like a pretty savvy business guy, negotiated for her to get equity in the brand. So that's sort of like a prime example of that where she wasn't going around saying that she was an angel investor, but I guess she was. yeah well yeah and poppy had had a stacked cap table of various personalities celebrities athletes etc also like private equity firms gross equity firms but yes um where uh give us a temp check on wellness right now are we did we are we past peak wellness in in many ways people are doing more wellnessy things but at the same time being unwell feels popular i did um i did like a chat with my readers the other day about alcohol consumption and people are all over the map with that there are people who are like i'm 35 i haven't had a drink in 10 years i'm perfectly happy life is great um and then there's definitely a decent cohort of oh somebody said um that people are who have been sober for years are kind of like strategically or strategically drinking now which was interesting to me we strategically drink we drink down pairing the king of wellness right now is glp ones like that's sort of the most exciting thing happening is that getting wrapped in different brands right now to appeal to different uh cohorts or yeah i just think it's changing people's chemical composition and changing the way that they think and work out and eat and socialize and that's very interesting to me with that what do you think that brands yeah do you think that brands are already being you know that the initial concern was that you know fast food brands were cooked because people just were going to eat less but then there's a downstream effect if if new york girlies are microdosing glp1s and then maybe they're like i'm not going to go to pilates today because you know i'm already you know happy with what i look you know whatever uh are you hearing any about any kind of like downstream business impact post glp1 um i don't think the pilates industrial complex is affected they're sort of the new weed store to me they're everywhere the pilates studios it's like people are buying these cheap made in china reformer machines setting them up and just like running crazy pilates businesses so i don't think that's going to be affected anytime soon um because even if you get really thin you still want to be shredded um you've got to be shredded but also with like fast food i'm I'm pretty sure people on GLP ones are still eating it.

1:41:26They're just not eating a lot of it. Like they're getting so fast and throwing it away. There was a celebrity that said the other day that he was just eating, eating through it. He was just basically powering through it. He was powering through it. I think that would be me. I'm like convinced that my hunger would overpower it. That's amazing. What about other downstream effects from GLP ones? I was talking to somebody who was like, they wanted to start a protein water company because they felt like with the GLP-1 boom, people would be eating less. They needed to supplement. Aside from the brands that have just wrapped GLP-1 in different landing pages, are there other kind of winners from the GLP-1 boom that you've seen?

1:42:07Adjacent. Somebody told me the other day that all these protein foods are not healing foods. And it changed the way that I think about people eating all this protein um any other trickle down things yeah i'm sure there will be some sort of like therapy blowback like people needing more therapy people women being anxious yeah what's the state of maha right now are people building new businesses around that or they're mostly just old line businesses they're beneficiaries of maha you guys follow ballerina farm oh yeah She just opened a little store wherever she lives in the Midwest. She feels very Maha aligned.

1:42:48Yeah. They like raw milk. They like farm stuff. She's an interesting business person though. Is any of that breaking through in New York? It feels like a lot of the actual products that are Maha aligned feel very New York aligned and yet the politics of the products could not be more disparate. You're right. It's so political. It's like there was a store in New York that was selling raw milk out of one fridge. It was like one of these. And it probably wasn't a right wing store, right? I mean, I don't know. But like, it didn't present as like, oh, yes, it wasn't an American flag. It presented as like rich women who buy$20 smoothies.

1:43:29So they could fall all over the place on the spectrum, honestly. But I don't think they sell it anymore because that's like bad for business if enough people call you out on it. um even like there's this guy at the farmer's market this amish guy who used to sell raw milk and he's not there anymore and i don't know where he went big big milk took him out yeah someone did brutal uh where where do you get uh what what's the the the most significant part of your scoop machine is it is it readers giving you tips is it walking around new york is it is it a combination one of your readers asked me to ask you that.

1:44:10I'm sure people, of course, they want to know where that is. I have some really good readers. That's the IP. That's the IP. Yeah, I have some really good readers who are super loyal with sending me scoops. And they could send them to New York Magazine or the New York Times or Puck or Semaphore or whatever. And they send them to me. And I'm super grateful for that. And we have a good thing going. I also have an anonymous tip line, which is, oh, I think people get a little confused about what falls on what I'm writing about. PR pitches, insane gossip. PR pitches aren't the bad. The bad stuff is like a politician walking out of a 3 p.m.

