TBPN | Wednesday, April 23rd

Heat, heat, double kill. Together we came to this world. We came to this world. We came to this world to reach the stars, to shape our future. [Music] You feel the music? Let me double it. It's ready. You're watching TVPN. Today is Wednesday, April 23rd, 2025. We are live from the Temple of Technology, the fortress of finance, the capital of capital. We're—we're sorry, we're starting late. We are under attack, folks, likely by a nation-state, likely by a state actor. That's why we're 6 minutes late, uh, 7 minutes late, but we got a great live show. They didn't. They didn't. They knew. They knew the lineup was too unbelievable. We're going to pull it up on the screen for you, folks. Today we got a massive lineup. We got funding announcements. We got Sam Lessin coming on, yapping about VC stuff. Uh, we got Perplexity, big announcement; they're—they're going head-to-head with Siri; they're launching an iOS voice assistant. Uh, Duran, physical intelligence, Northwood—we're going to be covering it all. But first, we need to cover Kevin Systrom, the founder of Instagram. He's changed, absolute dog. Turns out a billion-dollar acquisition isn't enough; he went to court and said, "You know what," testified and said his being mean, I guess. Um, so Alex Heath kicks it off with a post here saying, "Lord, give me the confidence to sell a company for a billion dollars, basically disappear for seven years." Not true. He built another app; uh, it didn't really go that well, and they wind up uh, selling it. But I used it for a while. I used it for a while, too. News app, news app, very cool. I wanted that to be like a modern incarnation of uh, Google Reader. I thought it would be very cool if it was like AI-assisted news reading. They had some summaries; it wasn't quite there, uh, and I think they didn't find their like early adopter. It wasn't clear if it was for like tech, text, tech people or like normies, uh, and so it didn't really go anywhere, but it was pretty cool at the time. Um, and so he basically disappeared for seven years and then reappeared in court to basically [__] on my acquirer, uh, and I went back and found a great Sam Parr from five years ago, basically uh, with some chat logs. You know, we love some leaked emails on this show. This is amazing. The chat logs from when Zuck was buying Instagram. The highlights: "Zuck, I can't get to $2 billion. Systrom, two billion was my absolute say-yes number. I'll have to think on it." Just two dudes negotiating the best $1 billion acquisition ever via Facebook Messenger. Uh, yeah, the deals go down in the DMs, uh, doesn't—as they do. Deals die in the data room, and they flourish in the DMs, I guess. DMs. Uh, Sarah Frier says, "Wow, after years of silence from Instagram co-founder Kevin Systrom is on the stand in federal court confirming what I reported in my book, including that Zuckerberg starved Instagram's headcount around safety issues, leading to major problems." And so uh, Alex Canowitz posts Kevin watching Zuckerberg saying Instagram would be nothing without him. No, I—I—I was onto something here. Uh, and so uh, there's another funny post from Will: Do you think Kevin Systrom and Mike Krieger envisioned this when they created Instagram? Chatting with AI, chatting with Walter White? Yep, or YN? I don't know who YN is, but 8 million messages. Wow, that's a lot. Uh, anyway, uh, also Kevin Systrom, apparently he's on the board of Walmart. Not bad. Pretty—pretty awesome. Post-exit founder mentality. Good place to land. Yeah, I grew up with a buddy whose dad was on the board of Walmart, and we'd go to Walmart, and—and he was just like, "Buy anything you want; it's customer—it's uh, research." I was like, "That's awesome." Well, if you think about it for Walmart, yeah, the amount of purchasing activity that Instagram drives is obscene. That does make sense, actually. Yeah, and—and that's why Ben Thompson was saying that uh, Walmart should buy TikTok, which seems weird, but it actually makes so much sense when you actually play it out, right. Uh, and so uh, this all comes from a report in the New York Times at trial. Instagram co-founder says Meta denied his company resources. Kevin said during testimony in a landmark antitrust trial that he believed Mark Zuckerberg, Meta's chief executive, viewed Instagram as a threat. Kevin Systrom, a co-founder of Instagram, testified on Tuesday; he said, quote, "Mark was not—investigate—not—not investing in Instagram because he believed we were a th—uh, we were a—heark—investing in Instagram because he believed we were a threat to their growth, but they'd already done the acquisition." It was interesting. So—so I think a high level uh, Systrom is basically saying that Mark wanted to own Instagram, and he knew it could be successful, but he didn't want it to be too successful because it would—it was basically—um, would have—would have been damaging to Facebook's metrics. Yes. Systrom called it uh, a buy-or-bury strategy to illegally cement the social media monopoly by killing off its rivals. The Instagram co-founder made millions when Mark Zuckerberg bought his company, but Systrom sharply contradicted Meta's defense during hours on the stand. And so what millions is sort of an understatement. Yeah, it was more like a hundred million, I think, a couple hundred million. Um, but—but I'm confused about this because uh, once it's a whole co-acquisition, like he owns the whole thing, um, was Zuck optimizing for like short-term earnings in the public markets? Because yes, if you funnel everyone over to a lower-monetizing product, this happened with Reels, too, when they started pulling people away from the feed, which was very—which was very efficiently monetized. Yeah, they moved people over to Reels; wasn't monetizing as well. Then they had to—they had to keep giving guidance to the market and say, "Hey, Reels is going to get there; it's going to get there; trust us; it's starting to really—well, the ARPU is going to be there; the retention's there; it's going to be additive to our business; it's not going to be destructive because there's always a fear of that." Um, no, it's very—it's very possible that Zuck at the same time was being, you know, wanted to be aggressive over the long term in terms of building out a new platform, but short-term, you know, wanted to basically be able to control the process basically. And I think that's a great take, being like, "Hey, we don't want to move a bunch of users over to this app that we're monetizing well on Facebook and move them over somewhere where we just can't make nearly—" It's just odd to—to sell your company to Facebook, and then you have Facebook stock or cash or some mix of those two and not just immediate—and still have it be like your baby. Like, I feel like if you're—if you sell your company to Facebook, you should be very much on like the Facebook team, and you should say, "Hey, yeah, I have my thing, Instagram, that I want to grow." But really, all I care about is the Facebook share price, and so like if—if—if what's best for the Facebook share price, because that's what I own now, is slower growth of Instagram, faster growth for in—for Facebook, uh, you should be fine with that. So I don't really understand the—the problem here. It does seem like he has a bone to pick. Lots of people do. I was listening to Palmer Luckey talk about the uh, Facebook acquisition, and he said that the thing that got him over the line was that Mark told him that yes, we're going to acquire you for—for uh, single-digit billions, I think three billion, something like that, but if you let us acquire you, we will invest $10 billion a year for a decade in VR and AR technology in Reality Labs. And—and Palmer was running the numbers and was saying, "Well, I'd either have to raise a hundred billion dollars to make that happen, or I'd need to sell you know millions and millions of headsets like every quarter to justify that type of R&D investment." And so in terms of like making VR a reality in the way Palmer wanted, this was a great option. Um, and of course, like Zuck did do that; he did actually back that up; it did wind up investing in Reality Labs very heavily. Of course, there was all the political stuff and the fallout from that uh, that—that left like a bad taste in Palmer's mouth, obviously, but then he built Andur, so it all worked. Yeah. I'll be interested to hear if Adam Mosseri ends up having to—is—is basically dragged into this because he was head of product at Instagram; uh, this was, you know, I think long after uh, Kevin had left, uh, but—but ultimately he was the VP of product management at Facebook during the period in which they were kind of integrating the platforms and then eventually kind of moved over and focused exclusively on Instagram. Yep. Uh, another quote from Systrom here: As the founder of Facebook, he felt a lot of emotion around which one was better, meaning Instagram or Facebook, and I think there were real human emotional things going on there. That's funny because the—the—the diss on Zuck during this—oh, he has no emotions; he's a robot—and now it's like, oh, he's too emotional. Uh, I don't know; it's hard to pick. Uh, it seemed like it all worked out, and it didn't—it didn't feel as a consumer like Instagram—oh, they're not investing in this; oh, they're trying to sunset this—it was like, no, they're trying to monetize this like crazy; they're putting more ads; they're adding video, stories, Reels; everything; they went crazy with the uh, uh, w—with that business, and it became very, very, very, very successful. So I—yeah, I mean, seems hands down one of the best consumer tech acquisitions of all time. Yes. Can you think of anything bigger or better? It's not like WhatsApp. Like WhatsApp was significant and was, but it was also 19 times the cost and doesn't generate nearly uh, the direct revenue to the strategy. But uh, I'm sure we will have uh, more to talk about on big tech strategy, but we have the founder of Perplexity in the studio, so welcome to the show. How you doing? Thank you. Thank you for having me here, John. Uh, I'm doing—What's going on? Can you tell us about the announcement today? What are you announcing? Who are you taking a shot at? Well, uh, we are uh, going to be partnering with an OEM. I think uh, Bloomberg's already written about it. Uh, Perplexity will be pre-installed on the phones of u—the—the Aram, and uh, we'll be able to like push notify all those users to set Perplexity as the default assistant on the Android phone. Uh, this is a pretty big deal because um, until now, like people would—OEMs would just not even take a meeting with you if you ever go to them and say, "Hey, like, what do you think about using an alternative assistant or an alternative search?" They'd be like, "Google's just paying too much money; I'm not going to do anything." So it took this long to actually build something pretty differentiated and new, which is obviously, you know, taking actions. Uh, we put out a Twitter announcement this morning, too, saying uh, we got most of it working on iOS as well. So I feel like this is the next stage, like, you know, moving all these AI chatbots to native assistants on the phone that uh, not just answer your questions, but also help—help you get things done. Is there any hope that um, there will be a more open ecosystem on the iPhone? Uh, I've used—I've used the shortcuts function on the action button; it's pretty janky. Uh, I'd love to just remap Siri. I love Apple; I love my phone, but I don't love that particular product. I'd love to swap it out with one of the more founder-led AI companies like yours. Um, is there any hope there, or is Apple just too dominant in that, and do they view it as too valuable? I would—I would assume the latter. Um, I think my hope is at least like, let's start with Siri being able to call multiple AI apps, sure. Um, you know, um, they let the user provide some preferences on what apps they like for what different things. Um, at least we can start with that, and um, I think calling other apps could be interesting. Uh, most of—actually, to be fair to Apple, uh, a lot of it is still exposed in the Apple SDK through the events kit, uh, so that's how we got things done, like calendar, mail, reminders; uh, these are all part of their SDK, so uh, you can actually call it—podcast, Apple Music, uh, that's Apple Maps; all these are possible to integrate into. You cannot do stuff that's more native, like alarms and volume or brightness, uh, you know, people want—and everything—assistant at the end of the day. Sure. So that's their advantage. Um, so I hope like, you know, what they can do is stuff that's so easy and obvious, like setting an alarm or making a phone call or sending a text message; they can just continue to be pretty reliable there. Mhm. But uh, anything that's more multi-step in nature, uh, they let the user invoke their favorite AIs to get things done. Mhm. Uh, what uh, what lessons have you learned from the antitrust history in big tech, and what are you watching today with the antitrust stories that are unfolding around Google and Meta on Capitol Hill recently? Yeah, so uh, one of our executives, Dimitri, is testifying today uh, for the Google versus DOJ case. Um, I've already wrote—wrote on X or core points. Uh, I don't think Google should be broken up for two main reasons: one is it's not even in the interest of America for Google to be broken up, and uh, number two is it doesn't actually increase competition; uh, it's just going to like uh, transfer from one monopoly to another. Yeah. Uh, and actually they've done a pretty great job at uh, helping other people build browsers, like Edge. I mean, if—if you want to use the word wrapper since I get accused of it a lot, uh, Microsoft Edge is a Chromium wrapper. Yeah, right. That's right. And Brave was a Chromium wrapper. No, that's great. Everything. Everything, every other browser that's even managed to take even a tiny bit of market share from Google Chrome has been built on technology that Google did, so we should actually credit them for that. And uh, I don't trust another organization to maintain that open-source repository uh, in the same way that they have. Yeah. And uh, the other—the—the other thing, honestly, though, that we're pushing back on against Google is their—how they couple uh, OEMs to keep them as a default and don't let the OEMs put in Play Store otherwise. M—so basically it's very simple to understand: if—if there is like an enthusiastic OEM who wants to ship AIs on their phones, u—what Google does is, okay, you can do whatever you want; it won't be an approved Android version of Google. Um, and—and if it's not—um, you're not allowed to put Maps, YouTube, uh, Play Store—the one that this most is the Play Store because nobody can see a phone where you cannot install other apps. And most developers of other companies are not interested in like maintaining versions of their apps for multiple Play Stores; it's a lot of work. Even Samsung couldn't really get Galaxy Store to work. And Meta, Amazon, all of them tried making their own phones and fail for this very reason. Even if you fork Android and try to make your own phone, uh, you have to convince everybody to actually ship to that particular new Play Store and maintain it and keep improving it, and you're not going to be able to share the uh, Play Store ad revenue and subscription revenue uh, with—with the um, OEMs that Google can. So this is the primary problem. And uh, but that said, like, you know, the assistant that they have, Gemini, is actually pretty horrible. And the model is great, but the product is not, and there should be no reason to uh, force people to keep that as the default when you have an inferior product uh, just because you have all these deals and lock-ins. So that's what we are pushing back on in our testimony. Yeah. Just because you build a great foundation model does not give you a guarantee that you'll win at the application layer. And so these wrappers, although they've been derided like that—that's the UI layer; that's what's so important to the actual consumer. Um, I want to talk about uh, social networking in the context of foundation models. We've seen the partnership between uh, the merger between X and xAI. Uh, there's been rumors that OpenAI might be thinking about doing some sort of social network. Obviously, Zuck has been able to vend Llama into all of the Meta core products. Uh, what do you think the future—does every—does every AI company need a social media dance partner? Are we going to see uh, Pinterest and Snap do things? I know some of them have—have experimented with AI features, but haven't brokered a huge partnership yet. Uh, what does the future look like? Are these two technologies like intrinsically linked and destined to be part of one organization, or can an AI company operate independently forever? Um, I mean, I—I don't—I don't know how social and AI go hand in hand; it's not—it's not that straightforward. Like—like if it was, then Meta AI would have already taken off pretty massively, right? Even though it's not the leading model, uh, you could imagine most of the uh, like say, let's assume ChatGPT, um, you know, get like a billion queries a day or something. Um, I don't think like—like 60 to 70% of them are probably going to be super simple enough that Meta AI is going to work for it totally. Uh, you don't need the fancy models; in fact, most people using ChatGPT in the world don't even know there's a model called 01 or 03 and—and don't even know what the difference is for GPT-4. So um, I would say the main problem is—it's not—AI is still very single-player. Uh, the only experiment that I feel took off on social uh, is this uh, thing that we did first called Ask Perplexity Bot on X. Yeah, I remember that. And um, of course, all the time in the comments. Yeah, it's great. xAI also, you know, implemented that with at Grok, and they have way more advantages because they own the platform, no rate limits, and like they can drive installs of their app through that, so it's—it's like a pretty good uh, experiment that worked. So I could imagine Meta doing this like, you know, on Threads; that could be like a—add Meta AI, and—and or it could automatically reply; you could do all sorts of cool things. Y—so it's more like the other way where a social network can help you grow usage of an AI app pretty—pretty immensely uh, compared to like an AI app benefiting from uh, social layer so far at least. Uh, but if you could figure out social for an AI app within the app itself, uh, it can definitely make it more daily usage and high retention, which is what most AI apps struggle with today. Can you talk about how you're thinking about your own models going forward versus leveraging the existing models? You guys have Sonar, and then our uh, 1776. I'm curious how you're—learn—where—where you're looking to uh, focus going forward. Um, I think we'll continue to uh, keep the same strategy, which is um, have a version of our product that can run with our own models but not hinder or disrupt the user uh, from this best experience that we can provide to them. If we're not able to do it with our own model, we'll just use other people's models; we have no problems with that. Uh, my—our belief is that nobody's going to have a lead in uh, the—having the best AI model for more than a few weeks. The—the pace at which the field is moving—like Anthropic did the 3.7s on it, uh, and then uh, within a few weeks, like Google did Gemini 2.5 Pro, which was way better. Grok 3 came out, and then OpenAI has 03 and 04. There's always some debate on like what is the best model, but they're all looking the same, and they're all good for certain specific sets of things. Um, do you think in—do you think in five years the average consu—internet, you know, user will have a favorite model, or they just won't even know the underlying model that they're using, and they'll just be sort of um, you know, the—the software application layer will just be serving, you know, the—the—the most effective model for the task? Uh, I think it's going to be more like the second case. Uh, the main reason I believe that's going to be like that because um, everybody's just chasing the same benchmarks. Everyone has the same set of benchmarks: LLMs, LLaMA, arena, GA, GPQA, and—and Humanity, last exam, uh, and—and they're just trying to like show their AI is the best, and so they do all the same sort of things that you do in RLHF. Um, and u—it ends up being making the models like look similar. Yes, there are some like tasteful things some model developers do, like some people like the way Claude responds, uh, some people like Grok's satire, but these are all like easy to build; it's not that hard. Uh, the difficulty—which is why everybody's working on the difficult problem—is making these truly smart and like better at all these reasoning tasks, and so that's going to end up making all these models roughly similar-looking to the average user. That makes sense. Uh, when do you and the team decide when something is ready to ship with the—the voice assistant today? Uh, one of the massive advantages that I think you guys have is uh, the culture of Apple specifically is about perfection, right? It's about pixel-perfect design; you know, it's not—it doesn't feel like the culture is sort of comfortable making uh, mistakes. You can see the uh, even iMessage summaries being very embarrassing to them, right? Whereas like as a uh, you know, big company now, but—but still relatively young, you guys have the advantage of being able to, you know, move quickly, try things, and—and iterate. Yeah, so I'll literally read out to you the chat I had with—with our engineer who worked on this: "Be good to launch and announce today?" Question mark. And then he's like, "Yes, still making improvements, but let's launch." And me: "We want to address some of the onboarding concerns people had, like all the—which calendar or which maps to use." And he's like, "Yeah, they're valid concerns, but I'm worried about waiting too long to launch." The Apple engineer in me says to wait until it's perfect; the Perplexity engineer in me says that's why Apple has never launched anything. So it's great. And then I said, "Okay, let's roll." Do you have a sound effect for us, Jordy? Founder mode. Okay. Anyway, uh, I—you posted uh, what—what's a social app that doesn't exist yet, but you wish existed? Doesn't need to get to 100 million or billion users. Uh, what did you learn from that? What was exciting, and what do you think uh, you might stay away from? Yeah, yeah. The major learning was the OG Facebook, mhm, and—and Usenet. Uh, these were the two main ideas. Marc Andreessen actually talked to me after that; he's—this is like his pet—like—like, you know, idea that he always keeps coming back to, apparently. And uh, Bloomberg Terminal is another interesting thing where, you know, nobody thinks about it as a social network, but it is what it is today. People are not leaving it because of the network it has, right? And it's very elite because you have to actually pay that much money to get into it, but then because of that, the quality of like exchanges are also high, or at least it gives you this feeling of exclusivity, which is what Facebook uh, leaned on to in the early days where they were only in Ivy League colleges and stuff. Um, and so I—I think that concept can be explored. Clubhouse definitely tried that and failed, so it's not like, you know, bulletproof, but that can certainly be explored again. If you want to build an app that doesn't have like too many users, you need to gate it in some way, mhm, whether it's by pay-to-play or like some other kind of uh, eliteness criterion; it's not clear. But then what—after you get in should also be uh, thought about. Yeah. Uh, the quality of exchanges should be high, like in—in some sense, Elon has turned X into something like that. Uh, there are a lot of trolls, a lot of random accounts, and all that stuff, but by forcing people to have the blue mark and paying $8 a month and um, and all that stuff—

I think he's already made it like much better than it used to be. So, uh, a more aggressive version of that could be explored. He could explore it himself, like if there is a super premium version where, um, you only get to like respond to some people if you're paying even more. Um, how how would that make the quality of exchanges on the platform? It's not clear. There's the other side of the story where some people like Bill Aman come and respond to random people on X, and and the reason he gives us is like, "I like the attention; I like I like, you know, I'm even on a vacation, I want to come and like keep replying; I like the dopamine hits of like getting notifications," you know. So it's not like if you get billionaires, uh, they want to stay with the other billionaires or stay exclusive; I think they actually like using a product like X to stay connected to the normal people.

