Sam Altman LIVE on Sora, Hollywood, & the Future of Ads | Bill Peebles, Dylan Patel, Elad Gil, Robby Stein, Morgan Housel, Misha Laskin

28:04There's an article in the New York Times today on a Singaporean firm called Megaspeed. The CEO has socialized with NVIDIA's Jensen Huang. Now the company is being scrutinized by U.S. officials for its ties to China. So on a humid June night last year, Jensen Huang, the chief executive of NVIDIA, held court with several of his company's major Asian customers at a bar with sweeping views of Taipei. They stood and toasted the booming artificial intelligence industry. Next to NVIDIA chief was a woman named Huang Lei, or Alice Huang, an executive of Singapore-based data center company called Megaspeed, great name, which was poised to buy two billion of NVIDIA chips.

28:47Though Ms. Huang and Megaspeed are little-known players in the AI industry, their association with NVIDIA and its CEO has recently become a preoccupation in Washington. Commerce Department officials have been investigating whether Megaspeed, which has close ties to Chinese tech firms, is helping companies in China sidestep American export restrictions, according to more than a half dozen current and former officials and other people familiar with the companies who spoke on condition of anonymity to discuss an examination that is not public. The inquiry, which is active, calls into question how closely NVIDIA is tracking where its AI chips end up and highlights the possibilities of American export laws easily being sidestepped.

29:23Megaspeed is also facing scrutiny from Singaporean police, who told the New York Times in a statement that they're investigating the company for breaching local laws without elaborating further. As the dominant provider of AI chips, NVIDIA's annual revenue has soared. I'm going to skip over this because I know you already know this. While you skip over that, let me tell you about graphite.dev, code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Continue, Gordy. Megaspeed illustrates the challenges facing U.S. government officials trying to keep China from assessing powerful AI chips.

29:54After splitting off from a Chinese gaming company in 2023, Megaspeed set up a subsidiary in Malaysia that quickly snapped up nearly$2 billion worth of NVIDIA's most advanced products. Most of those chips came from the U.S. branch of a Chinese company that has already been sanctioned for providing technology to the Chinese military, according to records obtained through Import Genius, which is started by Ryan from Flexport. Yes, that's right. And his brother. That was his first uh that was his first yeah his first company uh it's so funny that uh the whole like pivot to ai is not just a silicon valley american meme like it's definitely happening in china too they're like yeah we're doing gaming but like you know this ai stuff is way better let's just pivot to ai or like uh high flyer being like we gotta we gotta get on the large language trading is cool yeah it's cool but like let's just uh let's focus on ai yeah so anyways this this story is evolving I'm not surprised to hear that I mean I think it's actually good that the commerce department is trying to figure out what's actually happening here I think there's been a lot of rumors floating around of like data center companies in Malaysia and Singapore having an extreme demand for chips and this sort of uncertainty of like okay who's actually paying remember one of our first episodes we read a Wall Street Journal article on Malaysia the Switzerland of AI, because everyone was building a ton of data centers there so that they could sell inference or build token factories that would sell to both American companies, Singaporean companies, Chinese companies.

31:30And it was kind of like this neutral territory. But as the trade war heats up, people are going to want more answers, more analysis. They're going to want to understand exactly, you know, are any of these things, does every deal fit perfectly within the current restriction framework, right? Because obviously, diversion is a natural thing. I mean, we've heard stories of this in the Ukraine-Russia conflict, where companies will go, you know, entrepreneurs will go buy DJI drones and then go deliver them. And what are the actual restrictions on that? It all gets, when the economic stakes are high, and things are trading at a premium, you can literally, I mean, we've heard this from, where there was one account of someone loading the training data onto hard drives, flying to another country, doing the training run, saving the weights onto hard drives, and then flying back.

32:22And it's like, that's kind of hard to predict. That definitely is a runaround on the chip controls. It's not the stated goal of those chip controls. But when there's huge dollars on the line and training runs are extremely valuable, we'll figure out a way. It's fair to kind of try to understand Miss Huang, the CEO of Megaspeed's background. She spent much of her career in mainland China, including working as a television reporter for Chinese state media. Let's hear it for television reporters. One of us. One of us. One of us. From being a television reporter to creating a new hyperscaler. I love it.

33:00Do anything in this world. Who knows? I think we'll stay out of that game, but good to see that you can start content. We'll leave it to our friends. Good post here from High Yield Harry. Okay. It says A16Z and Sequoia go leading around. It's the Crips and the Bloods. Is that what this is? Uniting. Uniting. Is that what's happening? This is about Kalshi? Referencing Kalshi, who followed Polymarket, who announced their$2 billion financing earlier this week from ICE. Not the immigration. No. branch of the government. And not the International Commodities Exchange, the Intercontinental Exchange. That's right.

33:44That's the name. But a call sheet raised$300 million, backed by Andreessen Horowitz. And the prediction market wars are heating up dramatically. Coinbase, Google's Capital G, A16Z, and Paradigm were investing in this new uh as well as sequoia we're investing in this new round um and so by this point everybody's sort of picked a picked a side yep um and uh i would not want to be a net new prediction market company getting started today i think there's going to be an insane uh the capital war is just sort of starting in the category i mean do you remember the capital war between Lyft and Uber, there were like three or four, probably 10 other companies that were in the ride hailing industry.

34:40And they aren't public companies today. They either sold or pivoted or found some other niche. But when you're in this duopoly world and this capital fight, and there's maybe one that's running away one way or another, it's painful to be number three. Kalshi is on track to do$50 billion in trading volume. And they actually did$300 million last year. So a pretty wild jump. Competition is certainly heating up. This will be an interesting one to watch. Brandon Jacoby is putting another app in the truth zone. What happened here? Brookwell launched an app, I guess. And Brandon Jacoby says, hey, Brookwell app, great design.

35:30I was proud of it when we designed it for Capital in 2022. Yeah, so I guess this company, Brandon and I obviously worked together to design the app that you can see here. Brandon is an incredible designer and spent hundreds of hours developing this experience. and ultimately I think Brookwell seemingly pretty much copied it to a T and then even doubled down and said like, yes, and. Wait, really? Like the founder replied to the brand. It was like, yeah, yeah. Basically saying, yeah. I mean, I think it's like. It's sort of fair if it's not like. It's fair game. Yeah, fair game to pull a design off the shelf that's not actively being used as much.

36:14But it's nice to go further back in time. You can still bring your own ideas to it. I think Brandon's frustration is probably just that you didn't bring many new ideas. You just kind of copy and paste it. Yeah. A lot of that going on. John Titor is quoting a post that says, This is a camera. It's 200 by 200 pixels, 30 frames per second. I didn't know they made cameras this small. John says, There are certain things you just can't learn about if you're prone to paranoia. This is actually crazy. Yeah. So the community note gives a little bit more context here. It does say that this is the camera module.

36:54It is capable of producing an analog video output, but it needs a power supply, battery, or storage. Yeah, and the concern here is the implication is that this camera could be built into effectively just like a dot on the wall, and you could store the power supply and battery. Oh, it's only 200 by 200 pixels. So I was saying, should we get these and put them all around the set? And so you just walk into the TBP and Ultradome and you don't see any cameras moving. Like we have a slider cam now. We have a lot of different cameras around. And you kind of trip over them. And, you know, you've got to cable manage them.

37:28But imagine if we could just put one little dot on that turbo puffer fish there, one little dot on the microphone. You can just get any image from any place. Big debate on the timeline over whether you should have shoes on or on in the office. This Ben Lang over at Cursor says, no shoes at Cursor NYC. Will O 'Brien says, if Ulysses ever ends up like this, you have permission to shoot me in the face. Very aggressive response. I can see why companies that are just, you know, want to be cozy, want to have, I think, wearing shoes in your home is insane. Yeah. I'm very against walking around the house with shoes that you wear out in the world.

38:13Sure. Plenty of studies that just show you're just tracking in a whole host of things. Makes sense. Do you do slippers in the house then? I enjoy slippers. I feel like if you don't do slippers, then you need to keep the whole house warmer. You need more carpets. But I wonder what else went into the plan to make sure that Cursor HQ in New York City is fully cozy. Because you don't just want to take off your shoes and be around. We have concrete floors here. We are shoes on facility. No one takes off their shoes. But if we were to go to shoes off, I think we would need to get some carpets, keep the place a little bit toastier.

38:53Or maybe give everyone a pair of slippers. What's going on here? this account toys xyz is sharing a nano banana watermark have they how do you know how their watermark works functionally yeah so um you have to like turn like uh do you have to basically increase the saturation of the photo exactly so this pattern that you see are are subtle changes in the saturation and if you go to the next image you can see what it looks like on an actual nano banana image, not just the watermark. And so this is a black and white image. This was generated as black and white, but nano banana uses slight variations in the colors to just have a little bit of saturation.

39:36So normally if you're looking at like a grayscale image, basically the saturation is turned down to zero and there is no color whatsoever. There's only brightness, right? You're going from zero to one just on the black scale. Nano banana, even if you ask for a black and white image, it will output an image that does have a little bit of color, and it will vary this pattern. So there's little bits of red, little bits of green, little bits of red, little bits of green. And so this is being put in every nano banana image. Of course, as soon as you discover this pattern, there's probably some ability to remove it.

40:13Even just a slight edit might change this. you could just go in and actually reduce the saturation to zero and then the watermark's gone and I'm sure people will do that but yeah there will be somebody will make an app where you just upload an AI photo and then it figures out which model generated it and then figures out how to remove whatever hidden watermark is included totally you have a clean image totally there's already a ton of Sora watermark remover tools out there it's I think this is still just more useful for having a reality check on like, was someone dumb enough to not even remove the watermark?

40:50At least you can just automatically flag that. And at least it injects an extra step, an extra cost into generating spam images or whatever you would want to do that's malicious that you'd want to discover the watermark. But fortunately, I feel like even with Nana Banana, you can still just look at the image and tell. But if you want to generate some generative media, Head over to Fall. The world's best generative image, video, and audio models all in one place. Develop and fine-tune models with serverless GPUs and on-demand clusters. Let's fall. Jared Kushner? Jared Kushner. Matt Stieb said in an article a while back, Jared Kushner claims he can solve Israeli-Palestinian conflict because he's, quote, read 25 books on it.

41:38And, of course, seems like that was hopefully... The real reason is what Will Manitis said. The New York real estate. Yes, yes. He's been doing New York real estate deals. And so he's ready to broker another real estate deal, effectively. But, I mean, we haven't really covered the peace deals. The journal's writing about it a little bit. I guess there's going to be a vote. But there's plenty of... Another news. Barry Weiss asked everyone across CBS News to send her a memo by next Tuesday explaining how they spend their workday and what's working, not working. Is this a what did you get done this week, basically?

42:19It's a what exactly do you do here? You think it's that? I think, I mean, it's great asking, hey, what's working, what's not working. Yes. But as a manager and you're coming into a new organization, or a leader you're coming into a new organization, you want to get a pulse on, okay, what are people actually doing here? Because I think the question is with everything that David Ellison is doing, is he trying to turn these into media companies that can create massive cash flow? Or are they strategic enough? Like, I don't think when Jeff Bezos was buying the Washington Post, he was thinking, I'm buying this to make money necessarily.

43:00and so the question is I think Max Taney here the way that this has been written I think the question is like are they going to let a bunch of people go? That's kind of how I would read into it. Yeah, I mean at the same time if you're coming into if you're coming in as CEO or editor-in-chief of a new organization and you're like my view on it is that they're actually severely understaffed and the first thing I'm going to do is triple headcount, it's still reasonable to ask what everyone does. So you understand like, oh, okay, this person's doing five different jobs and they're working 200 hour weeks.

43:40Like maybe we need to get them some extra support. But I agree with your general assessment. And it certainly would be a little bit stressful to get this email from Barry. But, you know, she says, please be blunt. Just break it down. Recommendation, don't use AI. Any M dashes? There is an M-dash, but we know Barry uses M-dashes. She's been using them for decades. So there's no, maybe not decades, but for her career. It is, the reason that it's in the model is because it's a popular writing tool. Anyway, Cloudflare did a new rebrand, powering 20%. And I guess Ty, a former member of the Party Around team.

44:19Oh, really? I guess was behind this. Oh, no way. Why else he'd be sharing it? Wow. And he also was co-founders with Dylan on CTG. Oh, no way. That's very cool. A little bit of lore. Ty is a legend, extremely talented designer. When he had reached out to us at Party Round, and I talked once with him. I remember correctly. We didn't immediately make an offer. We were just kind of like, yeah, let's keep talking. and then he built an entire game, like a simulation of a Game Boy game that you could play on your phone and he just sent it to me. He built it. This was pre-Vibe Coding. He just built it and sent it to me.

45:03It was web-based basically? No, it was mobile-based. And I just immediately called him and I was like, okay, we're bringing on the team. This is amazing. Very cool, very cool. I mean, it seems like a good response. A thousand likes. No one's, people are very opinionated about branding, launches, launch videos, all sorts of stuff. Yeah, I found the original post. It was a Pokemon-style game where you could build like a cap table, basically. Perfect. Collect investors. That's exactly what you want, for sure. Great stuff, great stuff. If you're just tuning in, Sam Altman will be joining in about 10 minutes.

45:41He's on for half an hour, and then we're talking to Elad Gill. before we move on let me tell you about turbo puffer search every byte serverless vector and full-text search built from first principles on object storage fast 10x cheaper and extremely scalable um you see this reddit yes circulating that lists open ai's top 30 customers by token consumption and apparently duolingo number one uh second is open router but yeah that sort of doesn't really count yeah routing uh routing yep uh consumption uh surprising to see indeed at the top of this list. Do you see number nine, baby? Number nine. Let's go.

46:20It's ramp.com. You see WarpDev on there, Shopify. Sure. Also. So when this came out because of their, I don't know if there was this table is like a leak or something, but at Dev Day, OpenAI put up a list of names of people that had been using over a trillion tokens, right? And people kind of reverse engineered that to understand what companies they work for, right? And I was trying to make a meme that I was sending you, and it wasn't quite hitting, but it was like the Atlas holding up the world meme. And it was like the entire global economy, but instead of like OpenAI, which is the current meme, it's like 30 companies in wildly different markets that are all using AI.

47:07And so you would be more worried, I think, if this was more circular. And this is the Martin Screlly take, which is that if you looked at the top 30 customers and it was all like AI, chat, GPT, wrappers or something, something very circular, you would be worried. But instead, it's companies like Indeed and Duolingo and Ramp and Shopify and all these companies that are touching very different pieces. These aren't competitors. Like Duolingo is not a competitor to Shopify. And so it feels like evidence that this is less circular. I don't know. Yeah. I mean, it is the Grok, C of Grok, G-R-O-Q, recently said that 35 to 36 companies are currently responsible for 99 % of token spending in AI right now.

47:59Even among those 35 companies, two are by far the most significant spenders in their open AI and anthropic. So it's crazy. That's generation. that's buying inference. I'm talking about companies that are buying from open AI and buying from Anthrop. Right, right. That is certainly... So one more step. One more step. Because it's like, I'm fine with, like, it's scary when you hear, like, open AI is buying 99 % of AI stuff, but then when you look under the hood and you see, like, Duolingo and Ramp and Shopify, it's like, okay, well, that's actually, like, pretty diffuse in the economy. Yeah. So, yeah, there's two significant spenders, but in those, spenders are thousands of other companies.

48:40Exactly. Yeah. And I wonder just this idea of like, there's a, there is a trend. We're in a trend right now, the AI wave, and it's holding up the global economy. Like has, have other trends held up the global economy successfully? Like if we go back to industrialization, was that, was that holding up the global economy? Did it successfully hold up the global economy or like global trade, the containerization that held up the global economy. And, and like, there were certainly like ups and downs, but like it kind of successfully held up the global economy. Yeah. And certainly the, I mean, uh, going back to that clip from Ken Griffin earlier this week, he was basically saying that in 1999 and 2000, it was incredibly obvious that the internet would change everything, but that still took 15 years.

49:30Right. So I think even I would say most of the bears, like true AI bears still believe in the potential of the technology. They just believe there's going to be some sort of winter. Like a hiccup. Yeah. Winter. Yeah. We haven't heard a lot of talk about AI winters. There's been many of those. And we I had a hot take that 2023 was an AI winter, which is like, of course, like the most bullish, insane time possible. but uh i but my my my riff on it was that like chat gpt launched in 22 waymo launched in 22 and then 23 was more just like the adoption of chat gpt and at and i think at dev day they launched uh maybe gpts and it wasn't like a wild use case we didn't have reasoning models yet and so it was like this magical technology that launched in 22 and in 23 everyone was kind of like processing it but there wasn't like a massive jump in 23 i mean I guess we did get GPT-4.

50:28What do you think, Tyler? LG, by the way, is quizzing you in the chat. Oh, yeah. John Coogan, how many total tokens have been used on OpenAI year to date? Name every token. Name every token. Well, it's odd. It's$6 billion a minute. I think we did the math on this. It's in the trillions. It's a lot. So, yeah, when I calculated it, I had like around one quadrillion a month. One quadrillion a month. That was also the number. I think Demis a couple of weeks ago, he said, uh, one quadrillion. Um, but yeah, cause open AI was like a six million per minute. Uh, but that's just on the API. So if API is like 25 % of the company, exactly.

51:06Yeah. Yeah. You have to do a lot of like dancing, but yes, in 2023, I mean, GPT four, I think is like, yeah, I don't think you should be following product releases. You should be following, but GPT four was, uh, was trained in 22 and was released in bang in 22. So it doesn't count. I'm kidding. Of course, 2023 was a phenomenal year for AI. Not a winter at all, but it is interesting to see where the growth spurts come, how much of it is just a smooth curve of adoption and diffusion into the economy. What's the rate at which it's diffusing? Where are companies and individuals actually getting value?