1:44:50screening of baby girl alone. Like, I don't, that's like, let them rock. That's not news. Yeah, that's pretty funny, though. And then I go out a lot. like i i go out four or five times a week i go to parties i go to dinners i meet people i meet readers um yeah so i it's like my life i feel sort of like a machine right now but it's working really well so i can't um get keep oiling the machine then well congratulations on building last question do uh do copycats bother you i'm sure you've had hundreds of people by now be like I'm going to make feed me for XYZ or feed me for that or just blatant?

1:45:29Do they get under your skin? They used to, but data is a very calming force in my life right now and especially in my business, and the numbers calm me down. Numbers don't lie. It's fantastic. There we go. We're hitting the size gong. Well, congratulations on everything, and thanks for stopping by. Thanks for coming on. You're always welcome. Take it easy. Yeah, we'd love to have you back and talk more. Cheers. Talk to you soon. up next we have Leonard Heim calling in from the RAND Corporation we're going to talk about AI models we got crazy range today it's a whirlwind tour wait let's check on Tyler Tyler how are you doing?

1:46:06I'm done the fury has been assembled I think my final time was somewhere around hour 15 so double hour 19 so yeah I mean pretty shameful I think it looks good I mean, yeah, send the electrolytes my way. Okay. I'll start chowing down. I think Dylan grabbed a cup and some water. If you guys want to grab it, let's get some electrolytes going up in here. It's a very beautiful model. It looks great. Yeah. It's really cool. Very cool. I'm glad we have that in the studio now. Well, enjoy your horse electrolytes. Hope you have a good rest of the stream. We will hop on with our next guest now. We'll check back in for a taste review.

1:46:44Yeah. Yeah. Yeah. But when we finish this next interview. Anyway, welcome to the studio, Leonard. How are you doing today? Let's bring in Lennart and chat about artificial intelligence and geopolitics. How are you doing? Hey, good to see you. I'm doing great. How are you guys? We're great. Would you mind kicking us off with a little introduction on yourself and what you do day to day? Sure. Yeah. Lennart, I'm calling in from Washington, D.C. I'm a, what do I call myself, researcher working on AI policy, a governance background, electric engineering. I think part of my mission is to bridge the technical world and the policy world.

1:47:18and that's what I've been doing before AI was cool, the last four or five years, actually. And now AI is cool, and since then I've been ever busy. It is extremely cool. It's driving the news cycle every single day. Yeah, pretty much the coolest technology. Do a little level setting for us. What is the dominant story? How is the race between the US and China shaping up on the AI front? What are the interesting threads to pull on? What are the most important companies to be tracking right now? Yeah, well, that's a tough question. Where to start? I think it's like still the deep-seek trauma. It's like still there, still dominant, right?

1:47:55I think that was like this initial story, the US is like so far ahead. There is no Chinese model being closed. And then deep-seek came out and definitely had a big impact in DC, but I guess also in a broader world, as we saw in the stock market. And I think it's pretty dominant here still in DC. You know, like in DC, you always talk about things which Silicon Valley did half a year ago, even a year ago. So that's still part of the dominant discussion. And then export controls, I guess, is also a big topic right now, right? You've got the US-China trade talks going on right now where they're negotiating beyond export controls, which impact AI, but export controls famously had a big impact on AI in China.

1:48:30Yeah. So let's talk about DeepSeq. One of the narratives that came out of the DeepSeq story was all the American labs are completely cooked. and then we got Jevons Paradox and NVIDIA popped back up and all the labs seem to be making tons of money. OpenAI crossed$10 billion in revenue run rate and things seem to be going well. But what are the wrong lessons from DeepSeek? What are people getting wrong about the DeepSeek story? How relevant is DeepSeek? We're still hearing about it as being important geopolitically in some of these jump ball nations that aren't quite allies, aren't quite rivals, and they might be interested in building an open source stack on top of Huawei Ascend, DeepSeek, Manus, that type of stack.

1:49:18What are the wrong lessons that people are learning from the DeepSeek story? I think DeepSeek made such a big noise because we're just lacking a good reference class. They just came out, pretty impressive model. It was openly available, and we had a paper attached to it, right? And then it came out also saying it cost us six million. Does anyone of us know how much it costs OpenAI or like Anthropic or any of these other companies what to do? We don't, right? And I think this just really made headline news. We like sometimes get like Sam Altman going on stage saying, I think at some point he said it costed them a hundred million to do GPT-4.