Yeah, okay. The idea yet, I want to pitch you an idea that we've been kicking around here might be terrible, but uh, in terms of like the LLM research AI-driven social network, I find myself doing a lot of very niche deep research reports. Sometimes I did a whole uh deep dive on deep research report on Thoma Bravo, and then I did another one on the the story of Johnny Carson, the the the original host of the Tonight Show. Uh, fascinating. I don't know how I wound up down that rabbit hole; I really enjoyed it. Uh, if I'm on Perplexity and I find some and I ask some really interesting question because asking interesting questions is often more more challenging than getting a great answer, uh, what if I just could just uh click a button and share publish it to an internal network and then Jordy can follow me on Perplexity that he can see my most interesting questions? Not not going to see the whole feed because maybe I'm asking about a me a healthcare issue, but if I choose to share it, I'm just sharing my result, and then they can click and, "Oh, John was interested in you know this specific thing; maybe I'll take a tour down his research result." Is that an interesting option, or is there something I'm missing that's like that's actually a terrible idea?

No, it's not a terrible idea. We've exact we've considered this exact idea, and uh, you know, obviously we the more I keep working on features, I people ridiculed me that uh, "What are you doing when your core product is failing and buggy and like going down like you're going down the drain and like you're taking the company down?" So all I I read all the stuff too, so don't let the haters get to you. Don't let the haters get to you. Oh, look, I think some some of them have a valid point though; like more features you keep introducing to the product, you just lose sight of the core focus, why the product even exists in the first place, and making it like a pretty bad experience, and then you and you win on neither of these things. But I do agree with you, like it'd be great to share uh the the Perplexity threads or like that you create. Uh, we have this thing called Discover, uh, and uh we're trying to increase the volume of content on Discover. Mhm. And once we do that in an automatic way with curation to users' interest, like we're going to let some of the users p try to like publish it to Discover on their own, just like a creator on YouTube does it, and um and then if a few people engage with it, like the TikTok algorithm, we're going to try to surface it to more people. That is the long-term idea; that's why I'm even having to discover a lot of people even ask me why why does this thing even exist. Yeah, uh, because I I want to keep it as the optionality to like have a social product within Perplexity.

Yeah, that makes sense. Uh, if you have another minute, I'm curious how you're thinking about shopping; how much time you and the team are are putting into this. It's obviously a very exciting opportunity, uh, and I know you've shared some about what you guys are doing uh before. Yeah, so last year, towards the end of last year, we launched Perplexity Pro shopping where you can once once you do a shopping query, we're not just giving you the the answer; we're giving you the product cards, and we could even let you buy directly from there. Some of the things we surfaced from Shopify with their API and like let you buy with ShopPay, and some of the things we ourselves integrated with some merchants and like uh did the check-out flow ourselves, and um you wouldn't even have to go and check out on the merchant side; you can just do it directly on Perplexity. We thought this would be the best uh innovator's dilemma angle against Google because Google uh even if Gemini gives you shopping answers, they have no incentive to like let you transact uh because like the they lose they make money by sending you to merchant sites. That's why Google Flights doesn't let you book flights on Google; uh you have to book it on Expedia or Booking.com; hotels, same thing. So uh I think that's that's why we worked on travel and shopping where we let people try to book directly. We have some ideas on how to incentivize it, like free shipping or like x% off on hotel bookings done on Perplexity, uh, and but but what we learned the hard way is people actually want really good results first; that's why they're even coming here; they don't care if they transact here or not; it's it's a secondary thing uh that is only if if the result quality is such so good are they interested in the last step, and we uh didn't quite get that right in the whatever we did end of November. So that's what we spend time on the first quarter is to just improve the UI, improve the latency, improve the relevance, make sure the cards are like up-to-date, high quality, uh filter all the low merchants uh like low-quality merchants, uh make sure like even images for every card is present; like this is all a lot of boring work that LLMs don't solve. Same thing with hotels; TripAdvisor data is great; we work with them, but like there are so many other places on the web where there are good reviews that people want to know. Mhm. And you got to like aggregate all of that, and this is why I think like model companies are not guaranteed to win the application layer because you have to work on all these boring things, and then sometimes when people book a hotel on Perplexity, like when you actually go to the hotel, like they say the booking never came through. Uh, so there's a lot of pro real-world problems you hidden; like packages might not get delivered; you don't have a way to track the package because the merchant still does the shipping; we don't do that. So TikTok, if you notice, TikTok has a shop tab uh which literally looks like Amazon now, and I heard like they even have their own fulfillment in Seattle; they hired people from Amazon and like they're doing their own shipping; that's crazy. So that's my lesson from this is you want to go to a vertical; you have to go all in and like nail like like be prepared to like uh play the long game there.

How do you think about uh manufacturing viral moments? You guys had a cool experiment with your uh Super Bowl uh activation; I'd love to get a postmortem on that, but then you're also not afraid to go and you did this sort of Squid Games campaign that I saw getting a lot of attention as well. Um, how do you how do you think going forward sort of balancing these more like scrappy kind of organic activations versus, you know, going big and and working with, you know, global celebrities to build the brand?

I think we should keep being scrappy. Uh, the thing it's it's by the way like uh he's a pretty global star, like well-known star uh like people recognize his face pretty quickly, but he's also not like as expensive as Hollywood people, and so uh that's why we decided to work with them, and uh the other other thing is like it's the concept that matters at the end, right? Like coming up with the right concept is more important uh than like who you work with. Um, and we we we'll still try to keep doing these one-off viral moments; it's all it's take about taking bets; not all of them land. Um, the Super Bowl was good; like in fact like I the retention from people who came to the Super Bowl exceeded my expectations, uh, and um again like whether it's better than doing Instagram ads is not clear to me yet; we we have to explore all these different platforms. Um, but I do think there's like, you know, two things you benefit from, right? Like we did this thing with Ben Shapiro where in his podcast like he pulls up Perplexity and asks questions. Um, I think like uh that doesn't actually convert; like you cannot track performance, but uh it it it it does lead to more brand awareness; like if they're like he has a lot of listeners and like it's a different way of doing ads on a podcast where you're not actually like having him say, "Perplexity is awesome; go and install it." It's more like watching him use it, and then you learn what the product is and how what it's meant to be used for. Kind of inspired by Joe Rogan's "pull it up, Jamie," where all the time Google's being used there.

Yep. Um, yeah, we do that with Polymarket; we pull up Polymarkets all the time and recommend that people go and download and install. Yeah, that that's pretty awesome. Yeah, it works really well. Uh, I'm I'm curious; you mentioned earlier uh, you know, uh people sort of yapping on X, you know, trying to trying to uh sound in on on your product strategy. How do you how do you balance like, you know, on X it's like an echo chamber of people that are in Silicon Valley working on AI, but like Perplexity in theory doesn't really care about people on like, you know, in San Francisco uh or or New York or these hubs, right? It's like you care a lot more about like, you know, uh some you know random person in in um Arkansas's like, you know, you know wanting to wanting an answer about something and using the app. How do you think how do you personally kind of like uh...? Yeah, everyone in Silicon Valley could switch to DuckDuckGo, and Google would be unaffected; that's like the lesson from the last era, and so you could do the same thing, but yeah, I'm interested to hear your take.

I think this is both my uh like weakness and strength. I I I spend a lot of time on this platform, so I understand it better, but then uh like we are living in a bubble; like there most people are only using ChatGPT, and and uh they don't they haven't even heard about like most of the other apps. I think Elon has managed to break that a little bit because he has 200 million followers, so essentially he reaches a lot of people. Um, but that's kind of why we want to do more of these uh Instagram-related things, um the commercials, uh Super Bowl; like these are our attempts. Actually, if you notice the Super Bowl tweets uh didn't actually matter uh on X; like most people made fun of it or didn't even engage with it, but it did help us get a lot of like mainstream normal people uh become aware of the app and use it. Uh, same thing with the Legion J commercial; like it helped increase our usage in other countries outside uh, you know, countries that don't even use X. So uh I believe the platform that has the most people in the world is Instagram, uh for good or bad; like if you go to any city outside the Bay Area, just watch which app like people are having on their phone; it's mostly Instagram or WhatsApp; it's not really uh X.

Do you spend a lot of time thinking about AGI, or is there just enough uh consumer application uh features to build that that it's it's almost a distraction?

I do spend time thinking about it. In some sense, it's a unique opportunity to have like uh the ability to have the front end which people can feel the AGI, let's say. Yeah, uh uh so you do want to think about experiences where people can be made to feel AI, and that creates that jaw-dropping magical end-consumer experience. So so that's kind of why we worked on all these pro searches, this voice assistance agents; we're working on the browser browser agents; the you can only do all that if you're like making some predictions of where these models could potentially get to and try to build before even they get there. You don't want to be late; you want to build at the right moment where like kind of like how we built Perplexity around the GPT 3.5 time, not GPT-4 time; it would have been too late then. You need to pick your moment, so you do need to think about AGI. That said, I'm not a believer in like, you know, um fear-mongering. I I just think uh it's already happened; like the genie's out of the bottle. China versus America; that race is already going on; nobody's going to slow down. Um, and everybody wants glory; no nobody cares about other people or something; everybody wants glory and and and some kind of power to control the future. So uh let's just uh like like accept the truth and move on and try to like make sure everyone knows how to use the AI so that their livelihood doesn't get affected.

That makes sense. You're I I was thinking about it; I don't know of another 10 billion-plus uh AI startup uh founder who's not constantly promising that AGI is like two months away to get, you know, you're just like, you know, "Let's focus on our users and focus on the opportunity." That's probably the right thing. Uh, I mean, speaking of the the the foundation models on a more practical level, uh where are you seeing the most interesting vectors of optimization? You know, people are focused on large context windows, uh huge pre-training runs, but there's been some debate about are we hitting a pre-training wall? Are we hitting a data wall? Uh RL's very hot right now. Um, yeah, where do you think the foundation model labs need to go, and and what are you specifically excited about? I imagine that maybe a code generation model not super important to your business, but something that's more knowledgeable and hallucinates less about facts, extremely valuable. So what do you want to see?

I think RL is the uh place where most investments are going to go to, um especially with with models like GPT-3 that are able to do tool calls pretty natively uh rather than being prompt-engineered to do that. Or like, for example, before GPT-3, the way we built like our agents is there would be one model that came up with a plan for the query, another model that would execute the plan by converting the plan into like smaller queries, filtering links, calling searches, and then another summarization model that actually takes all the results of the planner and the router and like summarize. Now it's all like one single model; uh that's great; like that that that means like you have to like rebuild it; you have to throw away lines of code and rewrite it, but we've been doing this uh since the beginning; like people think like Perplexity has remained a stagnant code base or something; it's always changed as soon as like models became more capable. But the interesting thing about the native tool call kind of models is that if we want like a sonar version of the product that runs on our own uh setup, uh we also see need to start doing post-training beyond just uh training for instruction following and summarization, but uh like also like tool calls and and and completing tasks using RL, uh and and we hope to collect all this interesting data through the browser uh platform where like people are giving tasks on on our browser, and then we obviously see some of some of the agents will fail there, where we'll collect all the data and like try to uh create like positive trajectories, create like eval suites, and then uh do a post-training on that. And so the nice thing is a lot of open-source code bases exist on how to replicate GPT-3 or PO and post-train these models, and we've been doing that work already. So uh that's where we plan to invest more resources into for this year.

How are you thinking about advertising in the context of search? It's historically been an ad-driven business, but we're seeing a lot of AI product companies just charge 20 bucks a month, $200 a month, $2,000 a month; who knows. Um, is there a future where if you search for "what's the best corporate card," Ramp is going to show up at the top if they bid bid on that?

Hopefully not. I I think that's that's the main reason why people like using the AIs; they they think it's giving them something that Google doesn't offer, and and um so I think the subscription revenue is very healthy and positive; like OpenAI is making 10 billion a year or something like that, right? Or 7 billion something like that. So definitely uh that's only going to grow, uh and and if AI starts doing things not just answering questions or or writing code, uh people will pay even more because they kind of think about it as hiring somebody. Uh, if you look at the amount of people people pay for personal assistants, chiefs of chief of staff, personal chief of staff, estate managers, nannies, like you know, it's a lot of money, and there are like a lot of people who can afford all that, and we're talking about uh doing something 100x cheaper and also economically 10 to 100x more valuable in terms of end output. So uh I I think the subscription uh TAM is way bigger than what it is today. Uh I I believe like you can do a lot of interesting things with memory where once you understand the user deeply enough, uh the user can probably trust you if you show them relevant sponsored content as long as it's super uh personalized and hyper-optimized for that user. Instagram has shown some stats where the engagement time on their platform reduces if uh they remove the ads because that's level of personalization of the ads. So if if uh any of the AI companies can do that, uh I think that could be like a thing where brands could pay a lot more money to advertise there. So that's yet to be explored, but in order to crack that, you need to crack memory properly; that's kind of one of the other reasons we wanted to build a browser is like we want to get data even outside the app to better understand you uh because some of the prompts that people do in these AIs is purely work-related; it's not like that personal. On the other hand, like what are things you're buying, which hotels are you going, where are you, which restaurants are you going to, what are you spending time browsing, tells us so much more about you that uh we plan to use all the context to build a better user profile and and maybe, you know, through our Discover feed, we could show some ads there.

Makes no sense. How quickly do you want to launch Comet? Do you have a you have a launch date? I'm assuming you want to launch this afternoon.

Yeah, it was supposed to be out by now; we got delayed. Uh, I think partly because we underestimated the difficulty of the project and partly because we tried to do multiple things, and so uh we tried to scope it down, and we're aiming to get it out by mid-May.

Awesome. Good luck. Um, last last question I have: you posted, "ByteDance of America is worth building." I'm assuming you're you're building the ByteDance of America.

I hope to be able to, but we need to earn the right to do that. So uh let's first like, you know, I want to succeed with the Comet browser; I think that'll be like the real second product of Perplexity; everything else we launched like the assistant or the like like apps, platforms is all like just different versions of the same thing; the Comet will be the first really truly different product, and um I think if we can do that a second time, um I believe like we can and and Discover that's the other product we're trying, though it's the app right now, but you could imagine spinning it out as a separate app too. Uh, you could imagine us like earning the right to do that. I think there's a lot like I've spoken to the founder of ByteDance, and one thing he told me is they they're structured in a very different way; it's not like uh like CapCut or TikTok have like different growth teams; uh they they do have their own growth teams, but there's one team at ByteDance that takes care of infrastructure for all their companies, one team that takes care of growth for all their companies, one team that takes care of front-end mobile development for all their apps, and so they share knowledge across apps very quickly; it's insane. So that sort of structure doesn't exist in u in in the US; like take Google is actually the closest to ByteDance of America if you think about it; they have so many different apps in one umbrella, but they they don't share lessons; like the YouTube team is so different from the Gmail team, sort of from the Chrome team, and that's why it takes so much time to collaborate. So you need a very different leadership structure and a culture to make it happen in America.

That makes sense. Uh, every time I ask you one more question, I get one more question, and this I promise is the last uh without Perplexity.

Yeah, yeah. Um, without uh violating any NDAs, what what's going on with with TikTok, speaking of ByteDance? Uh, we there was a big flurry; everybody was submitting, you know, bids, and then it's been quiet. Is there any do you have any insight that is uh not necessarily confidential?

Yeah, what is...? I really don't. We we we've uh submitted our bid; we never expected to be the leading candidate or anything; we're a very small company compared to them; our bid was more interesting in the sense that everybody else who bid for them did not want to do anything to do with the algorithm, uh but I felt like the core problem is um the fact that like the algorithm is controlled by China in some form or the other. Yeah, even if they say, "Okay, the Chinese app is separate," apps running outside the US and outside China like European countries are all sharing the same code, and so they can get to like use that data and influence u you know what feeds people in America see. Uh, so I think like that's where we wanted to do some real work, and the search bar is another place where we wanted to do some real work. We thought our proposal was pretty interesting, but uh there are some, you know, we're not a data center company; we cannot guarantee them security and all that; Oracle can. So uh we'll see what happens. I think they've delayed the decision, and it's probably going to be coupled with the tariff situation too.

Yeah, that makes sense. Well, thank you so much for coming on. Uh, thank you for wearing a suit as well; you look fantastic, uh and uh you guys are you guys are killing it, like uh...

Thank you. Technology Brothers; I like the name.

Yeah, it's awesome. Well, it's great to have you on, fellow Technology Brother, and come back on anytime when you have news.

