transcript

Speaker 1:
[00:00] You're watching TBPN.

Speaker 2:
[00:03] It's Thursday, April 23rd, 2026. We are live from the TBPN Ultra Down, the Temple of Technology, the fortress of finance. We have a gigastream on our hands today. We will be interviewing one, two, three, four, five, six, seven, eight, nine, ten different members of the new Teal Fellowship Class, working on everything from ultra-fast logistics companies, to investigative journalism, to AI recruiting, and compute infrastructure, all sorts of different projects, and I loved last year- Ten minutes each. Ten minutes each. It'll be quick. It's a lightning round of lightning rounds. I loved last year. It's always super inspiring to see what young entrepreneurs are working on, and the Teal Fellowship class always impresses me. We've had three of these folks on the show before, so we'll be checking in with them, seeing where progress is at. The Teal Fellowship, I feel like it's grown. I don't know that it was always $250,000, but it is now. So the Teal Fellowship gives $250,000 grant to young people to build startups instead of going to college. So everyone here has either not graduated or dropped out or deferred or declined to attend college, and instead is building a technology company.

Speaker 3:
[01:20] Is this a smaller class than usual?

Speaker 2:
[01:23] We probably don't have the full class because of scheduling and not everyone can be everywhere all the time, and at times we will be having them, all the Teal Fellows will be joining us remotely, but they are all in the same place up in San Francisco, and we will be calling in to them, and we have Nick and Tyler up there managing that end of the show. This class of fellows includes young founders building hypersonic highways, a research lab for building foundational models for robotics, a fraud bounty hunter, and a simulated brain, and you probably know Dylan Field is a Vitalik Buterin, he's also a Teal Fellow alumni, and we are big fans of so many different founders that have gone through the project, and Wilmonitis, also a Teal Fellow. So... Anyway, we're going to get that sound ready because we got some bad news. Thoma Bravo is taking a massive write down on a software company called Medalia. Thoma Bravo is reportedly handing over the software company Medalia to creditors after restructuring negotiations failed to materialize. This is a $5.1 billion equity wipeout for the firm who bought the business for $6.4 billion in 2021. And there's some background in Reuters that I think I should read through, and then you can sort of give me your analysis and what you're seeing on the timeline around takes. So, from Reuters, exclusive, Thoma Bravo nears agreement to turn software firm Medalia over to creditors. That does not sound good. Private equity firm Thoma Bravo is nearing an agreement to hand over software firm Medalia to lenders, wrapping up months of restructuring negotiations. The move will wipe out $5.1 billion in equity. Medalia has struggled in recent months under the weight of $3 billion of debt, which it owes to Blackstone, KKR, Apollo Group and Antares Capital. Thoma Bravo, Blackstone, KKR declined to comment. Apollo and Medalia didn't immediately return quest for comments to Reuters. So, like other software companies, Medalia's valuation has been hit in recent months over concerns that its services will eventually be supplanted by artificial intelligence. What does Medalia do? Medalia provides software that collects and analyzes customer and employee feedback for companies. We've had some startups, some of them backed by Sequoia Capital, that are using AI to do this exact thing. That's a potential disruption. And then there's also Roll For Your Own.

Speaker 3:
[03:46] Their main rivals, Qualtrics. But yeah, and maybe you said this already, but you know that Sequoia was one of the big backers of Medalia. When they were a private company.

Speaker 2:
[03:56] Oh, interesting.

Speaker 3:
[03:57] So they're like, I mean, they're still private.

Speaker 2:
[03:59] There's a long history of that with, what was it? The firm that was before Zenefits was called Success Factors. Lars Dahlgaard did that deal.

Speaker 3:
[04:11] Sequoia is a real winner here. They did $35 million in 2012. They did another $50 million in 2014. And then they later led a $150 million round.

Speaker 2:
[04:20] Wow. And then eventually, did they take it public or did they sell it for $6 billion to Thoma Bravo directly from the private? Was Medalia ever a public company? That would be interesting.

Speaker 3:
[04:29] I think they were.

Speaker 2:
[04:30] You look that up and I will keep reading from this to give some more backstory. So private equity firms invested heavily in the software sector when interest rates were low following the peak of the COVID-19 pandemic. We all remember 3% interest rates. It was the golden era of startups and growth. All the DCFs were massive and then of course the valuations came down once the interest rates went up. So investors have become increasingly nervous about the sustainability of high valuations assigned to some of those assets and the debt raised to buy them if you're on a floating rate. These firms are not necessarily backed by a 30 year fixed mortgage like your house. Your interest rate went up and the debt payments ballooned. And if the cash flow from the business has not also ballooned, you could be in trouble like this company potentially is. Blackstone KKR and Antares hold some of the debt in traded and non-traded funds. FSKKR Capital Corp marked the debt at 79 cents on the dollar in its last quarterly report. And of course, debt is senior to equity. So what does that mean for the equity? Not good. And that's why this deal is happening. Apollo Debt Solutions marked it at 74 cents on the dollar. So a question about how can they even recover the full value of that debt? The equity is obviously in deep, deep trouble. So Blackstone's global head of private credit, Brad Marshall, set on a conference call in February that Medallia had been, quote, underperforming, not because of anything related to AI, but due to what we believe to be execution driven issues. So there is a question of, well, if they restructure and this company gets in the hands of creditors, maybe they can roll out AI features and become an AI winner and accelerate the top line and restructure the debt.