1:49:49But a hundred million what? Just a training compute? Is it just like buying the hardware? Probably not. Is it paying the staff engineers? So like the reference class is like pretty important here. Well then it costed them a hundred million what? like there's not two years ago even more yeah and now deep six six million so i think people just like we're lacking it's like wow this sounds impressive but like they had just no reference class of what's actually going on within the companies and i think a couple of weeks later and frobix ceo dario put out a blog post saying like actually we can train to the same efficiency and i think this was like for people monitoring the market not surprising we just like know the trend lines i mean no it gets like vastly cheaper every single goddamn day to train such a system So to that extent, it's not surprising, but you look at this, you don't have a reference class, and then it just looks really impressive.

1:50:34Yeah, I think a lot of the more technical analysts in the AI world were excited to see the open source results, excited to hear about FPA training, for example. We talked to Dylan Patel. He was telling us, oh, well, OpenAI was doing FPA training years ago. They just didn't tell anyone because it was kind of like their edge, I guess. But all the labs had kind of figured that stuff out. They just hadn't told everyone exactly how they did. DeepSea came out there and told everyone and it sounded like they really had jumped ahead. Instead, it seemed like they just jumped to the frontier, had not fully advanced beyond the frontier necessarily, but then done it in an open source way.

1:51:13What was the read on the strategy to open source? It felt like there was a very deliberate, like that launch, if it had been closed source, $20 a month, you know, very difficult to access, rate limited, lots of safety harness around it, that we would have been telling a very different story about DeepSeek if they had said, oh, this cost a trillion dollars, it took us forever. Instead, it seemed like there were like a number of superlative and viral hooks that really captured the American news media and drove a big cycle. How much of that was intentional versus how much of that was just like a random byproduct?

1:51:52Yeah, I think as you're saying, there were just like so many checkboxes that ticked. It was during the first week of Trump. It was an open release model, right? It was surprisingly so cheap. It's coming out of China. Oh, and it was like the first public reasoning model. Yeah. So just like all of the things happening at the same time. So like wherever you're coming from, DC excited because it's out of China, but we thought we fixed it, we did export controls, right? Silicon Valley, oh, it's an open source model, right? And everybody looking at like, actually, how does this reasoning work? Because OpenAI did it before, but it didn't tell us, right?

1:52:25So again, just ticking all of these boxes, how much of this was deliberate? I don't think this is like a state coordinated effort that they said, like, this is how we're going to do it. I think DeepSeq had like a pretty open commitment to like releasing models publicly before, right? So they've just been following that. It was just like to some degree, if there's a marketing strategy behind it that did it right, I just expect most of the things are just like how the tech works, right? And they were early on in reasoning. They were like one of the first companies doing it publicly. And then it just made a ton of noise.

1:52:56And again, justifying so. Even though ChatGPT had a reasoning model available, it was only available at a premium tier. And so for a lot of AI consumers, I think their first interaction with a reasoning model was DeepSeek or the DeepSeek app, which was just a very interesting kind of like go to market strategy, essentially. And also seeing the chain of thought, right? Yeah. Like the reasoning, which we didn't see before, right? And even the cost thing. It's like this changed a ton. Yeah, even a cost thing. A lot of financial folks, I mean, you mentioned that Silicon Valley was excited about open source.

1:53:29Of course, Washington was interested in the story because of the China angle, but also the financiers, the Nvidia investors were interested in the financial impact of this. So yeah, it was fascinating. I got it wrong, right? Yeah, yeah, for sure. That's at least my take. How much of the existence of DeepSeq do you ascribe to China's positioning around high frequency trading? Oh, that's an interesting one. I mean, like DeepSeq fits in the hedge fund, right? Yes. And they were initially buying 10 ,000 A100s or 5 ,000 A100s in 2022, I think, before the export controls. Yes. And tons of GPUs always still go to hedge funds because they, well, we don't exactly know what they're doing, but they definitely like accelerated computing.

1:54:10For sure. And partially machine, yeah, machine learning, a bunch of other stuff. Are they all training large language models? Probably not. Nope. And I think then one theory I've heard, I'm not sure how true it is, there was like the crackdown in China on the tech sector, right? Including high-frequency trading, hedge funds, and more. this made them then pivot to AI it's like well if we can't trade on the market anymore what should we do with all of these GPUs what turns out AI is this new hot cool thing and then I think the other thing which is important here when you have these hedge funds they do like really low level optimizations like writing on kernels they don't do like just high level PyTorch they go down there and try to get like every single point of utilization out of the models, out of the hardware and this helped them right you know like DeepSync has pretty correct engineers I think this was like pretty obvious once we saw like the code and just then help them to then train a pretty decent good model.