Yeah, we'll talk to you soon. Thanks so much. Cheers. Bye. Uh, on Polymarket, who will acquire TikTok? AppLovin has shot to the top of the charts, but on Lowval, uh they have about $10,000 in a $2 million market uh in terms of volume, but AppLovin is at 20%; Oracle/Larry Ellison combined if you consider them one entity, 21%; Oracle's at 12%, Larry Ellison's at 9%, Microsoft's at 8%, Amazon's at 8%, Tim Sweeney at 8%, Frank McCourt at 8%, Alexis Ohanian at 6%, uh Perplexity is down at 4% right there next to MrBeast at 4% as well, Walmart at 3%, who we discussed earlier uh would actually be the the Ben Thompson choice. We should get ourselves in the mix, and I don't submit a fake bid just to go viral; we're not allowed to uh bid on anything on on Polymarket, but uh but but we should we should throw our names in the hat, TikTok. Yeah, we should just bid on TikTok.

It's just going to be. We're going to. No more slop content. It's just. We considered doing a press release around it, but it never uh never hit the wire. It played out pretty quickly anyway. Uh, let's do some timeline while we wait for our next guest.

Tyler uh says, "Scoop I found TBPN's hidden warehouse. Nice try," and he finds a picture of the brothers' supply. I wonder where this is, but we always love when fans share fun photos. Found us. Tyler caught us. Uh, we are in the in the market for a new studio, hopefully moving into one soon, and uh this this brick building looks like it's uh fireproof. Been in it's Lindy. It's been around for a long time, probably safe, probably great to record a show there. We love a fireproof.

Luffy says, uh, there was this news yesterday in TechCrunch: Ex-Meta engineer raises $14 million for Lace AI, a revenue generation software startup. And this was shaking up the timeline. Everyone was quote tweeting this. Uh, Luffy says, "Hey Lace, make a 100 million AR software for me. Do not make any mistakes." Fantastic prompt, by the way. Use it, use it on any of any your preferred model; it'll it'll work. Why don't people? Why don't more people do this? I don't I don't know. It It seems like one of those obvious tian sort of secrets secrets I suppose. Yeah. Um, no, but Lace uh Lace I guess is is focused on maximizing revenue for your call center with zero extra investment. It's more just like the TechCrunch like headline went weird, but again, I think people wanted to joke. Yeah, a lot of startups don't generate much revenue. Yes, yes. So they're like, "Hey, we will our business is revenue generation." Um, but uh yeah, TechCrunch is back. I don't know if this is accidental, but I feel like again and again we're seeing more TechCrunch headlines post-acquisition, so they're doing well. Great to see. Great to see.

Patty, I just wanted to give Patty a shout out because he's a friend of the show. Patty uh he says he's proud to announce he's starting a profitable startup. VCs, my DMs are open. Uh, so yeah, if you want to get in touch with Patty, he's free agent. Could be picked up at any moment in Japan live streaming. Could be picked up by a venture capitalist or company, but uh we love Patty on this stream. So this post from East Village guy is: "32 late for me to lock in my brother. Ray Croc was a 52-year-old traveling salesman when he met the McDonald brothers. Get a grip. Awesome." And I thought this was worth highlighting. There's so many great stories of this. Uh, Enzo Ferrari is another; he was 45 when he started uh Ferrari. Granted, he had spent a couple decades in automotive racing. Founder of Zoom, founder of Workday, both 60s, 50s when they started their companies. Red Bull founder too, right? Dietrich, I think he was pretty old. Uh, founder of Monster, very old, uh which was very funny because it's a very young brand. Estee Lauder, there's tons. Yeah, there's tons of these examples. Never too late to start doing your life's work and start a generational company. Just do it. Why don't? Why not? Yeah, Estee Lauder was 38 years old. Why not just start a power law company? More people should do that for sure.

Anyway, we have this bizarre TikTok or YouTube video. It's the third most viewed video on YouTube this week. Is an AI-generated short of a pug that saves a baby from a plane crash, and then they try surviving on an island. Can we play this? It's it's uh this is the future, folks. This is this is entertainment now. It's great. Parachute. It does have a compelling narrative, you know, inciting inciting element, inciting action, you know, turn of events. Will he save the baby? The sound effects are brutal. I know. It's really well-designed. It's so optimized for that 400 million views. 400 million views. Yes, 400 million views. Feeds the baby, starts cooking over the coconut, cooks fish over the coconut. That is adorable. Uh, feeds the baby some sort of fish stew and then writes SOS in the sand and their voice, and then for some reason the military shows up and they like have like guns and stuff. Overnight success. And they overnight success and they rescue the baby and the pug, and then it just ends, and it's like the ultimate like you you don't expect it to end, so you watch it again. Fine art. That's the future. It's the future again. Uh, you know, clearly some some human element in there figuring out what's viral about it, but uh pretty sloppy. Pretty sloppy. Little sloppy. Well, you know what's not slop? AI Grant Batch One was absolutely insane. Jeff Huber, who we had on the show, shared this, so I didn't realize this was a throwback. Yeah, so going off uh the initial batch was in there. Curs. Uh, I know Chroma, Wombo, who else do we know in here? Pretty cool. Pixel, Cut, Dust, Forefront, uh just so early. I wonder when AI Grant Batch One was. This is the Nat Friedman project, correct? Um, very very cool uh little like under the radar, I guess. Was it structured as a grant? It wasn't even it wasn't even YC style, or was it? Did they take? No, it's an investment via no cap, no discount, MFN safe. Okay. Yeah, yeah. Pretty standard. So pretty standard investment. Uh, and you can imagine they've done pretty well. Yeah, I think some other cool I I'm almost sure Julius went through a later batch. I've saw some other folks go through. Uh, pretty pretty cool. Julius was in Batch Two. We have our next guest here. Let's bring him in. You guys told us you yearn for the minds, so we brought the man himself, Miner. What's going on?

Doing great. How are you, John?

Fantastic. What's happening? You're looking great in that suit. Welcome to the show.

Thank you. It's Technology Brothers. You need to need to dress up. You got to dress for sure. And I love the map too. That's the only market map that I care about personally. This is this is our target market, at least the market map. Uh, what what's actually going on with that map? Uh, what do the different colors represent?

Yeah, this is a a geological map showing kind of the common rock types in uh a given area, and you can see the central and uh eastern US are not quite as exciting as the western US, and that's where all the minerals are.

Interesting. Uh, can you give us a brief overview of you and your company just to kick us off?

Yeah, absolutely. So Ted Feldman, founder of Jiren, started the company about a year ago. We are building and operating automated diamond jewelry rigs used in mineral exploration. So I'll start out kind of high level. What, how does mineral exploration work and kind of get into what is the actual problem we're solving? So basically, building a mine is in many cases a billion-dollar endeavor. Incredibly expensive. In order to justify this capex, so you never have a really good understanding of what's actually going on underground, and so you have this kind of decade-long exploration process where you'll start out maybe geologists walking around; they're seeing some interesting rocks. You can do geoysics, look for a magnetic anomaly, a gravity anomaly, or do some seismic surveys. You can do soil sampling, kind of pick up some uh some kind of tiny holes, see if there's kind of trace elements of what you're looking for, but really in order to understand what's underground, you got to drill. And the main type of drilling used in neural exploration is called diamond drilling or core drilling. This is basically where you're collecting cylindrical core samples of the rock anywhere from a few hundred meters to a kilometer or more um underground, generally a few inches wide. You pull it up in 3 m intervals, but really once you collect these core samples, you send them off to the lab. The lab will tell you exactly what the composition is. You do that every foot or every meter across hundreds of holes. You plug all this data into a 3D model, and then you have a kind of. Yeah, 3D model of the the subsurface. You can visualize where is the mineral deposit, see how large is the resource, what's the grade, the kind of percent of the metal you're looking for within that deposit, and then make a determination on whether it's economic demine or whether you need to collect more data in order to make that determination. And the problem here. Go ahead before you go into the next uh segment. Can you talk about the value chain? Is there different groups that are doing you know the core of like research, and then do they sell the rights, basically say like, "Hey, this is worth spending a billion dollars, but like we're not going to spend a billion dollars, like you should do it," and they sort of like sell access to it effectively?

So, so the way this will normally work is you have an exploration company, a junior explorer, junior minor. They either own the land outright, they have rights to lease the land, or they have some claim on federal land, but they have the rights to to mine in a given area, and they basically have a hypothesis uh written in geology that there is some valuable deposit underground, and they raise capital primarily from public markets. Many of these companies go public extremely early on on the TSXV or the ASX, and they raise capital and spend most of that capital on drilling to actually collect more data. And so then this constant cycle of raising capital, spending most of it on drilling, analyzing those drill results, updating the geological model until they can either raise enough capital, raise a billion dollars in equity debt to actually build the mine, or what happens more often is sell the mine or sell the deposit to a larger minor that will actually um have the money on the balance sheet in order to develop the asset. And so you. Why public markets uh versus private? Is it just because you need to raise so much money? You need to be able to basically effectively market the. It sounds like biotech, like what happens with biotech companies going early earlier out.

Yeah. Are these like typically like penny stocks where they're just trading you know.

Exactly. Mining is historically a penny stock. In history, we used to have kind of small exchanges in Denver or Phoenix in the US, but in North America, this is dominated by Vancouver and Toronto, where a lot of these exploration companies are based. Canada um has a lot more capital flowing into mining than the United States does today, and it's really just how it's been done historically. These investors, largely kind of smaller retail investors, historically they want liquidity, and they they can't uh can't can't just be called up for a private placement if you're um Joe with 100 bucks to throw into a gold project.

Did you ever, speaking of mining, did you ever join a call with a VC early and have them be like, "Sorry, I thought this was like crypto mining?"

That that has happened a lot. It has happened less now than a year or 18 months ago, which I think is a very promising trend.

Got it. That's good. Uh, h how much of the business is kind of just you need to take off-the-shelf technology and just go do the thing uh versus you need to build new technology, and and is that more in like the hardware world or the software world, or are you just like going and doing the actual exploring? Like, can you concretize what you're.

Yeah, so I'll get into what is Jiren actually doing. So we are a drilling contractor. We are building rigs from scratch and uh and operating them for exploration companies. We're starting our first pilot with the Gold Explorer in Nevada in a little over a week. Um, we'll be out in the field for that. We started building this rig about 4 months ago, and so we get paid basically per meter that we drill for our clients, and so we're not taking on the geology risk. Let's just focus on the tech. Yeah. And why are you building the rig yourself? I imagine that is that just built. You're buying different pieces of equipment and then piecing them together. I mean, you've raised money, but not that much. Like, I imagine you're not reinventing the wheel, plus it's probably like there's good drills out there. I imagine you don't need to build a new drill, or do you?

Yeah, so the initial idea was let's retrofit a rig that maybe save us a lot of money, and there this is the avenue I was pursuing for about 6 months. We really had two options: We could buy a Chinese rig for less than 100 grand. We need to add a whole bunch of sensors. I talked to a lot of the operators; these things fall apart. They are not high quality, so we rolled that out, or you could buy a western rig for half a million bucks, and you either need to hack into the firmware and probably void a warranty, which uh you do not want to do on a half-million-dollar piece of equipment, or um partner with a manufacturer and figure out um how to actually pull data from it. We tried that with a couple of manufacturers; they were incredibly slow, and all the manufacturers they sort of have their own half-hearted efforts on autonomy, and so they didn't seem particularly eager to partner to partner with us. And so almost out of necessity, we had to design our own. And now the way this rig operates is fairly similar to every other sort of rig. All all these drill rigs are basically two hydraulic rams pushing what is called the drill head, which is what provides a rotary motion to the drill rods into the ground. So all drilling is really you have a bunch of pipes that you screw together and then push and twist into the ground, and you have a bit on the end um to make sure you get a a clean cut. And so we the off-the-shelf parts that we're buying is the drill head. U that's what grips onto the rods and rotates them. Um, you have other part called the foot clamp, which just clamps down onto the bottom part of the drill string to make it easy to load a new rod in, and then ramps to push it into the ground, and then the structure is ours. And then in order to actually retrieve the core sample from the bottom of the hole, think about it like this: You have your drill bit and then you have um which is a c cylinder um that's what our logo is actually, and then you have drill rods coming up from that, but inside of that bottom rod you have another tube called the inner tube. There's a latch on the top of that, so basically once that inner tube fills up with rock when you've gone down five feet, you can grab a tool called the overshot, lower it down on a wire line, it latches into place, you could pull up the inner tube containing the core sample, and then that's what you send to the lab to be analyzed.

Uh, how how mature is the venture-backed mining market? Like, are there market maps that exist, and are are there's a few, and and were you were you surprised? Everybody wants to go to space; nobody wants to look you know beneath our feet and and go down. Uh, has it been surprising how how little investment has gone into the actual technology side of the industry?

It's there's been very little surface area between mining and particularly Silicon Valley historically. Australia has a bit of a more mature mining technology ecosystem, but not nearly as much capital as we have over here. And so there's been a few um kind of billion-dollar, multi-hundred-billion-dollar mining tech companies over the last decade or so, but just a few. And like mining overall is a two-to-three-trillion-dollar industry, kind of yeah, overall market size. Mining tech is a small portion of that, but uh it's a difficult industry to sell into. Is really why it's difficult to start a tech company here. Building a mine is incredibly capex intensive, and you're basically making a bet on technology early and then utilizing that equipment or software system for a decade plus. And so these companies are risk-averse; they don't want to bet bet the farm on a on a new new technology, where I think we come in um there's a few other companies with a similar model is as a contractor, so we're just replacing a separate service provider, and any sort of innovation that we're creating is done internally, so we are a more efficient, safer, and more cost-effective drilling contractor because of the autonomy that we're building, uh but it's really no risk to the customer. We get paid just like any other contractor to them.

What's been the industry's reaction to the trade war? China very early uh came out and said that they were restricting access to various rare earths. Is that um you know what's been your read on the situation? How are how are US players kind of reacting?

Yeah, uh rare are very close to my heart. Prior to starting Juran, I work for MP Materials, which operates the only rare earths mine in the United States, about a year and a half. Their mine is in California, about three hours away from where we are right now in LA. Um, they empty produces about 15% of the global rare earth supply. Um, they refine about half of it domestically right now. They announced last week that they're seizing shipments to China, so keeping at all domestic. We're going to Japan um or Korea, which they have some offtake with um through Sumitomo. Um, it's it's I think it's certainly a tailwind for the western producers, but there is fear because we don't have everything in the United States. Rare earths, we would be almost self-sufficient on, but we really need to ramp up the the processing. The problem is that within this group of rare earths um about 15 different elements, you have some rare earths that we have a good chunk of at mountain pass, like neodymium and praseodymium, but then you have heavy rare earths like terbium or dysprosium um which we do not have enough of. Uh, MP Materials, it's very heavily weighted towards these light rare earths, and there's very little of these heavy rare earths, and so we really need to partner with countries like Brazil or Vietnam are the two that I point to for rare earths that are potentially um allies, and maybe we'll get closer to them over the next few years. Um, we're going to need to import. There are no other uh decent railroads deposits in the United States that we can just spin up production at that we know of. You never know. We could find us as a long history. I see a bunch of gaps in that map in Ohio. We'll find rare earths there. I'm sure that's that's where we come in. Is that in order to find a deposit you got to spend everything. Every suburban neighborhood just drop a Juran miner in the back of my backyard and start finding stuff. What we'll do first is fly planes over to do a some sort of magnetic survey. The USGS has a pretty cool program called Earth MRI where they're mapping a lot of the country with tighter spacing than they have previously. These geophysical surveys, but we need to be doing a lot more on that because that's really the top of funnel that narrows down the uh potentially uh minable sites.

Sorry. Uh, c can you talk a little bit more about autonomy? I'm seeing the first prototype rig uh can core 300 meters, around 1,000 feet deep, 2 and a half inch diameter, um and these rigs can run unattended for something like 2 to 3 years. So what is the math on that? Is the rig moving around, or does it just take that long to get a single core sample out?

Yeah, so basically, I'll tell you kind of how it's done today, what we have now, and then where we're going to be in 12 months. So today you have a rigged these things weigh you 10 tons. You're normally track mounted, basically on tank treads. You have you put them on a truck as close to the side as you can, then you have a guy with a remote control driving it in the rest of the way. You generally have three operators on a rig. You have the driller; he's the guy listening to it and looking at a bunch of gauges, and he's really sort of interpreting what is the kind of rock that we're going through that through um at that time and from that adjusting RPM, the amount of weight applied to the drill bit, the pressure of the fluid that you're pumping down hole to clear the cuttings and keep the bit cool. She's kind of constantly adjusting these parameters. Then you have another one or two guys called helpers, and they're doing a lot of the manual work; they're loading the rods into place because you normally they're using five or 10-foot rods; they're grabbing the overshot, lowering it down to pull up the inner tube, tapping on the inner tube with a hammer to actually take the core out, putting it in boxes for the customer, tracing the rods, adding additives to the the mud mixture, because you got to adjust the viscosity of the fluid that you're pumping down hole. And so you really have this this sensing problem, this controls problem of adjusting these drilling parameters, then you have this more so robotics problem of just grabbing the rods, putting them into place, grabbing the core samples, etc. So on this first rig, we have uh basically the ability to collect a whole bunch of data, but everything that's going on on the rig at any given time. We're going to use that data to build a sort of V1 autopilot system, deterministic at the start, move to machine learning when we get enough data. There really are not data sets available for this online, and so we've got to got to do a whole lot of drilling before we can actually get to full autonomy. And then we have a rod head. How does it how long does it actually take to drill 1,000 feet deep?

Yes, so a good shift is about 50 feet in a day, and so 12 hours, 50 feet hole, 20 shifts, 10 days traditional like you know human operated.

So that's like a month basically, like a month on one hole. All right. Wow. Yeah. And you're looking at about $100 a foot. Yeah. Okay. Yeah, that's a lot. Interesting. Uh, how are you so really what what what we're trying to do is we're we've built this first rig manually operate at the start. We'll have some automation in terms of rod handling and kind of a V1 of this drilling parameter adjustment system, but then on the next version, which we start building in just a couple months, we'll we kind of add add on to the rod handling system to actually be able to cover the core sample, design our own mud mixer that can dump in the additives separately, and then get most of the way there in terms of kind of removing these jobs on site. And it's really it's it's yes, I think we can save a lot of money because labor is incredibly expensive here, but it's more so about availability. We are in a workforce crisis um in the mining industry right now. I heard someone say a while ago, we don't have an an autonomy problem in the mining industry, but a labor problem. We just don't have enough people. And so is Minecraft not providing the pipeline that one would.