51:42you. Should we move over to the huge news in creatine? Yes. Dan McCormick says, work with your brother, go into debt if you have to, because Paki McCormick is giving a shout out to Dan, the founder of Create. He says, my brother Dan has been writing the weekly dose of optimism for three years. That's Paki's second newsletter, aside from the one that he writes. Last week was his last. Dan McCormick is stepping down from writing the weekly dose of optimism because his company creates absolutely ripping 85 million dollars run rate i swear i swear gummy vitamins are just an infinite money glitch there's a couple categories that are just doing so so well stick packs and uh powdered supplements i don't what's what's hard hard is hard is liquids right hard is do not try to sell liquids on the internet that's hard but gummies must have a great uh I invested in Create, I think, at a six cap.

52:41I met Dan before the launch. He was raising it even less than that. I didn't see the vision right away. And his execution in the first five months was absolutely wild. And I capitulated. And it's been a wild ride. The execution is insane. Well, Dan, if you want to keep a hold on that creatine monopoly, You got to get on Profound and get Create mentioned in ChatGPT. That's right. Reach millions of consumers who are using AI to discover new products and brands. This post from Frog is a banger. Please make sure you are only drinking as much water as you really need. We need that for the data centers.

53:25If you're thirsty, Grok is thirsty too. Completely agree. I completely agree. reply here installing a low flow shower head out of concern for grok i love it i saw someone else was trying to put the yes this is it andy masley a couple slides later he said every day i find a new way of trying to get across just how ridiculously fake the problem of ai water use is we've talked to a number of neocloud ceos data center builders on this show who have built data centers and they've told us, yeah, we use it. We use a decent amount of water, but once we have the water, we actually have figured out how to just recycle it in, in, in the system.

54:08And so, uh, we're, we're not actually using that much water. Even the new iPhone has a single drop of water and they sell a lot of iPhones. So I mean, millions of drops of water, you got to answer something, but it just proves a point that, Oh yeah. Yeah. You can recycle it for heat. Yes. Yes. That's a great point. I hadn't even thought about that. Um, it's not like you're, you're, I got to go refill my iPhone. Can you imagine how upset people would be? It was like, yeah, I need to charge my iPhone. Oh, I forgot to refill it with water at the gas station. So he shares some data here. So this is a quote post from someone else.

54:43Yeah, but the form of AI that uses the most water and electricity by far is Chachipiti. You can start somewhere. The whole, I can't do it because I got cut off is just an excuse I don't care for. And so here is some evidence, some new ways to think about the amount of water used by AI. Have you ever worried about how much water things you did online used before AI? Probably not because data centers barely use any water compared to most other things we do. Even manufacturing most regular objects requires lots of water. Here's a list of common objects you might own and how many chatbot prompts worth of water they used to make them.

55:19Leather shoes are 4 million prompts worth of water. Smartphones are 6 ,400 ,000 prompts of water. Jeans, a single pair of jeans is over 5 million prompts of water. This is such a silly metric, but I love these silly metrics. T-shirt is a million prompts. A single piece of paper is 2 ,550 prompts. If you want to send 2 ,500... Imagine trying to give up all the things on this list because they use water. But you'd have Grok. You'd just be naked using Grok. But you'd need a smartphone, and that's six million prompts. Chad is going crazy right now. We're being told to not ask him anything about Trump or crypto.

56:01They've got money on it. It's because there's prediction. Oh, no. Okay. We are not paying attention to the prediction markets. We will not be steering the conversation if you happen to have a bet one way or another. Well, I believe we have our guests in the Restream waiting room. But before we bring them in, let me tell you about Julius. What analysis do you want to run? Chat with your data and get expert level insights in seconds. Julius is the AI data analyst that works for you. Connect your data, ask questions. Data super intelligence. It is. And we are joined by Sam Altman and Bill Peebles.

56:36Sam, Bill, how are you doing? What's going on? Hey, guys. Hey, guys. Great to see you. Congrats on all the progress. I've been enjoying Sora a ton. personally I've been enjoying making them I had a ton of fun making the collab post yesterday and I was wondering prompting your cameo feature John made it so that he always appears as a bodybuilder if anybody is cameoing him so you guys got some some chaotic results do you have favorites or a post that you've been coming back to or they've you know stuck out to you as particularly creative uses. I mean, definitely all of the ones of like me stealing GPUs or doing other crazy things to get GPUs have been funny.

57:22In the last few days, at least in my feed, there have been these very beautiful sort of fantastic scenes that are just not things that could have ever existed without something like Sora or wouldn't have been easy to make. And watching people build those and watching the trends flow through that has been pretty awesome. What about you, Bill? Any favorite uses of Sora so far? Oh, man. Mark Cuban came on the platform a few days ago, and there have been some hilarious Shark Tank memes. Those are probably my favorite. Pitching some Sora features to Mark. Also leveraging the prompting function to always include an ad for cost plus drugs I thought was especially hilarious considering he's been one of the most vocal opponents of advertising in AI.

58:07He is leveraging the feature to the max. Yeah, I think they're going to be all of these weird new dynamics that we see emerge that just weren't possible in previous kinds of video. and this is like a fun period because it's all going to be so different every few days. Man, I'm watching this Polymarket ticker go by and it's so tempting to say things to us. Yeah, yeah, yeah. To be clear, don't worry about the ticker. I don't think we're including, I don't think any of those markets are being featured in the ticker. But yeah, again, this is the new world we're in. Yeah, you can move the market live on TBPN.

58:44But today, people I'm sure will be happy or disappointed we're here to talk about Sora, of course. Yes, yes. So none of the other topics. But I mean, I also want to know about ads. Why no ads in Sora on day one? I feel like you've laid out a really great, you know, mental model for how you think about ads on Stratechery, on the Andrews and Horowitz podcast. I'm bought in. Is it a technical thing? Do you need scale? Do you need to think about it more? Why no ads on day one? This is like a 10-day-old product, right? Like it's hard to get anything to work at all. And we don't assume success. We got to like go hard-earned success, and then we can think about monetization for it.

59:30But this is like, it's gone great so far. It's still very early, and there's still a lot of work to build something that a lot of people are going to love first. What about surprising capabilities of the model? You mentioned that you've seen some fantastical scenes. I'm interested to know about specific breakthroughs that you've noticed that Sora 2, the model, is particularly good at. I noticed one about reflections being great. Obviously, people love the cameos. But what has surprised you in terms of just like technically the model can do something now that it couldn't do before? This model is a huge leap forward in terms of physics IQ.

1:00:06So pretty much all past video generation models really struggled with prompts that involve like backflips, gymnastics routines, etc. And this is really the only model that exists today, which can reliably handle these kinds of really complicated dynamics. One of the big features that people have really loved on the app so far is the steerability of the model. So, you know, if you give it like a really simple text prompt, that's maybe even only a few words, this model is really good about kind of telling a coherent story with like a beginning, middle and end and doing this like automatically in a way that doesn't require like a lot of direct steering from the user.

1:00:38If you want to like go into, you know, a ton of detail about exactly how your prompt should be laid out and how the story should unfold. It supports that too. So it can kind of meet you wherever you're at in the creative process. But really this model is just like so hyper steerable and it's like just vastly higher physics IQ just makes it able to do things that were like not possible a few months ago. Is that all within the model or is there some sort of like reasoning step where you're hydrating or unpacking my prompt and writing a bigger prompt or breaking down the problem in some way? Can you share anything about that?

1:01:09Yeah, it's a good question. So, you know, the intelligence for these text conditional video models kind of lies both in the core model itself like Sora and some amount of it also comes in through the text prompt. So, you know, where however the user decides to kickstart a prompt, you can have like a language model under the hood, add some details in. But for example, you know, when it comes to things like, again, like doing these backflips or any kind of physical interactions, how refraction is modeled, you know, when you're pouring water into a glass, all of these details have to be captured by the core video model itself.

1:01:38So that's intelligence, which is really innate to Sora. And certainly you can supplement it with intelligence from a language model as well. But it's not necessarily a prerequisite to get kind of amazing results out of these things. Are there any areas on the physics where you think that the model falls down and you want to improve? I mean, we went through the era of like six fingers. It seems like reflections in water are solved. But someone was saying something about doors being hard. I haven't noticed that one personally, but a lot of the stuff's great. But what have you noticed is like the next version is going to be even better at?

1:02:06this is still very early yeah i think bill said that i appreciate it is this is this is the this is like the gpt 3.5 moment for video i agree and if you went back to use the actual gpt 3.5 you'd be like okay signs of great promise can do the occasional impressive thing but it was really not until gpt4 where these text models started providing real value for people and we know how to go make the gpt4 equivalent of video models and we will do that and then a lot of these things that are currently annoying, like doors or, you know, once in a while, something goes through something else it's not supposed to.

1:02:41In the same way that the world, you know, loved to complain for a brief period of time about where 3.5 fell down and, oh, it's never going to be useful. It's never going to do this. It's never going to do that. And then we were able to just keep making it better and better and better and better. The model physics IQ is certainly the best I've ever seen, but it is nowhere near as good as it will be in the future versions. And I think, I hope, we'll see a similar thing to what happened with the GPT text models, which is people will always demand more and better, and they will always find new and better things to use it for.

1:03:15And the world will just make ever more amazing videos. And how quickly, oh, sorry, how quickly, go for it. for video here. Yeah. Like GPT-1 really was Sora 1 for this modality. And the progress we've made kind of in the last 18 months getting to this 3.5 moment, right, is really compressed compared to how long it took to go from GPT-1 to 3.5 in the language domain. So we're really expecting progress to continue to be meteoric here in the near future. How quickly do you expect the Cameo feature to be cloned? That feels like an equally important part of the, you know, the models made a leap, but the product and the experience and the experience of creating these assets is wildly innovative.

1:04:00We saw stories get cloned. We saw, you know, algo, video short form feeds get cloned. I expect many other platforms to be looking at this functionality and realizing that this might be the future. You guys certainly believe that it could be important. and so how quickly do you think? We're actually totally okay with a world where we do the product innovation and everybody else copies and I don't think it works for them as well as they think it does. Like the, you know, a lot of people have tried copy in ChatGPT and you can go look at some of our competitors apps and they even copy the mistakes.

1:04:36They even copy the design decisions we really wish we hadn't made and maybe it's worked well for them. I guess I kind of hope it has but it's been fine for us. I think like the key to this is not any one innovation but it's repeatedly putting them out again and again and being first to come up with them and put them into a cohesive offering. And, you know, that's what we want to be good at. And if other people want to clone the stuff that works, we also sometimes clone stuff that works. That's fine. But mostly we want to be able to drive the innovation. And I think Bill and his team have done an incredible job of figuring out how people actually want to use these video models, what the models need to do.

1:05:16Really, they've approached this as a full stack problem from how do you train the video model to how do you make this enjoyable for users. But cameos are one out of many ideas they have from here on the journey to like the product that we hope to eventually build. And so if people take some inspiration from us and copy us along the way, I'm sure they will. It's fine. How do you think about the like popular claim that we want AI detection, I want AI content flagged. Is that a stated preference that's not a revealed preference? Because personally, I don't want bad AI content, but I don't want bad human-made content either.

1:05:58I want great both, and I'm fine when someone comes up with something genius and they instantiate it with a video model. How do you think about it? I think that is the real thing is you don't want slop. You want great content. Different people, one man's slop is another man's treasure for sure. But what you care about is like good, original, thoughtful, new, helpful, whatever content. And whether that is generated entirely by a human or entirely by AI or what I expect will mostly happen in the future, which is tool-assisted, human-driven generation. I don't think you care that much if the content is great there's a lot of like you know stuff that is technically written or drawn or filmed by a human but is completely derivative and much less original than what an AI has generated and I think that will be what people really care about long term you just want great content now I also do want some human connection with it when I read a great book first thing I want to do is read about the author that wrote it and what life experience went into that.

1:07:07I don't think that'll go away. But if they're using an AI as a tool to help them make the writing better, sign me up. That sounds great. Similarly, I would rather watch a video about someone I know than some random AI-generated character, which is part of why I think this was cool to offer. One design decision the team made that I thought was really great, and I was actually pushing them in a different direction earlier on, and then I decided they were totally right, and I thanked them and dropped it, was the fact that the feed is AI only and not a mix of AI plus some uploaded videos, I think is a subtle but extremely important design decision in how people are relating to this.

1:07:46Yeah, it was a very weird experience for me. I was thinking about the collab post that I was making announcing this interview. And my initial thing was like, well, I'm going to have to think of a script or I'm going to have to think of what I say or I should record a piece of this and then I'll use it. And it was like, no, I just typed the prompt in and then I get the front facing videos. Remarkable. What are you what kind of indicators are you guys looking at as Sora can transition from what it was the second it launched, which was a creative tool into something that's more of a like a consumption platform, traditional, you know, social media platform.

1:08:22Like what talk about kind of what you guys are pushing forward to because obviously you're seeding seeding the network with with the tool. But it's it's certainly much harder to turn it into something that people are spending hours a day and purely consuming content and not creating content. You know, we really wanted to design this from the ground up to be centered around creation. And a lot of the metrics that we've been focused on optimizing here are really aligned with making sure as many people as possible are actually like getting their hands on the SOAR2 model itself and, you know, able to create content with their friends and like for the rest of the world.

1:08:59One metric that we're really proud of with this launch so far is that 70 % of our users are actually creating content, even to this day, you know, a week and a half after launch. And that's like vastly higher than on any other social media platform. And I think it really speaks to just how fun creation can be with the right tool set, right? If you look at any of these kind of legacy platforms, there's just like so much friction from like getting off the feed and like into some creative flow state, right? You have to like put the phone down, you have to go get like a camcorder, start recording yourself, find your friends, like do a dance, etc.

1:09:32It's just like a lot of work, right? On Sora, like you can just pick up your phone, find like any video you like in the feed, remix it, you know, cameo any of your friends. And I think one insight that was not obvious to us at first, but we've kind of clearly seen as an emergent behavior of this product, is just like there's all these people out there who would not necessarily want to be like, you know, influencers or something or have like a big social media presence. But the fact that like all of their friends can just access their cameo, right, put them in all of these crazy situations, actually like kind of gets them into the playing field in a way that felt really high friction before.

1:10:06And so, you know, we're closing in on close to like 2 million weekly active users now. We're really excited that such a huge percentage of that user base to this day is like still creating with Sora. And we're going to continue pushing on that direction and making sure people have even more powerful tools in the future. Yeah. So 70 % of Sora users are creating content. The typical benchmark that people kind of quote randomly is like 1 % creation, 99 % consumption, something like that. And that certainly feels like my experience on Instagram. I post a photo every once in a while, but most of the time I'm just kind of scrolling.

1:10:39And I'm wondering if you think that that 1 % will be much higher on Sora in terms of actual time in the app, time prompting versus time scrolling. If you have any data, that'd be super interesting. But then also, does that make it more of like a competitor to video games than traditional social media because it's such a lean forward experience versus just lay back? What do you think? Yeah, it's a great question. we still need to study this more exactly how creation versus consumption habits kind of change over time for folks on the platform. It's still pretty early days. I do agree with your point, though, that I think over time, this is going to feel much more immersive in a way that like video games kind of do like you have more agency when you're actually using the platform, you know, not just kind of like mindlessly scrolling a feed, like hours a day.

1:11:26And like one interpretation of this product, which I think is kind of interesting, especially from the research perspective, right, is cameos in some ways like the simplest way where you can kind of like inject yourself into the model right so it's a very low bandwidth communication channel right now you know you're only giving like a few seconds uh of video footage uh of like any given individual like into the app uh but like over time right you can imagine like these models know more and more about your life they really like deeply understand your friends how you want to like show up in the world and like over time this can almost become like a little mini like alternate reality right so like you're not just generating like videos of yourself with your friends.

1:12:02Like you actually just have like digital copies of yourself running in the model on the Sora platform, interacting with other people with agency. And so I think over time, we're really going to see this platform evolve into, you know, something that feels kind of familiar today and just something that really leans into like the full intelligence of Sora 2 in the future and like really leverages all the world simulation capabilities that we're working on internally. Yeah, I would add to that that if if you think of this like spectrum of the kind of entertainment you can have in front of a computer at one end you have like watch a two and a half hour movie and you hit play and then you lean back and you don't do anything at all um and then at the other end you have like a very intense video game and you're like you know sweating and your heart's racing and it's like super super active um ai is going to push things to be more in between there so you'll have maybe you're still watching that movie, but now you can like, say something a few times throughout the course of it, and it changes what happens as the movie plays out.

1:13:00Or with Sora, you're seeing this amazing new phenomenon where most users are creating in a in a world where traditionally only 1 % of them did. And so you're, yes, you're like watching a video feed, but you're, you're doing a little bit more. And it, at least for me, really changes how fun the whole thing is, and how I feel about it. Then maybe you'll do what Bill said, and you'll have like, you'll be way more actively participating in a Sora feed. And I think you're just going to see that continuum blur a lot more. Did you see Bandersnatch by any chance, Sam? Have you seen this Netflix? It's like a Netflix choose your own adventure.

1:13:30And it was really cool idea, but ultimately people, it never really took off and became like something that they do again and again and again. And I'm wondering if it was because it was like not customizable enough, um, or people just want to just sit back and see a director's vision. I don't know. Anyway. I never heard of that, but it sounds cool. Yeah. How do you think, question for Sam, how do you think about allocating compute to Sora versus the rest of the business? I imagine Bill is constantly in your ear every other hour, but how are you thinking about it? You know, my real answer is I've entirely changed my focus of how I spend my days to just go get more compute rather than have to make the compute allocation decisions.