1:54:59Let's talk about Llama. The Llama 4 launch was a little rocky. Obviously, there was allegations of some benchmark, just over-optimization. It seems like a lot of the big frontier labs or the independent labs kind of just don't even care about benchmarks anymore. It's all about big model smell and just the actual performance of the products. At the same time, I've been really interested in this narrative of LAMA as a very important American export, as a counter to the Huawei Ascend, DeepSeq, Manus stack that other countries might be comping? Because yes, they might be interested in doing a deal with Stargate and buying a whole bunch of NVIDIA chips, but what are they going to put on top of it?

1:55:44Maybe they don't want to go all the way with OpenAI or Anthropic. LAMA seems like an interesting wedge there. How important is that? What is your reaction to the Meta-Llama strategy and how they're being perceived in Washington right now? I would be curious on your guys' takes on this. I'm still unsure how I feel about this. Let's just start with DeepSeek. I think it's an interesting moment in time when all of the leading US models are just behind closed doors and the best open models coming out of China. I'm worried about it because this model is partially biased and it lies about the CCP. It has some Chinese propaganda in it.

1:56:18Why is this bad? Imagine you're like some developer in another country. You want to build a new education app. You're using this model and then you teach your kids Chinese propaganda. I think that's great. I think it's even worse if you think about sleeper agents. people know about this paper from a topic, right? You can basically have like sleep age and sitting this model when you talk about a certain topic, it starts giving you a cold with vulnerabilities, but it can apply to anything else. And it's like really hard to look for it because you don't know what to look for. All right. So it is worrying.

1:56:47I don't think that's the case of DeepSeek right now, but going forward in the future, the case is just like, if you use an open model, you don't really know what's in there, what's hiding in there, right? This goes for all companies, but I think you've got more reason to mistrust certain other companies than, for example, Meta. Yeah, it kind of makes me think that Anthropic might have a real business on their hands with AI safety. Obviously, they branded it all around, you know, oh, like, you know, we want to prevent AGI doom and fast takeoffs. But if you just think about like, hey, we're going to launch this model in our organization, and we want to make sure that the code that it writes doesn't have random vulnerabilities, that feels like a very important AI safety product almost to evaluate these models, test them, and essentially root out sleeper agents.

1:57:35And so I've all of a sudden flipped to very excited about the work that Anthropik's doing on that front. Jordi, do you have any questions? I think this could be a real benefit to them. Yeah, totally. And that's kind of how it started. Forced learning from human feedback gave chat to be the rise. It was like some people trying to make the models more useful, but also less harmful. and turns out to be a bunch more useful. And I guess we will see more about this going forward, and particularly for code, or even if government wants to adopt these models. You just better be goddamn sure. My question was around news from the last 24 hours, basically.

1:58:08And specifically, people talk about the application of AI in warfare, but it's always at this sort of very generalized, high level of like, oh, we're using AI. And I can easily imagine, you know, simulation scenarios, data collection, processing, all that stuff, like kind of going up to the point of a conflict and then autonomous, you know, drones and things like that. But what are the ways in which you've understood AI to be used in an actual like conflict scenario? Sure. I mean, look, I don't do narrow AI or drones. To some extent, I expect there's like a bunch of just like computer vision applications and more.

1:58:49But if you look at like frontier, like if you look at large language models, I think one big application is just intelligence, right? Literally my job on a day-to-day basis is like open source intelligence. I want to know what's going on around the world. Nobody's telling me, well, except DeepSeek, they sometimes publish a paper. But like, I don't know what the US companies are doing. I don't know what's going on in China and the semiconductor industry. So just a bunch of data, I feed into my LLMs and ask them what's going on, right? And for intelligence operations, it's the same. We saw in the last 24 hours pretty targeted attacks.

1:59:17Israel is famously known for pretty good intelligence operations and I think they exactly knew who they were hitting based on these types of operations right and if you just have like a nice vector on a smartphone and you just like feed all of the chat transcript and an LLM can make sense out of this right you can do it at a larger way bigger scale to identify the targets and also identify patterns which again might be harder for humans to do at least you can like augment humans Yeah, I mean, they used to do that with like keyword search. Okay, like let's, we have a ton of data. Who's talking about bombs?

1:59:49Who's talking about attacks? And people would use code words and LLMs are really good at deciphering that type of stuff. Fascinating. Give us a state of the union on export controls for AI chips. What is, what's expected on the horizon? How has the landscape changed over the last, you know, couple months at least? Great. I mean, when Biden left the office, there was like two big rules coming out. The one rule was, I think many people are tracking, the foundry to diligence rule. It was a big hiccup where Huawei produced a bunch of chips over TSMC in Taiwan, which they were not supposed to. They did it via a shell company.