No, I think the the problem here is you play Minecraft when you're like 8 to 14 or something uh or or maybe later, and then uh then you have this kind of 8-year gap between start of high school and finishing college to where you're not involved in the mining industry. We need Minecraft. Okay. Uh, I have to ask, how are you so goated at marketing? You have like one of the most unique sort of like hard-tech brands. Every time I I I think like it's very it's very common for people to copy what Anderoll does, and it's very hard to do something that feels like new and fresh and like super opinionated, and I've loved your out-of-home ads uh and this segment is brought to you by AdQuick um but uh you I I think you've been extremely strategic. I saw this hiring billboard that you had uh that you can't have rocket science without rock science and sort of like placing these strategically in in Hawthorne. Where does that come from? Uh, when did you figure out that you had a a knack for this? Because I'm assuming it's I'm assuming your your team helps, but oftentimes marketing this good typically is is very founder. It's we've got an incredible team. U my head of operations has uh um really helped a lot with the marketing. We have a few incredible designers um that help us out part-time with all this as well. Hiring is a life lifeblood of any any company, particularly a company with as as bold a vision as what we're trying to accomplish, and so we need to get the word out and.

Have kind of this massive funnel of people that have heard of us; some of them will be interested, and some of them we are going to want to hire. And so we, I think we need, need a cool brand for people to hear about us, both in kind of the—we have the billboard across from SpaceX; obviously, want to pull, pull incredible talent from SpaceX, but also like there are very few companies in the mining industry that really care about—I think that care about the brand. There are very few that are particularly good about creating a cool brand. And so like we get, like we, we get good, great engineers in LA reaching out to us because they, they like our stuff, but even more so drillers and geologists in the mining industry who are like, “You’re the only cool company in the mining industry,” and we’ve hardly put anything out yet. Um, and so both kind of in both of those sides, the, the engineering and for kind of the mining customer and talent acquisition, uh, we want to be very intentional about it, and there is much more to come. Once we get this, uh, first rig in the field in about a week or so, there’ll be some really cool videos of us actually drilling and showing some hardware in the field. Love it.

Uh, can you talk us through, um, the path to full autonomy? I imagine it’s like a walk, crawl, or crawl, walk, run situation. Is tea operation a big deal here, um, as you get towards a fully autonomous autopilot system?

Yeah, so it’s—we need a whole lot of data for this, and so it’s mainly controlled at the start. You can have a driller there with an iPad or laptop, um, the infrastructures there, but it’s really like a Tesla full self-driving, uh, sort of system. Like you build the, the software to actually control it—a drill-by-wire system, um, is what we’re calling it, um, and then you need data. And so we’re going to have a driller on site operating it, um, initially, collect this data. We need them there to actually test the robotic side of this in case anything goes wrong. And, uh, I think rod handling we should have done over the next few months, in the extraction system by the end of the year. There’s a bunch of little stuff like, how, how you going to grease the rocks? We need to figure that out, um, or the additives into the, um, the mud mixer. And so we’re, we’re going about this incrementally. We’ll have three guys at the rig initially, go down to two, then hopefully have one can operate it from a self-safe distance, and if something goes wrong, bring another guy in. But we’re never get to a point where you don’t need at least someone on site, at least not over the next five years, because you need a flat surface to drill on, so you have guys out there with dozers clearing these drill pads. You need to dig a sump pit to put the, the water into after you use it, um, you need to deliver your consumables, your water, your fuel, you need to pick up the core samples to deliver to your customer, the exploration company, um, and so there’s always going to be things to do, like you need someone to get the rig there and then maintain it. But the point is like we don’t need someone there listening to the rig and loading rods into place, like, uh, they’re better use of their time operating a fleet of rigs.

Two weeks ago, uh, it came out that that China was, uh, limiting rare earth exports. Chimoth, uh, chimed in; he said, “This may be a good moment to let everyone know that I control one of, if not the largest rare earth supplies outside of China. At full capacity, it could be 25% of what the world needs, all located within countries allied with the US and the US itself.” A lot of people push back on this; they said, okay, what’s the company? He said it was in stealth. Uh, I want to give him the benefit of the doubt and, and say, you know, it sounds very, it sounds almost unbelievable that a stealth company could control, uh, 25% of the world’s rare earth, uh, supplies. Is that, is that possible? Could, could there have been some sort of like roll-up behind the scenes? What was your reaction to that exchange?

Perhaps I haven’t heard anything about Chimat’s involvement in any Rare project beyond like he helps back campaign materials five or so years ago. Um, I haven’t heard anything; maybe he does. I hope he does; that would be great. There’s some mastermind that, that—yeah, myself and my friends who are analysts, uh, haven’t heard of. And so I want him to surprise us, but I haven’t heard anything.

Sure, cool. Uh, can you talk about the challenges of putting a sensor a thousand feet underground? Like, I just imagine if there’s like dirt and grime and crust and whatever else is down there, plus like a diamond-tip drill churning everything up, like even putting like a thermometer down there is going to be difficult. How are you actually collecting data? Uh, is, has, does, does this exist off the shelf? Is this something like oil explorers do? I imagine it’s mature.

So the short answer is we’re not collecting any data underground at this point. Everything we are measuring you can measure from the surface, just like the human does, because the human’s listening to everything and collecting that data from above.

Got it. Okay. Not even using vibration? We’re not even using vibration right now. So you can measure your, your weight on bit—basically how much force are you pushing into the ground with your RPM, how fast are you spinning, your pressure going into the hole, your rate of penetration—just how quickly you’re moving into the ground, your torque, um, you can do that, um, through a secondary pressure reading, um, that there are—you can put sensors down hole. There’s a whole kind of field of study, um, within the, in the gas industry primarily called MWD or LWD—measurement while drilling, logging while drilling. You can put sensors down hole; it’s something we’re going to be looking into. One of the, the things that you’ll do when you’re doing this sort of drilling, mineral exploration, is you can send on a survey tool that can tell you sort of exact position, because your hole—you want to drill it straight; it’s not always going to go straight—might deviate a degree or two every hundred meters. And if you’re trying to hit a 50-foot wide target 500 ft underground, um, it’s—you might miss. And so making sure you know exactly where the, um, tip of the drill is at all times is paramount, so you can lower a tool down either every run or every 50 meters, really whenever you want, um, and that’ll tell you your exact position—basically a gyro probe. And so if we can integrate that into our core barrel down the hole, that would be cool. Not doing that yet, um, or, or doing some sort of EM survey or XRF to scan the core, um, or the sides of the walls as you’re actually drilling, that’d be great. We’re not doing it yet; it’s going to be done soon.

Are cave-ins a problem? Like as you dig down, you, you get, you know, you have a 2-in hole that you’re digging, and then if you pull up the rods to send something else down, and then all of a sudden it collapses on itself? That seems like a problem.

You’re you’re thinking like a driller; that is a huge problem in this space. Basically what you do as you’re drilling through the first, call it 100 ft of, uh, kind of unconsolidated sediment or looser rock through the surface, what you’ll do is you’ll drive casing. And so casing is basically a wider diameter pipe that you’ll drill down with and then drill through that hole, and that basically prevents that. There’s some sort of collapse near the surface; it goes around the casing, and maybe your casing gets stuck, but at least it’s not around your drill string, and that continue drilling through that. The problem is what if you don’t try the casing deep enough, or there’s some sort of fracture downhole, um, like 500 feet down or whatever, that does collapse in, and you’ll basically—what you try doing then is try a whole bunch of different RPMs, pull up with maximum force, but sometimes you get stuck. There’s really two options: you could try to, if try to guess where the, uh, caven is, actually tools to cut the steel above that and just recover everything above that. You lose some, some steel in the ground—oh well, you’re out 10, 20 grand at least, um, or you can of say you’re stuck, but you can actually drill, drill through it. There are tools that they can literally drill through your previous tools that were down hole at a smaller diameter, and so you get some steel in the, steel in the inner tube, which is pretty cool to see, um, let you continue the hole; you don’t have to abandon it. You’re decreasing the diameter—not as valuable data, and you are still going to lose some equipment. And so you basically don’t want to, don’t want to push it too hard to, uh, to avoid, uh, risking your cave.

Last question on my side, just curious how has the, uh, have, have—I, I’m assuming throughout history the, the sort of drill bits, diamond drill bits were just actual diamonds that people pulled out of the earth, and then at some point they were lab-grown, or were they always—what, what’s the history there?

Yeah, so the way these drits work is you have a—it’s primarily iron, the soft iron matrix with diamond particles or impregnated into it. And so you can like, kind of see—I, I should have brought a drill bit over here to show you all; that would have been cool, um, you can see these tiny diamond particles—it looks like glitter, um, on it. And then as it, as it cuts, the diamonds—the hardest material on earth—but they do wear, they dull, and so the bits are designed such that as the diamonds dull, the iron wears at approximately the same rate to reveal new sharp diamonds.

Very cool. Very cool. Last question for me: Armageddon—in the plot of Armageddon, they say it’s too hard to teach drilling to an astronaut; they got to teach going to space to the drillers, to the miners. Uh, realistic or unrealistic based on what you know about how complex mining is? Uh, is—would you have trained the drillers to go to space, or would you have trained the astronauts to drill?

I think we’re doing both here; we are, uh, we’re teaching, uh, SpaceXers how to drill and, uh, drillers how to, how to participate in a high functional, high-functioning engineering organization. And so I think you got to do both. Um, they’ve been trying to, trying to build, build a large organization and manufacture thousands of these rigs over the next five, six years—become a leading drilling contractor. Yeah, they really didn’t bring any space, any astronauts on that trip; it’s all drillers. No, that movie—I don’t think that’s realistic. Okay, uh, anyway, this is a fantastic conversation. Thanks so much for—I’m very excited for you and the team, uh, it’s been awesome watching, uh, you guys, uh, over the last few months, and, uh, congratulations on all the progress in the new round. We’re excited to be on—just a quick message for everyone: we are hiring, jobs, brilliant engineers, please apply or reach out to me. Thank you all for having me. Beautiful. Have a great rest of your day. Talk soon. Talk soon. Bye. Cheers.

Uh, let’s go back to the timeline. Duran did run a great out-of-home ad, uh, brought to you by Adquick. Uh, there’s also news on the timeline about a Series A that was announced by Artisan—apparently a Y Combinator company—they announced their Series A with a massive billboard, uh, and if we can pull this billboard up, it’s the previous slide, I believe, uh, Selene is just commenting on it, but, uh, Tee says, “What’s going on here?” And so it’s the two founders there, and are those the actual founders? I thought they were—those are the founders. Those are the founders. The founders are Jasper Carmichael, Jack, and, um, his co-founder Sam Stallings, a former IBM product manager. But the way they’re looking at each other in this photo makes me look like they’re in love or something—maybe they’re married—but, uh, I don’t know. We, we often look at each other like that, uh, longingly and fondly during, uh, during photo shoots, so it’s understandable, uh, but very funny to put out a billboard about your Series A, but with your own face on it—with your own face on it. But I guess, you know, there’s no information about what the company—have the whole team there, presumably, or are those the AI SDRs? Because I think this—representative—yeah, I thought these were just two AI employees. Those are the founders. Those are the founders. They’re stoked. Yeah, I, uh, this, you know, every once in a while the timeline starts talking about taste. Yeah, this is a very specific kind of taste. I mean, also, look—founder that they absolutely jacked, the guy’s diced—we love it. Let’s hear it. Let’s hear it for Jasper. Guy is looking, looking peel, activate, gold, retrunch. Y Combinator, Day One Ventures, HubSpot Ventures, Oliver Jung, Fellows Fund participated as well. They closed a $12 million round in September, and now they got a $25 million Series A. Uh, they’re, they’re running a marketing campaign—“Stop Hiring Humans”—very viral. They, I, I think genius. They get it. I think they get it. They’re think they’re playing 5D chess. Yep. Yep. They’re going to frustrate a lot of people, and this—they knew this billboard would go viral. Seline post about—Yeah, look at this—Selene, CEO of Loyal, says, “I cannot believe a startup bought a billboard on the 101 to advertise there…” dot, dot, dot, Series A, and 671 likes. Um, and so yeah, out-of-home advertising consistently underrated. We’ve said this—go to adquick.com—out-of-home advertising made easy, uh, and go—yeah—and measurable, and go, go viral, uh, as long as you can come up with something that’s, uh, going to infuriate enough people on the timeline—will get attention. Andrew McCallip says, “I think company towns are going to make a comeback.” We’ve been exploring this. We’ve been exploring a town. Yeah, I was actually thinking about this, so I think if, I think if things go really, really well for TVPN, what we’re doing here in Los Angeles—people, a lot of people say, oh, why are you building in Los Angeles? Like you should be in Silicon Valley. And I think that we could potentially make Los Angeles like the Silicon Valley of media—like a Silicon Valley of television almost—and, and we could create a whole boom. And I think in, I think in a number of years you could see, you know, some massive media companies, entire film and television kind of—it’s totally possible, right? It’s totally possible. And I think it could all start right here. And so what I would say to folks is like, let’s check back in 5 or 10 years, see the market cap of all the media companies that are built in Los Angeles, and you know, maybe it’s in the hundreds of billions if we add it all up. I, it could, it could become possible. It could become possible. Uh, yeah, we are—I think John’s on to something here. Yeah, we are in a company town—Los Angeles. I call it the Silicon Valley of news. Yeah, it’s interesting. I mean, it, it actually is very, I think the experience of living in Los Angeles has become, uh, worse as the entertainment industry has struggled, right? Yeah, um, there, there’s, you know, the, the number of restaurants that can be supported has just gone down. Um, and the interesting thing about this post is, is that he’s saying specifically company towns, so there’s obviously industry towns—you know, New York is a finance hub, uh, Silicon Valley is a tech hub, LA is a media hub, Miami is like an NFT hub—but, uh, doing this for Foster City—they are—remember—oh, I didn’t know that. Oh, yeah, yeah, yeah, yeah, that’s right. They left SF; they went to Foster City; they were like, “We’re going to focus—it’s nicer to live here.” Meta and Facebook were kind of doing that in Palo Alto to some degree; they were like building housing for employees and stuff, uh, but I think, I think what he’s getting at is that things like Starbase, where a company can’t just be the hub of a main city, so they go somewhere really bizarre and offgrid, and then the infrastructure builds up about around that. We were talking about like the Stargate facility—like if $500 billion really does pour into one data center, there’s going to be infrastructure around that. Welcome to Abilene, Texas. Abilene, Texas is going down—there’s going to be people that are like, like, yes—it’s going to be low headcount because it’s just data centers, but like you’re talking about so many data centers, there’s going to be people that are working on the air conditioning, working on the power—they’re going to need an Arby’s for sure. They’re definitely going to need an Arby’s, um, and so yeah, I, I was, I was talking to a real estate guy; I was like, maybe you should just go to Abilene, Texas, and then just start doing deals because there—like the construction is going to be going on for a decade, there’s going to be a lot of people that, that settle there after the fact—they’re going to need an Arby’s—you could sell them the property for the Arby’s—they’re going to need gas stations and all sorts of different—there’s going to need to be a hardware store, right? They’re going to need hammers, all these different things, so, um, it’ll be interesting to see, um, if, if this happens. I think it’ll be a long, long journey to get to new company towns, but, um, I love that, uh, we have a post from Adam Rosenlum: this is my nomination for the next TBPN AI deep dive—live from the forge of finetuning, the MCP monastery, the eye of AGI—weights are sacred, the context is cash, the vibes are autoregressive—tune in now. Uh, I messaged Adam, uh, I think he’s over at Cal; I said maybe you should write for TBPN, so, uh, open invite, Adam. You absolutely cooked, and we love being live from the eye of AGI. I like it. I like it. We, we—Yeah, we need to, we need to spice up the intros—keep iterating on them. I think people enjoyed the Institute of Iron, the Palace of Pump, the Hall of Hypertrophy, uh, but there will be more. We’re doing a crypto day; we’ll have to come up with some new jingles for that—that’ll be fun. Anyway, uh, Google co-founder Sergey Brin, uh, pulled up in downtown Miami in his $450 million—founder—no, $450 million, 465-ft yacht—that’s less than a million dollars a foot at that price—they’re giving these yachts away, to be honest. I mean, it’s a lot of money, but it’s a lot of yachts. It is a lot of yacht, and you can’t sell a house or a midsize Series B company. That’s right. Uh, Mike Salana says, “I honestly can’t figure out why everyone is attacking him for this—sorry, yachts are awesome; pirate wires will absolutely have one in a few years; will be laughing there, champagne in hand, before the waving flags when you come for us—enjoy your bike ride.” One of the greatest posters of all time—he, he cooked them—John—generational poster for sure—fantastic—good friend of the show, um, uh, yeah, I love it. Uh, beautiful, beautiful yacht, and why not? I mean, Google co-founder Sergey Brin has generated probably a trillion dollars of value for the economy, right? So enjoy your little yacht. I get a—I support big tech; I love big tech; I would take a bullet for big tech. I, in theory, would take a bullet for big tech; I would scale a castle wall for big tech. Yeah, I would fight in a trench war; I would dig a tunnel under a castle with a spoon for big tech; I would work years in the mines for big tech. Yes, I would fight a decades-long trench war; I would enlist in the Hundred Years War. We got our next year to, uh, join us.

Welcome to the studio. How are you doing? We are doing—Hey, we got both of you; that’s fantastic. What’s happening? I’m so happy. Um, welcome to the show. Tech—you can ask him about how it was. Yes. Yes. I mean, maybe that’s a great, uh, way to start. We’re huge fans; we were just singing big tech’s praises, saying that we would do anything to support big technology, and yet, uh, you left big tech, uh, why did you leave, and what are you building now?

Uh, we are building physical intelligence. Um, we want to build a model that can control any robot to do any task. Mhm. Uh, this is something I did explore in big tech before. Mhm. It’s just much more fun to do it in a startup—way more fun—also a little quicker—a little quicker—a little faster paced. Yeah, I mean, anything’s become clear in the last few weeks—we need the robots now. Yes, we can’t really wait 20 years. And, uh, if big tech was, you know, fully responsible, we probably would have to wait something like that. Yeah, for sure. Uh, so can you walk me through the most recent announcement? Uh, I, I saw the video, uh, fantastic, but, but how would you break it down in terms of like what the milestone represents?