Speaker 3:
[06:14] I saw someone was sharing that there were some salespeople doing some anonymous reporting and saying a lot of the reps were struggling to hit quotas, basically delivering 20% of quotas, something like that. So just basically getting out competed in the market. And just to go back to the earlier point, the S-Medalia went public in 2019 on the NYSEE. It was taken private July 26, 2021. It had been trading around the $5 billion mark prior to the take private at 6.4.

Speaker 2:
[06:48] So Thoma Bravo, after taking it private, installed a new leadership team in early of 2025. Marshall, the head of Blackstone's Global Head of Private Credit, said that they were working on a turnaround plan, and we expect there to be discussions around the capital structure. And so people are blackmailing on the timeline. Brandon says, yes, I am sure this is the bottom in software and we won't get worse from here. Not good signs. This is potentially one of those cockroaches that Jamie Dimon is worrying about. There's another post that seems like it was deleted, but we can see some of the screenshot here.

Speaker 3:
[07:22] The tough thing is that everyone involved here has been on a kind of a press tour, saying like everything's fine. Yes. We're all good.

Speaker 2:
[07:31] Yes.

Speaker 3:
[07:32] AI is going to be an accelerant. But even ignoring the AI question, a lot of these businesses no longer have our founder led. They're competing against other companies that are founder led. And yeah, it's just kind of credit to Jamie Dimon for his comments, basically saying, I don't think this is over. Yeah, when you had first brands and some other shutdowns.

Speaker 2:
[08:04] Yeah, in the previous startup era, there were, I mean, I feel like it would have been hard to raise for just like a vanilla Qualtrics or Medallia competitor and just saying like, pre-AI, like we're just going to build SaaS and we're just going to build a crud app and write software.

Speaker 3:
[08:23] Yeah, a new player, is that company Aru?

Speaker 2:
[08:26] Yeah, that's a very different approach. Simulation, yeah, that's a very different approach.

Speaker 3:
[08:31] And then we had another one, I'm blanking on the name, it was a YC founder, he had gone through YC a few years ago, but his company is taking off, doing something similar. So people have been well aware of the opportunity that Medallia and Qualtrics have owned.

Speaker 2:
[08:52] But in other debt-related news, Xi Jinping wants to use the International Monetary Fund to rescue the array of distressed Chinese loans around the world. This is an interesting piece from the Wall Street Journal opinion, Noah Smith is laughing, he says, lull, Belt and Road failed so hard, Xi Jinping is incompetent. This is a very narrative violation.

Speaker 3:
[09:20] Do not go to China, Noah. Do not go to China.

Speaker 2:
[09:23] Yeah, this is an interesting thing, but maybe a potential bargaining chip in all of the other discussions. In tech, Tyler always points out that tech is like too focused on chips when it comes to Chinese diplomacy, that there are many, many other questions around trade and what Apple's doing in rare earths, and there are so many, and just the broad history of the Chinese empire that play into what their Taiwan policy is, how they will interact in the Middle East, and we tend to focus it all on AI. It's all about the data centers and the chips, and perhaps there are more things, and this is one example of something else that is a key factor in a broader negotiation that includes chip exports, and also rare earths, and also talent movements, and whether or not Manus will be able to move over to Meta and a million other things. And so that is the job of these world leaders, is to swirl around all the different trade-offs and get to, hopefully, a good deal for both sides. Anyway.

Speaker 3:
[10:27] Some breaking news.

Speaker 2:
[10:31] Merriam-Webster.

Speaker 3:
[10:31] GPT 5.5 is out.

Speaker 2:
[10:33] Oh, it is.

Speaker 3:
[10:34] Is out. Introducing GPT 5.5, a new class of intelligence for real work and powering agents built to understand complex goals, use tools, check its work and carry more tasks through to completion. It marks a new way of getting computer work done now available in ChaggBD and Codex.

Speaker 2:
[10:52] I like this demo of the Rubik's Cube. That's very cool. GPT 5.5 excels at writing and debugging code, researching online, analyzing data, creating documents and spreadsheets, operating software and moving across tools until a task is finished. GPT 5.5 delivers this step up in intelligence without compromising on speed, matches, GPT 5.4 per token, latency in real world serving. There's the model card there and there are some other posts.

Speaker 3:
[11:22] We will wait for the reactions to come in. We will gather them, we'll summarize them, but the team cooked.

Speaker 2:
[11:28] Cooked. Congratulations to everyone that worked on this.

Speaker 3:
[11:32] Let's head over to Merriam-Webster. John, what do you got?

Speaker 2:
[11:34] Merriam-Webster added a new word to the dictionary, I believe. And I feel like the pace of Wait, they actually added this? I don't know.

Speaker 3:
[11:44] I mean, it's in merriamwebster.com. Okay, so they have a slang section.

Speaker 2:
[11:47] Oh, okay, slang and trending. They must be working overtime over there, because the number of wombos, if you're not familiar with wombos, these are word combos, things like quech, lorraine, the lore plus explain, or quech.

Speaker 3:
[12:03] I could tell John, can you lorraine GPT 5.5?