2:00:27Three million chips, that's quite a lot. That's way more than they produce. That's what they got their hands on. And the government reacted with a new rule basically telling TSMC, like, goddamn, you've got to do better to diligence and check who your customer is. and make sure they don't end up with Huawei. Because they broke two rules. No chips for Huawei at all. And please, no big AI chips for any entity in China. So this was a big move. That's still the case. I applaud that. I think that was a pretty sensible move to just make sure they can't, again, get their access to that many chips. And the other one was the diffusion framework, which I guess you probably discussed at some point before, right?

2:01:03Like broader controls to some degree on all of the world, deciding to some degree dealing with the Gulf states, dealing with a bunch of in-between countries where people were worried about chip smuggling, but also data centers being built there. Malaysia is being such a country with a bunch of data centers being built there. Really interesting reporting today where I think the Wall Street Journal reported people were like going back and forth on an airplane with hard drives to train models over Malaysia. I saw that. We were going to read that. Yeah. Wow. Which again, speaks to the efforts they're going to to get access to AI chips and how creative they get.

2:01:36And also from my point of view, I'm just getting surprised. every single time of how creative they get to get access to AI chips. Anyways, that's what I did. But this one is not being enforced right now. It's not officially revoked yet. We're still waiting for it to officially be gone. But at least Commerce, Lutnik, and Kessler put out a statement saying they're not enforcing it. They will come up with a replacement rule. And this replacement rule then needs to deal with, well, yeah, basically all the countries except the clearly competitors, China, Russia, North Korea, they're all controlled.

2:02:04They will continue being controlled. But what about all the other countries? like the Gulf states, like countries with a bunch of smuggling going on. And yeah, me as somebody from outside the government is waiting to see more moves there. Yep. I have one last question. Good, Jordy. Today, a bunch of technologists joined the Army Reserve in the newly formed Detachment 201 Executive Innovation Corps. Can you tell me a little bit about how you're seeing technologists and the tech community plug in in D.C.? What is needed there? What do you want to see on the horizon? This seems like a step in the right direction, but it feels like a job's not finished moment.

2:02:44Yeah, I think that's the right audience, like a call to all the techies to come to DC. The pay is not better here. It's also pretty damn hard with living on a swamp here. But I think there's a strong case for impact. So as a techie, I think like when I talk to like young undergrads like, oh, in DC, like, I just did like a bachelor's in machine learning. What do I know? It's like, it's plausible you're the smartest person in the room. and many times in DC, which is way harder than in San Francisco. People don't know policy, but sometimes can be a benefit to be a bit ignorant about how the game is being played.

2:03:16Just like, look, here's just a math. Here's how it works out. Right? Yeah. So I love techies coming to government, just crunching numbers. I think they should sometimes be less opinionated. Just like, look, here are the numbers. We crunch them. Could suggest this, could suggest that. But just grounding it like an analysis and objective things. When I think about China and AI chips, I can crunch the numbers on how fast the chips are, how many they're producing, and in contrast, how fast American chips are and how many we're producing. And this is just a good analysis to be run. How this then impacts the broader scheme of the AI ecosystem and who's winning the AI race and what?

2:03:51It's a different question, but we can then take it from there. right so if i see stuff like this be it in again in the military but also just generally in government particularly anything around export controls anything around just understanding of what the hell is going on just like having somebody explain like hey deep seek is a reasoning model what is everybody talked about distillation like really just explaining the basics is like yeah things people should do and i think people with a cs like an undergrad degree in computer science they can do it and i would love to have them there yeah i'm hiring by the way so if anybody's listening, please join me, apply.

2:04:26Yeah, in general, I think it's good that we don't just have, you know, these sort of high-profile technologists only joining Doge. There's other things that are important in the government besides just, you know, taking a sledgehammer or scalpel to different parts of the government. So it's positive. I also discovered getting stuff wrong. I think the most famous example is the initial Xbox controls 2022 were trying to control AI chips. Yeah. And they used two parameters, technical parameters, which probably 99 % of DC won't understand. total processing performance, how many flops it got, and then also the interconnect band with how fast they can talk to our chips.

2:04:59And they messed something up there. Then NVIDIA could basically design a new chip sitting right at the threshold but still having high flops, which is basically as good as the other chips. And that's what DeepSync used. And then it took them a damn year to fix it. And I can confidently say they knew a month later they should do better there. And that's just a thing where like techies and you didn't need to be the smartest person. I've never trained a really big model. You can just tell by the specs, like guys, something's off here. and can make a tremendous difference. Or even just asking one of the foundation model labs, like OpenAI, I'm sure they could tell you, like, hey, do not restrict these three parameters.