1:14:11I still do have to make some short-term compute allocation decisions, but I hope we're heading into a world where I am instead telling people, you got to find a way to use more compute. And we're going to be very aggressive here. It feels like you're doing a great job of bringing things within your control, within the supply chain. What is outside of your control at this point? I mean, most of it. But I feel like you have great partners all up and down the stack, multiple partners in different parts of the chain. When I think about scaling up Sora, I feel like it's crazy to bet against you. You're going to get the chips.

1:14:49You're not going to be GPU 4. You know, you go try to buy like 10 gigawatts of power for delivery next year. It's not so easy. It's funny. How are the conversations going with Hollywood? Oh, yeah. Oh, actually. Yeah, you take it. Yeah, I was going to say, we've been chatting actually with a few very notable folks in Hollywood over the last week. You know, I think people's first reaction to this is like very understandably going to involve a lot of trepidation and like anxiety. When we've gotten to just sit in a room with these folks, though, you know, and really explain what we're building. I've actually been pretty struck by like how excited folks in Hollywood are about this.

1:15:25You know, we were chatting with with one actor recently who mentioned that, you know, on Twitter like a year ago, saw like a deep fake of her generated with one of these like open source models, which really had like a lot of nasty content created. And when we really like walked her through kind of all of our safety mitigations, right, how we're making sure that we have this like very well-defined model spec, which dictates the behavior that we allow on this platform and how we are really leaning into like full control of likeness, right, more so than any other platform. Like you have to come in through the Cameo process.

1:16:00you can't just like upload an image of yourself and just like generate a video of it like any person you have to come in through cameo um i think it became clear that you know we're really setting the right standard here uh in terms of making sure people are in full control of their likeness in hollywood i think that's where like a lot of this anxiety comes from right it's this feeling that you know some random person can just kind of take videos or images of you and do whatever they want with them and create all of this like like like terrible content that's like outside of your purview. But we've really been like designing Sora from the ground up to put users in full control of their likeness end to end.

1:16:33From the moment you sign into the app to, you know, needing Cameo permissions to like access any of your friends' generations. So, you know, I think we need to engage more with Hollywood and we're going to continue to do that. But once we really explain the story of Sora, you know, they're very receptive to it. Do you think there's a world outside? To add something to that, like, you know, the team asked me before launch if they could put my Cameo in there, open access and I of course thought about it for a second and said absolutely yes. I had all these Hollywood celebrities then messaging me on the first day being like, you're absolutely crazy.

1:17:04This is insane. This is like the dumbest thing I've ever seen. And then by about the third day, they were like, hmm, that was really smart. You got like a lot of, you know, free publicity. Maybe we need to be doing that. And I think you're now seeing actual celebrities say, okay, I'm going to do this. And I expect a lot more of them will. Similar thing on other kinds of characters in IP. I can totally imagine a world where our problem in a year or six months or maybe even less is not that people don't want their cameos or their characters appearing, but they think we are not fairly having their characters or cameo appearing often enough.

1:17:40This may turn out to be a really big thing for fan connection. Now, it may be that kind of the previous generation of celebrities don't want to do this and the influencer celebrities all do. I don't know how that's going to go. But I bet this will be like a pretty deep kind of new connection. Yeah, it seems like it's been good for DiCaprio in the memes. Like he's not directly monetizing those when you show the champagne meme or him pointing at the TV. But like, you know, it builds his aura in some way. A friend of ours posted something yesterday. This is Jeremy Gaffan. He said, the reason we're so upset about slop is because it's obvious we're all going to love consuming it in two to three years.

1:18:17It's not going to be slop for long. Do you agree, Sam? I mean, some of it will be slop to some people and some of it won't. I remember, like, there was a real reaction like this in the early GPT days where people were like, I can't believe anyone reads this. It's like total crap. It's full of hallucinations. You know, it's like it's not useful to anyone. And then it became more useful to some people, but they said, I can't believe anybody like ever thinks this thing writes a beautiful sentence. That's insane. And then with GPT-5, you have authors saying like, wow, this is a useful tool. It sometimes like writes a beautiful sentence.

1:18:51Yeah. And I kind of think it'll follow a similar trajectory. What do you think about the fact that people feel at least, I don't know if they actually can, but it feels like you can still clock GPT-5 writing, you know, it's not this, it's that, the M-dash. Like, will we still see these artifacts in three years in Sora 5 that people are like, oh, if you know, you know, you can tell, but most people can't? Yeah, it's like, what's the M - of video? Because I don't think it's like six fingers. No, no, definitely not. That's the typo, which doesn't happen anymore. Yeah, I think right now the M - is like this like slightly wired speech pattern in Sora where it likes to say a lot of words very quickly.

1:19:31You know, these generations definitely have like a style to them. I think analogously to GPT, we really want to give users a lot of control over exactly how their videos show up right on the platform. Like if you really want kind of like a very soothing experience, right, not a lot of shot changes going on. We want to give users the ability to generate that. We're going to continue to give more optionality to people. So, you know, there'll be some default kind of behaviors and quirks of Sora for sure. But we definitely want all power users to be able to be in full control. Random question. Where did the name Sora come from?

1:20:04Yeah, this is a fun one. So the original Sora came out in February 2024, the OG blog post. We did not have a name for it. I think like up to two days before we like revealed the model to the world. We just could not agree on the team what it should be. Did you at least have a code word or something? We just called it like VideoGen. And so at some ungodly hour, I like just started pumping a bunch of crazy ideas into ChatGPT. And then like we basically ran out of like English words. So then we switched to like Japanese words. Wow. And then Sora came out. I was like, wow, that sounds really nice.

1:20:41It means sky, you know, linked with like imagination, like all the possibilities of creation. And so then we just like last minute ship Sora. So yeah, it was kind of a mad dash. Okay, speaking of Japanese stuff, Sam, you said you were looking for an Acura NSX a while back. It's kind of this throwback car. It's not a Waymo. What do you think the piece of content or format will be that remains loved in an age where everyone's taking the Waymo of video, the Sora video generation? What do you think is like the way to do that? Well, first of all, I got that NSX, and it lived up to all of the childhood hype.

1:21:16I mean, just incredible. That's amazing. that car is so fantastic um and i i don't know i kind of think there's going to be a lot of stuff like that for people that generated or not where you still you want the real thing you want the thing that you have the kind of childhood connection to uh you know someone like a kid today is not going to want the nsx but whatever the a cool car like that is they will want and And at some point, the fact that they can have a crazy VR experience, they'll still want the real thing and the connection to it and everything they have. So I think there will be a huge amount of that.

1:21:55In fact, I think the future looks like much more of that kind of stuff, not much less. How quickly do you want to create an economy on Sora? It feels like there would be a number of ways that you could create incentives for creators to create things, for IP holders, for individuals to just be passively monetizing their likeness.

1:22:18Bill, what do you think for timing on that? I mean, this is like a top priority for the team. You know, there's clearly such an incredible value proposition for celebrities, for rights holders across the board here. We think Cameo is like a great entry point for this, right? You can imagine right now we have Cameos for people. Maybe you have Cameos for like, you know, your character or like your brand or something. And so we're actively working on the team right now, coming up with like the right monetization model here to get this rolled out. But it's really important to us, right, that our creators on the platform are rewarded and that there are clear financial incentives.

1:22:53They're like the incredible work that they're already doing. So this is like top of mind for us. And we'll have updates here over the coming weeks. This is like something we're actively working on. I will. I think it's super important and awesome. I will say I would like to know how many hours of sleep Bill has averaged for the last few weeks, but I bet it's not enough. So we got a lot of stuff. The team's got a lot of stuff they have to do in a short period of time, and it's going to take a little while. Okay, let me put one more thing on your plate, Sam. I mean, earlier, like years ago, you built Looped, location-based product.

1:23:24Have you thought about how AI and location-based content fits together? Like on most of these social apps, you can tag a location. That wouldn't even make sense in the current SOAR app, but what does the AI Maps product look like? I haven't thought about AI and location that much, but I've thought about how AI can really change the social experience for people. We don't have a for sure answer yet, but we have a lot of interesting threads to pull on. And I have thought back to my days running that startup there more. My instinct is it is possible to make a very interesting new kind of social experience.

1:24:10connecting you to people, helping you find people that is intermediated by AI in an interesting way. But, you know, we'd have a lot of exploration to do there. What advice are you giving to startup founders these days? I remember in the GPT 3.5, GPT 4 days, it was like, don't build a company that assumes model stagnation. How do you think about in the age of Sora? That's been really great advice. It really has. It played out exactly like that. There's a bunch of great companies that aren't built that way, and they've done great. But if you were just, oh, I have a special prompt that tunes up GPT-4, yeah, bad times.

1:24:48But how are you thinking about it now in the context of video and Sora specifically? You obviously do have an API. You have DevDay. There's people that will build on top of this. Is it a different shape of the problem? Totally. The reaction to the API has been nuts positive. at least the fastest ramping revenue I've ever seen for one of our new models in the API. I mean, maybe there was something faster than I'm not remembering, but the demand there has been just incredible and people are doing awesome stuff with it. Bill and I have not had a chance for a one-on-one since launch because it's been so crazy.

1:25:22We're doing one later today. But one of the things I was going to suggest to him was that we, given how much excitement there is to build on this stuff, that we do something we don't usually do and put out our intended roadmap of the things we're going to prioritize because I can imagine really cool new startups that simply were not possible, that will be possible at each of these new things we'll ship. So I had a question when you guys released the Sora 2 via API, which was that if Sora has the potential to be an Instagram or YouTube-scale business, why release part of your edge for the entire world that they can integrate into other creative tools and then use the model to generate content that doesn't have a watermark, that's not in your feed, that you're not able to get that feedback loop on that you guys do with the Sora app.

1:26:16For ChatGPT, we also put out a great model in the API, and people can theoretically compete with us on ChatGPT, and some try to, but we are willing. We're never going to build every cool use of the technology, and we want the world to get all that stuff. we're delighted to also get paid on people using our api but like we just want ai to flourish out in the world we're not going to build every great use of what you can do with video models either we'll build one and i think it's pretty awesome but people have a lot of other ideas of business and products to go build and we'd like to enable those okay last question back to cars what's wrong with the porsche 911 yeah you said earlier the timeline was in turmoil you said if you're worth somebody said if you're worth 5 million when you buy a 911 you said no you agreed with PG what did you mean by that?

1:27:08I mean it was maybe it was like a tasteless joke it was kind of like late at night I was you know whatever but I have an unfortunate proclivity for expensive cars and the response was like would you ever spend 250k on a car and I took that literally that's amazing Hit the size gong for taking it literally No time For 250k cars Not necessarily That was not my best tweet I enjoy it now We all enjoy it I enjoy it now that I have the context Congratulations on all the progress to both of you Thank you so much for taking the time to stop by the show Really appreciate the update And very excited to see where this goes Cheers I don't think anyone read it No one No one got it that way.

1:27:54That's amazing. 250K? How about two and a half mil? Anyway, we have our next guest joining in just a few minutes. In the meantime, let me tell you about Linear. Linear is a purpose-built tool for planning and building products. Meet the system for modern software development, streamline issues, projects, and product roadmaps. We have a lot of You all know that OpenAI builds on Linear. Yes. Oh yeah, that's right. They're a customer. Well, we have a lot of Gil coming into the TBPN UltraDome from the Restream Waiting Room. Let's bring him in now. Thank you so much for joining us a lot. How are you doing?

1:28:27Finally. Thanks for having me. Good to see y 'all. We have wanted to do this interview since probably the very first week that we did guests. Took us a while, but you have your own show. Yes. But thank you so much for taking the time. What this week has stuck out to you? I'd love to just get a State of the Union on how you're thinking about the market broadly, and then we can zoom in on individual startups and trends and subcategories that you've been focused on. But have you had a reaction to this like bubble talk that's going on? Have you been thinking about this? How have you been processing the information?

1:29:04How do you even research whether or not when something is going viral like that? Yeah. I mean, I've been looking at this stuff for a while because if you look at the 90s as a sort of precedent or antecedent, the 90s internet bubble, I think there was something like 450 companies that went public in 1999. There's another 450 that went public in the first couple months of 2000. And so about 2 ,000 companies went public. And then you ask how many of those are still alive? Like how many survived? And there's probably a dozen, two dozen that are still up and running. There's probably two or three of those that are really important to Amazon, as an example, etc.

1:29:40And then 1 ,980 out of 2 ,000 probably died. Went to zero. And they were being priced on eyeballs. They're being priced on eyeballs, not even account creation. It was such a different time. Like IPO-ing was like doing your Series B. I remember there was a guy, Bill Gross, out of Idea Lab in Pasadena, my hometown. And I believe he took 100 companies public. And he ran an incubator. It was like the Y Combinator of the day. One company, I believe, was AdWords that went to Google and became the backbone there. And he had a bunch of great outcomes. But it was like a bit of a machine. if the IPO is no longer the metric that we should be watching, is it these revenue ramps?

1:30:20Is it churn? How can we dig into understanding where true value, durable value moats are accruing versus froth? We often call it the barnacle economy. If you're a toilet cleaning startup, but you have Anthropic as a customer and they 10X their office footprint, you 10X revenue, that's not exactly the type of business we want to see long term. Yeah, I think durability is a great question, and it's a really hard one for this era. For two reasons. One is things are changing so rapidly from a model and underlying capability perspective that if you look at every prior technology wave, you look at, for example, Microsoft OS, right?

1:30:58They forward integrated into the Office suite off of using Windows OS. Or Google forward integrated into vertical searches, so they killed a bunch of companies or really hurt a bunch of companies that are providing those services. So we should see the same thing with the foundation model companies, right? It's most likely if they're going to forward integrate, they're already doing it in code. Maybe they end up doing it in customer support and sales, sort of all the big categories. There'll probably be some effort at some point. Now, they may or may not succeed with that, which is a different thing.

1:31:23But there's durability in the face of competition from the big folks. There's durability in terms of like, well, people keep using your product or we get subsumed by a startup. And then there's just things that are just running up that clearly are never going to really work. and the real question is, you know, how do you identify each one of those classes of companies and how do you think about them as either a founder or an investor? Right now, we've been noticing there's sort of like three types of, three buckets of companies that are trying to like, you know, craft an AI narrative around whatever market they're in.

1:31:53If we're talking vertical markets, smaller markets, not the foundation model layer. You have the legacy Fortune 500 company, you know, career CEO in the seat who's maybe paying a consulting firm for some AI transformation plan. And maybe the stock's not doing so well. Then you have the startups that are, you know, we go to YC Demo Day. We talk to five companies that are building the same space. And it's their complete greenfield project. Amazing place to be. It must be super fun to just be puppeteering 25 cloud code instances and codex agents to build your thing. But then we talked to a lot of founders that have, they started their company five years ago, 10 years ago.

1:32:32They have a serious business, but they're still in founder mode. And we always find it a little bit hard to bet against those guys who are like coming back in re-energized. They have the balance sheet. They have the customers. And they can kind of take a second crack at it. Like what nuance would you add to that framework? Do you think it's on a per industry basis? Is it all about the founder? How do you think about that? Yeah, I think there's basically three viewpoints on that. I think there's a very small number of singular founders who just make amazing things happen, and that's Elon Musk, right?

1:33:02Like, who else would go to space and build cars and do all these things? Honestly, I think Arvind and Perplexity is one of those, where in anybody else's hands, I think Perplexity would be a dramatically smaller business. And I think most other companies in his hands would do better. He's very good, right? But that's very rare. So I kind of put, you know, amazing founder aside, because even great founders with their terrible market tend to get crushed. And so I think there's a second piece of it, which is there are a bunch of these AI markets that have recently really crystallized where, you know, I used to say a year and a half or two years ago that the more I learned about AI, the less I know.

1:33:34And it was the only market that I ever felt that way because, you know, you learn new stuff and you know more, right? You do better. You can predict stuff. And I think that changed over the last six to nine months where suddenly, at least for certain areas, it's really clear who the finalists are. We may not know the winners, but we know who the final contenders are. We know that for the foundation model market, we know it's Anthropic, OpenAI, Google, perhaps XAI, Meta, a few others, Mistral, whatever. But, you know, it's a small list. We know for coding, it's Cognition, Cursor, and then the foundation model companies and Microsoft.

1:34:06And you can go through sort of vertical by vertical. There's a bunch of verticals now we know. A bridge for healthcare and maybe Commure. You know, there's a handful for each thing. But then there's a bunch of markets where it's clear the market's going to be important. And there's tons of players, but we don't know who the winners are. That's financial tooling. Maybe that's sales enablement. Maybe that's accounting. You can come up with a list. Legal feels like it's already solidified except a handful of these more vertical, specific. Specifically, we were joking. There's people doing personal injury, law, agents, ambulance chaser agents.

1:34:44But it feels like the categories are solidifying. I guess the question I have is you've backed winners and basically all these categories. Where do you feel underexposed from an investment standpoint? What do you think you didn't quite anticipate? Is it like energy possibly? I'm sure you have bets there. But where do you feel underexposed? That's a really good question. I feel like the two big trends of this era so far have based it. And by era, I mean like two years. It's not a year. eras of recent last two weeks, you know, modern millennia last two weeks. Yeah, exactly. The last two weeks. Yeah, exactly.