Speaker 2:
[12:07] Yes, give you the lore and tell you about what happened with 4.5, 4.0, 3.5, 3 da Vinci. Give you the whole lore, but then also explain what this model is capable of. That's the full lorraine of GPT 5.5. But we're not here to talk about lorraine. We're here to talk about choppelganger, which is a new word in Merriam-Webster, the Dictionary. It's basically the main dictionary as far as I'm concerned. Choppelganger is a term for a less attractive version of someone or something. So whoever's going out there and distilling 5.5 into sort of a chopped version of it, that will be the choppelganger of the official GPT 5.5. So stay away from the choppelgangers unless you're really down on your luck again. And you need a Chinese open-source model to do something for you. Then buyer beware because it might have some flaws. But separately, there is new news from Michael Kratzius that the government seems to be taking anthropic and open AIs messaging around distillation very seriously and is setting up a task force. We can read a little bit more about this later, but setting up a task force to actually figure out how to prevent distillation at scale. As the models get bigger, it's more and more of an economic impact. When you're talking about a $100 million training run and a Chinese lab can distill it and sneak the weights out and exfiltrate the data, that's a lot different of an economic impact from stealing something that, I don't know, cost billions to train or billions to put together. So good news there. Hopefully, they're successful. It was also a white pill to see a lot of the labs working together to understand, hey, we're seeing this weird amount of data go to this particular company. Are you seeing the same thing? Or it's a shell company and we have five shell companies in these areas that are asking for these queries and then the other lab says, oh yeah, we're actually getting something similar. And if you puzzle piece those together, all of a sudden you're getting a really solid map of what a frontier system is capable of. And so it feels like we, the industry are the victim of a distillation attack, even if it might not register among a single lab, because that single lab is only getting hit with like one third or one fifth of the total attack.

Speaker 3:
[14:29] So we can pull up this post from Kratzios. I'll read the letter.

Speaker 2:
[14:33] Yeah.

Speaker 3:
[14:34] Subject adversarial distillation of American AI models, the United States leads the world in AI technologies. That lead reflects decades of foundational research, bold entrepreneurial risk taking in hundreds of billions of dollars in annual private investment. American AI leadership drives economic growth, strengthens national security and advances the frontiers of science, medicine and human knowledge. The breakthroughs emerging from American industry raise living standards, expand opportunity and improve lives around the world. However, the United States has information indicating that foreign entities principally based in China are engaged in deliberate industrial scale campaigns to distill US frontier AI systems, leveraging tens of thousands of proxy accounts to evade detection and using jailbreaking techniques to expose proprietary information. These coordinated campaigns systematically extract capabilities from American AI models, exploiting...

Speaker 2:
[15:27] I don't like that. That sounds bad.

Speaker 3:
[15:29] I don't like that at all. Exploiting American expertise and innovation.

Speaker 2:
[15:33] Every AI researcher needs this button on their desk if they detect a distillation attack. Fire it off and the security team will come and make sure you're locked down. And it does seem like overall all the labs are being more careful about how the APIs roll out, how they're doing KYC, how they're doing different partnerships and stuff. And I think it's all a back and forth and a dynamic. But it's fun that this new model is available and people will play with it. And I'm sure we'll see some cool stuff.

Speaker 3:
[16:08] Yeah, and Quen 3.6 rolled out yesterday. People are like, wow, it's almost as good as OPUS 45. Yeah, because they distilled OPUS 45. Anyways, I'm glad that Kratios is on it. And I'm glad the labs can coordinate and work together on this one.

Speaker 2:
[16:28] Yes, yes.

Speaker 3:
[16:30] More breaking news.

Speaker 2:
[16:31] Yes.

Speaker 3:
[16:32] And unfortunate, but we had talked about this previously. Meta is cutting 10% of their workforce, which is 8,000 employees.

Speaker 2:
[16:39] But this has been telegraphed for a while.

Speaker 3:
[16:41] Eliminating 6,000 open rolls.

Speaker 2:
[16:43] Yeah. I saw Microsoft was doing something similar. They're offering early retirement to something like 7% of the workforce trying to lean out a little bit. I'm sure the debate will continue to rage over the underlying motivations. These companies can be big and sometimes maybe they're too big. Maybe there's AI gains. Maybe there's different investment strategies. We will have to see where they are moving chips around, what projects they are actually cutting, what projects they are doubling down on, because these companies are also hiring at all times, effectively. Well, Kristoff had some actual media ideas for the tech industry. He says there's not enough media in tech, you see.

Speaker 3:
[17:28] We need more tech-positive media.

Speaker 2:
[17:31] For sure. So he says, silent library with founders, winner gets investment.

Speaker 3:
[17:36] What is silent library?

Speaker 2:
[17:38] Silent library, so is this like a webcam that observes?

Speaker 3:
[17:44] Silent library is a television show, it had four seasons, it's a game show.

Speaker 2:
[17:49] Oh, it's a game show. Game shows are fun. I can see a game show being fun, at least different.

Speaker 3:
[17:54] Six friends vying for a cash prize. If only they can remain silent as one of them is forced to endure a bizarre stunt while seated in a library.

Speaker 2:
[18:03] Okay.

Speaker 3:
[18:04] What else? Jackass with founders.

Speaker 2:
[18:07] What does that mean? How do you make that tech related, maybe like humanoid robots? When I think of Johnny Knoxville, I just think of bull riding effectively. So I imagine riding some robotic bull would be it, but it's very dangerous. The team that worked with Bam Margera and Johnny Knoxville, they had some serious injuries from time to time. And I don't know if it's in your best interest to be a founder and be like, yeah, I need to launch my startup. So I got to go in the ring with a mechanical bull and potentially break my arm and not be able to code. And I don't know, seems rough, but would be entertaining and with the right twist, I believe it could work. Interview founders while going ghost hunting. I feel like paranormal tech content, Jesse Michaels doesn't get enough credit here.

Speaker 3:
[18:58] He's a technology podcast.