2:05:32And it just seems like there was not enough... If they were to answer. NVIDIA doesn't answer these questions. Yeah, NVIDIA wouldn't, but I mean, you could imagine that plenty of the labs that will compete with Chinese labs would love to answer that question. Because they have a huge incentive not to be out-competed by DeepSeq, right? Yep. But maybe they'd be talking their own book too much. There's always dynamics, but at least having a technologist in the room would be beneficial. Anyway, this was fantastic. Thank you so much for hopping on. We'd love to have you back for more news. We will talk to you soon.

2:06:00Enjoy the rest of your Friday. Have a great weekend. Cheers. You too guys. Bye. Closing out the news for the week. We have, let's take it over to Tyler. Oh yeah, let's take it over to Tyler. How are the horse electrolytes? Okay, so I haven't tried it yet. Scoop. I think, so it says a scoop. One scoop is for light exercise. Okay. So for, I think, probably a medium-sized horse, I assume. Okay. So I'm looking at the ingredients. There's an incredible amount. It's like all salt. It's all salt. Of course. Of course. You have to at least chat GPT that you're not taking a deadly dose of salt right now.

2:06:35I think it's probably fine. Probably fine. We'll go with probably fine. That's the horse mentality. It smells good. Does it have like 11 lime taste to it? Is it apple? Oh. It's strong. It's like I'm drinking salt water. Yeah. Amazing. I think you might have to water. There's lemon in there. There's some, there's definitely some flavor. I mean, it says apple flavor. I'm not really getting any apple notes. Yes. Okay. Well, congratulations. You did it in under an hour and a half. You did it. Well, you have a whole summer to finish the whole bucket. Yeah. Wait, what was the final time? An hour and 15 minutes?

2:07:09I think hour 19. Hour 19. That's still way faster than ChatGPT clocked it. I took a picture of that Anduril Lego set build time. I said, how long do you think it will take to put together this pretty simple 619-piece Lego set? I said a beginner would take three to four hours, an intermediate builder two to three hours, and an experienced Lego fan 1.5 to two hours. You did it better than they did. So I think you were only wrong in your own. You overestimated yourself. Which you should always do. But you outperformed the average Lego builder. So congrats for that. And enjoy the horse life. I'm sweating a lot.

2:07:44I think this will help me, you know, bring me back to. Yeah, you're sweating out the electrolytes. You need to refresh. This makes a lot of sense. This is great. Everybody in tech is always talking about going to the horse doctor, but never talking about getting horse electrolytes over the counter. Yes, yes, the horse electrolytes are key. That's what you want to be taking. Well, the scale deal has officially closed. Met is paying$14.3 billion to buy 49 % of scale. the industry's leading data dealer. Alex Wang wrote a note to Scale employees. He said, when I founded Scale in 2016, it was amidst some of the early AI breakthroughs.

2:08:20DeepMind had just released AlphaGo and Google had just released TensorFlow. It was still incredibly early. It was clear even then that data was the lifeblood of AI systems. And that was the inspiration behind starting Scale. Since then, the journey has been extraordinary. We've grown to over 1 ,500 people, become a trusted partner for model builders, enterprises, and governments, building and deploying the smartest AI tools and applications. Scale is now one of the most impactful companies in the world. He closes by saying, today's investment also allows us to give back in recognition of your hard work and dedication to scale over the past several years.

2:08:54The proceeds from Meta's investment will be distributed to those of you who are shareholders, invested equity holders, while maintaining the opportunity to continue participating in our future growth as ongoing equity holders. The exceptional team here has been key to our success. So I'm thrilled to be able to return the favor with this meaningful liquidity distribution. So congratulations to everyone. Let's give it up. Hit the gong for liquidity events, John. Lee Marie, Eric Torenberg. Lots of winners. Lots of winners. Lots of friends of the show. So Ben Thompson broke it down in a piece talking about Meta and Scale AI.

2:09:34The most obvious explanation for the structure is that Meta wants to avoid the sort of antitrust scrutiny that would attend to an acquisition. This explanation applies broadly. Big tech generally and Meta specifically are under massive scrutiny in terms of acquisitions, including Meta going to trial for acquisitions made over a decade ago. And narrowly, Scale AI is a supplier for not just Meta, but also Meta's competitors. The problem is that just Meta isn't acquiring Scali doesn't mean the deal can't be scrutinized. Indeed, Section 7 of the Clayton Act is explicit about covering only partial acquisitions.