1:35:22Um, I think that there's, uh, the two big things are basically defense and, you know, I was very early involved with Anderil and Leather D and of course, basically in every round of that company. And then I'm an investor in, um, Saronic and Helsing and then, uh, Kayla and Israel. Um, but very, very little defense actually over like a 10 year span of investing. I did an ender roll only for seven years because it was the only company that I thought was just going to keep going forever. I think it's a generational winner there.

1:35:51Then there's AI. AI means everything now. It means consumer. It means roll-ups. It means vertical fast. Natural gas, turbines are AI indexed. Exactly. Trains have to move GPUs across the country. FedEx, maybe. Throw that in there. Yeah, I love it. Yeah, FedEx is my favorite AI company. So I think those are the sort of two obvious trends now. Six or seven or eight years ago, they weren't obvious. Now they are. And honestly, obvious trends often go longer than you think. I remember with the social networks, there was a bunch that didn't work out. I mean, MySpace and Friendster, and eventually I had Facebook and LinkedIn and Twitter.

1:36:30And then at that point, everybody said it's over. Everything and social saturated. But then we had WhatsApp, and we had Instagram, and we had TikTok. And, you know, it just kept going. AI is in a much earlier version of that right now, where I think we're at the very, very early days of this massive wave. And so to some extent, I'm underexposed to AI in general, because I think it's the biggest thing that's happened in, you know, 20 years or longer. A lot of guilt says he's underexposed to AI. You're very, very humble. The chat says everything important, a lot is in. What is, what is, what's not AI?

1:37:04When do you get a pitch where they use the letters AI on their website a lot, but you're just like, hey, guys, this is just regular enterprise SaaS? Well, I actually bring this back to what's durable in the face of AI. And a good example of that is Rippling. Rippling is an amazing company. They cross-sell a dozen different HR products. And AI can make some stuff better, but nobody's going to do the AI first Rippling and suddenly win. right now the the main threat maybe to a rippling or deal or sort of related companies is if company headcount actually goes down because of ai then they have fewer seats they sell right and so that's maybe how there could be a headwind from ai for these companies but but maybe that means more companies maybe that means you just get a lot more companies smaller teams right yeah it's quite possible yeah so i just think like that's the when you see that you're like okay this is very durable in the face of AI, and that's great.

1:37:58And so a lack of AI means AI can't displace it. And so as long as it's working, it's actually more interesting in some ways. Yeah, the question I sort of ask myself is, using a company like Rippling, for example, even if companies start needing less people, I could see them transitioning to kind of value-based pricing around, what's the value of running your HR department? right is it a three percent of revenue is it five percent of revenue is it one percent of revenue like either way they're going to make money if they're providing like infrastructure for that function yeah it's a great insight because i think one there's two or three things that are underappreciated about this ai wave um i think that the first thing is that um the the capability set has shifted dramatically not just in terms of what these models can do but the fact that you can just ping them with an api and something's accessible to everybody and i think that's actually very under discussed right relative to the prior world i think a second thing is that the markets are oddly open like legal never bought any software it was really hard to sell into legal but because of ai suddenly harvey can exist right um and then uh the third thing is that uh a lot of this is about what you're saying which is the tams of markets are shifting from seat-based pricing or seat-based value to labor you're replacing human labor and so you're looking at for example customer support.

1:39:22It's not Zenda's, how many seats can you sell the customer support reps? It's how much can you augment and do work for customer support reps? It's the labor market versus the software market. And I think that's very underappreciated when you think about market size for some of these things. You're really going to miss the size of these markets and how big they are. The services economy that we looked at on my team in terms of where AI could intervene is about$5 trillion. dollars right so it's a lot of GDP is accessible to this and so then you ask okay is it is it a workflow that's specialized to customer support legal whatever is it a roll-up where you're buying assets and changing them is it a different approach like how do you sort of span all the change that's coming because of this uh Ken Griffin gave a talk earlier this week and he was saying that in 1999 and 2000, it was very obvious that the internet was going to change the world, change the way that our economies run, yet it still took 15 years for it to actually have an impact.

1:40:22And he was comparing that to today. The difference of today is that we have the internet, so we can deliver these products instantaneously to the entire world. Do you think that this time can be different and we can, as an industry, unlock the value of this technology on a shorter timeline than the internet took because we just didn't have, you know, the internet is the greatest distribution engine in history. Yeah, that's an excellent question. And I think both things can be true simultaneously, which is we're seeing real revenue for these companies, right? Cursor is rumored to be in the high hundreds of millions of revenue.

1:41:00You know, Azure added something like two or three billion of AI revenue per quarter from sort of a cold start two or three years ago. Right. That's amazing. That's like$10 billion run rate plus just off of AI revenue. Right. So it is working. It is being adopted. But the flip side of it is it'll probably take a decade. Right. And so I think both of those things are true. And I think the biggest impediment to adoption isn't the technology. We could do so much stuff with the technology right now. It's organizational process. It's workflow management. It's all the stuff that happens when a big enterprise uses anything.

1:41:33And they're like, you want me to change my tooling? You want me to change my people? You want me to, you know, my processes? And that's what's going to slow it down. And that's slowed down every technology wave. But to your point, we have massive distribution. It's already everywhere in some sense, right? I don't know that you guys probably know the number. You just talked to Sam Altman, right? What's the number of people using ChatTPT per month? But that's a huge impact already. Yeah, it's interesting they haven't said that. It has to be north of a billion because they're reporting 800 weekly active users.

1:41:59If you have 800 weekly, you have to have over a billion. And that feels like such a news cycle. But maybe he's just keeping it in his back pocket for when he needs a good news cycle. Bit of a wild card question. And I didn't plan this, so I didn't mention it beforehand. But no, it's not bad. But I just think it's interesting. If you couldn't be an entrepreneur and you couldn't be an investor, what hyperscaler would you want to work at? Be an executive at? That's interesting. Oh, that's so interesting. I don't know. I could make arguments for two or three of them. because I think there's such different problems to be had and different assets.

1:42:34Google, for example, just has such amazing assets relative to this era. They have the most data. They have the most compute. They have amazing cash flow. I mean, they're just in a great spot. Deep mind. It's a really crazy stack over there. Yeah, amazing stack. So there's amazing things they could do. Obviously, there's crazy stuff Microsoft can do on the business side, plus with GitHub and Copilot and everything. So they should really be driving a lot of the coding future, in my opinion, if they make the right moves over time. And so you can kind of go through one by one and say there's really interesting things to be done with meta in terms of social.

1:43:12I think there's opportunities everywhere. Give us an update on AI roll-ups. You have some investments here, from my understanding, but how do you see the category evolving? We see new teams coming together to attack various markets with this strategy almost daily now. But what's your view? Yeah, I think so. It's back to if you look at services in the US, it's three and a half to five trillion of it is sort of labor that to some extent could be augmented or displaced by AI. And so the idea is, there's certain types of companies that are going to be very slow to adopt software or AI. And so there's two things you can do with that.

1:43:58You can wait and build a software company that will take a really long time. You can actually buy those companies, implement the AI changes, and dramatically change their margin structure. And that doesn't mean letting people go. It could just mean you make people five times more productive for certain types of roles. So you can look at different businesses where 80 % of the cost of that business is repetitive white-collar labor. And so you can help augment or automate stuff for people. And so I looked at a few dozen teams or companies doing this. I ended up backing two of them. And really, you need three things to make this sort of strategy work, which is truly transforming a business with AI and then scaling it up.

1:44:33The first is you need a great AI person, obviously, right? You need to be able to implement the technology. Second, you need a great PE person. You need to buy assets properly, understand your envelope. Just like a SaaS company has its ICP or customer profile it goes after, you almost have your M &A profile. Like, what fits in my pocket of stuff that I want to go after? And then lastly, you need somebody who's operationally great, who can rework the organization against the AI, because that's often the hard part. You actually have to get people to use this stuff in order for it to be implemented.

1:45:01And so very, very few of these teams have all three of those things. And many of these teams are basically doing traditional PE roll-ups. They're not really using AI, but they're raising at AI prices, and then they're buying at PE prices. And so there's arbing. And so I've tended to avoid arbitrage. Yeah, it seems like a great deal with the founder. In another world, they'd be doing a private equity fund with 2 in 20, and then here they can go out and raise 20 on 100, and then they own 80 % of the business. Yeah, remarkable. Yeah, it's a very smart thing to do. And if I was a PE person, I'd totally go do that instead of doing AI.

1:45:36but most of these things aren't doing AI, right? But a handful of them that are going to be massive, right? Imagine an AI denner, right? It's just, it can really be transformative to big sectors. And it changes something from a services margin to a software margin business with software leverage. So it changes the characteristics of the business. Sort of flashing back in your career, one of my questions I've always had on my mind is that you kind of like created the solo capitalist idea. people have kind of, you know, put that label with you. Was that just a happy accident? Did you, were you deliberate that you didn't wrap what you were doing in a firm?

1:46:12There's obviously people that start with similar scales, but wrap it in a firm with a brand and, uh, like how thoughtful was that? What were the considerations? Do you like what, how that, yeah, I think, um, you know, for a while it really was just me. So it wasn't some strategic move to, you know, do something. It was just, I was on my own doing stuff. And I, you know, um, and then eventually I brought on people for back office and finance. Cause like, I think that's really important, right? You want compliance, you want things to be proper and all that. Um, uh, and then I got this moniker and I never asked for it, right?

1:46:44Like I actually am happy to be called whatever, as long as I get to be involved with the most interesting technology and technologists in the world, you know, they could call me, I don't know what a carpenter. I don't care what you call you big VC, big venture capital. That's the one thing I don't want to be called. Other than that, I think what we do is different from traditional VC, too. We do traditional investing and we do traditional venture, but I actually think we do a bunch of other stuff and we do interesting projects around things like one person on my team who's working as an investor, as a technical background, is actually driving AI -driven translation of the world's thousand most important books that are off of copyright.

1:47:23And we're working with a few really big foundation mobs on that. So we do stuff like that too, just because it's interesting. So I hope it's never just a venture fund. Do you think, when do you expect AI to transform venture capital? I think it's notable that the firm today and the activities of the firm look quite similar to pre-ChatGPT. Maybe you can write an investment memo faster. Maybe you can seem like you prepped for a board meeting better. if you just drop the deck in and ask for a summary. But it doesn't feel like it has changed the profession at all yet. It's still finding and winning allocation.

1:48:09Yeah, I think it can help with some aspects of diligence, to your point. Like it can pull competitors and things like that. To some extent, it depends on the market. There may be weird uses that nobody's done yet. So an example would be, have you ever done the prompts where you ask the AI to like cold face read somebody and tell you about their personality? No, you up for this? Yeah, I mean, just analyzing somebody's personality with a picture, even just any picture of their face. Yeah, it's pretty, I mean, this is like the schizo is called his physiognomy. Sure, sure, sure. Yeah, yeah, yeah. But there's stuff like that where I've done that just for fun with my friends, right?

1:48:46I'm like, hey, what is this person like? And my friend's sitting there with me, right? I'm not secretly trying to psychoanalysis or something. And then I'll ask it to give me a detailed breakdown of those characteristics and why. And I'll say, oh, this person looks like they have a genuine sense of humor and they're warm because of the way that the crow's eyes around their eyes exist in this way versus somebody who fakes smiles because it doesn't get to the eyes so there's no wrinkling. And you're like, wow, that's actually like super interesting, right? And so it'll break down the sense of humor.

1:49:13It'll break down how likely that person is to be loud or quiet, the likely aggressiveness, all this stuff. Yeah, you can imagine somebody could create an EQ co-pilot for the new meta display glasses. You can just walk around. And if I'm really high IQ, but I don't read people that well, I could look at John and say, which is probably inverted. But I could look at John and be like, oh, John's very interested in the conversation. And he clearly wants to be friends. So I think there's more to build there. Give us the update on your company with Jared Kushner. I know he's been very busy, but I'm excited to hear the latest.

1:49:56Oh, sure. Yeah. So we recently launched a company that's called BrainCo, which is focused on using AI as basically a platform to help transform the world's largest institutions. And so we've been working with a number of large enterprises, some private equity, and other firms around this. And so this has been a fun project to do with him and with Eric Wu and Louis Vittigray and a few other people. Eric was the former CEO of Opendoor. And then Louis is the former finance minister and foreign minister of Mexico. So it's kind of this interesting group of people coming together to try and solve really big AI problems.

1:50:32So that's been really fun. So is this realizing how much the Accentures and the McKinsey's of the world were getting paid to make pitch decks on basically, here's what you should know about AI for your business and realizing, hey, we could probably do that a lot better and then ultimately build software within these organizations? Yeah, it's much more focused on the software side of it. So basically, there's a common platform that's involved in terms of dealing with different forms of data, dealing with evals, dealing with a lot of the things that every enterprise needs to build in order to really adopt AI.

1:51:04And then we build vertical-specific applications on top of that, or in some cases, horizontal applications that can be reused over and over by similar companies in the same vertical. So you could imagine, for example, for a financial industry company, there's like a dozen things that every single one of them needs to build, and there's some customization around it. But it's kind of funny because if you look at very large deals, right, if you're Dell or your VMware or your Oracle and you do like a tens of millions of dollar deal with a customer, you're going to have customization against that customer.

1:51:33You can afford to do it, but also it's important enough to do that for them. And so it's similar in that regards where we have common infrastructure, common platform, the same vertical applications, but then there's going to be some customization per client just like any other giant enterprise company. Have you thought about slicing the target customer either across vertical, like we're not doing healthcare because of HIPAA just yet, or we're not doing defense because of FedRAMP just yet, or we think we have a lot in industrials that we can go after, or are you more focused on slicing by market cap or size of business?

1:52:09Like, yeah, we're not going to work with mid-market companies. How do you think about where the wheelhouse customer will be? Yeah, it's very much, the goal is to work almost solely, and there's going to be some counterexamples of this, with companies that are truly the world's biggest institutions in terms of revenue, market cap, and then potentially impact, which means sometimes you work with somebody a little bit smaller. But the goal is to ask, how can you use AI to really get leverage on things that are important at sort of a massive scale? Yeah, there was a report that JP Morgan is spending$2 billion a year investing in AI to effectively save$2 billion a year from automating.

1:52:49It's like when there's that much money. Yeah, yeah, they're breaking even, basically. I mean, presumably, like, some of those savings are durable, but certainly there's a lot of money. What do you – how good do you think you've gotten at clocking AI pilot revenue as non-durable? Because I'm sure you saw that headline. An MIT study came out that said basically 95 % of pilots are not actually delivering in value. This felt like the year of the pilot next year is maybe the year of reality. But I don't know how you see it. Yeah, I just view that as a standard technology cycle. I thought that was a very overstated article.

1:53:31So, you know, this happened with mobile. You do the kind of crappy mobile app. I don't know if you guys remember the first B of A mobile app. It was basically like a, yeah, it was just like a web page. Yeah, but everyone was just wrapping it. Well, the current B of A app is still pretty rough. I actually think it's pretty good. Like, you can go to an ATM, you can take out cash. I think I'm the only person who still does that. You can take pictures of checks. You can pay with Zelle. You can do all these things, right, that you couldn't do before. But you started off and you tried to log in and the thing would crash.

1:53:59So I just feel like we're kind of in that era of AI implementation. People will get it. It'll take a decade to fully sort of propagate to your earlier point. But it really is the big wave that we're living through right now. And I think it's truly transformative. Would you ever, have you ever considered an LBO of any of these companies that are in the SaaSpocalypse in the public markets that clearly have some super meaningful customer relationships and a ton of potential important place in their market, but just aren't evolving their business model quickly enough? Yeah, we've looked at that, actually.

1:54:36I think there's a few really interesting things to be done at scale. And to your point, I think part of it is just driven by, can you actually implement AI? And I think if you look at the private equity industry in general, they've talked a lot about tech transformation. That was a prior wave, right, like 10 years ago. And you had all these tech-based roll-ups, like Compass was supposedly a technology company. And it's a great company, but it's not really a tech company, right? And so I think we've kind of lived through this cycle before where people do almost like fake tech implementation, where they claim they're doing it, they get a higher valuation, and it doesn't quite happen.

1:55:09They still load up the company with debt. They still run it a certain way. The drivers are different. The CEO is the wrong person, et cetera. I think there's a good version of that to be done. I don't think it's easy, but if you do it, I think you can unlock an enormous amount of potential in companies that won't make it otherwise or won't go there. And so, yeah, I think that's a super interesting area. And we've looked at a few things over time. And part of the decision sometimes is like, do you want to try and do that or just fund a startup that you think will get there instead? And it's a little bit of that.

1:55:38It's kind of easier to just find a startup because buying a company and transforming it is quite hard. Yeah, it has to be a company that is at a sufficient scale. Obviously, we just saw that the EA LBO, you don't need to go that big, but finding something that's really a whale. Can't hurt to go that big, though. Yeah, we would like to see you go that big. I would love to see you. I feel like that's the next. I feel like you've done it all at the early stage. You've done it all in growth. if I can just get into the... How about an Activision spin out? Activision spin out, take Microsoft Game Pass, Xbox.

1:56:14Sorry I've been lagging. I know. I'm moving too slow. I know. You're like my mother. You're putting all this pressure on me. I know. That's our small ask is set the record. Wait, wait. Do you have any... We like to ring the gong for people around here when they share a number. Do you have any number you can share? What's your favorite number? Number of deals, number of AUM, number of... I don't know, anything. What's a number that quantifies your corpus of work? number of startups founded my corporate i've started um well i've started two companies directly and there's two i've incubated so far that's a lot of businesses relatively modest amount for the amount of ev created yes it's good good hit right well thank you so much for stopping this is super fun this is a lot of fun really really glad to do it again we'll talk to you soon have a good day really appreciate it take care uh before our next guest joins let me tell you about numeralhq.com.