Yeah, so the biggest challenge in robotics so far hasn’t really been agility or dexterity—what robots can do—but been generalization—y—kind of similar to what we’ve seen in, in language before, um, where it was really, really hard to get it to, to do tasks where you just ask it to do something and, and see if it can work. Um, and what we tried to do for the past six months or so is to get to the next level of generalization for these models for robots. So the challenge we set for ourselves is to take a robot to a completely new home—it’s never seen before—and ask you to do a complex, long-horizon task, like clean a bedroom or clean a kitchen. And there is so many details that go into cleaning a kitchen that you need to understand when you’re in a new home, and you only start appreciating it when you try to do it with a robot, where you know everything is different—like the countertop looks very differently, you don’t know where the drawers are, you don’t know how to open them, you don’t know where the objects are, you don’t know—I don’t even—when I try to wash dishes in my house, I’m always—I’m like, I can’t—where’s this—where’s the sponge? I don’t know where the soap is; where do I put this? So it’s a mess. So—not even—yeah, I guess really hard. And on top of like knowing all of those things, then you also need to connect it to motion—you actually need to get the robot to do the right thing. Um, and it turns out that with, with PI05, which we, which we just released yesterday, we, we can do that. And it, it doesn’t work all the time; it’s not that I can just give it to you and it will work in your kitchen every single time, but it works quite often—quite well. So we bring it to a new home, and it can do those things maybe like 50% of the time—sometimes 80% of the time—big increase from 0% of the time. But yeah, big increase from what we’ve seen before, where the, the previous state-of-the-art was basically—if you want to show a robotic demo, you need to collect data in that specific environment for those specific tasks, and that’s where you show it. But now we can, for the first time, bring it somewhere else, and it kind of works—it understands what it needs to do.

Do you think that consumers will generally be more patient around reliability with robotics? Because if I, you know, let’s say I have some type of robot in my home and I say, “Hey, do the dishes, right?” And 50% of the time it does it perfectly, and like, you know, the other 50% of the time I have to kind of interject, whereas if I’m like booking a flight and only 50% of the time it like books the flight, it’s like, well, I’m just going to do it myself, right? Because like it’s only going to take me a minute, whereas like the dishes could take 20 minutes, right? So, how do you think about kind of, uh, the sort of threshold of reliability in order to like really deliver value for consumers?

Yeah, I think the—the, like we don’t think right now about delivering value for consumers, and it’s kind of why we structured the company the way we did—we’re a research lab; we’re trying to solve this problem.

Of physical intelligence, we really like these consumer-oriented tasks, and I think people tend to as well. Like, when they think about laundry being folded for them, I think there'll be a point at which it gets good enough that we can deploy it to consumers. But it's not going to be like 50%; it's going to be closer to 98, 99%. And there, I think we can harp on self-driving cars where there will be a period of interventions, right? Like if it doesn't work, it's not that it just will stop and do nothing for a while; we could have a human tele-operator intervene and finish the task. But I think the other cool thing about the home and consumer use case is there's so much that could also just happen overnight. Um, there's like, while you're sleeping, your laundry is folded; your, your, your meals are cooked for like the, you know, they're prepped for the week ahead; um, your house is tidied. Uh, so consumers still a little far away, but making a lot of progress.

Can you talk about the, uh, just the path in terms of the underlying technology to go from a Roomba—we're pulling up the video here on the stream of, uh, what you've actually built—uh, what were the foundational turning points in terms of the, like, the different models and different breakthroughs? I imagine the transformer was really important, but there's probably a ton of other, uh, developments that excited you, and now, now is the time, like, we're ready to go.

Yeah, there's been a lot of things that we are, we are building on top of. Um, things like transformers, things like vision-language models, the concept of pre-training and post-training. Yeah, a lot of those things transfer to the robotics world, but they're, they're not as well understood. We are still in the process of figuring out what that recipe should be like. We, we kind of have to rediscover some of the steps that that language people had to do initially and see how we can map them onto the robotics world. We don't have the, the privilege of having an open internet full of data; we need to collect that data ourselves, which, on one hand, is, is a big challenge—like the data isn't there, you can't iterate nearly as fast—on the other hand, it also gives you more freedom in figuring out what kind of data is the most important and what data to collect for, for this particular PI5 advancement. What we had to do is one, collect a very diverse data set, large diverse data set that involves not only model manipulators in homes but also static robots in the office or data of the internet. And it turns out if you collect very diverse data across many different tasks from many different form factors, they all contribute to each other, and that they contribute to a better understanding for the model of what actually is happening and how to utilize all of the data to figure out what to do. Um, so that was a really, really big component. And then there's also a lot of architectural things, a lot of details that we need to get right to make sure that we take full advantage of that data. Um, interestingly, most of the data is actually not the, the model manipulators in many different homes; it's a very, very small percentage of it. So it gives us also a lot of hope that we can leverage the data off from the internet or from other platforms where it's easier to collect it to, to get to that kind of generalization.

Can you talk a little bit about, uh, simulation, uh, like data generation through simulation? I imagine it's very easy to procedurally generate like a million different floor plans or a trillion different floor plans, uh, to try and navigate those in, you know, kind of two-dimensional space. But at the same time, when you get into the manipulating of a sheet, uh, all of a sudden that's a physics calculation; it's probably harder to simulate there. It's not like with self-driving cars; there's, you know, Grand Theft Auto can train on—I haven't, maybe there's a game where you clean up your house, but I certainly haven't played it—how effective has simulation been in generating data, and is it useful? Is that a, is that a viable path here?

Yeah, it's been so far really, really useful for locomotion, for robots walking around. Yep. Um, and the reason for this is that for that kind of problem, the main difficulty is modeling your own body—like how do you place your foot, how do you, how do you walk—and that you can do once when you, you know, you model your robot really, really well, and then it works; it works across many different terrains. You can, um, easily, you know, randomize that and, and figure out the gait that is robust. It hasn't worked nearly as well for manipulating objects, for working with your hands. And I think the reason for that is then the difficulty isn't about like how do you move your hands; it's more about the world that you're manipulating, and that is much harder to simulate. You don't just do it once; like every object you interact with is different, and you have to model each one of those, and it's really, really hard to, to figure out all the different physical parameters to make it, to make it good. Um, so, and, and this is kind of the data that is the most important, the data of, of physical interactions, because this is the data that is not on the internet; this is the data that is not even described in language; this is something that comes really natural to you; you just, you just know how to do it; you don't even sometimes know how to describe it in words. So I think it's kind of like the a really bad combination where it's the hardest thing to do for sim and is the data that we need the sim the most for. And what we discovered so far is that basically looking at the past successes of machine learning, of, of AI, uh, it seems that the, the best successes are where you, you take real-world data and large diversity of that data and, and learn directly on that; you don't try to find some kind of proxy or some kind of simulated environment that reflects what you actually want to do; you just go after the problem head-on, and that's what we're doing here. So we're collecting a ton of data ourselves in the real world; we can collect very diverse data this way; it's also very easy to collect it across many different scenes, many different objects; we don't need to, you know, create them in sim; we can just buy them and bring them in and, and start interacting with them. And that so far has been actually easier than we had initially thought.

Yeah, I mean, you, you say you're collecting a lot of data, but I, I imagine like there's only so many of those robots in that demo video that you can manufacture; there's only so many houses that are like, "Yeah, come try your 50% robot in my house." Um, is, is, is that a key, uh, is, is that scaling as you'd like, or, or, or is this more like you're going to build a physical, uh, you know, demo, like demonstration unit, m, like, uh, and then be manipulating it in a warehouse, or is the plan to be more like, let's roll this out and just, uh, have beta testers kind of dog food it for us?

All of the above. Um, okay, basically, where we find there's benefits to scale at this stage, we'll scale it; we'll figure out a way, and whether that's producing more robots, giving them to people, whether it's scaling up our, um, our operations team and the folks that tele-operate these robots ourselves, whether it's going out and commercially deploying these into environments where they're doing economically useful or viable tasks as a training data set collection, we'll do it. We also think there's, there's so much though to do on just the algorithmic development that, that can make the data far more useful, that can, uh, reduce the necessities of scale. But we're structured such that we can go and pursue every avenue. What you guys both mentioned there is that was one of the big questions before PI05, where it kind of was unclear, you know, do we have to visit million homes? Do we have to visit, you know, hundreds of thousands? And at some point it becomes kind of not feasible or really, really hard, and maybe we need to find a different path. But so far we've been quite surprised by how few different environments you need to see to be able to generalize to a new one. It's awesome; we actually got really reassured that this path could, could really work.

Would you guys like to see way more, uh, early-stage, like, robotics companies? It feels like there's, you know, the Optimus, the 1X, you've got, you know, Figure making noise, um, but it feels like, you know, we just covered this, I think, uh, Monday, the Chinese humanoid marathon; I'm sure you guys followed that. They've got a lot of people working on this problem; it, it, it seems like, u, there's a tendency in venture to think that, okay, there's a bunch of heavily funded players now; I shouldn't go build in that space. But at the same time, the when you look at some of these TAMs, maybe we should have 10 times the amount of, um, you know, early-stage robotics companies, uh, getting started.

It's extremely early. We work with, I'd say probably most of the new robotics companies starting in the US and abroad. Uh, if you're starting a robotics company, reach out to us; we'd love to work with, uh, you can build the body; we'll build the brain. Uh, but yeah, then we need to see a lot more robotics companies, particularly in the US. That's awesome. Uh, I want to talk, did you ever interact with the Google arm farm?

Yes. Yeah, one of, one of my co-founders actually started that project.

I had a feeling. Uh, do you have your own version of an arm farm, or can you describe for people that might not know, uh, what was the genesis of the arm farm, what was the purpose, so what was the takeaway, and is that, does every robotics company need an arm farm, or is it just you, or, uh, and, and what will that look like in the context of what you're building specifically?

Yeah, so back then, that was few years ago, the idea was that for robots to learn to, to, to acquire those kind of skills to manipulate the world around them, you can't really prescribe it; you can't just code it all up; the world is too diverse; you can't, you know, have a lot of if statements describing what you should do in every single situation; they need to learn it the same way as, as we do. And, um, the idea of the farm, arm farm, was to set up many different stations where you have static robots, static robotic arms where they just practice and they learn from experience. So in that particular case, they were trying to learn how to grasp objects, so they just had a bin in front of them with lots of different objects that were very diverse, and the arm was just going down and trying to figure out how to grasp it, and over a long period of time it gathered enough experience to actually learn from it and get, and become really, really good at grasping objects—like remarkably, remarkably good, and way better than any kind of prescribed systems that people, that people designed by hand. And, on one hand, it was a big success because of that, because, you know, it was, it was clear that this learning approach is something that can truly work and, and understand the nature of grasping and truly nail that skill. On the other hand, it was also disappointing in that it took really long time; it needed a lot of data, and especially a lot of data at the beginning was kind of just like the arm wandering around and not knowing what to do. So it seemed like a lot of that time was just kind of wasted with the arm figuring out the simplest things. And one thing with IO5 that we are really excited about is that we are now at the stage where the robots kind of get the, the sense of what they should be doing in the environment, so they are no longer in this space where, you know, you just like arrive in a new home and you start with just like moving your arms around not knowing what to do and hoping that you, you do something that is useful and then you learn from that; you start at a point where you kind of know more or less what you can do; it works some of the time; you just now need to get it to work every time and really, really well.

Can you talk about the path to, or the importance of, end-to-end learning in the context of robotics? Uh, my understanding is that tele-operation is great, and as long as it's economical, we should do it, and then, uh, having a deterministic code, uh, like, you know, control system that's written in C++, that's also great as long as it works; it's sometimes more debuggable. Um, but the reason that we want to get to end-to-end AI systems is that then you're on the scaling law; then you're just data-bound, and the more you can manufacture, the more you can produce the actual robots, you're on this flywheel, and you're now bound by actual productive, you know, like getting the cars into on the roads, getting the robots into the world that will naturally create a flywheel, uh, that's what everyone's hoping for in self-driving. Um, but, but what does that path look like now, and how ridiculous is it to claim that end-to-end robotics will be here by the end of the year or something like that?

So end-to-end robotics is already here; everything we've shown so far is fully end-to-end, where you take camera input in and few other sensors and output actions directly. Wow. Um, I think there's another reason to do end-to-end learning, which is this is, I think, the only thing that has a chance of working. Like, if, if there was a way to just pre-program your robot and write a really good C++ code to, to get it to do all kinds of different things like folding laundry, we would have done it long time ago. Totally; it's not for the lack of trying; many people have tried it for a very, very long time; there's just the world is too complex; there's too many things that you will never see, you will never predict, and you can't really write that code. And, um, I think the only way to get there is to, is for, for AI to figure it out from experience. This is similar to what we've seen in language or in vision where people have tried to write chatbots with writing different instructions and prescribing logical steps of how you should proceed, but it turned out that, that the intelligence you need is much more messy; it just like you give it a lot of text and you let it figure out all the different patterns and analogies, and there's many, many more of them than, you know, the ones that we can express in code or in language. And I think something very, very analogous is going to happen here, and that's what we start seeing—like the demonstrations that we, that we've shown so far here at, at Physical Intelligence are of tasks that were not possible before, like things like folding laundry; you can't really, there is no program that I've ever seen that could, that could do that; the same with arriving in a new home and making the, that there's just too many variables there to do it any other way.

Can you talk about, um, some of the experiences that you both have had in your careers, uh, Google and Stripe, uh, in some ways big companies that maybe move slower than a small startup, but at the same time, both of those organizations, I feel like the time from, hey, we're starting the company, to we have a product, that it wasn't a research organization for years, um, h, what have you learned from those organizations, what are you taking into this experience?

Stripe really taught me everything I need to know about building a, a robotics research lab, uh, it's just lessons galore. I'd never worked in a research environment, uh, so I don't have priors; I don't know what a research environment actually looks like; I just know what our research environment looks like; I feel like we, whatever it looks like, uh, maybe it operates exactly like startups, like maybe grad programs are exactly like startups, but we feel like every other company I've worked with that moves extremely quickly and has a clear set of goals and direction and just has a bunch of people that work behind it and work extremely hard to solve whatever it is that, that we're setting our, our minds towards. And, and, uh, there's so much I learned from Stripe that informed that, but a lot of it's just the, the obvious stuff, right? It's like, comm, hire exceptional people, set a really high bar for it, don't compromise, set a very clear set of goals for everyone, really align people. Actually, I think that's one thing I really took from Stripe is you want an extremely aligned set of people, and I've never seen more alignment than I've seen at PI. We, we talk about the alignment tax, like when we're bringing someone on, how much work is there to align them around our mission, our way of seeing the world, and almost everyone that joins, there's like basically nothing. Uh, most of the people that work here have dedicated their lives towards robotics or robotic learning or AI in some form or another, hardware, whatever it might be, uh, and that just allows us to move so much quicker; it's like we need to communicate a fraction of what the average company needs to communicate to someone; we don't really need to inspire people or motivate them because they're so inspired and so self-motivated. Um, I think that's probably one of the things that's worked best for us to date.

How have you seen your customers react to all the news and chaos around the tariffs? I think a lot of these companies are not in full commercial production yet, so it's not like, hey, we, we're no longer making money, but how are they thinking, kind of like long term, just given how much of the supply chain is based in, in Asia, and are there opportunities for, you know, you know, new, uh, US, kind of like subcontractors and, and manufacturing companies to sort of service this new industry?

Yeah, the good thing is, I mean, good and bad, it's also subscale right now, like most of the money is being spent on R&D rather than scaled production, and so it's not as if there's 100,000 robots that everyone's buying and it's now just twice as expensive. I think the good thing is that given it's subscale, there's a lot of time to build out US supply chains, and it's putting a lot of focus on figuring out, can we get US actuators? Can we, can we start to create companies that are developing those and all the other critical supply elements of the supply chain? So it's actually just getting people into gear, and maybe it's the right time for it.

Can you talk a little bit about, um, what you're excited about on the data center side? Is there a moment where you're like, really pulling for Stargate to come together, and we need the, the, the 100-gigawatt data center to crunch, um, you know, all of the data that you've collected in, you know, kind of a GPT-5 class training run, or is that something that's like so far out that you'll always be able to just, uh, you know, uh, tag along on the residual capability from the large language model labs?

Yeah, we are, we are not there yet in terms of like having a full scaling law the same way as we've seen for LLM companies where you can just translate more compute to progress or to capability. We, we are searching really, really, really heavily for that; we, we're trying to figure out what is the recipe that would scale like this, but we are not there yet. We do, at the same time, generate a ton of data; I think that's one thing that, that I realized since starting the company is that robots generate a ton of data, and you don't need that many to generate data that is close to the levels that LLM companies use for, for their models, and there is no ceiling to it.

Yeah, right, not that we run out of the, the data that robots can collapse; it's not like the internet. Y, so I think over time it's quite likely that, that, uh, the, the places are going to switch a little bit where most of the models, including, you know, LLMs and BLMs, are going to be using real-world data collected through robots, because that's the data that has no ceiling and it's very active as opposed to just passive observations of what people wrote on the internet. Um, and I think at that point, probably the, the question about data centers and compute is gonna, is going to be a big one, but for our models, we are not there yet; we are not bottlenecked by, you know, if only we had 100 times more compute, everything would have worked so much better.

How do you guys think about, uh, demos long term? We joked on the show recently after seeing the, the Chinese, uh, humanoid marathon, like, I want to see humanoids doing, like, big-wave surfing, cliff jumping. You know, at what point is that, like, worth even doing or exploring, just because of the amount of attention that it would bring to the industry? What do you think we should demo?

I think I'd like to see one of your robots surf Jaws; I think that's, that's really—

I was saying I want to see, I want to see a robot do the 900 on a halfpipe, Tony Hawk style.

That was a really foundational moment in, you know, my childhood and American skateboarding culture.

Really life-or-death stuff.

Exactly; it's got to be high stakes.

What about with, with hands? What about manipulation?

Uh, juggling, for sure. Rubik's Cube, for sure.

Uh, juggling Rubik's Cubes; you can do that. I can do the Rubik's Cube; Jordy can juggle; we need to learn each other's skills so we can do both at the same time. Um, but yeah, I mean, these stunts, uh, when they're done right, they can draw a bunch of attention, although you guys have plenty of attention; I don't know if that's really what, what's—

No, but it's an interesting thing where it's like you have the, the intention of the entire industry, but then at some point, you know, to basically inspire the next, you know, however many thousand robotics companies, I, I, I would love to know about, um, obviously with any AI project, there's always the public perception of, like, job displacement, dystopia, AI doom, etc. But when I look at the demo that you just posted, I'm like, that thing is going to be fighting on my team in the singularity; like, this is a friendly robot that will be defending me. But how do you think about the, like, tuning the language interaction, uh, so that it, do you see a world where, um, yes, it's doing my laundry or, or making my bed, but if I happen to just also ask it, tell me about the news, I can just have a chat with it? Is that something that's even, uh, in your mind in terms of, like, human-computer interaction?