2:10:07So the FTC can still review this, but it'll probably be a little bit easier. And so that's what they're gunning for. These guidelines were reaffirmed by the Trump administration earlier this year. So a lot of people were expecting it to be complete game on. Anyone can acquire anything. That hasn't been the case. We heard about this on the campaign trail. A lot of folks in the Trump administration were signaling that Lena Kahn made some good points And they were not going to completely reverse course, even if they were going to replace her Even beyond that says Ben Thompson, however, and and perhaps this makes the regulators point It seems likely that this will kill scale scale AI's business with the big labs in particular Who would be concerned about their data ending up in the hands of meta?

2:10:48which is to say that this is a deal that would destroy the enterprise value that Meta is theoretically investing in. And so that will be something that will be debated in these merger guidelines when the FTC ultimately reviews it. But it's looking pretty good because they close pretty quickly. As it is, many of the labs, particularly OpenAI, have been bringing in more of their data work in-house, which helped contribute to scale AI missing revenue targets. Last year from the information they reported this, I mean, they still put up amazing numbers. And so obviously a lot of stuff was working, but you could imagine that some of the labs were starting to bring these.

2:11:22If they really are believers in, hey, we got to own our data collection and data processing forever, let's build that function in-house. And I think they started to do that or at least partner with other competitors in the space. And so Ben Thompson continues to write about Meta's reset. In fact, the more pertinent angle to discuss this deal is probably Llama. Llama 4 was widely viewed as a disappointing model, and a big portion of the original Llama team has since left Meta. And Ben Thompson thought that Mark Zuckerberg's media tour about AI seemed a bit forced and unfocused, and had heard through the grapevine that Zuckerberg was considering a wholesale reset of the company's AI efforts, with the biggest priority being the search for a new AI chief to take firmer control over the company effort.

2:12:03That, in the end, may be the Occam's razor explanation for the deal. This is a very expensive acquihire Alexander Wang, Scale AI's co-founder and CEO with the price softened a bit by virtue of paying Scale AI for work that Meta was going to have the company do anyway. Wang isn't a researcher, but he's an executive and leader who is familiar with the space and Meta needs leadership in addition to talent. He's a deals guy. He's a deals guy. He's going to go into Meta and continue to be deal-making, which will be, I think, a great asset to Meta overall. And so Ben Thompson closes with a little bit on sustaining versus disruptive innovation.

2:12:38The other reason to believe meta versus Google comes down to the difference between disruptive and sustaining innovations. The late professor Clayton Christensen described the difference, and we're familiar with that. So the question of whether generative AI is sustaining or disruptive innovation for Google remains uncertain two years after Ben Thompson raised the question. Obviously, Google has tremendous AI capabilities, both in terms of infrastructure and research, and generative AI is a sustaining innovation for its display advertising business and its cloud business. At the same time, the long-term questions around search monetization remain as pertinent as ever.

2:13:11And this is the question I want to debate with Ben when he comes on the show. There is this idea. So it came out earlier that Sunder Pichai had not read The Innovator's Dilemma. And people were saying, oh, like he should have because this is the classic example of Google being disrupted by a new technology, generative AI. Ben Thompson said, it doesn't matter. The whole point of disruptive innovation is that it's structural. And therefore, it doesn't matter if you've read the book. There's nothing you can do about it. That's kind of the point of disruptive innovation. You're being disrupted. But my question is, what if Google had chatbots before ChatGPT?

2:13:44What if they had launched the Gemini app first and been the first to market and gained all of that memetic attention? Well, the disruptive innovation framework would say that Google would be in a tough spot because they wouldn't be able to monetize Gemini as quickly as they were losing volume in search. And so search revenues would decline faster than they could make it back up on the new model. And so what we're seeing right now is Jatchypt is growing their 10 billion in revenue, but I don't think we're seeing a fall off in search. It seems additive. And so the weirdest thing is that if you ran back the simulation and you did put Google in a position, like let's just say Google owned 100 % of Jatchypt, like the combined enterprise value would not be, the stock would not be in the trash as they disrupted themselves.

2:14:33And so there's this interesting question of like, there's this fear that your earnings will drop as you pivot away from search-based display advertising, search advertising to a subscription chatbot model. But I don't know if that's actually playing out because the overall market, the combined revenues of Google and OpenAI seem to be not declining. Like we're not in this nadir. We're not in this trough of like disruption necessarily. Yeah, Meta and Facebook had the luxury of acquiring Instagram, starting to monetize Instagram, still being able to grow Facebook at that time. I mean, maybe there's a little bit in deceleration.