1:57:04Sales tax on autopilot. Spend less than five minutes per month on sales tax compliance. I hate to pick favorites. I hate to say it. I was more engaged for Alad than - Alad was great. Alad was great. But his book, High Growth Handbook - Yes. Fantastic. It's a series of interviews. Yeah, it is my favorite business book. High Growth Handbook, Scaling Startups from 10 to 10 ,000 people. It is timeless. A bunch of great interviews. And yeah, you should go pick it up. I need to re-listen to it. Well, without further ado, we have Robbie Stein from Google in the Restream Waiting Room. Welcome to the show, Robbie.

1:57:49How are you doing? Good to see you. Hi, can you guys hear me? We can hear you. We can hear you. I know people are going to be confused. So why don't we just get out in front of it and explain your role at Google relative to Gemini, relative to AI? Take us on a little Game of Thrones HBO intro of the map of Google and where you fit in. Sure. Yeah. So I work on the Google search team. So it's the search bar. You type things in. Yeah. I was going to ask, could you explain Google? Google. Yeah, it's a thing. You can type a question in. Sometimes you could use a camera. I don't know if you've used Google Lens.

1:58:27Oh, yeah. You can also Google that way, too. There's this app. It's also called Google. All that kind of stuff is where I focus my time. Consistent. And don't forget about the I'm feeling lucky button. Yes. It's still on there, right? Yeah. Yes, that's one, too. So how do you think about integration with the DeepMind team, integration with the Gemini team? How do you think about bringing AI to search? Yeah, so we work very closely with the Google DeepMind team, Demis, Cori, that whole group. And what we think about is we want to have frontier models right in search so you can really ask anything and have the ability to use all of search's incredible knowledge, real-time information systems, and context of the web to help give people this incredible information.

1:59:11And that's really where the two come together. Yeah, there was this question, this like, you know, chattering class being like, oh, well, Google's, you know, it's such a different paradigm chat versus the search, 10 blue links. But have you drawn on the fact that Google's search started with two modalities? The I'm feeling lucky button was a different way to interact with Google search. The 10 blue links was one way. How are you thinking about how you kind of level up or educate the consumer to use all the different tools? Since now there's not just search results and AI overviews, but there's so many different things.

1:59:50How do you think about ramping those up? Yeah, so there's always been many ways to use search. Actually, Google Lens is a good example. You can take a picture of something. You can ask, what is going on with this plant that seems to be dying? And you can get information from that right from the camera. And that happens in the app. So there's always been different ways that you could access and tap into this search knowledge base. But I think increasingly it feels like you want to go to search, ask whatever you have in mind. You only want to think about where you're going to ask that question. And then if you have something where AI is really helpful, you get this little AI preview.

2:00:19It's starting to show up now around AI overviews. And if you click into that, you're in this AI-driven experience. And we've now, through a new project we launched called AI Mode, allow you to follow up, go deeper, and have a full generative end-to-end chat-like experience right within Google Search. Yeah. Walk me through the international expansion. I'd love to know if legal or engineering is more rate limiting there. It feels like it's got to be incredibly complex to check all the boxes when you want to expand. You've obviously expanded really fast. So what's the secret to taking over the entire world so quickly?

2:00:57Well, our newest product, AI Mode, which is the thing that lets you really ask anything within Google Search using frontier, state-of-the-art models. We launched in the U.S. and then quickly in India, a couple countries, U.K., over the summer, May, June, July timeline. And then just a couple months later now, we're at over 200 countries, 40 languages. We moved really fast to ship that. And so we're now basically everywhere except for a couple countries. And there's certainly a bunch of considerations, policy and infrastructure. Honestly, they both affect the timeline. Oh, yeah, because there's actually inference going on.

2:01:33So you need local compute, basically, more than you would. There's also, we want to make sure to do quality checks because in every language, you have different permutations or you have multilingual people moving between English and other languages. You just want to make sure everything is doing what you expect it to do. You know, this is also a multi-turn conversational experience. So when you evaluate it, you want to make sure that it's dialed before you go. And so each of those just takes time. I know everyone's always frustrated, like, why can't I use it today? It's the main thing I see on X.

2:01:59But hopefully now almost everyone can. How are GPUs allocated at Google? You have TPUs. You have TPUs in the cloud. You're selling them now. Obviously, DeepMind wants a bunch for research. Every team member just gets a rack. Everyone gets one rack. You're split equally. But I imagine that, you know, is there a spreadsheet? Like, what's the process to actually figure out where to allocate GPUs or TPUs? Yeah, I mean, there's an allocation process. I mean, everyone needs TPUs. It's just compute is by far the most important thing in driving infrastructure right now. And there's lots of needs, but they're important things.

2:02:36We want search. The search is one of the largest ways people interact with AI. So that's a very important one. Obviously, we're doing frontier modeling work. We've got a bunch of things happening in cloud. So there is a process by which you do those. It's funny, I did an event and a Google team actually gave me like an old Ironwood, like prior gen single TPU in this little case. That's cool. And I was joking that if somehow I could like crack it open and like refurb it and like get it up and running, I could somehow like improve my standing on the team. But I don't think it's going to happen. How often do you guys come back to Google's mission when thinking about product decisions?

2:03:15We've talked about this on the show before. There's always, you know, I'm sure you wake up every morning, there's a new headline. What is Google doing in AI? Why haven't they released this faster, et cetera, et cetera? But I, you know, we were never, we never really stressed about it too much because when you look back, I have an AI overview here. I said, what is Google's mission? My AI overview says Google's official mission is to organize the world's information and make it universally accessible and useful. And it feels like LLMs are just so aligned. It's the perfect technology to carry out the mission.

2:03:51And so it's like, relax, everyone. I think this is like a natural evolution for the company. Yeah, I mean, we talk a lot internally, actually, about how the mission has never felt more relevant. And it is really incredible to work somewhere. I worked at Google in 2007 for a while, did some startups, and worked in some other companies, too. I'm back now. and it feels that same level of entrepreneurialism in 2007 where it's like you're building all these new things for the first time, except you're kind of building them again because AI allows you to say, what does search look like if you could really ask literally any question and create an AI that's the most knowledgeable AI out there that could understand all of Google's information, the context of the web, and you could talk to it.

2:04:32And by the way, you could talk to it live. We just announced Search Live that's available. So if you're driving, you could just talk to Google literally in your car. You could take a picture and have this multimodal conversation back and forth now. This is all stuff that I talked about a long time ago, but was limited because of technology. So that is incredibly exciting. It's very motivating. And I do feel like it's one of the reasons people on the team are fired up. How do you think? I mean, it's so interesting that as LLMs boomed, AI search overviews made a ton of sense, were baked in, adopted, loved.

2:05:04But now we're already in the next phase, which feels like agentic purchasing, agentic checkout. How do you think about evolving the product even further to go from knowledge retrieval to taking action? Yeah. We have a bunch actually active there. And we recently launched in the U.S. an agentic experience where you can book restaurants and local services through agentic. So it's actually really neat. Like through AI mode now, you can just have a conversation. Hey, what's a good date night place? It'll do all the thing. It'll tap into the knowledge of Google. It'll look up Google Places. It'll do research.

2:05:38It'll do whatever. but then if you want to start a task it'll also go bring back availability across talk and open table like right in the experience and you can just book it and it's awesome yeah i've been using that it's just a small example of what's possible you know people come to google not just for information but they get things done it just materializes itself as a query but you're not like i don't like want to just know restaurant reservation availability and like oh sweet i'm glad to know there's a table available i'm gonna go back to my day like you're trying to do something we think about that a lot and that's just the you know the surface that there's so much people are trying to do but they're just kind of getting started with those journeys on search what could we do to really help you shopping's a big one um so we haven't we just launched a new visual uh way to do ai which i feel like is one of the first ways where ai can be helpful with inspirational tasks so now in ai mode if you ask to design a bedroom or look up landscape lighting it actually finds you beautiful inspirational imagery and products in a grid you could click on it, you can imagine getting much more help finalizing those purchases, being reminded of price changes.

2:06:39And it's much more of this interactive version of search that can do things for you and you can really connect with versus in just a pure informational experience. Right now, my mental model is like Google.com, the search bar, then AI mode almost as a vertical product like flights, like images, like shopping, where it's sort of a portal or sub product that I get, you know, I go down a funnel and I wind up in. Do you see that holding? Or do you think that AI mode acts as a wrapper on top of all the different sub search products? Yeah, I think what's going to happen is you have this AI mode, which is going to hopefully be this most knowledgeable AI possible.

2:07:28It knows everything in Google, billions of products, millions, hundreds of millions of locations, all of the web, and it has access to all of it, knows how to use Google as a tool, and it's super powerful, but not necessarily the best thing for all things. Like if you just need a specific phone number, you'll probably get that in like 50 milliseconds right at the top of the page. You just want to know a sports score, what's literally the sports score right now? Just works. You just, you know, you put it in Google, or if you're just looking up a musician's name for the first time, you're trying to get a browsier experience.

2:07:55Like you actually want to kind of see images. You want to see what's going on on, what are people saying on X, which shows up in the search page. So I think what happens is you have this incredibly knowledgeable system that we feel like is designed for more complex tasks, more of this, like, how do I do this? What's my advice for this? I'm doing this trip. I need this restaurant. I'm trying to buy some jeans. How do I get started with that? And that if you have knowledge baked into that, that's really powerful. But then people need, how do you, how do you bring that to people? And there's basically two ways.

2:08:21One is you just search and through AI overviews, we will show AI where we think it's useful. And for many queries, it's not useful, actually, which is why it doesn't show up. The system learns that. But for these longer queries, we have a specific question. It typically shows up there. Then the other way is for power users, we feel like they kind of have this mental model of like, oh, I'm doing this planning thing. Oh, I'm really curious to know what a stock price difference between these three stocks are over some period of time. Usually, you wish you could just type that in natural language and have the thing generate a chart and look at use google finance as a tool which it will do and for that you can go right to ai mode so if you can do that through the mode you can do it on mobile through those direct um kind of buttons to go right to ai mode you can go through chrome now we announced like a way to just type and go right to the i mode in chrome and um you could also just go to google.com slash ai now which is kind of fun what anomalies are you seeing uh have you seen any of these screenshots of uh of Google trend data.

2:09:21People are posting, you know, people forever have been posting screenshots of effectively search data and using them to infer what's happening in the world, right? And so over the past few months, we've seen people posting search queries like help with my mortgage. And and it's just like a crazy ramp up and to the right. My intuition is that it's potentially a bunch of different keywords are seeing these crazy ramps because there's potentially agents sort of like leveraging Google search to drive a higher volume of searches on potentially a search that's happening maybe in an LLM. Have you seen any anomalies there?

2:10:09Is that something you're aware of? I haven't personally seen any anomalies there. I haven't really heard of that before. So I wouldn't buy too much into that. We do have protections for those kinds of things. I think in general what we're seeing is people who are using Google very differently and at a very fast growth. So people are asking very specific questions of Google. They're using Google Lens and asking multimodal questions. They're asking follow-ups. I mean, those are kinds of things that we're just seeing. We're seeing in such a broad way that I think that's the thing that we mostly focus on.

2:10:48But yeah, I can't speak more to some of the trends things that you're mentioning. Yeah. How do you think about advice for brands? I've run brands and, you know, grinded my way up the SEO rankings. I never did a ton of optimization. Most of my strategy was just try and make something eventually that people talk about and it gets written about. And those websites have, you know, ranking power. And then you kind of rise to the top of that keyword. Is any of that changing in terms of in the AI shift? What advice are you giving to brands that want to perform on Google these days organically? But what's really interesting is I think the core Google search ranking is more relevant than ever.

2:11:35Because it turns out that one of the best things that every AI model does now is they kind of like search the web. It's like, I don't know, like what are you trying to do? Oh, I don't know. I'm trying to figure out if I should go to this hotel. All right. Well, like it's not parametric knowledge. Do you like know the rates of those hotels? Like, okay, I'm going to search the web, right? And even AI mode is special in that we're Google. And so it uses Google really effectively and it creates these query panouts. It does dozens of queries, but effectively is Googling stuff, right? And everyone's Googling stuff all the time.

2:12:05So for a given question, what are the things that show up as highly relevant to a given question? Those end up getting absorbed into the context window and have a high probability of being displayed to the user. And so if you are just thinking, how do I build great original content that's trusted and authoritative for specific kind of query, turns out that's still going to likely be very valuable. And you can go read the Google guidelines on content and human radar guidelines. I think they're super interesting. There's lots of detail in there on how Google evaluates trustful, high-quality content and scoring systems that are used.

2:12:41It turns out that's a really good investment because every other system is kind of proxying for the info that's most useful for a question. So that's my main advice is to kind of like dig in more on understanding how those systems work because it's largely going to apply later. And then the second one is what are people using AI for? Those are the growing kind of needs. Oh, sure, sure. And there's a disproportionate amount of use case in different types of domains now with AI because it allows complex needs. And it allows for things like advice. It allows for things like really nuanced troubleshooting with stuff.

2:13:12It allows for these emotional needs to be satisfied differently and mental health. So those are areas that are probably growing markets of use cases. And I would be a student of those as well. Sorry, can I have one more on that? So what about content that was previously buried to Google that is now available for search rankings because of artificial intelligence? I'm thinking about like literally this show, it will be a three-hour video on YouTube that, you know, a buried mention previously probably wouldn't be perfectly translated, indexed, et cetera, et cetera. But I would imagine that the strength of like video content and audio content gets relatively better over time in the AI era.

2:13:59I mean, honestly, you guys have been doing this for probably a decade. But take me through a little bit of that. Is that a reasonable thesis? Actually, it is a reasonable thesis. We have been seeing an increased diversity of the kinds of pages and sites that show up within AI. Because people are asking these nuanced questions. A lot of times there isn't even a single web page that has this information. But if someone, I don't know, sometime in the future, this video gets segmented and scanned and understood by Google. And someone asks for, I don't know, an interesting conversation on Google's AI journey, recreating search for the AI era.

2:14:30Oh, sure, sure. Maybe it's like, here's a cool conversation you could check out. And this clip will show up now in a way that would be very difficult for that to happen unless you search for like TVPN interview of whatever, which was like very specific keyword. Or we'd have to go manually create a transcript and then get that into like the text SEO world. So yeah, very different era. Sorry. What are the plans around giving, if there are any giving publishers, control over whether or not their content is used in various AI functionality? Yeah. We have a bunch of publisher controls we have. So there's an overall opt-out training side.

2:15:11You know, in search, you can opt out of crawling. You can opt out of this other thing called snippets where you can show up in Google, but you won't show up in these rich experiences if you want. and so there's a bunch of things that publishers can do and can look at there but I think ultimately the belief I have is that to the prior conversation AI should be this massive discovery engine over time because if you think about it, these complex needs we're getting growth, like we're seeing 10 % growth for instance in large markets like India and the US for these really specific questions, which at Google scale is like enormous, enormous number where if you have a really specific question, people are doing that more and more and more and more because you can get AI to go deeper.

2:15:58Multimodal, right? You're asking, you're taking a photo of something and you want to shop it. I'm seeing 70 % year-over-year increases in those kinds of questions. And these are billions and billions of queries. So these are huge numbers. Wait, how many queries? Billions? Billions. Yes.

2:16:18The sneaky. I'm sorry. That was totally worth it. I'm glad you did that. Thank you. So like these are big numbers and each of those generate AI experiences with links to go deeper on stuff. Every single one of them. And so theoretically, you should have this like unique opportunity to say, oh, did you take a picture of your bookshelf? What book should I read? Oh, well, here are cool reviews of other books that are like things you've liked. It's like I could never Google that before. And so theoretically, you should have, because it's an expansionary moment, we're seeing that this is an expansion more than anything.

2:16:52Like the way people Google, they're Googling, and they're using it in all of these new ways. And so hopefully over the long term, that produces growth. I have a lot more questions, but we're coming out of time. Yeah, we could go way deeper here. I would love to even just dig into how TBPN shows up on Google. It's such a fascinating content because we create so much content all over the web. But thank you so much for taking the time to hop on the show. Thanks for having me. This was fantastic. And congratulations on the progress. We'll talk to you soon. Thanks so much. Cheers, Rob. Have a great rest of your day.

2:17:25Before we bring in our next guest, let me tell you about Fin.ai, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bake-offs, number one ranking on G2. And we have Morgan Housel in the Restream waiting room. We're going to bring him into the TPPN Ultra Dome. We're very excited to catch up with you again. And one of our earliest and just most enjoyable guests. I had such a fun conversation. One of my favorite conversations ever across a thousand plus interviews. Remarkable. So thank you so much for taking the time to hop on the show. Especially on a huge, huge week.

2:17:56Yes. Congratulations on the lunch. Big shoes to fill. Thanks, guys. Nice to see you. Yeah. Give us the update. Set the table. What's going on this week? Get that gong ready. Get this gong ready. Yeah. So my, my, my third book, the art of spending money came out this week on Tuesday. You know, it's a tough thing with, uh, there it is. Thank you. Thank you. It's a weird, it's a weird thing with books. I think, I feel like it's almost like a startup where I've been a writer for 20 years, but you get like three shots on goal when, when a book comes out and when, when, when, when you're writing a blog post, if it sucks and at times they do, there's always next week.