Speaker 2:
[19:01] It's in the technology charts and it charts high. It's up there and he has an incredible pedigree, great investor and also has some of the best reporting on aliens and paranormal activity and conspiracy theories. And I guess he just hasn't cracked the way to bring a founder along. But if your friends with Jesse Michaels are in touch, maybe you just tag along for a little cameo on one of his videos and that's all the PR you need.

Speaker 3:
[19:33] Founders give their pitches while skydiving. Can you talk while skydiving?

Speaker 2:
[19:41] With the right headset.

Speaker 3:
[19:42] Have you gone skydiving?

Speaker 2:
[19:43] No, but I like the idea and I like the videos. I like that it's always in like the Vibriels of like the... What's that movie? Is it Keanu Reeves Point Break? They go skydiving and he goes skydiving and there's only one parachute, so they have to fight it out for the parachute. You haven't seen this movie.

Speaker 3:
[20:00] Have I told you the story where I was on an overnight flight back from Europe?

Speaker 2:
[20:05] No.

Speaker 3:
[20:06] I'm sitting on a plane and I'm awake, I can't sleep. And out of the shadows, John Wick is walking up the aisle.

Speaker 2:
[20:20] Keanu Reeves.

Speaker 3:
[20:21] Yeah, the real actor. And I'm just sitting there. It's effectively like 2 a.m. And Keanu Reeves is like walking at me on this plane in the middle of the night. We ended up talking for like five minutes.

Speaker 2:
[20:32] Really?

Speaker 3:
[20:32] It was cool.

Speaker 2:
[20:32] That's cool.

Speaker 3:
[20:33] And then he went back to his seat.

Speaker 2:
[20:35] And yeah, well, he stayed on the plane. He did not skydive.

Speaker 3:
[20:38] He was a super, super nice guy.

Speaker 2:
[20:40] What else do they have here? SF Party, I'm Schmacked Videos. I don't know I'm schmacked.

Speaker 3:
[20:45] Wow. Unc. You're unc. You guys know. I kind of remember, but... You remember? Basically, this was this whole like, you could call it a media company, but it was really like a guy that would make party videos.

Speaker 2:
[20:57] Okay. What era are we talking about?

Speaker 3:
[20:59] For every college. So it was really popular probably 2010 to 2013.

Speaker 2:
[21:04] Okay, okay.

Speaker 3:
[21:05] And I will say a lot of people use these videos as a sort of college guide.

Speaker 2:
[21:10] Okay, okay, okay.

Speaker 3:
[21:12] So I'm sure they have millions of views at this point.

Speaker 2:
[21:16] Yeah, yeah, yeah. Last one, VC Price is Right. Is this a good idea? I feel like people don't... Like there's always a huge adverse selection problem with anything related to sharing financials or sharing your company. I mean, this is a twist on Shark Tank to some extent, but they actually did try to do Shark Tank for software. It was called Planet of the Apps. Gary Vaynerchuk was a host. We can ask him about him tomorrow. He's coming on the show. And I'd love to know his reflections on tech media. I think Gwyneth Paltrow might have been a host, co-host. They had a pretty stark... Planet of the Apps. Look it up. Planet of the Apps instead of Planet of the Apes. Yeah, Gwyneth Paltrow was there. Gwyneth Paltrow, and Will.i.am. They had a stacked line of Jessica Alba, Jessica Alba, founder of the...

Speaker 3:
[22:06] On each of the episodes, software makers have 60 seconds to pitch their idea in front of the advisors on a slow moving escalator for the visual idea of an elevator pitch. Why not put them in an elevator? New episodes were released on Tuesday. Did it get renewed?

Speaker 2:
[22:25] Did any of these apps go anywhere?

Speaker 3:
[22:27] They had Silo, Focus and Study Timer, Companion, Mobile, Personal Safety, Pair, A Showroom to Your Home, Dote, The Mobile Mall, Traxx Battle Squad, Twist, Live Events with a Dating Twist, Scooch, The Word Game. Yeah, these sound like interesting ideas. Not as interesting as a cooler that has a gas-powered engine that you can ride around, which I think of as the quintessential Shark Tank product.

Speaker 2:
[22:59] Yeah, exactly. Shark Tank works because the products are highly visual, and you get to see, oh, we made a new mop. Okay, we're gonna watch Mark Cuban try and mop, and this is going to be physical and visual and interesting. It is very difficult. Are we making progress on our Shark Tank idea? I feel like this is very, very important. We'll have to get the update soon. How's the Shark Tank idea going? Are we making progress? We're on the way. We're on the way. Okay, good stuff. We're very excited about that. Anyway, lots of opportunity in media. Also, lots of opportunity in simulators. We have talked about simulators, TBPN simulator, we made Jeremy Gaffan simulator, but the simulators are getting less advanced in some weird twist. You would think they get bigger and bigger and you turn them into real games. We're going the opposite direction because the simpler the game, the funnier. This is Coconut Simulator. Let's watch this video.

Speaker 4:
[23:52] Coconut Simulator.

Speaker 5:
[23:53] This better be worth 99 cents, okay? I'm not gonna lie. Okay, what do we do, do we move?

Speaker 2:
[23:58] It looks good.

Speaker 4:
[24:00] I can't move. Why can I not move?

Speaker 6:
[24:03] Am I like dead ass a coconut? Are you serious? I spent 99 cents to be a coconut?

Speaker 4:
[24:08] There's no way. What could be the difference between arcade and realistic? Coconut has no eyes, so you can't see anything. In realistic game modes, you're like, you got it.