It's not a big focus of ours right now, really. We're so focused on, on manipulation and economically valuable tasks, and, and more so than that, the fundamental building blocks that we think gets us from here to physical intelligence. I think it's inevitable that everyone has robots in their homes, their workplaces, uh, just like in their lives, and I think they'll want robots that are more useful than just doing things, whether it's companionship or, like, it's the best Amazon Alexa that can actually then go, like, cook the recipe that you ask it for. Uh, it's a place we focus, u, around the interaction, but right now it's more understanding the intent of, clean my kitchen, and then breaking that down into tasks, but it's pretty straightforward to go from, do the thing, to tell me about the thing, to let's have a conversation about the thing. And so it's, it's on the, the horizon, but not the greatest priority. And then one thing there I would say is with the models we've been releasing, they're actually built on top of vision-language models, so these are the models that are truly what they call multimodal, where you talk to it, you can ask it what they see in the image, and every now and then you can ask it to perform actions too. That's cool. And what we, what we start to realize is that all of these different data sources contribute to each other; they give you just like a bigger picture of what the world is like and better understanding, and it just—

Turns out that robot actions is just like yet another language that these models can speak, and they just need to learn it and see enough examples of it. So the model that we have already is the model that you can talk to, and it works, you know, just as well as as open source BLMs, um, but on top of that, it can also have that understanding be very embodied, and it and you know, it's it understands what it sees in front of it, and it's a much deeper understanding when it knows how to how to move its arm to actually accomplish a task.

We were kind of joking before the show about the obvious comparison between uh robotics and self-driving cars, uh, but can you explain to me like I'm a 5-year-old or like a venture capitalist like why is that a bad analogy? Why don't you love that that analogy? I feel like it's it's it's not that you don't love it, it's just that it um it can put you down a bunch of wrong directions. There's a lot of parallels, but uh, I mean, even like the Waymo Tesla thing, right? Like Tesla has this incredible advantage with how much data they're collecting and passively, yet Waymo is so much better so far, and it has so many fewer cars on the roads. There's useful things about it, but there's also aspects that that don't transfer in the analogy, and it uh I think the like the reason why we were joking about it is it's the number one question we don't talk to many 5-year-olds, but investors and VCs ask us, and so we have to go down this rabbit hole where we're breaking down all of the assumptions and correcting some of them and validating some of them.

I mean, is that a is so so should VCs, if they're looking at the robotics market, just throw out that analogy entirely, or should they be saying like there are a set of robotics companies that are in the Waymo category, and there's a set of robotics companies that are in the Tesla category, and those are reasonable uh like an an ontology to map to?

Yeah, no, it's a there's a lot of very useful stuff in the analogy. Um, I think uh I I think one thing that's interesting is that there are all these self-driving companies that have died uh over the past 15 years, and one thing that we actually like to remind people is that this is not coming tomorrow. You log on to Twitter, and you'll see all of these crazy robotics demos, most of them teleoperated or most of them being like robots doing backflips, which is a much easier problem than actually a robot folding laundry. And the thing we really try and and remind everyone that looks at investing in us or is thinking about investing in us is this is not a problem we're going to solve tomorrow. There's fundamental research breakthroughs that that we need to make, and much like self-driving had a it's what like a 15-year arc at this point, there is a very high likelihood that robotics is the same way. Um, like we we think our greatest competition is science itself. It's not like this company or that company; it's just maybe we can't pull it off in our lifetimes. We think we'll be able to; it's looking more and more likely, but it's uh it's not a tomorrow thing.

I have one last question. I know you enjoy food and cooking. What is the final eval, the Mount Rushmore, the Mount Everest of cooking that you expect will be the last the last dish that a robot will be able to cook? What's the hardest dish for a robot to cook?

The Don Angie lasagna. Oh yeah, okay, very difficult. So so when when when when they cook that, AGI achieved, it's game, it's game over. That's amazing, amazing. Uh, there was actually a recent AGI a AGI benchmark someone sec shared a screenshot, and uh it was a very old definition of AGI. It said it'll be able to describe a sheep, tell you three things that are larger than a lobster, and all of and AGI is here by that definition, but one of the one of the things that it can't do is bake you a cake, uh, and and we just thought it was funny that like that was the that was the last thing that the that the computer can't do, um, but maybe soon, maybe future. But thank you so much for coming on the show. This is a fantastic conversation. Best of luck to you, and and and thank you so much for for building this this really important technology.

Yeah, we're excited. Put a put a robot in the studio when you're ready. Send it over. It's a mess. I have clothes all over here that need to be folded, uh, so we'd love to have one awesome live demo. Bye.

Next up we got Sam Lessin coming in the studio, venture capitalist yapping about venture capital. Uh, we will bring him in when he's ready. Uh, in the meantime, I will do some ads. We'll talk to you about Wander. Find your happy place. Book a Wander with inspiring views, hotel-grade amenities, dreamy beds, top-tier cleaning, 24/7 concierge service. It's a vacation home, but better, folks. We can also tell you about Ramp. Time is money, save both. You heard me slip it into the Perplexity interview. I'm going to try and slip more ads in. We've been hearing a lot of feedback that there aren't enough ads on the show. We are working hard to remedy that. But we have Sam Lessin here in the studio. Welcome to the show, Sam. How you doing?

I'm doing great. How are you guys doing?

We're doing fantastic. Uh, we just had a fantastic conversation with Carol Hueman and Locky Groom over at uh Physical Intelligence. Did you see the demo of their uh their cleaner robot, the the room?

I love Locky. I haven't seen him in a bit. I know. Is this the folding robot? They've been folding they've been folding [ __ ] in like a warehouse for a while.

Yep.

Well, they're doing it in the real world now. The demo was they they they sent the robot out into the field. It cleans someone's house. They say it's about 50% accurate. They're getting ready to deploy it once it gets to 99% accurate. What does a 50% accurate cleaning robot do? Like half the time?

Yeah, it it can fold half of your shirts prop uh properly and half of your shirts improperly.

Look, in fairness, that's probably better than I could do. I'm not much of a folder myself.

I'm terrible at it. I'm terrible at it. So you know, I would be a terrible laundry robot personally.

Yes. Uh, great to see you, Sam. Always fun. You guys are crushing it. I'm loving the vibe. You You're both full-time now?

Like ads?

Yeah, ads. Yeah, yeah. We can never never side hustle. That's our never side hustle, always full hustle. Um, what's uh what's going on in your world?

You know, I don't know. I've been traveling a bunch, but I'm I'm back and uh traveling as a venture capitalist. How does that work?

Oh, just for fun, guys, not for work, please. No, I just But I would imagine such a prestigious career path is so demanding that you would never be able to take a day off.

Never. No, I I am a slave to Zoom. I just sit here and zoom back to back.

You seem to have cracked the code. What's your stance on uh Zoom? You do you invest much purely over Zoom or are you just meeting everybody at this point?

You know, it's a really interesting question. I I personally think Zoom Zoom's had like a really interesting impact I think on venture capital, cuz initially people were all bowled up on how Zoom and this like no meetings in person was going to open the funnel, and people would like invest all over the country, and we break down walls because all of a sudden physicality like there was like this kind of euphoric Zoom will be democratizing type thing going on, um, and I it's interesting. I do think that Zoom means that people like me and venture capitalists are willing to take more meetings than we otherwise would be on like interesting topics with people that like again like it's just like the barrier is lower. You'll meet with more people now. Does that actually result in more investments? Unclear. Um, I I think it might invest in like broader sets of first meetings uh just cuz the barrier is lower, and like if the meeting gets boring after 15 minutes, I can just do my email and say ahuh, right? So like it's like there is some breakdown in like access, but it's not clear to me how much that resolves to like actually broader access means. In fact, I think that's one of the big things that's interesting about AI broadly right now is you know, there's this narrative like with autoscripting. I get the number of pitches I get that are like half-written written by AI or written by AI is like out of control, right? I don't know what people think they're doing with those cuz they're all just going to get deleted, right? And if anything, the irony is the fact that the barrier now is so low to those emails means that even the ones that actually are legitimate just get archived because like on the margin they're probably spam, and so there's this interesting thing where AI is actually leading making venture capital I would argue more insular right than it was before, not less insular, right? So there's all these like unintended consequences going on of Zoom, of AI, of all this type of stuff in VC.

Yeah.

Email, you just got to use it like text messaging. Just no no subject, no body at all, just whatever you have to say, just put it in the subject and send.

Look, I actually am a hu I'm like probably I'm old, I'm like 41, right? So I'm like a huge I love email. Like I think it's great, and I'm an inbox zero guy, but I also like run aggressive filters on my email and like am fine not responding. Like I don't consider email a contract that because you sent me something I'm required to send you something back. I think you just have to treat it differently.

Um, what about on text? Are you an inbox zero on text?

Text I take pretty seriously. Um, actually, so that is like I I kind of have a pretty quick SLA on text, and I I tried to make sure that not everyone knows my actual phone number.

Yeah. Let's put let's put Sam's phone number up on the screen for everyone uh so you can text him if you have to get

Well, I'll give you my plug, which is you know, one of our companies is called OpenPhone, which is a great company and that I use an OpenPhone phone phone number for a lot of things where you want to put it up on screen, and it's great. It's like a second text inbox. It's not really the purpose of the company; it's more sophisticated than that, but I I I like it for that.

That's great. We should We should set an OpenPhone. We should.

Yeah, yeah. For me, I I I like to think of the hierarchy of like you know, inboxes, right? Email, uh, but then now it's like okay, you have XDMs, uh iMessage, WhatsApp, Signal. If you want to get in touch with me, show up and grab me by the collar, shake until I hear what you have to say. That's the only I mean, I I so I have like my stack is text is incredibly important and serious. That's like a one-hour SLA, but it's limited. Email I take very seriously. Like I really care about email, um, but then people are like you should be in my Discord server or like Slack. I'm like abso fuckingutely not. Like I think those things are disasters. I refuse to engage with them. I hate everything about them, despite the fact slow one of the seed investors in um in Slack, and so I have to thank Stuart for making us some money, but like I just like can't deal. Uh, I refuse to deal.

Let's Yeah, take us through that. Should we talk about uh venture capital? I actually want to start and go back to your 2023 piece, uh, which is on the timeline, uh you called out uh this is October 16th, 2023, the shutdown of the VC factory line and the death of the factory-farmed unicorn narrative, um, the awkward crowd into seed investing by multi-stage firms, the mirage of AI and LLM startup investing, the post-pandemic fundamental cultural change impacting startups, um, what which one of those grade yourself? Yeah, how would you how would you grade your I obviously grade myself excellently on all the um No, I mean, I think like look, I try to pull up every few years and just especially in times of uncertainty like what is going on, like where should we be spending time and attention. I think you know, two years ago, you know, those were the big themes for me is one we were very used to for 10 years as a fund effectively operating this factory line. We take in companies at a certain stage; we know what metrics they have to hit; we give them the money; we then package them and send them on to our friends at Series A, who then send them on to B and da da da, and the whole thing works beautifully because at the end we pop them out into the public markets, and retail investors buy them, and life is good, and I was just saying like after the pandemic, you know, people wanted that to come back, and I was like this is not coming back. Like this whole the market is it's a mixed-up market. You know, we've basically produced a bunch of these things which are on paper unicorns, but like they're not fundamentally important businesses, and more importantly, the there's been this huge release, which is the biggest platforms in the world can just keep getting bigger. Like I remember a time when like a hundred billion dollar company was a huge company, right? People thought there was a limit to how big the biggest could get, and so there was this constant hunger in the public market for what's the next $3 billion company that's going to grow really fast, and there was coverage for that. People cared about it, and now the obvious answer is just like put more money into Amazon, put more money into Meta. Like there is no upper bound, and so I just think the markets have shifted. Demand from consumers has shifted at the at the public has kind of rippled back to the ecosystem. You know what we have now, and you know, I have this in the 2025 version is what I'm calling zombie corns, right? So there's all you know, people thought people friends of mine were like, oh, we're going to see this mass extinction of these unicorns; they're all going to die because you know, they're going to run out of money; no one's going to fund them; the liquidation preferences are huge; they can't go public. They actually mostly didn't die, right? What they did do is they basically sacrificed growth; they cut their burn a lot; they kind of got marginally profitable; they can kind of exist, and they're kind of zombies; they're just they're out there; they're not going anywhere, but they're also not going public. There's no market for them; no one knows how to buy them; the liquidation preferences are set up so that no one wants to deal with them, and they're just going to kind of exist. So so the the factory line is broken at the late stage because there's no off-ramp, but in many ways, in many ways, the sort of early-stage preede to seed to Series A is still accelerating as though there's an off-ramp.

I don't think I don't think it is actually. I think this is like one of the I think the part of the 2025 deck are trying to think about what's going on. I actually think what we now have weirdly is effectively several markets for companies that are pretty decoupled from each other and kind of have their own logic and exist in their own vacuum. So like there is a public market, like the public market exists; it's the biggest, right? Um, there is a private market now. There are companies that are private and will like probably never go public or ne for various reasons. The way that companies are valued by the latestage private and the public are actually just different. Like what's valued is different; how people think about them is different. Large LPs actually invest in both, so they don't care; they like, but there's basically like running two parallel universes that don't have a lot of operability. Now at the early stage, I actually the same thing is going on, right? Which is like there is a kind of seed to preedeesque market that exists, and people compete, and people get excited, and there's kind of a market-clearing price for startups and invest sure, but then when you say well what do you what does it take to hop from an early stage, call it like preede seed maybe sneaking into A to like a legitimate B and beyond growth round, there's no more like magic numbers you hit and like a valuation framework that's consistent. It's actually much more about belief. You know, I I say in the deck it's a lot about this kind of new math of people want downside protection and then an option on infinity, right? And so like what's the average of infinity and zero? It's infinity, right? So the way people are backing into valuation is so the entire factory is predicated on $1 million in ARR equals Series A at price blah blah with 30 you know, 30% growth and $10 million in AR like triple triple double double. You can come up with all these frameworks; everyone kind of agreed on [ __ ], and there was just like market-clearing action in prices, and now I actually think there's just like distinct markets of belief um that are really hard to move between.

U yeah. Well, uh, does this necessitate like a different model around growth? We kind of saw this with like the crossover investors like Tiger, but a lot of growth funds have this downside protection mandate. No zeros, but let's underwrite to a 3x, uh, and and yes, I mean, I hear the I hear your infinity thing that does happen every once in a while, but I think in general a lot of growth investors are just saying no zeros and let's triple our money uh over this deal, uh, but should you have more later-stage growth investors that are thinking more like portfolio construction at the seed stage?

I I look, I think that the answer to that basically is I don't even know. When they talk about we're underwriting to a 3x, yeah, who's buying, right? Like everything is about the marginal buyer, right? There is no value on anything, right? It's all about like a DCF. I mean, you could you could justify the cash flows, right? You could comp to the public market. The pro the pro, but the problem with first of all, comping to the basically nothing at late stage is trading comp to the public market really, right? And we can get we can get into like why and how. I think I think that look, the DCF comp to the public market that way that is the factory model, right? It's basically saying like, hey, I have a late-stage thing; I put money in; it's going to triple; the DCF looks like this; it's has this much profit margin; like this is the story; package; sell to the public market; the public market buys on that same story; like that that was the mentality that persisted for a long time, and it was a great system for a lot of for moneym for a lot of people, right? I actually think that again, the public market now is like, well, if I kind of just want those types of metrics, why don't I just buy more of the mag 7? Like I don't I don't even want to dick around with your subscale offering. Like I don't care, right? Like I think, and there's reasons for that. It's because the big LPS are bigger; it's because of meme stocks; like there's a whole bunch of stuff going on there, right, that like kind of makes that happen, but the net outcome is there is no off-ramp to the. So then the question is when you're underwriting at a late stage, you're not you also have to underwrite to someone buying from you, right? And like the question is what are they buying and why are they buying it, right? And I think this is where it gets a little bit squirly because I do think you know, I'm not the only one saying this, but like private to private transactions are going to happen way more, right? Like you're going to look just as it happened in private equity, like funds will sell the funds. Then the question is well what is the buying fund paying, right? They're going to pay; they need some margin; they need to have a framework in their heads about how they're going to sell, right? So then they're going to you're going to be pay you as an early-stage fund are no longer underwriting to some late stage to some DCF to the public market; you're really underwriting to who's going to buy from you; what's their narrative on buying; like why do they want to hold this; right? Like what's their time horizon? What's their purpose? And like the irony of the whole thing is like honestly, those prices are going to probably be much lower than what the DCF might otherwise imply, right?

How do you think about how do you think about funds, you know, selling an entire fund, and I'm talking about venture funds maybe selling to other um, you know, kind of like not continuation vehicles, but just other secondary buyers versus trying to sell off and kind of like prune the portfolio and say like well, at the end of the day, look, I I think it look, the fun side of funds, you're just going to take some massive discount on that [ __ ], right? Because in the end of the day, like if someone's buying a fund from you, they're really only buying the winners, and the rest is [ __ ], right? And like so in an ideal world, they just carve out the pieces they want, like I want this position and this position and this position because I care about these companies or I have an infinity thesis on them. Like everyone has to have I actually think the infinity thesis really matters in terms of how people are thinking about how big things can get and whether they matter or not in the world, and then everything else, it's like it's basically worth zero, right? Um, you know, we were joking at our firm about like we were joking about starting right off CO, right? Because that's the other funny thing that happens, right? Is you just kind of give up on positions, and you sell them for a dollar to take the tax the tax advantage, right? On it. Every once in a while, the hilarious part is you sell something for a dollar because you give up on it, and it turns out being worth something, right? So like there's a whole business up hoovering up you know, irrelevant positions, but I don't know. I like it. Does it happen? Yes. Do GPs sell piece of the GP? Yes, but that's complicated because the reality is at least the public market only really values the fees, right? Which mean like so it's basically I've been having trouble because I wrote this 2023 thing that I was pretty proud of and I think was honestly pretty accurate about what was going on. Understandably, it's been two years. I wanted to update it and be like, well, where are we now? And like there are things I think I got mostly right; there are things that I think are wrong, but I think that the real story is end of factory model; the factory's over.

Mhm. What the the the the yada yada yada on is we're now I think entering this period where like you don't even think about it as one integrated capital system; they're just like different parallel universes. Like everything in the world is regionalizing and fractionalizing. This is happening with globalization; we're having a del-globalization moment; this is happening with all sorts of things; I think it's happening with capital too. There are just literally distinct ecosystems, and people play in multiple of them, right? But they have their own valuation logic, etc., and then I think like the way people value companies as a result has changed; what you should be looking for has changed; the types of CEOs you want to back has changed; um, it's just a new world.