2:18:32It's not, not that big of a deal. When, when, when, when you write a book, you really got to be like, this is, this is it. You really got to put your best foot forward. So it's always a stressful thing to go through. Was the process different for this book from the other ones? Do you, we've heard from authors who like go lock themselves in a cabin. Like what's your process like and has it changed? It's usually roughly this. It's about a year of very informal noodling where it's like, I'll be going for a walk and I'll be like, oh, that would be a cool chapter and I could use this story. It's a year of that, like just no effort.

2:19:02And then three months of part-time writing and three months of full-time writing. during the last two weeks of the full-time writing, the world does not exist outside of my keyboard. That's usually how it works. Yeah. And are you like combing like front to back of the manuscript like every day? Are you focused on like a single chapter for a full day? Like how do you actually like chop through like what becomes a full book? It's usually the latter. There's not much going through everything end to end until the very, very end of the process. So it's usually just focusing on one chapter. and very roughly, this is not a hard and fast rule, but when I was writing it, it was like, I wanna focus on one chapter per week.

2:19:39And normally, I'd say if there's one thing that I've gotten a little bit better at over the years, just a little bit better at, not perfect by any means, is that I think my first draft is closer to my last draft than it used to be. Not necessarily because I'm a better writer, but because I'm better at knowing what writing is not going to work and then stopping it very quickly. And so I think when I go through a chapter, by the end of it, I can go back and scan it and be like, oh, this is pretty good. I'm gonna set that aside. And then when I do my little self-edit at the end, that's usually where most of the work happens.

2:20:08So it's much easier to write a book now that we have AI? Yeah, what was your prompt? I'm kidding. I'm kidding. No, but here's the thing. Share the prompt. I've talked to you. Just share the prompt. Just give me the prompt. Yeah, come on. Share the prompt. Share your prompt? Come on. Just give me the prompt. I was talking to a guy the other day who just finished his manuscript. And he said without ChatGPT, he would not have been able to write the book. And I'm like, I love hearing that. I love that if we can have a tool now, we're going to have more books being published. Because I think there are people who have very good ideas, who have a story to tell, but are too intimidating to write a$50 ,000, a 50 ,000 word manuscript, which is not an easy thing to do.

2:20:48So if we can just get more people out there because they have ChatGPT to get them through writer's block, awesome. I think it's wonderful. I'm still old school because I've been doing it for long enough that I want to write every single one of my words for better or worse. Even if I could have done a better job with some LLM getting me through those blocks, I'm still going to try to power it through myself. I'm sure you saw David Simon, the creator of The Wire. There was a screenshot from an interview. Someone asked him, OK, you've spent your career creating television without AI. And I could imagine today you thinking, boy, I wish I had that tool to solve those thorny problems.

2:21:20David Simon says, what? and the interviewer says or saying dot dot dot david simon goes you imagine that and the interviewer goes boy if that had existed it would have uh screwed me over simon says i don't think ai can remotely challenge what writers do at a fundamentally creative level and then he says i'd rather put a gun in my mouth there you go yeah just get right to it here's what i think i've always thought that writer's block is actually a symptom of your idea sucks sure and It's not working. And the reason you can't find the words to get through is because you know in your soul that this idea is not working.

2:21:56Yeah, you fundamentally don't care about what you're writing about, right? And I think when I think about the points where I've been writing that I cannot get myself moving, it's like writing that essay. I had to take a class in college about dinosaurs. I had to write a paper about dinosaurs. You studied dinosaurs in college? It was an actual class that fulfilled some sort of science. And now your son is obsessed with dinosaurs. Apple doesn't fall far from the tree. That's incredible. But you're writing about a topic that you don't care about. The words do not come easily. You're not if you don't care.

2:22:30Of course. Jordy's secretly a PhD paleontologist. I'm finding this out for the first time. Sorry. I love it. No, I would say the flip side of that is when you know you have a good idea and it's the right idea, the words just tend to fall right out. And it's no problem to get them on the paper. And so I bring that up because I think if you have writer's block and you're like, oh, let me use ChatGPT to get me through it, it will find a road through, and it'll help you onto the next paragraph. But then you're probably ignoring the signals of your idea not working. And you're much more likely to get to the bottom and finish the essay or whatever it is, even if your idea sucks and it didn't work.

2:23:04Yeah. I've noticed - Writer's block is just this divine presence that's telling you, you don't need to make this. Yeah. It's not needed. It's not needed by the world, and you can force through it, But what did you actually do? I've been writing a... I think there's a... Oh, sorry. Oh, sorry. I was going to say, I think there's a similar analogy with music where I've heard from many musicians, I am not one of them, of course, that their best songs were the easiest to write. And like the tune, the lyrics just flowed right out. And if they are struggling to figure this out, figure out the tune, figure out the lyrics, it's probably because the song's not working.

2:23:37It's usually the same. Yeah. I've been writing a daily newsletter for this and to prep the show, about 500 words. And I've noticed that I haven't been using AI for anything other than knowledge retrieval. If I have to look up what's the market cap of this company, of course I go to ChatGPT. And then I also notice that a lot of times I'm kind of just reiterating like a discussion that I had with Jordi earlier in the morning. And so I'll use dictation sometimes to just get some of the words down and then I'll kind of write from there. But I'm never crafting a prompt for what I'm writing, which I think is just interesting because the models have gotten so much better.

2:24:11and yet there's still something about like, I got to come up with the own idea or like the seed of the debate. Yeah. What do you think? Yeah. It's also very difficult to really make it right in your voice. Yeah. So you can prompt it and say, write it like Jordy, like do it exactly, but it's not. It hasn't gotten there yet. I feel super grateful that it's still easy to clock AI generated text because if I'm scrolling on social media, maybe I see a post and then I'm in the comments like interested to see what other people think about it and I can just easily clock, like, okay, AI generated, I'm just going to skip it because it's probably just like summarizing it and asking a question.

2:24:46But I worry we'll lose that ability to filter like in the next iteration of the models. It's definitely made me a sloppier writer. And I'm like deliberately not trying to craft the perfect sentence structure constantly because I feel like the rigidity is just an extra layer that I impose on myself. And if I take that away, it just kind of feels more stream of consciousness and is just actually a better product. I don't know. Shifting gears. Yeah, yeah. Sorry. Go ahead. I was going to say shifting gears to the book. Yeah, yeah, yeah. It's solved. Do you think, well, maybe a kickoff question. Do you think tech has figured out how to spend money?

2:25:27Because there's always this critique that tech people don't know how to spend money. And I think part of that was the tech uniform was not maybe like the East, you know, the West Coast tech uniform. jeans and a t-shirt and sneakers people sort of came you know participated in this industry in a very sort of plain way patagonia jackets like 200 bucks okay it's reasonable well let's let's compare tech spenders to wall street spenders the major difference is most tech wealth is not liquid and wall street was paid in cash every year so so i mean that's that that's the that's one of the biggest like tech wealth is a lot of paper wealth relative to wall street wealth that was so much more liquid.

2:26:09You can go buy the Rolex. You go buy the house in the Hamptons cash. And so I think that was probably part of it. Yeah, there's a culture of buying the watch based on a banker deciding what watch to buy based on the size of their Q4 bonus. It's going to be some percentage of that. So the watch I get, maybe it'll be a Submariner, maybe it'll be a Daytona. I don't Yeah, I do think there's something to be said that on Wall Street, your value and your success was how much money you made. And in tech, it is much closer towards the product that you built and the intelligence that you have and whatnot.

2:26:46And so there is less desire to show off wealth in tech because that's not what people get valued for. And I can't think of hardly anyone in Wall Street outside of maybe Warren Buffett who is very well respected and admired and wears a T-shirt and lives in a modest house. It doesn't happen. But you can name 100 of those people in tech who do that because they're valued for their ideas, not just their annual bonus. On that illiquidity question, Paul Graham had a take recently that a lot of tech people, they're locked up. And then by the time they can afford art, they struggle to get up to speed on the art world or something.

2:27:23Does that resonate with you at all? Just the fact that if you come on Wall Street and you can afford the Rolex and then the AP and then the FP Journe eventually, it just ladders you up the luxury ladder easier than just having all this money and being like, well, do I really want to jump to the top of this luxury ladder that just put the hedonic treadmill on 15 miles an hour on the first run? Yeah, I think it's very difficult for people to know what they want. They're very good at knowing what they need and they're very good at knowing what they don't want. Knowing what you want is actually very difficult.

2:27:55And part of the reason is because what I want might be totally different from what you want. People have completely different needs and whatnot. And I think if you are thrust into wealth very quickly, there is knee-jerk reactions of what you think you should want. I want a mansion, I want a fast car, I want art, I want the plane, whatever it might be. And sometimes it's true and sometimes it's very not. And I think if I can predict all of your spending habits based off of your income, there's a very good chance that you're not doing it right. Because there's a very good chance that you are just buying and spending your money in the way that society told you to do.

2:28:25If you earn this much money and your net worth is this, you should have this house and this car and this art and travel this way and whatever it might be. And most of the time, it just doesn't work that way. People have very unique spending preferences. The people who I've seen, the wealthy people who I've seen, who've done the best, have a lot of really extreme quirks in their spending where they spend, they're very wealthy, but they spend no money on cars or no money on travel, no money on food, whatever. It's very unique to them, whatever it might be. But I'm not a wine person in the slightest.

2:28:55I'm not even necessarily a travel person. But a lot of people would be the exact opposite, and that's fine. So I think it takes a lot of looking in the mirror, so to speak, to figure out who you are and to go down the path of trying to figure out what you want, which is not easy. I feel on a personal level that my relationship with money has been very distorted because the business that I started building in college ended up, you know, as cash flowed every month for my entire adult life, which has been a tremendous benefit. But it's also completely, it's just distorted the way that while in tech, the primary way you generate wealth is you get a measly salary for a long time and then you get a whole lot of money at once or maybe you sell some secondary along the way and get some liquidity but it's been this weird dynamic where it's like i remember t-shirts well so so the first time i took a meaningful profit share from from my company i bought a nine i took the entire amount and i bought a 911 because i was just like well i'm gonna get the same amount next month and then i'm gonna have this 911 for a long time and so So I had to kind of like, it took me a few years to kind of like learn.

2:30:08Such a fascinating psychology. Learn that lesson, whereas I wouldn't think like, oh, that's a lot of money. I would just think of money in increments of basically months of cash flow. Yeah, I think there's good mental accounting there of like, rather than thinking about the 9-11 costing 120 grand, you think of it as one month. It cost me one month or something like that. I think that's actually a smart way to do it. And even I value a month even more now, even though monthly income has increased throughout my adult life. Because as a kid, you're like, oh, I got all the time in the world. Now I'm turning 30 at the end of this year.

2:30:43I'm like, oh, time is a real thing. It goes by. You don't have an infinite supply. How do you think that makes me feel when you say time's running out because you're turning 30? How does that make everybody else feel? I would kill almost 30. Well, are you spending a lot of money on your health because you look younger than the last time you came on the show? Maybe I just hadn't showered last time I was on the show. I don't know. But, you know, I do think there is a real thing where it's a fine balance between spend for today, live for today, and save for tomorrow. Both of them are really great ideas.

2:31:22And it's never as simple as YOLO or, you know, save and save for the future. It always just comes down to what are you going to regret? What are you most likely to regret at some point in your future? And it goes both ways for people. You could easily imagine looking at yourself 20 years from now and regretting the trips you didn't take, the 9-11 you didn't buy, the house you didn't buy. You can so easily imagine, too, looking back at some point in your life when you're tired and your career is not working out, whatever it might be, and saying, I am so grateful that I saved the way that I did.

2:31:53It gave me a sense of independence that I value more than anything right now. What are your quotes? on the... Oh, sorry. You said a lot of successful people have quirks in their spending. I imagine you understand your own at this point. Yeah, I mean, this is a trivial and small thing, but I love this heuristic by Rob Henderson. He's a great academic. And he says, rich people food looks better than it tastes, and poor people food tastes better than it looks. And on that spectrum, I love cheap food. I love Taco Bell. I love Jimmy John's I can't get enough of it and I've had some very expensive meals and whatnot and I have enjoyed them marginally at best the best meals I've ever had tend to cost the cheapest and so it's such a trivial thing but never would I want to be like oh I can afford to eat this way so let me abandon all the food that I love and go eat some food that is subpar because I'm supposed to like this stuff better when I don't Did you cover private aviation at all in the book.

2:32:55We had a funny conversation with a friend this morning. He basically said, my house is going to be paid off by the end of this year, and after that, I'm giving all my money to NetJets. He's not getting a second home. Which is a quirk in itself. He finds commercial aviation to be very dehumanizing. I would, if you had a choice between two houses and commercial or one house and NetJets, a thousand times out of a thousand, I would do the latter. Absolutely. Totally agree. Two houses are a giant pain in the ass anyways. But to answer your question, I do cover private aviation in that because I made this observation that having a private plane is like the ultimate luxury.

2:33:36If you talk to wealthy people, they're like, that's the only thing that you get pleasure out, that the house, the yacht, the car doesn't do you much. The plane will change your life forever. And I think part of the reason we love it so much is because the vast majority of people in that situation remember what it was like to fly commercial. Now, here's the observation is nobody thinks it is an ultimate luxury to have a private car, but virtually all of us do. And the reason we don't think about it is because we've always had private cars. And so there's nothing to compare it against. Now, you could imagine that if you spent your entire life on a train, on a public train or a public bus, and then you got a quote-unquote private car, it would feel like the ultimate luxury.

2:34:12But we have nothing to compare it to. Well, another example of it. So much of their lure. Yeah, if someone's lucky enough to be born into a family that only flies private, can you imagine their experience of flying? They're like, I don't want to fly. You can imagine they don't even have a positive feeling associated with flying private because it's all they know. Because if you actually, you know, it's like, okay, I'm going to be up in the air and like the bathroom's small and I don't have that much access. There's not that much. There's some food, but it's not my favorite food. Can you get Taco Bell delivered?

2:34:45Yeah, exactly. Exactly. Right. Right. Do you think people, I want to hear your reaction to Jordy's take that buying physical things is actually an experience. So I grew up. False dichotomy between I just want to spend money on experiences. So I, growing up as a kid, I always, I always loved brands and I would get obsessed with different things, whether it was mountain biking or snowboarding or surfing or anything. and I would get fixated and obsessed with like a certain kind of surfboard, right? As a kid, I wanted to surf mayhems, but I couldn't really afford them. They were super expensive and I worked at a surf shop.

2:35:27I could get these other boards for a lot cheaper, so I'd always get them. Now as an adult, I only surf mayhems. And I always, we grew up in this era, our generation was told constantly, don't spend money on things, spend money on experiences. I never understood that because in my view, I was like, when I put on a jacket that I love and it's a thing and I wear it for the day, it's an incredible experience. Or if I take this surfboard that's incredible and I go out surfing with it, that's an experience in itself. So what's the verdict on things? Can things make you happy? I've found that things have and continue to make me happy in my life.

2:36:08Yeah, totally. Because I think what you broke down there is if nobody were watching and nobody could see your surfboard, nobody could see your car, you would still buy them. And therefore you're doing it for something that actually makes you happy rather than trying to signal for the attention of strangers. If you use like a fancy sports car, for example, there are some people who own Ferraris because they love the artistic engineering of it. They love the line. They love the beauty of it. They love the growl. They love the engine. They work on it themselves. They wax it themselves. they love the art of owning it.

2:36:37They love the acceleration. They love driving it. That's one group. Another group just wants to get the attention of strangers and they just rip it down the road, just trying to turn as many heads as they can. The former is going to get way more happiness out of it than anyone else because they would still do it if nobody was watching. I think that's the framework is like, if nobody was watching how you lived, what would you spend your money on? And as you just described it, I know you would still buy that surfboard. And so it's the right thing to do. I think the stuff versus experience tends to go astray too, because particularly in the social media world, a lot of experiences that we want to spend money on are literally just to impress other people.

2:37:15It's where should we go on summer vacation that's going to generate the best Instagram pic? This is John's thesis on Europe. He's like, I don't need to go to Europe. We have lakes here. We have oceans here. We have mountains here. Why would I leave the great United States? You know what my example of this is? Bali. I don't know if you've ever been to Bali. It's a dump Thank you My wife and I took our honeymoon to Bali And it's a dump So I went there On multiple surf trips And I only cared about the waves But the funny thing is people go there on surf trips And like the actual best surfing In Indonesia is really not in Bali There's a handful of solid waves The downside is there's Actual trash in the water You were just swimming and it's like supposed like idyllic reef break and you're just swimming through plastic bags.

2:38:08It's just absolutely disgusting on a half the aisle and the water is just like brown and dirty. And people surf in America too. And it's like, do you want to say that you want to take the picture and say, look at me, I'm in Bali in my trash water. I think that's it. I think Bali becomes a popular tourist destination because people love to say, I went to Bali and post that on social media. The whole time I was there, my wife and I kept saying, we could have flown to Maui, which is a five-hour flight. Instead, we flew 20 hours to Bali, and it's 97 % worse. But I think, honestly, we were attracted to it back in the day before we knew better because it sounded like a cool thing to do.

2:38:45Like, oh, we could tell our friends we're going to Bali. We didn't know anything else other than it was a cool – it sounded like a cool thing to do. I don't want to say too hard on Bali. But part of why it's not as great as the hype is that there's too many people there. So the hype sort of like created the problem. I think it is, you know, probably is as nice as Maui just physically. But the challenge is, again, the infrastructure, the traffic is a number of times have almost died on a scooter in Bali, too. I'm surprised. Oh, yeah. I'm surprised I'm here on this podcast. Safety is real. Right.