Speaker 2:
[24:17] It's just black because the coconut has no eyes. It's first person view.

Speaker 1:
[24:20] This is impossible.

Speaker 2:
[24:21] You were in third person view, but it gets better because there's a sequel. There is Coconut Simulator 2, which is now, I believe multiplayer. Let's play this one.

Speaker 1:
[24:31] What game mode? Arcade.

Speaker 2:
[24:35] It's the same game.

Speaker 1:
[24:37] So what's the difference between Coconut Simulator 2 and 1? Coconut Simulator 2 is a lot more in-depth. The other Coconuts, it's multiplayer. I thought you guys said there was a story to this one. There is. You're just not seeing it. It's a beautiful story about growing old with your friends around. This game is definitely game of the year. It's beating Crimson Desert. No chance it survives. Trust realism mode is different in this one. I'm telling you right now, if realism mode ends up being the same, I'm going to crash out. Coconuts have no eyes, so you can't see clearly, oh my god.

Speaker 2:
[25:11] Coconuts have... It's such a funny prank. I love a simulator prank. What were you saying? Who's in the chat? Ty?

Speaker 3:
[25:17] Ty's has games as a bit.

Speaker 2:
[25:19] Yeah, no, it is true. There's actually a third simulator that we need to review. This is Bananas Simulator.

Speaker 6:
[25:26] Bananas Sim.

Speaker 7:
[25:27] You are the first person in the whole world to play Bananas Sim.

Speaker 3:
[25:30] What do you mean?

Speaker 7:
[25:31] Remember when I showed you Coconuts Simulator and I added this to my wishlist?

Speaker 6:
[25:34] It's not supposed to be out yet.

Speaker 7:
[25:36] You're the first person to play it ever. I'm kind of jealous.

Speaker 3:
[25:39] How did you get it then?

Speaker 4:
[25:40] The developers liked our last video and they sent me an email.

Speaker 6:
[25:44] They gave me this one for free. The developers of this game sent you an email.

Speaker 7:
[25:49] I have it on my phone too if you want to see it.

Speaker 3:
[25:50] It's the same email, just smaller.

Speaker 8:
[25:53] So you didn't have to pay real human dollars to play this?

Speaker 7:
[25:57] Not this time.

Speaker 8:
[25:58] This one doesn't even have sound?

Speaker 3:
[26:00] It's the beta.

Speaker 7:
[26:00] Technically you're the number one Bananas Sim player in the world right now.

Speaker 9:
[26:03] You have 391 points.

Speaker 2:
[26:06] How do you get points?

Speaker 7:
[26:07] 395 points now.

Speaker 2:
[26:09] Just for playing? Just for sitting there? Diet Coke Simulator.

Speaker 3:
[26:12] Diet Coke Simulator is good.

Speaker 2:
[26:14] You just play Diet Coke is good.

Speaker 3:
[26:15] You just sit there and every few minutes.

Speaker 2:
[26:17] I mean, we have Data Center Simulator. Maybe the next one is just GPU Simulator, and it just sits there and it just hums.

Speaker 3:
[26:23] Yeah.

Speaker 2:
[26:23] That's it.

Speaker 3:
[26:23] Sometimes you overheat.

Speaker 2:
[26:25] That's the end.

Speaker 3:
[26:25] Sometimes you got to be plugged back in.

Speaker 2:
[26:27] Yeah. Well, let's move back over to tech. Kevin Kelly, the founder of Wired, had some incredible predictions in 2016. Let's read through them. Summary of the inevitable understanding, the 12 technological forces that will shape our future. This was Kevin Kelly's book from 2016. According to Kelly, much of what will happen in the next 30 years is inevitable. The future will bring with it even more screens, tracking and lack of privacy. In the book, he outlines 12 trends that will forever change the ways we work, learn and communicate, becoming, moving from fixed products to always upgrading services and subscriptions. That has definitely happened. We're moving away from even seats. Everything's consumption based now. Cognifying, making everything much smarter using cheap, powerful AI that we get from the cloud. Nailed it.

Speaker 3:
[27:19] That is a fantastic prediction for assuming he wrote it. He actually wrote it probably in 2015, published in 2016.

Speaker 2:
[27:30] But he's the creator of Wired. He's been tapped in to attack his entire career and has a ton of interesting reflections too. He's written a more recent book about reflections on life. That's very good. Depending on unstoppable streams of real time for everything, for sure, turning all surfaces into screens. That definitely happened. Your toaster can watch TV now. Shifting society from one where we own assets to one where instead we have access to all services at all times. Collaboration at mass scale. On my imaginary sharing meter index, we are still two out of 10. Filtering, harnessing intense personalization in order to anticipate our desires. Remixing, unbundling existing products into their most primitive parts and then recombining in all possible ways. That's definitely happening. Coconut simulator, unironically an example of that. Interacting, immersing ourselves inside of computers to maximize their engagement. Tracking, employing total surveillance for the benefit of citizens and consumers. Sounds scary. Potentially a good outcome if things are done properly. Promoting good questions is far more valuable.

Speaker 5:
[28:32] You're out of answers.

Speaker 2:
[28:33] What?

Speaker 3:
[28:34] You think total surveillance is a potentially good outcome?

Speaker 2:
[28:36] Well, he says total surveillance for the benefit of citizens and consumers. So if there is a black box where my Netflix activity exists, where no Netflix employee can see it because it's encrypted, but it can make great recommendations and recommend me the next great show that I will actually enjoy, I'm cool with that surveillance. That's surveillance, but it's potentially good for me, and I have a better experience.