Did you dabble much internationally? This was a very 2021 2022 thing with funds thinking that you know, they they win a deal and like you know, like 10,000 miles away, and and then like in hindsight, it's like you have to think like why did you win this deal? Why didn't you know uh why why are you uh you know, so lucky to have the opportunity to back this company, and then a lot of them just you know, are derivative and and

The only, yeah, the only, we've always invested in Israel, um, when it made sense. I think the the US-Israel relationship is strong. I think there's a lot of good tech; there's a lot of reasons that makes sense. And so that's always been a thing we've done; we know it well enough to like be confident investing there. I again, I think that's the thing for us historically is yes, in a Zoom world, you'll take the meeting in Europe, right, cuz like it's interesting and like, you know, no one's accounting for your time as a venture capitalist. So if you're interested in something in Europe and they really want to talk, like sure, you'll do it. But we're so lazy, right, that like I don't want to deal with… there's always an exception; like we have done a handful of deals that are kind of outside the wheelhouse, but it's of of looking at… but I kind of believe that there's really something to, you know, be New York, be San Francisco focused, pick a few geos, understand them, understand how they fit into the global capital world, etc. Um, so no, we didn't get drawn in too much.

Europe is… uh, what what is your interpretation of the rumor around OpenAI potentially buying Windserve for three billion? It's the it's not an AI cherry-on-top business; it's a rapper. Is it bull market for rappers now? Are we going to see tons more acquisition slow is going to FOMO into a bunch of rappers? Yeah, FOMO into every rapper because they're just going to get hooed up. I mean, my my take was maybe, you know, OpenAI buys one, then Anthropic needs one, then Amazon needs one, then Google needs one, and all of a sudden you have like seven unicorns getting bought; everyone's making money; everyone's generationally wealthy; that's the good ending, right? I don't look, I really am pretty cynical/don't think that we're going to see a lot of acqui-hire AI type stuff in this era. I think there's a few reasons that one is like what are you really buying, right? In some of these things like you can are you buying talent that is not unreasonable, but you know, and we've seen that; we've seen people like effectively quote-unquote buy companies that are literally just for like the one person they want to pay $200 million to cuz they really think they're special because they need an ex… look, some people are going to get massively overcomped; that will happen. I don't think it's an investment thesis; it is what it is, and I think it won't happen that much, but we'll see. I think there's going to be… are you buying technology? It's like the thing about AI and a lot of where we're going is like why, right? Like software is getting commoditized; like what is it like? If you're buying technology, you got to be buying some really important technology, right? And I think that's like an… the third thing you in theory can buy is just distribution, right, which is like if someone really has, you know, for whatever reason got their hooks into a few key contracts or like, you know, they have a tailwind, like fine, you buy distribution, and that that can be worth something. But look, honestly, it just seems like all very squirrely to me at this moment. You know, the thing I'd say in venture capital is there will be random walk; some random stuff will happen, right? And like I think you can't get too, you know, twisted around that; you certainly shouldn't be chasing random walk. Um, but no, I don't I don't personally see it. And if anything, I'd say like look, I was part of this, you know, my first company was acquired by Facebook in the era of acqui-hires, right, and um, I think the and I say this with some humility but also perspective is like I think when you go back and look at like what was really bought there and was that a good use of capital by most of the acqui-hiring company? I think the answer is probably not. You know, like that era is kind of over; people aren't that special. You know, the big companies just because they have such incredible access to capital and distribution at this point, they can kind of just build whatever they need anyway, right? So anything that does happen will be highly bespoke as opposed to like industry-wide trend is my personal view.

How do you think right now about the dynamic between 30 to $50 million early-stage funds versus, you know, 30-plus billion dollar AUM funds? In in some ways, they uh, the the the the tiny, you know, uh, upstart funds benefit from these big platform funds coming in and sort of marking up deals, right? The returns look good, but maybe I just think they're complet… like they're two completely different business models, right? Like I think is the thing; this goes back I pull about like fra… regionalization and fragmentation of what capital even is, right? If you're running a $50 billion venture fund, you can't possibly be deploying that well early, right? And actually, you're paid to move gross dollars; the problem you're solving for LPs is you have some massive [ __ ] LPs; they're like, I want exposure to private markets; it's really hard for me to go find how to do that; I would love you to deploy as much capital as possible. And like, effectively, the way you get paid is on fees; you're not getting paid on carry; like that's, you know, if you have a $50 billion fund making 3x on that, so you're making money on carry is like extremely difficult to impossible; like the numbers just don't add up; you're getting paid to deploy, and that means your business model is attracting more capital; tell you have to return enough to justify it, right? But like you're not actually shooting for maximum DPI or actual returns; you're just shooting for that. And by the way, just further doesn't the big, you know, I just see this all the time where I have friends with funds that that are maybe sub-$50 million funds, and they're investing into like even if the big platforms just periodically dipping down into seed when when they have a, you know, really like the founder or whatever, and then suddenly the round is like, you know, six on 40, and then tiny fund, and then the tiny fund is like you could get a bunch of bangers, and you just do the math, and you realize like they're not making they're not going to be making DPI either, and they're not getting the the latest stage… what happened after 20 in 2023 era, which I I wrote about then is like the latest stage capital allocators who again are paid fees to deploy gross; they're like they're weight deployers; they're mass capital deployers; they got they they couldn't deploy, so a bunch of their junior people in particular are like, well, I need to do something to justify my paycheck, so they started dipping into seed, right, cuz like they're bored, right? And they're like, we can't deploy big checks, so we might as well deploy small ones. And by the way, no one cares, right? Like it's such small amounts of money; it's irrelevant either way; that completely messed up the seed markets cuz it got super undisciplined, right? And like it did because it's candidly we do the same thing at Slow to like the angel market, right? Whereas we we it's a recursive problem; like we will write $100,000 checks off a meeting cuz it's kind of irrelevant to us, and it's just relationship building and like whatever, but there's some poor angel who's out there trying to price it properly, and we don't care, and then we [ __ ] it up for them. So like it's it's a recursive problem, um, that did happen. I think mostly honestly the late-stage guys with AI have a narrative where they can put billions of dollars to work and do their actual jobs, so they've mostly pulled out of [ __ ] up the early-stage markets because they have better things to do with their time that's actually what they get paid for, right? Um, and just to make a finer point on that, you know, a lot of these late-stage public platforms they really are like setting themselves up to go public. Here's the thing about that: when when they go public, the actual way re… like the public markets value these funds has absolutely nothing to do with returns; it is 100% the fee base, right? And so their their structure and their incentive structure is 1,000% about earning fees and just making enough returns to justify the fees they charge and raise more money; like that's what they do.

Then there's the early-stage market. Here's the thing about those $50 million funds, right? Um, ultimately, you got to eat; you got to actually deliver DPI, not just marks, right? And and so marks are nice; like they're fine, but I think what we're going to find in a lot of ways is the the tr… the the gulf between I have on paper made a bunch of money or these deals look good versus like, oh no, I actually returned capital; I like made you money; you should give me more money, and I made myself money doing it; that's a pretty big gulf. And I think what we're going to find is that you know a the market most of those funds are going away because they don't have that, and they're not going to b… there could be a world where late-stage funds start saying, okay, at some discount to the last round, I'll buy out these seed funds effectively and give them some DPI, etc. But then the problem for the seed funds is that mark they were using to be like, look how smart I am; that's not what they're getting paid, right? That's like that's like the high-water mark; some investor invested later for primary, and when you come around and say, hey, by the way, would you like to buy my shares? I and they're like, well, we'll take more, and it'll lower our average cost base; they're not paying what they what they paid for the primary, right? And they're looking at their portfolio and they're like, I need to do this for 80% of my bets basically in order to like actually… and then it's like, you know, yeah. And so look, I mean, the upshot, the really simple way to think about it is like if you're an early-stage investor, you have to make money; like that paid for people are saying, "Hey, I'm going to allocate a small amount of money to you." By the way, it's not efficient, right? Cuz if you're even a medium-sized LP, someone's running a $50 million fund; what are you going to give them? Like a few million bucks; you don't care unless they make you a [ __ ] ton of money, right? And so like if you make them a [ __ ] ton of money, you're doing your job; you get to keep playing; if you don't, forget it. And that's just in a completely different game than what it means to be a late-stage capital allocator in the private markets. Yeah, I have uh, kind of a random topic, but there's there's two early-stage kind of publicity stunts going on this week: one is by Roy Lee; he launched Cluey Clue Lee… cheat on everything. I'm not sure if you saw this, but it was very controversial, and he's kind of like a troll, almost like a Nathan Fielder type, really kicking the bear. And then there's also this uh, artisan uh, company announcing their $25 million Series A with a billboard on the wall. I love billboards, as you know. Yeah, we love billboards here; we're sponsored by Adquick; we love billboards. Um, but uh, you know, the the the positive take on this is that, hey, like they're breaking through; they're getting attention; attention is valuable; distribution's important. The counter to that is uh, should they even need to do that? Shouldn't they just be heads down building? Uh, where do you sit on that continuum? I I guess the question I would ask is what percent what is the track record of companies that started with marketing stunts that ultimately were important or successful, right? My sense is the track record off the top of my head is zero, right? Well, I mean, Facebook was very viral; I'll give you an example; I'll give you an example. So the challenge with going super viral early, and I had this with Party Round, is that people get a fixed idea; a lot of people get a very fixed idea of what your business does, and then you run into this like product-marketing challenge, which you know, people are aware of your business, but they're aware of it for something that you may not even do anymore. And that's why I was talking with Cli founder yesterday of like you need to be committed to like iterating and basically burning the whole brand down because you might find in two months that the real opportunity is something else. Yeah, I think that's a really good point, and like I I'll do a step further, which is I in my experience, really successful things you actually want fairly high barriers to entry so that the people who show up as your early customers are like deeply in need of it and true believers, right? Cuz if they're deeply in need of it, they're going to put up with a lot of crap to get what it is you're offering out of it because they're deep because they really care; like they showed up first, and they like have a real stake in it. Um, and then they become true believers in that cult that advocates. I think if you have too much attention too quickly from a not fervent enough audience, you get distracted; you have to deal with a bunch of the wrong stuff; people are flighty. Like so I think there's this irony which is like it all… how you get your first 100,000 10,000 people and the barriers to entry there are like really… and I'll give you a kind of counter example which actually kind of is a marketing stunt if you get into it, which is quite by accident, you know, uh, and I kind of started this Jelly Jelly Memecoin blew way the hell up; went crazy, but was supposed to be like a component of the app Jelly Jelly; we've been working… the app's super cool, but like the app wasn't ready, right? Like when… and what's been really interesting to watch is because the app wasn't ready, you got a bunch of people; in most of them bounced; they're like, "This isn't ready; this is weird." Whatever, but you did attract a kernel of like crazy true believers that are really engaged with it, and then it's kind of like a fire; like you kind of blow on the coals of that, right? And you kind of keep iterating and working. So I guess that's a long-winded way of saying I think the history of companies that start with a marketing stunt and blow up big is pretty poor; there probably is a way to like be very inefficient and like blow up something big or say something… funnel out 99% of the noise; y… somehow find that kernel 1%; work with that 1%; and like treat it like kind of the embers of a fire, right? And grow up. So that's kind of the the mental model; it's like how you handle it. Got it.

Uh, last question: how cooked is Tesla? I mean, look, I I I've been in the camp of like Tesla's a meme stock for a long time, right? Um, and I think Tesla's a meme stock, right? Um, you know, and so I um, yeah, you posted uh, maybe it was yesterday; no, it was this morning: if Elon can move Tesla stock up by 7.5% by saying he's stepping back from Doge against the backdrop profits and revenue they did, then yes, he probably deserves the $56 million difference as a pay; that is what the markets… his attention is worth. Uh, I thought that was pretty on those. Yeah, look, it's it's it's… Elon is the greatest marketer of our generation; um, he's the greatest capital uh, raiser of our generation; you know, he is the greatest I think storyteller; I mean, there's a lot that he's really really really good at, right? And um, you know, I think he's the ultimate cult influencer in a lot of ways, right? And he's built a lot of cool companies doing that, but it is so belief-driven, and I think this is kind of the thing where it's like, you know, what does Tesla work from a DCF perspective? We talked about public markets and how you value these things; not a fraction of what it's traded at, right? But it is absolutely… he is great at the infinity story, right? The infinity story is so big, and infinity, you know, plus zero equals big number; everything's about the marginal buyer, and it's incredibly loved because retail investors want something to believe in; like they want something they can that can go to infinity; it's the same thing with the Mars thing; it's like, look, I again, I find the whole Mars thing in SpaceX so frustrating; I love SpaceX; it's like, you know, I think it's an amazing company; like what they do is incredible, right? And there's a lot I love in the whole nine yards; the Mars narrative is so frustrating because it's so disingenuous on one hand, right? Like it's just like the the predictions are out of control; like it doesn't make any sense from like a fundamentals perspective, but my god, people need something to believe in, right? And so believe in something… well, I think Tesla's coming back; I think they're going to put a naturally aspirated V12 with a gated manual in a new car, and they're going to sell 700 million cars in a single quarter. I [ __ ] love that; if Tesla did that, I would be even… I would buy Tesla stock just because that would be awesome. There we go; we cracked it; we cracked it; it's going to happen; you heard it here first. Uh, thanks for stopping by, Sam; this is fantastic; we will talk to you; thanks for coming on. Got it. Next up, we have Bridget Mendler of Northwood Space coming into the studio; very exciting; I believe $30 million Series A uh, from Adrien Horowitz in partnership uh, I think Founders Fund and a bunch of other folks got in the round, so we'll talk to her about that. Bridget, welcome to the show; how are you doing? Hey guys, what's going on? What's up? Um, I haven't been on a podcast before, so am am I on? You're you're on; you're not only on a podcast, but you're also live, so there's no there's no post uh, post-editing, but hopefully we got the facts right, but you can break it down for us; tell us what does uh, Northward Space do, and tell us about the $30 million funding round that just was announced. Yeah, um, we're we're building the ground network for the industrialized space economy. Um, you know, we view it kind of as the third critical pillar of infrastructure for space where you need to get things into space on rockets; uh, you need to have things to put into space, which are satellites, and then you need a way to actually communicate with them and use them once they're up and operational. And so we're focused on that last third part and building the the shared infrastructure that the whole industry can take advantage of, really drawing parallels to the cellular industry and to the internet where shared infrastructure is just a big enabler for being able to push um, technology forward. Talk about uh, what companies have had to do historically; you know, we've heard a lot about satellite companies that send a satellite up, and they're like, it's working, but we don't know where it is, you know? So it's like, you know, kind of critical aspect of uh, you know, you know, maintaining… Yeah, what was the status quo prior to you starting the company? Oh yeah, um, yeah, I mean, it's not just prior to us starting the company; it's like ongoing. Um, you know, we talked to companies I think like last week that are just not getting enough coverage, and so they're endeavoring to build their own ground stations themselves, um, and you know, our our co-founder Char… it was actually interesting during our first fund raise; he was still working at his old company, and he was woken up two times in the middle of the night; there were a total of four ground failures just in the course of one evening while he was um, manning their operations uh, at all different locations; all different ground networks; uh, one was a site that had already been down and just like not even notified the company that they weren't going to be able to make their contact. Talked to another company last week that had been out of uh, contact with their satellite for 28 hours. Um, it's like, you know, you're not just tossing like a $50 piece of equipment up there; it's like tens or hundreds of millions of dollars, um, so it's very stressful, and uh, we're we're excited to, you know, pursue setting a new standard there. Yeah, people get stressed out when Slack's down for like five minutes, and imagine having like, you know, this hundred-million-dollar billion-dollar device you don't have contact with… What what is… I'm curious what does scale look like for Northwood? You know, how many different, you know, ground stations uh, you know, do you hope to kind of get to within the next call it decade? Oo… decade; that's a long horizon, but that's fun. Uh, we are looking at scale both from like a network level and a site level. So when you think about like why do you need to have a global network to begin with with space, like the reason why you need to have a global network is because satellites orbit the earth, and so maintaining contact with them um, requires having uh, you know, locations all over the earth to make sure that you can be in contact all the time. So think of it kind of like when you're using a cell phone and you're driving on the freeway and you're passing different cell towers; you need to have maintain contact with cell towers in order to maintain connection; same with space. Um, and so for us, there's like two verticals: one is coverage, so you want to have enough coverage, so basically like a cell tower, like you're always in contact no matter where you are, and then the the other one is throughput; that's like density. And so kind of gold standard for this is SpaceX where they have hundreds of ground stations to um, not just have global coverage but to be able to serve millions of users in different regions. So when they're wanting to like service a region that has a lot of customers, they need to put a lot of ground stations in that region in order to support that much capacity. And so we want to be able to offer that to other folks so that they can have like that kind of gold standard of uh, connectivity through space. So um, we're going to be putting um, you know, ground stations in different regions as well as ground stations densifying in the same regions. Um, and one of the things that we think about with scale is really like how can we put as many ground stations to support as much capacity as possible at a single region? So our kind of um, you know, sub near-term goal is 500 sites.

Can you take me through some of the can you take me through some of the history of uh, these ground stations? Maybe explain it in really really simple terms; like maybe like I'm a venture capitalist or something. Um, something like uh, how do we communicate with the Hubble telescope? Is this like a big satellite dish like what I saw in contact with Jodie Foster? Uh, is that how we communicate with the Hubble, or is there a different network of ground stations? What what what was kind of the gold standard 10, 20 years ago? Yeah, um, I have to just say like us ground nerds in space, we do not often get asked these questions, so thank you very much; like the ground is just generally like the not sexy part of space, um, so it's very fun. So yeah, I mean, what you're doing is uh, you know, generally using RF to contact a satellite that is like hundreds of kilometers away; um, you need to concentrate enough power to be able to do that, so that's why you see like the big parabolic dishes; it's concentrating power. Um, and so when you're, you know, further away, that requires more tow… more power. So actually, um, if you're, you know, in the Palo Alto area, you go by the um, the Stanford Dish is kind of a well-known one, and I imagine like some folks uh, if they're watching might know of that; it's massive; a giant dish that's used to make that contact. And so it's interesting because like the legacy of space; it's more exploratory; it's more research-based, where like booking an antenna is kind of more like booking a telescope, you know, where it's like an individual piece of equipment where um, you know, they're located at these different locations around the world; you book the time; you're kind of in control of how that functions and how it operates; uh, but as the space industry…

Has been scaling, um, that's not really a sustainable model to think of like, as you're needing to coordinate, you know, tens or hundreds of different sites to think about that individual booking and coordination, um, and with that we kind of like to analy-analogize to network routing for the internet, where it's like you're not thinking of every single, you know, router and network switch; you kind of trust in a network that can reliably deliver that data. Um, and so that's something that we're starting to think about, about, you know, how the space industry is going to evolve in terms of communication, going from uh booking an antenna like you book a science telescope to having the outcomes through um a global network where you can um you can have a lot of observability and control into that network, but it's much more like software-defined um and controlled, kind of like modern internet infrastructure.