2:39:22And so then it's like, if, if nobody could, if nobody got to hear where you're going on vacation, you can't post it and you can't tell anyone else. Never would I want to go to Maui. I'd be like, ah, let's just go to Santa Barbara or let's go to Maui or something like that. It's so much better and easier. Yeah, totally. Uh, I want to talk about like the, the, I mean, maybe the frame is like, do people ask you more questions about, or how do you see the job to be done by the book? Is it more about how to, is it more about money or just happiness? It's definitely more about happiness. And one of the big points in the book is there is no formula for how to do this.

2:39:57And so that's why when you have a formula like spend money on experiences, not stuff, it tends not to work because I'm different than you are. And people from different generations, different countries, different backgrounds totally want different things. And I think it is immature to say that because I like spending my money on this, you should too. Or because I don't value this, you shouldn't either. It's not – but it's an innocent mistake to make, and a lot of people make it in finance, the assumption that there is a right answer to earning, saving, spending, investing. It's a very individualistic endeavor.

2:40:29And so a lot of this – the question you ask is like, do people ask me for advice on this? The advice I have, and people don't like to hear this, is like you need to figure it out for yourself. And so the book is about the psychology of envy and contentment and social aspiration, which tend to be universal. But there's nothing in this book that says you should spend your money like this. And so it's not called the science of spending money because I don't think that exists at all. Yeah. What about what kind of things that you can spend money on have the worst value? because when I talk about, when you look at luxuries, like maybe staying at an Amman property, Amman is maybe 10 times as expensive as like the average four seasons.

2:41:11But I think it's probably like five times better, in my opinion. So it's not like a perfect trade, but it's at least, if you only have a limited amount of vacation time a year and you want to have the best possible experience, I think it's a trade worth making. But where do you think maybe is on the opposite side of that, things that are 10 times, do you think it's, you mentioned food. Is there anything else? I'll tell you one little spending quirk that I have. This is slightly off topic, but I think it's, when I grew up, I was a ski racer. And I always felt that everyone else on my team had better gear than I did.

2:41:53They had nicer skis. They had a nicer jacket. They had all of that. and it drove me crazy. And so when my son, he's nine now, when he started skiing, to kind of make up for the hole in my soul that I had when I was a kid, I was like, I'm gonna buy you the best of everything. You're gonna get the nicest skis, the nicest everything. And the quirk is he could care less. He could not care less about any of that, about having the nicest stuff. And so that too was like a realization of like, that would have meant more to me than anything that I could have ever had. And he could not care less because everyone has their little spending quirks about them.

2:42:28Totally. Last question for me. I obviously we're here. I want to celebrate the book. I want to promote this book, but I'd love to know about another book that you think, uh, serves as, uh, just something you enjoy, something that you, an author that you respect, maybe someone from the 20th century or 21st century author, uh, someone who you've pulled influence from, or just have respect or just a book that you keep coming back to? Yeah, I mean, two nonfiction authors that I think are the greatest of modern times, they're both still living, still writing. One is Eric Larson and the other is Robert Kersen.

2:43:05They've both written some very famous books and some very successful books. And I think their ability to craft a sentence and tell a story is unparalleled. And even if I were to compare them of all the writers of the last 200 years or so, I'd put them near the top. there. It's just so it's effortless to read their, their work. Never do you have to reread a paragraph and say, what are you trying to say here? You can just kind of glaze your eyes over the page and completely understand what they're saying. And what I love about what Eric Larson in particularly does. So he's a nonfiction writer, writes books about like world war two and all these, these, these different events.

2:43:38Some of his chapters are half a page. And so some of his books can have 200 chapters through each a page or two, because he's so good at just being like, here's my point. I'm going to make the point, use an example, and boom, I'm done. I don't need to ramble for another 17 pages. I'm just going to move on. And that, to me, is the key of good writing. It's like the person who can say the most in the fewest words wins, and he's the best at that. Yeah, I read Devil in the White City a while, maybe a decade ago. Fantastic book. Loved it. Very good. And perfect example of that. Thank you so much for coming on the show.

2:44:08So fun. It's always a great time. Let's do it again soon. This time, we're just going to send a recurring slot on the calendar. You can move it if you want. but we love talking. It's always fun. Hit the gong again for being number one in the business section on Amazon. Business decision-making. I'll see you at number one and I'm sure many other charts to come. Thanks, guys. This is fun as always. Great to see you. Before our next guest joins, let me tell you about Adio, customer relationship magic. Adio is the AI native CRM that builds, scales, and grows your company to the next level. CRM, super intelligence.

2:44:45And we have another gong-worthy guest. Let's bring in Misha from Reflection. How are you doing? Boom. Hey, guys. Good to see you again. Good to see you again. Good to see you. You've been busy raising billions. Jordy has this habit of telling people when they come on and they do a great interview, we'll see you soon, but you delivered. I think you're the first person that if we roll the tape, Jordy probably said, we'll see you back here. We got to roll it back. I bet I said, knowing the progress you've made and the progress you've will make. I bet you'll be back on here with more news. But quickly, give us a reintroduction to the company and, of course, give us the news.

2:45:24A quick reintroduction to the company. I'm Nisha. I'm the co-founder and CEO of Reflection. Together with Giannis, we started the company about a year and a half ago. And we were formerly at DeepMind. Giannis was one of the founding engineers at DeepMind, contributed to a lot of projects like AlphaGo and Gemini and recruited a team of about 60, I would say, researchers and engineers from Frontier Labs. And the charter of the company has opened up since we last spoke. We've raised this capital to really build out the frontier open intelligence based in America and exported to the rest of the world.

2:46:07So that is kind of the focus of the company. Yeah, before we get into the details, how much did you raise and who did you raise it from? We raised a total of$2 billion from a syndicate of investors. Oh, this has a gong hit. From a syndicate of investors. It does. Let's go. Everyone's excited. Let's give it up for a video. Disruptive. DST,$17 billion. There we go. B Capital. A bunch of existing investors as well. like Lightspeed, Sequoia, and CRV, and so forth. So it was quite, you know, we're very grateful for the support from the syndicate. Okay, get even more kind of granular with what the focus is today.

2:46:53My read on it is open source is the focus. Putting DeepSeek out of business. That's what I want to hear. Finally. That's right. Yeah, maybe the short of it is that it's U.S. DeepSeek. Yes. That's what we want. Thank you. Yeah, Frontier Openweight Models. that we train here in America and export to the rest of the world. So this is meant to be a global technology. We have a big presence in the UK. We have a team there. And so this is not just, you know, even though it's American built, it's really built for the world more broadly. Can you share anything that you think you'll be able to do to Outfox, DeepSeek, and GPTOSS?

2:47:33We're having Dylan Patel from SemiAnalysis come on next, and he's talking about Inference Max. He's benchmarking and I learned so much that, you know, it's not just the model. It's how you run it the batching the Different GPUs sometimes in video is better sometimes AMD is better Like how are you understanding because I imagine it's not enough to just say it's American deep. It's got to be better So what's your plan to actually beat them? It's a really good question and I think it actually falls in two parts first actually just having an American compliant DeepSeek would go a really long way because a lot of enterprises are basically locked out from using those models because of various legal marketing provenance, data provenance risks that are associated with Chinese models.

2:48:20So from a commercial standpoint, just having something that is as good, but kind of compliant and built here would be really powerful. But of course, there is, you know, an aspect that you want to leapfrog and really be the leader in open intelligence across the world. And we do have some tricks up our sleeve. But we can't share them. Yeah, but I mean, hopefully eventually they'll be out. Yeah. Yeah. Yeah. But, you know, we have some great work happening on reinforcement learning within the team. The other thing that, you know, the Chinese labs don't have access to is obviously the same level of chips that American companies have access to.

2:49:01And so what I think DeepSeek did really well is co-designing their algorithms together with the chips they had access to. And so there's some really interesting stuff that you can do with co-designing algorithms with frontier chips that are going to be accessible to us as well. Interesting. Quickly, talk about the business model. I can imagine this turning into sort of like a Red Hat Linux play where there's an open source model, but you're implementing it, working with enterprise, working with the government, and there's a contracting piece, a SaaS layer on top. Is that logical? Or are you thinking more like you nail open source and then you can do a closed source model, sell API, you could go and own the whole token factory, the inference stack?

2:49:44Where do you see the business looking in a couple years? I think that the primary thing you need to set first is how do you build the kind of open intelligence, open models and the sets of tools around them for, you know, you partner with some inference providers, you, you know, set up, you know, make it easy to customize things, you make it easy to build agents out of these models and ensure that that kind of spreads like wildfire. So I think that having some sets of open models that are really fully permissive is really important. But the pull for this kind of model really comes from large enterprise.

2:50:27That's when does it make sense for you to move from closed to hybrid to open models? It's really once you're a very big consumer of intelligence. And that's basically large enterprise, sovereign, and scaled-up startups that are spending crazy amounts of money on closed APIs. Yeah. And so the way you kind of commercialize it, yeah, you want you want them to be building on top of your models. And there's all sorts of services and products that you can build out on top of it to effectively solve their problems end to end, because just providing an open rate model is not enough. These things are very hard to customize.

2:51:03These things are very hard to build evaluations around. They're very hard to do anything useful with if you don't help a customer end to end. So I think that there's a lot of opportunity for commercialization, but you really need to be the core intelligence that others are building on before you can really be useful at the next layer as well. uh why uh why do you think the dialogue around open source open ai models went from you know up to a fever pitch people demanding it and they release it and then now uh you don't hear it talked about really uh at least online in the timeline much at all

2:51:46clearly clearly and and and and i would say like what i'm trying to understand is like clearly there's massive demand for open source models, but I have a feeling that developers would like to be leveraging the technology of a company like Reflection who's wholly dedicated to open source and dedicated to commit, you know, and really committed to it. Whereas it's hard to, you know, we had Sam on today, we've had a bunch of people on from OpenAI. It'd be hard for anyone at OpenAI to say like open sources are a top three priority, right? Maybe it's in the top five. Exactly. Exactly. It's really hard for you to both be the world's open model and open intelligence provider and for it to be the number two, number three, or number five thing, right, that your company is focused on.

2:52:40And the reason is that what matters is capability. You want highly capable open models. And the only way to get that is if your commercial incentives are fully aligned with open intelligence as the first and primary thing that you're doing. Now, when you release something like this, you can't just release the model. I mean, I think that that's one part of it. And then, you know, inference providers can take that model and optimize their stack around it. But these models are so big and hard to do anything with, unless you are an expert, that you really need to help with that as well. uh the models seem to be exhibiting spiky intelligence where are open source models particularly best or demanded to be best like the the customers of opening of of open models what did they want to do that uh might not be as relevant in a closed source ecosystem i could imagine that agentic payments is maybe not the hottest thing in open source models or or IMO level math that might be, maybe that's really important in open source, but what is unique about the customer of the open source model?

2:53:50What do they want it to be best at? Yeah, there are basically two things that as customer, you are looking to do and achieve when you adopt an open model. The first thing is, suppose you have good performance on something from a closed model, but it's ludicrously expensive, which is very common. Then you want to drive down the cost while keeping the performance. So you want to customize the model for those tasks. The other way around is that, yeah, you have some finicky data distribution that was not represented when the closed model was trained and the closed model is spiky, but not on the data that you need it to be good at.

2:54:29And so then you want to drive performance on that. And so then you want to post-train and customize for that. So it's really your customizing for driving extra performance or you're customizing for driving down the cost. But it's really important to have control over both. Yeah, that makes a ton of sense. Well, thank you so much for coming on the show. Congratulations on the huge raise. I'm sure we'll see you back here in a couple months. Yeah, I'm so curious. I imagine you guys are thinking about how you can create your own deep seek moment. So when the time is right, come back on. We'll pump it.

2:55:03Play that eagle sound. Thank you. Have a great night. Yeah, well, thank you so much for having me. Of course. Great to catch up. Talk to you soon. Congrats to the team. How did you sleep last night? I woke up way too early. I woke up at 4 in the morning. I'm physically unable of a 3-peat. Oh, no, this is so bad. I have not gone three days back to back with a 90. What you got? I got a 79. I got a 69. You beat me. There we go. You beat me almost every day this week, or maybe the last two or three. But I got cocky. So you got your 3-peat. You got your 3-peat. and got cocky yesterday. So play some sound effects, do something.

2:55:37Let me also tell you about public.com, investing for those that take it seriously. They got multi-asset investing, industry-leading yields, and they're trusted by millions. Please take it seriously. Did you see that Marginal Revolution is calling for Vitalik Buterin, the co-founder of Ethereum, to win the Nobel Prize in economics? Really? I think that'd be very, very cool. It's a wild card. Tyler Cowen's mentioned it a few times, that Vitalik would be kind of the outside-the-box pick. But in terms of advancing economic theory, designing Ethereum. I mean, it would be remarkable. And he also advocates for Robin Hanson, the father of prediction markets, to win the Nobel Prize.

2:56:14I don't know. We'll see. It'll be fun to track the Super Bowl for economics grads, I guess. I want your take on this. Luke Kawa is quoting the Pepsi CEO. I think fiber will be the next protein. Consumers are starting to understand that fiber is a benefit that they need. We put creatine and everything we put protein and everything is fiber the next thing we put caffeine and everything i mean you why not just add them in so the thing here is that uh fiber gummies is that a thing olipop already like leaned heavily into fiber okay so it might already be happening maybe the pepsi ceo is a little behind the times the game he might also be did pepsi buy olipop's competitor i don't know i mean kyla scanlon in the reply says didn't we already do this with the rise and fall of fiber one.

2:57:01So I don't know. Maybe it's too late. So PepsiCo acquired Poppy. I think Poppy includes fiber. Yes. Oh, I have a reaction for... You do. You want to keep going on that? You have anything else? No, go ahead. I have a reaction to Rune, who put us in the truth zone. So on a previous show, we said that, we said incorrectly that during the AlphaGo game between Lysadol and DeepMind, um alpha go dropped the 37th move move 37 that iconic moment that kind of scrambled lisa doll's uh brain uh we told the story such that move 37 happened lisa doll was so racked by it that he stepped outside to smoke a cigarette apparently that's not true apparently he smoked a cigarette before move 37 and so it's just more cinematic to tell it that way I think it's actually maybe even more dramatic because potentially Move 37 was so crazy that he couldn't even bring himself to smoke a cigarette.

2:58:01Do you think that's what happened? Maybe. No, I don't know. But thank you, Rune, for doing the fact check. Obviously, you are correct. You know the story more. And we always appreciate the truth sound. Well, we have our next guest, Dylan Patel from Semi Analysis with some massive news. Dylan, how are you doing? That is a cinematic shot. Is this AI or something? Where are we? That's called aura. You're aura farming. I'm literally in America, bro. That's amazing. Out of the back of the truck. You're the bald eagle. You're the bald eagle. He's not a China hawk. He's a bald eagle. This is proof of work.

2:58:34You're out in rural America at the back of a truck bed. You're inference maxing. You're podcast maxing. Thank you so much for taking the time. Give us your breakdown quickly on the launch today, inference max, the launch yesterday. I went on your Zoom call at 10.30. I was laying in my bed listening to you. It was very interesting, but I'd love to hear you kind of break it down first. Wait, first, can we get a – do you have cowboy boots on? We need the full fit check. We need the full fit check. The full fit check. Oh, okay. There we go. Looking good, but cowboy boots next time. Okay, anyway, sorry.

2:59:13Please, give us the high level. So this is a fire station behind me, by the way, just so you know. But in Tennessee. Anyways, yesterday we launched Inference Max, which is a humongous release for us. It is a benchmark that's doing cost per million tokens and tokens per megawatt across all major AI infrastructure, AMD, NVIDIA, all the newest GPUs. and on all the latest models, GPT open source, LAMA, DeepSeq, et cetera. And so the reason why this is so important is throughout the industry, people are always like, oh, our chips are great at cost this way. Our chips are more efficient that way. Well, it turns out to actually measure inference, you have a variety of different metrics.

3:00:04You can always just cherry pick something. It's some vendor saying some BS. And so there ends up being a cherry picking. And then it's also on some super hyper-optimized software stack that's not real, right? It works for that one specific cherry-picked use case. But guess what? When I'm running inference at a major company, sometimes I have big requests. Sometimes I have small requests. Sometimes I'm outputting a ton. Sometimes it's like an agentic workflow. Sometimes it's this model. Sometimes it's that model. So what really matters is the real software that people are running, and it's on the latest drivers, the latest open-source PyTorch version, latest VLM, latest SGLang.

3:00:39All these things matter because at the end of the day, software changes every day. Performance changes every day. Models change all the time, right? And to actually get, hey, there's trillions of dollars of infrastructure investments being made over the next few years. How do you actually measure what's the best hardware? What's the most efficient hardware? What's it cost? And that's what we're aiming to do with Inference Max. And so we're supported by NVIDIA, AMD, Microsoft, OpenAI, Oracle, CoreWeave, Dell, Supermicro, HPE, and all sorts of vendors that I can't remember off the top of my head.

3:01:12Well, you're not making any money on this, right? It's all open source, but there was a ton of capital that came together, a ton of people that did put up money. What's the scale and the scope of the project? Yeah, so Semi Analysis has multiple engineers that I'm paying full time. So I'm losing like, you know, a million dollars a year on this or a bit more, obviously, because engineers are expensive. But on top of that, it's, you know, the vendors are contributing and the cloud companies are contributing tens of millions of dollars of GPUs. Congratulations. That's fantastic news. So, you know, the thing is, I'm not necessarily sure how I'm going to make money on it, but I am more of farming as I am with this background, right?