Speaker 3:
[29:00] For the benefit of John Coogan.

Speaker 2:
[29:01] Yeah, yeah, there is a surveillance bulk case. Constructing a planetary system connecting all humans and machines into a global matrix. Okay, that one we're still waiting on, but a lot of good, interesting predictions. Jason Schuman says, wild how accurate these predictions were. And they were, in fact.

Speaker 10:
[29:19] Wow.

Speaker 3:
[29:19] Without further ado.

Speaker 2:
[29:22] We have our first guest of the Teal Fellowship Gigastream. Welcome to the show. How are you doing?

Speaker 10:
[29:28] Great guys. Nice to meet you.

Speaker 3:
[29:30] Nice to meet you too.

Speaker 2:
[29:31] Thank you so much for taking the time to come on the show today. Introduce yourself in the company.

Speaker 10:
[29:37] Yeah, I'm Victor Boyd. We're building autonomous forklifts. The real goal of all of this is get anything anywhere in just a few hours. And we're starting with what we think is right. And we're going to do everything it takes to get there.

Speaker 2:
[29:50] How vertically integrated do you want to be on day one? You want to retrof, is this the comma AI of forklifts? Or is this the Tesla forklifts on day one? What are you thinking?

Speaker 10:
[29:59] Yeah, we actually started thinking we'd be the comma AI of forklifts, but you know, cars have CAN bus standardized since 2008. It's a software problem to get control over a car. You know, the difference between cars is just software. The difference between forklifts, even if it's the same forklift, same year even, the internals will look different from each other pretty often. So if you want to build like a kit that goes on any forklift, it doesn't actually make sense. So we tried that, we did, and then we realized, okay, that's not the way to go. Then we decided, let's just retrofit one forklift, one model, we'll deal with the differences throughout the years, but at least it'll be okay. We tried that, that also sucked. It was just unreliable, it wasn't fast enough for the customer. We were where everybody else was in the market. This isn't the most original idea in the world, everybody knows.

Speaker 2:
[30:48] Yeah, construction.

Speaker 3:
[30:49] Well, there's also big, there's Applied Intuition, like there's big companies in the category that have been running at this problem for a long time. So you do, you kind of probably have, you got a speed run trying a bunch of different approaches to figure out what works for you.

Speaker 2:
[31:03] But the same thing happened in the car industry where Ford and Toyota were like, yeah, we're doing self-driving too. And then it was like, okay, they can do some lane keep assist and some adaptive cruise control, but I'm still waiting for even Tesla level FSD from like three years ago to roll out into like the major American car manufacturers. So there's clearly an opportunity. So where are you now? Did you build that first autonomous forklift? Like what went into that?

Speaker 10:
[31:31] Yeah, I mean, I think what Tesla did right was that they got control over their platform, right? We realized that that was going to be the thing for autonomous forklifts. If you wanted to make it viable for customers, you actually had to make the platform viable for autonomy first. So we had to build our own forklift, and we did. We built our own forklift in Q4 last year through a manufacturing partner, and it worked great. We've done some iterations since then, and now we are deployed, and we're deployed in a very difficult environment. We're doing better than anybody else in terms of throughput and reliability, which is really all we care about. We plan on continuing this year. What we want to do this year is we just want to make it a super scalable product, because to be honest, it's not right now.

Speaker 3:
[32:15] Sorry to interrupt, but has there been a company in the last 10 years that decided to build a new forklift from the ground up, or is this a category that has been generally overlooked, as everybody has wanted to build new EVs and platforms that are maybe more exciting to some?

Speaker 10:
[32:35] There's some people that got halfway there. They would go with somebody else's design and make a few changes and think that that was enough. But realistically, you have to be in the process from the very beginning. Otherwise, there's just all these trade-offs that are made in the design process that really affect you. It makes your product horrible. So we did it from scratch. We have some off-the-shelf parts, but that's a motor, right? Obviously, I'm not going to design a motor from the ground up. I don't need to. I'll just use a forklift motor. But everything else needs to be me. Everything else has to be my design so that we can have full control over the system, iterate faster than anybody else, and build the actual real product that actually works.

Speaker 2:
[33:22] Can you talk about the environments that forklifts operate in? You mentioned that you're deployed in a difficult environment, but what's the standard environment, and then how does your test case differ from that?

Speaker 10:
[33:37] I mean, the standard environment, you can think of a warehouse where they just ship a bunch of dog food. Let's say 400,000 square feet, just lots of racks. It's not that difficult, right? If you cause product damage in a dog food warehouse, it just stinks. It does stink really bad. Don't get me wrong, they're not going to be happy, but we went into a pharmaceutical warehouse. Whereas in the dog food warehouse, okay, I damaged the pallet. Let's say I destroyed everything on the pallet. That's a couple thousand dollars. In a pharmaceutical warehouse, I destroy a whole pallet. Dude, that's hundreds of thousands of dollars. So I think our idea behind it was, this is an incredible forcing factor. This makes us actually build a reliable product that's not going to cause damages. And then when we are successful here, we get to show this to all of our other potential customers. Like, look guys, I know you've been burned by the industry before, but look at how good we're doing in this place. If we destroyed product in their place, we wouldn't be there anymore, because we would have already... You know, it's a bad thing. You don't do it. Yeah.

Speaker 2:
[34:39] Talk about teleoperation. Did you go down that path at all? Are you compatible with teleoperation? Is there any value to teleoperation?