Uh, question for now: I think there's an obvious opportunity to serve existing space companies. What kind of companies do you think uh are potentially enabled by your technology and network that uh may not have been smart to start five years ago if you didn't have kind of the resources of of SpaceX?

Yeah, uh, there's a company that we were recently talking to that I get really excited about. Um, you know, in LA we had the wildfires a couple months ago, and uh absolutely devastating, um, you know, really difficult to figure out where to route resources with a really fast-moving fire. Uh, if you're trying to get a sense of like the scale and and the direction of that fire with a heli-a helicopter, um it's often like not safe or not even permitted to go into those regions because there's just so much debris. Um, and so satellites are a really interesting application where, um, if you're able to have uh enough revisit rate, which is what, you know, in the space industry call like being able to go over a region again, if you're like a low earth orbit satellite where you basically just need to have like a bunch of satellites that pass over and take turns because it takes time to orbit the earth, um, so having enough revisit rate to where you can actually like regularly track the movement of a fire is pretty revolutionary; like you can you could stop fires much more rapidly and be able to um detect the movement.

Um, the challenge with with that is if you don't have um your your latency down low enough to um to be able to give the information, it's pretty much useless; right, like if you deliver information like an hour later then you can't um you can't deliver anything actionable and helpful towards firefighters on the ground; like they're gonna go into, lives are on the line. I remember John and I—I live in Malibu, John lives in Pasadena—I remember the watch the watch fire watch duty went down for like an hour, and I was like, I was looking I was looking at the mountain behind my house just like being like, if the fire comes over the hill, I just want to have eyes on it quickly. Um, and that was like, you know, very brief that it was down.

Uh, I I I have a follow-up question, um, again maybe a stupid question, but uh what is going on in the various uh different orbits? We talked to Albido about VL LEO; obviously LEO is kind of the hot one with Starlink, but uh do you need different ground station technology or scale to hit something in high earth orbit? We talked to Astronis, which is maybe partnering with Impulse to kind of boost to higher orbits. Uh, what are the challenges in uh or or benefits to different orbits when you're thinking about it from a ground communication perspective?

Yeah, no, that's a great question. Um, you know, V Leo, like you're getting closer to Earth, so you're able to get um more like high fidelity imagery or like sensor um things like that. If you go up to LEO like, um, that's useful both from that perspective but also from like a latency perspective when you're talking about like trying to hit internet similar kind of latency timelines; just the time it takes to go with those altitudes. There's also operators in like MIO, which is, you know, middle orbit, that are supporting internet use cases. Um, and then if you go out to geo, the benefit of that is like it's geostationary—that's what the name means—you're fixed at a certain location, um, and you're able to have really continuous coverage over a wider area because it can see so much of of the globe. Um, this is a super interesting area and an area that we're actually like really enthusiastic to be working in is is servicing multiple orbits, um, and yeah, I think it's it's both of interest on the commercial side and also on the government side; you know, they have a lot of assets um that stretch up into higher orbits, and uh they're looking to have, you know, more capacity, more coverage, more resiliency; like there's really a shortfall of ground assets in the higher orbits actually. Um, and so that's something that we didn't enter into the business planning on, but that's something that has been like a a very large driver of activity in our business over the past like 18 months of existence. Um, and yeah, I I think like the more dynamic movement is also a really interesting point where it's not like you're just hanging out in one orbit, right? Like that's kind of Impulse—Impulse's really exciting uh proposition with prop—um, is that they're able to uh you know maneuver between orbits in um in new ways and uh you know in space economy, rendezvous, proximity operations; like that's that's something that's definitely going to be picking up and and very significant in the coming years. I feel like Middle Earth Orbit is like super ripe for a Tolkien-named startup, some found back thing.

Um, uh, I I mean, I would love to hear where the name Northwood came from. Um, yeah, tell me where where the name come from, and then I do have a follow-up question that's more serious.

Yes. Um, that's a very serious question. Um, the name came from the lakehouse where uh we first did our prototyping of antennas during pandem-pandemic, and it was the very origin of becoming a ground nerd. Um, and so yeah, that's that's kind of the history of the name. I mean, that that lakehouse, my great-grandparents got it in 1945; it's just a little shack in New Hampshire, and uh it's been uh kind of the all all the companies that I've had have had some affiliation to it. Got it.

Uh, so on the business side, um, can you walk me through where you're playing in kind of the value chain? I imagine that there's a fair amount of equipment that's available off the shelf; you might, or I correct me if I'm wrong, but do you need to build the equipment from scratch? Is this a project where you're going to be building like a gigafactory like what we've seen for the Starlink units at some point, or is it more about um assembling different components and then being really strategic about placing them and then building a network on top of that and really like the services side of the business?

Great question; something that our head of manufacturing, Thomas, thinks about a lot. Um, we're definitely going to be leveraging outsourcing in early days; we we are a vertically integrated company; like we we design all of the different uh components, um, we have those outsourced and then um you know brought in; we're not we're not um you know like making our own circuit boards at this point in time. Um, there's certain things that are like more efficient uh efficient to insource versus outsource, um, so largely um leveraging outsourcing um initially. I think, you know, we're really focusing on modular units and making our our units designed for manufacturing, so um making sure that we can parallelize development, making sure that um we can have things ready to be integrated at like kind of the final hour is is the thought process, just to accelerate our manufacturing uh capability, and then gradually over time, when we figure out like what is actually cost-efficient um and time-efficient, we bring it inhouse entirely.

I want to know more about actually the mechanics of setting up a ground station somewhere. Um, I imagine you could do it, been looking…

Yeah, yeah, we're we're thinking about doing my backyard, you know, I have some extra space. Uh, are you going to pay… there's a lot of ham radio amateur folks that do that. Yeah, I I have a I have a f-a family friend who their family has some land in Napa, and I think they monetize it by uh selling a cell phone tower right on top of it. Is that is that is that the the the one of the folks that you'd buy land from, or um or is it on federal land? Like how do you think about placing these? Do we need them to be equidistant across the United States and beyond? Are there other countries? Are you placing them in allied countries? Like how do you think about the the coverage map? The Ver—I want to see the Verizon map with all the different coverage points, right? How does that grow over time?

Yeah, no, it's it's real; we do that modeling in-house, so um yeah, we we think in terms of like the coverage mapping and um the metrics that we're prioritizing hitting for customers. Um, also shout out to Christian at Astronis who has his own amateur radios; we had a fun time talking about tune-in. Um, yeah, but in terms of where you put sites, it's a great question; basically comes down to three things: land, fiber, power. Um, so you just need to optimally be able to um make your sites uh you know, we we prioritize making our sites as uh generic as possible really so that like we have the most optionality possible. Um, we are also prioritizing like really high throughput backhaul, so um data centers honestly become like a good spot to to put them at because they have the power in the backhaul um already uh set up, um, and you know generally try to just make sure that the land is like easy to deploy. Uh, one advantage of the way that we're building our systems is we don't need to lay like a concrete pad, which can add weeks to months to your uh time frame, especially when you need to like do permitting and all that, so our goal is to make it so that you know the uh the tech bros of the world can just you know have one in their in their backyard, very uh easily deployed, easily. Um, and yes, it is it is a global effort that we're… I remember and did something similar with the uh with the sensor tower; they didn't want to pour the concrete pad because of permitting, so it's on wheels and saying it's completely unnecessary; you could just drill it in the ground, but then it's way more complicated.

How do your uh how do your timelines work? Uh, a lot of, you know, I'm assuming a lot of your customers are kind of planning around like launches, and that means those are kind of busy moments for you guys I imagine, but at the same time you can serve a lot of existing companies that have assets in orbit already. How do you think about uh kind of the advantages that you guys have of like being on the ground and not needing to plan your entire business around SpaceX?

Oh yeah, it's very convenient; we can work on our own schedule. I mean, we have different challenges because you know you're going to different countries and they have their own local regulatory regimes and all that, but we we are not constrained by launch schedules, which is great. Um, and then the first part of your question was what was the… no, you answered it; you answered it already. Uh, we… I have a I have a I have another somewhat random question. Um, we ask a lot of artificial intelligence founders about their P Doom; we ask a lot of space founders about their P moon. Uh, what is the probability that you will visit the moon uh in the next 30 years? Let's call it. Would you go if the capability was there? Let's say there's been 100 people or a thousand people or 10,000 people up; are you going? And then what's the likelihood that you think spa-the space economy and the flywheel that gets us to the moon happens based on your insider knowledge of the industry?

I would absolutely go to the moon if I had the opportunity. I I did hear from like an astronaut one time just how life-changing that experience was, and yeah, I mean, I I feel like uh that would be definitely a thing for the bucket list. As a mom now, I think that's honestly like the only thing that holds me back from… We got to bring the kids; you got to bring the kids. It's going to be Disneyland on the moon; that's the first economic… We talked about this too; like we need to go to… Do we go on a Blue Origin flight? It's 250k. Do we just go and podcast in space? John was John was all in; I was like, "My wife will absolutely kill me." Uh, I don't know if it's worth it, so it's definitely part of the calculus. I know to be to be young and wild and free and like the the lunar economy; I'm very bullish on it; I think um yeah, I think we'll hopefully see that within our lifetime.

Uh, somewhat related to space tourism, uh, the Blue Origin flight did just happen last week. I want to know uh specifically what are the challenges with uh with again uh connectivity, because it seemed like we lost the video feed while they were at the apex of their kind of trajectory. Uh, it was only three miles or three kilometers up or something; it wasn't that high, and yet we still lost the live video feed. Is that something… it's a moving object, but satellites move too. What does it take from a ground station perspective to uh you know be able to watch Netflix on your Blue Origin flight consistently?

Yeah, that was actually something that we talked about like in the very early days, like pre-forming Northwood, was like being able to watch Netflix in space, and we're like, "Oh, wouldn't that be like such a cool feature?" Um, I mean, to accomplish that, like there's multiple different kinds of the communication going on; there's like how do you actually make sure that the rocket is going where it's supposed to go and um it's safe and like, you know, we talked to someone the other day who was concerned about like a rocket trajectory not going the direction it was supposed to and winding up like landing on another country and needing to deal with kind of like the uh catastrophe that falls out of out of that and managing that, so like you really need to know where your spacecraft is going, um, because yeah, the the consequences um that fall out of that are um serious. But then yeah, having um actual, you know, humans on board, needing to have some kind of communications on board, um, yeah, it's it's going in a different trajectory than a satellite that is just kind of conventionally like orbiting, um, and that's something that we're excited about with our technology as well is um being able to vary our beam width. Um, so if you think about like uh you know the the signal as you get further away is kind of like a if you were to shine a flashlight on a table and like the the the area that the flashlight um spotlight covers changes depending on how far away the um the flashlight is; it's the same thing with an object going up in space; like it's changing the actual um signal propagation depending on um how far away the spacecraft is, and so you need to be able to have like some way of of tracking that, um, and we're excited through uh you know the the tech that we're developing to be able to to track the beam with as as it changes for more dynamic trajectories.

Can you talk a little bit about the long-term uh mix of customers? I mean, we've all been following Delian's uh trajectory with Varta; it was he was talking about ZBLAN at one point, then it was far pharma; now there's some DoD mixed in there, some government contracting. It feels like a lot of these companies that are doing stuff in space or doing stuff in hard tech; it's dual-use. Um, is there a government angle here at some point, or is that just something you're thinking about in the future?

No, it's it's very near-term; it's very real term, um, very real term, um, very very real, very real, very real. Yeah, across a number of different applications; I mean, they're dealing with the same challenges, like if not even more so, um, where like they have a aging assets that um are kind of infrequent; like there's, you know, certain networks that just don't have a lot of assets, and they're old, and they're vulnerable to outages, whether it's like an intentional outage, you know, by somebody targeting that site or not. Um, and so there's been a lot of interest in how that they can leverage commercial to get um sites deployed quickly; like for us in the conversations we're h-having, we're really emphasizing like we can deploy capability quickly, um, and we can serve up capability that's like quite scalable, so if one of those outages happens, you'll have that that backup and that resiliency. Um, and so as you know, government use cases, um, like so much of our world runs on space in a way that I think people don't really realize, and so um it was, you know, that's been a refrain that you're hearing more and more through government stakeholders where there's this concern on, you know, if anything goes down in space or or through the ground connectivity, it has ripple effects through like a lot a lot of our critical infrastructure. Um, and so for us to be able to uh you know deploy capability that can enable resilience there is something that's definitely resonating.

This might be a silly question: are you guys already making hardware that's actually on satellites? And if not, is that something you would do at some point? Because I imagine when it comes to, you know, reliable communication, you're somewhat reliant on the technology that's actually on the craft.

Yeah, that's a it's a great question. I think like so far we've been pursuing partnership there, but like if the need presented itself to stretch onto that side, we have amazing engineers that would be uh very capable of of doing something like that. Um, but yeah, there's kind of a decomposition happening right now where there's a there's one company that just makes the satellite buses, and so you could imagine that there's a different company that makes, oh, just downlink connectivity, and then you kind of vend all that together, and then you do the important thing, and you can focus your company a little bit more…

Exactly, exactly; that's the vision where, you know, in the same way that a developer doesn't need to think about like their you know networking or any of it, it's just like you just focus on building and… Yeah, the rest is is kind of focus on that key value creation.

Uh, when did you initially start researching or like catch the space bug? When did you get into this?

I mean, honestly, it was around that that time of the um you know prototypes that we were making. Uh, my husband, Griffin, is our CTO. Cool. And so, you know, we were just uh working on those prototypes during the pandemic, and um I feel like I don't pursue things as one does as one does as one does… Some some people were, you know, baking like, some people were podcasting, sourdough bread, but you know, rocket… saw a lot of like new founders coming out of the pandemic too, just like, yeah, too much time on your hands. Um, we were fortunate to be in that position, but yeah, just can't do something casually; I was just like, all right, and then after we did that, we kind of like, you know, wrote a white paper with commercial folks, and then we did another one with uh some government stakeholders, and um like, damn, like this is a really critical vulnerability in the space industry. Um, it kind of took off from there. So five years of work to get here. Can we play the overnight success sound? Overnight success! But congratulations on the on the funding round; really really milestone, and congrats to you and the whole team. Come on next time, you… What are What are you guys drinking over there? Is that a yerba mate?

That's a yerba mate; I am I am gui big uh yerba mate guy, but John's… you know, we have both. I'm having the third… Wait, this is like endorsement or something, but um yeah, we have the trifecta, the holy trinity of energy drinks. What are you What are you drinking?

Red Bull.

Red Bull, a classic. Very Lindy… Lindy… holy Trinity, the holy trinity of of energy drinks. Anyway, have a have a great rest of your day. Thank you so much for coming on the show, and uh love to have you back when there's more news.

Thank you so much. Cheers. Bye.

All right, have a good one. Bye-bye. Uh, let's close out with some timeline. Um, what else we got? Oh, uh uh Mike N, former guest on the show, uh, has released the results from OpenAI's 03 model, which everyone's raving about on ARC AGI, and so his takeaway is that 03 medium—because there's a million different varieties now for how in how intense and how long-running these uh these reasoning models can run for—but 03 medium is the industry-leading AI reasoning system by a large margin; 2x the score and 1/20th the cost compared to the next leaning leading chain of thought system as measured by ARC v1 semi-private set scoring, uh 57% for $1.50 per task. And that's interesting because we talked to Sean um Swix uh about how Google was dominating in this paro frontier of model capability versus cost, and they'd really—we talked about this with Logan too—how Google has been dominating in these benchmarks and then cost, but a arc AGI is this completely separate benchmark from MMLU and LM Arena and Humanities Last Exam and all these other things that are uh ARGI; it's so simple; it's these puzzles, but it's in some ways harder to game or harder to optimize for apparently. Um, and so um he says his key question for released 03 is it more like 01, slightly better than pure LLM on novel tasks, or more like 03 preview? I love OpenAI's naming scheme; keep it keep it simple, guys. Uh, it makes this really hard to do my job. Uh, qualitatively new capability to solve problems outside training data. And so um we are going to be following uh the ARC AGI development very closely. He closes by saying ARC v2, which is the latest uh puzzle eval that he released, uh ARC v2 still has a long way to go even with the great reasoning efficiency of 03; new ideas are still needed. He called this when he came on the show; he said, "We haven't evaluated the new OpenAI models yet; we've heard rumors about them; we think they're great; obviously very economically valuable; obviously amazing tools; we love them, but uh in terms of ARC Arc V2, uh they're not solving that fundamental problem, and it raises questions about is it AGI? Is it 10-minute AGI? Is it it's it's AGI that can do like IMO level math, but it can't solve a puzzle that a kid can solve; it's a different type of intelligence, and I think that's great; I think it's amazing for the economic impact, but we still got our edge; we still got it; humanity is not done yet." Um, but anyway, let's move on to uh some news. Uh, the United States banned artificial dyes from all food products effective yesterday. Yeah, this is big; we got to get Cali co-founder on to come break this down; I don't have full context. Uh, it seems like it's going to be incredibly disruptive to big CPG. Can you imagine trying to reformulate M&M's? Like M&M's, they've been making this for 100 years; get ready to have some gray M&M's, folks. M&M's, they're still going to taste the same. Uh, actually, we'll see if they taste the same if they're not, you know, the color of the rainbow. Yeah. Uh, but I I I think this is good; there's plenty of evidence that these different dyes like have really terrible impacts on health, and especially considering that kids consume these and they don't have the same ability to reason. I have a different take; I think natural immunity of the human body is incredibly resilient, and so as long as you build up a tolerance to the poison, uh you're going to do fine, so uh I would say just start slowly; microdose the M&M's, build up your tolerance, and then you take a ton of artificial dye. No amount of artificial dye could do anything to me at this point; I have consumed so much Celsius and so many processed foods that I…

Am invincible. Some people say hubris. Yes. Thank you. Thank you, everyone. Thank you. Yes, I'm unkillable by the American food industry. Uh, uh, let's end. Let's end the show there. We've got to get on with uh, Taipei. We do, but we will see you guys tomorrow. A great show. Thank you for tuning in. We'll see you tomorrow. And thank you to the incredible corporations that make the show possible. Thank you.