3:01:55All that matters is how do we deploy AI efficiently across the globe? And perhaps by aura farming in this way, we'll figure out how to – people will buy our other stuff, right, is the hope. Not exactly sure, but this needed to exist, and there was no way for it to exist unless we did it. What's your life been like the last few weeks? how often are billionaires calling you, asking you specifically for financial advice saying, hey, I'm thinking of putting a billion into this one. What do you think? Did I do it? Do you find yourself having to push back and say, I think you're having the second trillion dollars of debt to flow into this.

3:02:34What do you think? The crazy thing over the last few weeks is that companies that you would have never expected to need debt are in the debt markets. Right. You mentioned that, you know, people like Meta and Oracle, you know, who three years ago, you've been like, these are the most profitable companies on the planet. Well, they're in the market for debt because they're they're building as far as like how often are people in the DMs or calling me? You know, that's what the company does. We provide services around this. So, you know, I'd like to say the company and business is taking off like a rocket.

3:03:05And so, you know, the whole point is inference max aura will increase the aura of like other people like, you know, you know, doing this. But I will say it's just like we've hit terminal velocity, right? It feels like we're building a motherfucking like Matryoshka brain. You know, like I don't know what people are trying to invest in besides God. That's great. What's the biggest debunk that's come out of the results of Inference Max? Is there some narrative out there on the timeline or in the AI community that you feel like you've kind of you're able with this data to turn things around? yeah so i mean there's there's tons of people like oh amd's best oh nvidia's the best oh this is better that's better um it turns out like everyone's statements are sort of like you know there's got to be a lot more nuance to it um and so my favorite thing is yesterday i saw twitter war between two accounts of like 5 000 followers each so these weren't like small accounts per se and they were going back and forth posting data from inference max saying no you're wrong you're cherry picking.

3:04:06No, you're wrong. You're cherry picking. And it's like, the reality is it's a little bit more complicated and they're debunking each other. Um, but I think what's relevant is that, you know, Nvidia is not the only game in town. Um, a lot of people, uh, thought that they were, um, you know, between the open AI AMD deal, uh, that happened and then the results that we've shown, and we've been working with AMD and Nvidia on this for many, many months. Um, it's clear open AI. I mean, Nvidia is definitely ahead, right? But there's certain use cases where AMD is better, right? If you're running GPT open source, that model's exploding in usage, then, hey, guess what?

3:04:42Actually, AMD may be a better hardware on a dollar basis. It's not better on a watts basis. And those are the two things, right? So why maybe I'm in test is there's a lot of watts here that we could put on AI infra. But it's a challenging sort of thing is sometimes your capital constraint, sometimes your power constraint. And what you should do, maybe you do actually think, maybe you should deploy AMD, right? Maybe you should deploy NVIDIA. The default is NVIDIA, but actually in many cases, it makes sense. And the software works. The open source software works. It's not buggy completely. It is if you're training and doing other things, but if you're running inference on specific models, it works.

3:05:22What's the biggest risk to the overall buildout? Is it energy capacity? What's top of mind for you over the next 12 to 24 months? All these deals have been announced, but a lot of people are asking, where's the energy going to come from once you, once you start talking about, you know, gigawatt scale clusters? Yeah. So it's, it's, it's not even, it's not even, uh, you know, like you've got all of these like dudes in suits like you, you know, in, in their little cushy little offices, signing these big checks of, of thick money on bank accounts. But the reality is, is like, Hey, um, I can buy the GPUs.

3:06:03I can get them made and import them from overseas. I can buy, uh, you know, like the sheet metal, I can buy all these different things, but you know what you can't do is there's not enough motherfuckers in cowboy boots, um, deploying and building these things. Right. Like it turns out electricians wages are skyrocketing, right. It turns out like plumbing wages are skyrocketing because data centers need liquid cooling. And, and so like how, I think that's the biggest risk is, you know, where's the skilled labor going to come from in, in the West? Interesting. Um, because the West has not built at this scale before.

3:06:36Yeah. Uh, chat says that, uh, Regalvin, the chat says you're the only 10 IC. So he's having fun that you're out in Tennessee. Uh, is there any, uh, is that, uh, what does the long-term vision look like? Is it relevant to think about adding TPU, Grok, Cerebrus, other system? Like, like what is the shape of the, roadmap. Are you sharing that yet? Or is that even relevant? Yeah. So Inference Max is amazing because it runs every single day on the latest software. But right now, we've only got tens of millions of dollars of GPUs. We've got to hit the 100 million number to actually get everything, right?

3:07:13So what that means is more models are being supported, and we got that in the works. We've got adding TPUs and Tranium. This is a real big, difficult engineering effort. Google and Amazon are excited. We'll see how long it takes us, but it is a difficult thing. But they've got to put up the capacity. We've got to get the capacity somehow and add those chips. And then if you do that, over 99 % of the flops are around the world. Maybe we add Huawei. Maybe we add Grok or Cerebris. It's really a lot of engineering that's going to be required. And so that's all on the roadmap is to add more hardware.

3:07:48Quality. It turns out there's a lot of innovations that people are doing on a model inference beyond just quantization, right? You can do 8-bit or you can do 4-bit, but let's say everyone's doing 8-bit. You can still do certain innovations that make performance better, but quality worse. And so there's these tricks that people are implementing that, you know, actually it's, it's completely unknown to people. And so, you know, measuring quality as well is really, really important. Um, and, and, and, you know, continuing to run it every single day in an automated basis and continuing to get more people pushed behind it so we can get TPUs, traniums, GPUs on as many models as possible with quality as well measured.

3:08:30Well, that's fantastic. Jordy? One more question from my side. It seems like the debate is heating up around depreciation schedules for GPUs. What's your framework on that front? a lot, you know, NeoCloud wants to say five to six years, but maybe that's not realistic. How are you thinking about it? So every company, a major company in the world, the Googles, the Microsofts, Amazons, et cetera, do six years, right? That is the industry standard. But that may also be erroneous. And the reason why it may be erroneous is because the reason it got pushed up from, you know, four or three years to six years over the last decade was CPUs, storage, you know, that sort of stuff uh was not advancing that fast now we've got it advancing like a rocket ship um and it's faster than ever and so the question is there's two points on useful life right one is does the thing still work in six years and the other one is is it even useful to run it in six years right and these are two very very different questions um you know for will it still run in six years, the answer is most likely, but these things are running super fast.

3:09:41GPUs, TPUs, et cetera, are way less reliable than CPUs and memory. So it's a very high likelihood that it may not work in six years, whereas CPU servers, they'll actually run like 10 years. It's fine. But GPUs may not last the full six years, especially the new ones that are super hot, liquid cooled, et cetera, right? A lot more complexity, a lot more likelihood it could break down within the six years. The other side is economically useful. Well, if NVIDIA is releasing a new GPU that and it's back for 50 % more money every year and a half. Well, then in six years, you're at like 20 X improvement in performance.

3:10:15Right. And it maybe only costs like three times more. So, so you're like, okay, well, yes, the old GPU still, even if it still works, is it even useful or should I throw it out? And with that power, should I feed the new thing? Right. Um, and so that's the big question is, you know, is it useful to keep using the newest GP or the old GPU, or should you buy the new thing? You know, as Jensen says, the more you buy, the more you save, right? And so maybe his argument is correct, right? Yeah. And does that present a real risk? A lot of people are levering up and raising debt against GPUs and assuming that five, six year useful life.

3:10:55How big of a risk is that is what I'm trying to understand. It depends entirely on the company, right? So for example, Oracle's raising debt and they're building out Stargate and all these things, right? You know, they've got this$300 billion deal with OpenAI. Their biggest risk is not that, hey, you know, our depreciation schedule is six years and OpenAI's contracts are five years, right? And they still make money if they don't, they aren't able to last a year, right? Because they've got the contract with OpenAI. The real challenge is where the hell is OpenAI going to pay$300 billion, right? You know, I'm a believer, I'm a believer.

3:11:29I think you guys are believers, but a lot of people aren't. But for other folks, it's like, hey, I'm out here deploying GPUs. I'm just putting them out there, right? Hey, anyone want to rent them? Please rent them. And maybe you only sign a six-month contract. Maybe you sign a three-year contract. That's where it gets more risky because at the end of the term, I haven't paid off my GPUs. I haven't paid off my debt. Where am I going to sell it? Does the price fall? Where does the price end up being in that after a year? And so we saw that with Hopper GPUs, right? The people who signed the long-term deals initially weren't making as money.

3:12:01Because they were selling them at$2, whereas other people were out there like, oh, yeah, six months, I'll sell it to you for$3. That was amazing money for that first six months. And then on renewal, it's like, oh, shit, it's only$2.50. And then on renewal, it's like, oh, wait, now I'm selling it for less than$2. And who knows, as NVIDIA's BlackBell comes out, as Rubin comes out, as AMD's new chips come out, as Google starts selling TPUs, all these things keep driving down the cost performance and how many tokens you can get per dollar and per watt. So then all of a sudden, is a hopper still worth$2?

3:12:27Is it worth$1.50? Is it worth$1? For the people that are locked into a five-year contract, that's one thing. For the people who are just yellow and don't have a long-term contract, it's very possible that the GPU works, but it's not able to produce economic value worth what you actually put into it. So that's the big risk. That makes sense. Oracle sold off earlier this week based on reporting from the information. What was your immediate reaction to that piece? There was a lot of pushback on it. Yeah. So they said that Oracle's margins were low. Oracle's margins are not that low. They're higher than that.

3:13:06At that exact point in time, their reporting is accurate, right? But what they deduced based on the reporting with the numbers they saw were not accurate, right? Which is that Oracle's margins are low for the deals they've signed. That's not accurate. What's accurate is that NVIDIA's GB200 NVL72 has a lot of issues, right? They're mostly being solved and have been solved, but there are a lot of issues. It can be unreliable because of how complicated of a thing it is. So much power, liquid cooling, it's got the backplane. So there's a lot of difficulties with the hardware because of how complicated and how fast it is that are being solved slash solved already.

3:13:42The other one is, hey, Oracle has to rent these massive data centers before they fill them up with GPS. So Oracle's paying all this money for these data centers for Stargate, right, like in Abilene, Texas, that aren't necessarily generating revenue yet. And when your revenue goes from like this to rocket ship up because you've got Stargate, you know, what happens is you've got all this cost right before the revenue comes in, right? You bought the GPUs. You're trying to figure out how to make them to work, you know, because they're a little bit unreliable. You're replacing things. You're building out the data center.

3:14:09You're renting the data center. All these costs are hitting their books, right? That doesn't necessarily mean that they can rent them. I might be getting kicked out. You're all good. This has been a pleasure uh dylan next time you're in los angeles we'd love to have you at the tbp and ultra dome in studio uh everyone's a huge fan here congratulations and thank you so much for stopping by yeah massive launch excited to see uh how it plays out all right see you folks see you so much for having me today uh we got to get uh dylan uh nick reach out to dylan get his shoe size we'll get some uh we'll get him some cowboy boots that sounds great i'd love that good uh use out there There's some big news going on.

3:14:55So one is About 30 minutes ago Trump put out a truth. He said a hundred percent tariffs on China starting November 1st. What? So since then I think broadly today 250 billion has been wiped out from crypto. Okay. Yeah, Bitcoin is down 5 % Yeah, we have to check in our on our retail trader Another white pill, though. Yes, give us the white pill. Demis just said they did, last month they did 1.3 quadrillion tokens on Gemini. Found it. Congratulations. That'll fix the global economy. That'll fix the trade work. Congratulations. Absolutely wild. Our retail trader in residence has probably put a hole through the wall by now.

3:15:44Thankfully, the markets are closed for the week. but my portfolio on public is looking rough. It's a rough day, but we're happy to see that. Hey, it's a rough day, but Monday will probably be rougher. Maybe. We'll see. It might be a white suit day. Everything can get resolved over the weekend. You never know. We have to figure out what our bear market suits are. Maybe suits that actually look like bear kind of like fur. Ooh, fur suits. That would be good. Yeah. Yeah. You know, in Hollywood, there's a certain synthetic tears. It's like propylene glycol or something. That's what's in vapes. That's not it.

3:16:25But there's some sort of eyedropper that you can put in your eyes. They put in eyes of actors when they have to cry. And so maybe we should just be applying those the entire show so that we're just crying constantly. Well, in some much better news, of course, if you want to get away from all the chaos, you can you can book a wander you can find your happy place that's right book a wonder with inspiring views hotel great amenities dreamy beds top tier cleaning 24 7 concierge service it's a vacation home but better uh joey in the chat says uh fart coin is down uh 73 percent i don't know i don't know why no no no that's got to be a joke that's impossible i looked it up i mean i'm looking at a chart right now it says today so maybe it's rallied a bit today yeah that's I can't believe that something that's like a household name in tech is actually down that much.

3:17:13That's absolutely crazy. Well, in some more serious news, we have a new partner at TVPN. We're partnered with Gemini, Google AI Studio, the fastest way from prompt to production with Gemini. AI-powered coding apps, ease of use. It's built for everyone. You've been a power user. We're friends with Logan. We're very excited about Gemini. Logan is a dear friend of the show and works around the clock to make this product work. He's done a fantastic job over there. For startups. Supercharge your creativity and productivity. Chat to start writing, planning, learning, and more with Google AI. So you'll be hearing more about Gemini in the coming weeks, in the coming shows.

3:17:57Not tomorrow, because tomorrow's a Saturday. We're staying in the week. But I can't wait for Monday, John. Yeah. What else? Oh, Sheil had a take on Sora, which is interesting. He says, he personally got bored of Sora after a day, but a surprising number of my friends are still posting 10 plus videos a day. Maybe this AI slop thing has likes. I haven't been consuming a lot of Sora. I've been generating a lot, but I haven't been posting a lot. But I started posting. And once I got over that threshold of like, yeah, I'm just going to put my cameo. Yeah, I'm just going to post. It's going to be fun.

3:18:31I think I'll be experimenting with the actual feed. I only have 20 followers now. But I think that there is some fun, creative stuff that you can do there. It's definitely a different tool to pull off the shelf than the video camera or the text post on X or the email. But I've been having fun, and I think I will. Apparently, I didn't catch this when you said it, Tyler. The 100 % tariff on China is on top of all current tariffs. Wasn't it already at 100 % or something like that? We had Ryan Peterson on a few times, and the tariffs were like up and down and up and down. Get to the ultradome, Ryan.

3:19:04Seriously, we've got to check in with Ryan for sure. Monday, I'm back on. Buko Capital says, I legitimately cannot believe we are doing this again. Tremendous nukes have hit the cryptocurrency chart, says Joey in the chat. Did you ever read that? It's very funny. Thank you, Joey, for your service as the resident crypto bro. One last note. You know what hasn't changed with a tariff? Out-of-home advertising in America, baby. Adquick.com. Out-of-home advertising made easy and measurable. Say goodbye to the headaches of out-of-home advertising. Do not think about tariffing our billboards. Do not tariff our billboards.

3:19:38That is my line. You are on my line. Yeah, single issue voter on billboards, honestly. Liquidity says there won't be a second date, but at least she now understands why the AI circle jerk deals are quickly becoming a major systemic risk. Wait, wait. Can we pull up this picture of this eight-year-old building in Kazakhstan from offsite? This post made me laugh out loud. It's page 70. Do you see this, Jordy? Look at this picture of this eight-year-old building. At the bar? An off-site that quote posted and says, that MF on Roblox, left hand on AWSD. Because if you look at his hand, it's like caught gaming handed for sure.

3:20:25Very funny post. But that would have been me at eight gaming. And, you know, who knows? If he's on Roblox now, maybe he'll build a Roblox game. Maybe he'll build a startup. But yes, eight years old. It might be a little bit young to take a company public unless you're a 23-year-old. And then apparently you can SPAC a nuclear company. Be fun. Anything else you want to cover? Any other breaking news before we log off for the weekend? I'm going to miss being here at this table with these mics, this team, with this chat. We miss Dave Portnoy's mansion. Dave Portnoy made the mansion section in the Wall Street Journal.

3:20:56I'll read you one line of it because it is iconic. Dave Portnoy adds record-breaking home to his$95 million property portfolio inside the Barstool Sports Founders collection of luxury houses. They break down all of his different houses. They're all beautiful. But this is what made me laugh out loud because it's printed in the Wall Street Journal in the mansion section. Barstool Sports Founder Dave Portnoy has been dogged by champagne problems at his Miami mansion, from construction delays to losing his coffee because the waterfront home is too big. Last year, a more serious issue emerged, and we've talked to some founders about this, mold.

3:21:32And as he often does, Portnoy turned to social media. Quote, I need the best mold company in the history of Miami to come look at my moldy ass house. Any wrecks? He posted on X in August of 2024. I just thought that was hilarious that one of his X posts made it into the Wall Street Journal. So you can post your way onto a private plane. You can post your way onto the Wall Street Journal. You can post your way on the show. So enjoy the weekend and stay locked in on the timeline. Satrini says, can we change the channel? I've seen this movie before. Uh-oh. Our retail trader just posted his portfolio.

3:22:09He's down 12 % on the day. He says, I'm so cooked. I'm so cooked. You're cooked and chopped. Cooked and chopped. But we love you. Get the Dylan cam. Get the Dylan cam. No, he doesn't want to be on TV. Thank you so much for tuning in. We're going to set up. We're working on setting up a retail corner. Yes. We're thinking about it. We're going to give Dylan, not Dylan Abruscato, the other Dylan, we're going to give him a public account, some funds, and he'll just trade. This is just crazy enough to work. He won't be cooked or chopped. Thank you, everyone, for tuning in. We will see you on Monday.

3:22:42Have a great weekend. Have a great weekend, folks. Bye. Bye.