Speaker 3:
[34:47] Yeah, I feel like this is such a unique environment because you can figure out autonomy, but there should be a relatively easy way to take over the system and just use an Xbox controller as you get to full autonomy. I don't know if that's what...

Speaker 10:
[35:03] I'm glad you said Xbox controller. That's what it is. So when we do teleop, which is about 50% of the time right now, it is with literally an Xbox controller. There's some things where they have to click on the screen to select what palette they need to pick up, for example. But when there's some scenario where the robot's lost, like it delocalizes, or we need some extra training data on some workflow, then it's like, yeah, just pick up the Xbox controller. It's literally, guys, they're not on-site. They're somewhere thousands of miles away just getting it done. It works perfectly. I think it's extremely valuable. I think it was a dirty word in the industry for a long time to say, like, remote operation, which is really stupid because customers don't care, guys. Customers don't care if you're fully autonomous. Customers care if the work is done. So, if you're over here, like, you know, stressing about, oh, dude, we're only, you know, 10% autonomous, but you're getting the job done every single time and you're profitable, like, I mean, obviously, keep going for more autonomy, push those margins, but, dude, the customer could not care less.

Speaker 3:
[36:12] Yeah, totally. We've said that to some humanoid founders on this show, and they're like, no, it's got to be fully autonomous.

Speaker 2:
[36:19] Yeah, it is like a less sexy narrative. So I think people want to deny it or something, but yeah, and work product makes the most sense. Talk to me about the interaction modality for the human or the manager or the owner of the warehouse in a basically full autonomous mode, because I imagine at some point you have to have a system that decides, okay, we actually need to dispatch an order to get that dog food off of the top shelf, and there's a lot of different pallets that are stacked up at different levels, and what systems are you plugging into, or do you want this to be something where there's still a human in the loop managing and dispatching orders, and then they're sitting in an office, maybe overlooking the warehouse, just sort of giving orders, but what is the path of the workflow?

Speaker 10:
[37:17] Right now, it's basically like the robots have this task that needs to get done every day, pretty much all day. This is very common in warehouses where there's only a few things that you need to do every day, and it just takes a really long time, and you need to have it done by the end of the day, but it doesn't really matter how fast it is. In those scenarios, you just give the robot instructions. We have a map, and you can set up zones or shelves, and you say, this is what's in this zone or this is what's on this shelf, and it needs to go to this other zone or this other shelf, and then the forklift just does it. It can be that simple. Eventually, the idea is that I would like to be able to run the warehouse for the customer. I don't want them to ever have to think about it at all. I'd rather it just be like, okay, we are literally your palette movement within this warehouse. We run your whole business anyway in terms of physical movement. We also have your inventory management. We can make decisions based off of that. We know your truck is going to be in here soon. For example, we'll just set up the staging for you. You don't have to tell us that has to happen. That's the goal in the next year.

Speaker 3:
[38:25] Forklift is the wedge. It's very smart. We were talking about simulator games earlier. The chat is asking, have you thought about making a forklift simulator where normal people could play the game online, but in reality, they're actually helping you train your own?

Speaker 2:
[38:44] Yeah, getting training data.

Speaker 10:
[38:46] Yeah, that's actually funny. I did think of that because I was looking for a game myself. I've always been into logistics, but I got into warehousing a few years ago. I looked everywhere, dude. I even looked on Roblox. I was looking on Roblox. They don't have any good warehouse games, man. It's such a shame. Somebody needs to do that. Maybe somebody in your chat will do that for us.

Speaker 3:
[39:09] How did you make your first dollar?

Speaker 10:
[39:12] How did I make my first dollar? Yeah. I think I'm pretty sure it was when I was selling. I sold candy on Easter when I was very young. And I actually made money there, which is funny, because the kid I sold it to was literally walking with a basket full of candy. Definitely more candy than I got that year. But I sold him this egg and I was like, dude, there's something really good in here and you won it, but I'm not going to tell you unless you give me $20.

Speaker 2:
[39:36] A random award.

Speaker 3:
[39:41] Did you drop out from anywhere or are you coming out of high school?

Speaker 10:
[39:47] Yeah, I dropped out of UAB. I went to a state school in Alabama. I grew up there and then I didn't have the money to go anywhere else. I didn't want to go to school in the first place. The second I got the opportunity to fly out here to SF, I was like, okay, it's over. I'm leaving.

Speaker 2:
[40:07] It's over. I love it.

Speaker 3:
[40:08] Just getting started.

Speaker 2:
[40:09] Well, Victor, thanks so much for coming on and breaking it down.

Speaker 3:
[40:11] Fantastic to meet you. I'm sure you'll be back on this year.

Speaker 2:
[40:15] Good luck.

Speaker 10:
[40:15] Yeah, guys.

Speaker 2:
[40:16] Have a good one. Up next, we have The Antifraud Company. If you hate fraud, you're gonna love this company, Alex from The Antifraud Company. He uses AI.

Speaker 3:
[40:27] We gotta ask him about those bears.

Speaker 2:
[40:30] Oh yeah.

Speaker 3:
[40:30] Fake bear attacks.

Speaker 2:
[40:31] Yeah.

Speaker 3:
[40:32] This is a big opportunity.