2:15:11I remember when Meta was in the process of somewhat being disrupted by TikTok. They launched Reels, and there was questions about Reels monetization. because if people spend more time on reels than the normal Instagram feed and there aren't as many ads in reels as there are in the feed, well then, even though the user time might be going up, your revenue might be going down. And so that was a fear. That was something that was weighing on the stock. Ultimately, it was not borne out. But I'm just not seeing the data, but I want to dig into it more. I want to talk to Ben about it because there's clearly, I think I'm wrong, but for reasons I'm unaware of.

2:15:47And so I've been digging into that. So Ben goes on to write meta, however, does not have a search business to potentially disrupt and a whole host of ways to leverage generative AI across this business. This is what I was talking about with Joe Weisenthal talking about even if meta doesn't get like a banger AI first consumer product out into the market. They just have so many AI generative AI workloads, whether that's sorting ads, generating ads, internal profanity filtering, you know, all their different content moderation. I think also, I think it's easy to see how Meta will benefit from more, in 10 years from now, more content will be produced, which means better, the more good content will be produced, even if it looks like AI slop right now.

2:16:35And when you think about the business holistically and Meta as effectively, you know, an entertainment company, will people want more and more entertainment? This is what Ben Thompson was writing about. He said, meta strategy is to commoditize your compliments. A generative AI is a compliment to Meta's platforms. And so, yes, if people are using generative AI to generate slop, that's still fine as long as people are enjoying it. But ultimately, people will use it as a different tool for video editing and sound generation and drop out the background noise and replace my background with a cinematic Vista or something like that.

2:17:15So for Zuckerberg and company, Ben Thompson thinks AI is absolutely a sustaining technology, which is why it ultimately makes sense to spend whatever is necessary to get the company moving in the right direction. It's also worth noting that Meta doesn't really have any alternatives other than continuing to invest. Google is a competitor for advertising the most financially compelling use case today. OpenAI is a competitor for consumer attention, Meta's most important scarce resource. I suppose Anthropic would be a potential partner, but per the point above, that seems like the ultimate culture culture clash if anything the most compatible partners for meta would be microsoft and apple but the former is obviously tied up with open ai and the latter well never say never forever but definitely never for now great ending to ben thompson's update that's right well we have a video that we can pull up uh that was that hit the timeline earlier this morning.

2:18:06Let's review it. I was able to confront PT. Oh, yeah. And I think we should play it. You saw him on the street? Let's play it. I saw him on the street. Mr. Teal, is it true that competition is for losers? Sir, is it true that you encourage young people to drop out of university? The background is really true. Mr. Teal, is it true that your funds have double-digit GP commits? sir how can you say you're contrarian when everyone is contrarian mr teal is it possible for there to be too much venture capital never mr teal have you been a victim of other vcs gaming the midas list absolutely sir why do you refuse to pick a favorite renee's gerard book mr teal what do you have to say to people that think they're a lottery ticket Mr.

2:19:00Teal, must all venture returns be power law distributed? Sir, why do you think why questions are overdetermined? Oh, man. Well, really asking the pressing important questions, the questions that he clearly doesn't want to answer. But I'm sure over time we'll have him on the show and we'll try to pull some of those answers out. That's a fun... I mean, the last story we should cover before we drop off. Chinese AI companies dodge U.S. chip curbs by flying suitcases of hard drives abroad. Leonard on the show mentioned this. Engineers are carrying data to countries where NVIDIA chips are available.

2:19:41Frustrating Washington's aims. No one really anticipated this, but apparently Chinese engineers are transporting hard drives with AI training data to Malaysia. So they take all their training data, they load it onto a bunch of hard drives, throw and suitcases, fly to Malaysia, and then do the training run there, and then take the weights back. And so that's like, the chips aren't even moving. The whole idea of exporting chips is now irrelevant if you can do the run locally in a different country. So this AI diffusion story is going to become so much more complex. It's just going to be so complicated forever because there's going to be shell companies.

2:20:16And if people are willing to do this, anything's on the table. The next Deep Secret Run is going to be a movie. It's going to be a movie. Anyway, we will be following it here. We hope you have a fantastic weekend, and we will see you on Monday. I cannot wait, John. Cannot wait for Monday. I cannot wait to be back at this table. Two more sleeps. Two more sleeps. Two more sleeps. Or is it three? We're not going live on Sunday. We might. We might. We might. If there's an emergency, we might go live. You never know. But until then, enjoy. Thank you for tuning in, folks. the rest of your weekend. Have a great rest of your Friday, and we'll see you Monday.

2:20:52We'll see you Monday.