Speaker 2:
[40:34] Yesterday on the show, we were talking about an insurance scam where individuals dressed up in full bear costumes with fake bear claws and vandalized a Rolls Royce Ghost from 2010 and then filed an insurance claim, which was then disproven by, I believe, some sort of biologist who knew about bears and what they look like on camera and was able to debunk it. We were able to debunk it too because it looked ridiculous. Anyway, we have Alex from the Antifraud Company back on TBPN here for the Teal Fellowship Gigastream. Welcome to the show, Alex, how are you doing?

Speaker 3:
[41:09] Great to see you. Suited up. Suited up. Looking sharp.

Speaker 11:
[41:13] I cannot look like a fraudster when I'm catching fraudsters. Yeah.

Speaker 3:
[41:16] Have you been dailying a suit?

Speaker 11:
[41:19] Not every day, only when we're doing formal things. Yes.

Speaker 2:
[41:22] Yes, indeed. So you've been on the show before, but reintroduced the company and the shape of the business and how you think this will play out. Then I want the update on where things are, what traction looks like, how you're actually applying the technology, and then I'm sure there'll be a ton of follow-on questions.

Speaker 11:
[41:39] Yeah. So it's very simple. We're building AI models to detect fraud when it happens, and then we go and sue the fraudsters and we only make money on contingency when we actually get a recovery for the government. We're able to keep 15-30 percent under these whistleblower laws.

Speaker 2:
[41:55] Interesting. Yeah. And have you received a whistleblower bounty yet, or is that something that's a goal for the next year or two?

Speaker 11:
[42:05] That is a goal. Maybe not for the next year or two, because the US legal system is quite slow. So it might take three to four years before we see our first dollar of revenue, actually.

Speaker 2:
[42:15] Okay. How much can you pull forward? I know that the plan is to use AI, but how much can you pull forward by just rolling up your sleeves and doing it yourself or hiring humans to sort of follow the same process? Is this a problem that can only be attacked with AI, like a recommendation feed? You could never curate the TikTok feed for every individual on that platform without AI, but you can certainly go and take a photo instead of using AI. How dependent is this on AI from the very first project?

Speaker 11:
[42:52] Right, so I would say that people have always been able to find fraud manually, but usually either they luck into it or they had a tip from somebody on the inside. And we're sort of trying to eliminate that bottleneck, essentially. We want to find the fraud using AI because we're essentially building the palantir of fraud detection. We're tapping into all these sources of data. And once we have a good idea of what's going on, then we'll send out human investigators again, go talk to sources. But we want to be going outbound. We don't want to have to wait for people to come to us because that's just less efficient. When there's $600 billion in fraud, you're not going to get tips for most of it. And if you want to solve it, then this is sort of the way you have to go.

Speaker 2:
[43:33] I remember during the COVID stimulus checks, there was someone who got a check for a company. What were those called, the stimulus checks? It was specifically for business loans.

Speaker 11:
[43:45] SBA.

Speaker 2:
[43:46] SBA loans, something like that. But those loans, one of them was just called like Ford Raptor LLC. And then someone dug into it and lo and behold, like someone set up a fake LLC just to buy a Ford Raptor. And that one seemed extremely obvious. But I'm wondering how much is available in terms of public records versus stuff that you plan on doing FOIA requests for to get new sources to comb over.

Speaker 11:
[44:13] That's a great question. So this sort of stems off of the OSINT community, Open Source Intelligence, which is where people go around and they look at publicly available data. And when you're doing pure OSINT, you're only using, again, open source stuff that's publicly available. That limits you. So we do get closed sources too. But I guess we want to start there. We view it like a funnel. So starting with the open source information lets us cast the widest net. Of course some of it we're going to need to do other things. Like you mentioned, FOIA is one way that you can get private information. But it's really like you have to know, again, what you're requesting from the government or what sort of private data source you want to buy or what human you want to go out and talk to. And so to get the high level view, you do need to be essentially just ingesting a bunch of stuff. So one analogy we like to make is the story of the blind men and the elephant, right? Is there's like five blind men and they're each touching a different part of the elephant. The guy who's touching the legs thinks it's a tree. The guy who's touching the trunk thinks it's a rope, right? Fraud is kind of like this. It always leaves traces, but it's somewhat difficult to put them back together. And that's why we just want to have as many different essentially like sensors on the government databases, on contracts, on all these data sets. And that gives us sort of a good look into what's going on. Then we then we ontologize it, run it through the models, make graphs and see if it matches on to any known patterns of fraud that we're searching for.

Speaker 3:
[45:41] How many projects are you guys taking on at any one point? Because I'm sure every single day there's like new ideas, new opportunities, sounds like a very, very big market, unfortunately.

Speaker 11:
[45:52] Quite a few projects at a time. And again, we're just looking everywhere. I mean, like you mentioned, small business administration is a great place to look with all these loans, defense, just anywhere the government spends money. There is undoubtedly going to be people who are doing it. In an unethical way. Oh, absolutely healthcare, yes. I mean, like we spend trillions on healthcare. We spend trillions on defense. There is a lot of stuff going on in these places.

Speaker 3:
[46:20] And what's going on in Sacramento right now? Is there some new legislation that's trying to make it harder to do journalism around fraud? Did I see that?

Speaker 11:
[46:30] I believe so. I think this was a response to Nick Shirley, I believe.

Speaker 2:
[46:34] Oh, really?

Speaker 3:
[46:35] I think so. I think it's called the Nick Shirley Act.

Speaker 2:
[46:38] Wow, okay. Right?

Speaker 3:
[46:40] I don't know if it's actually called that, but that's what people on the internet are calling it.

Speaker 2:
[46:43] I know it went super viral over the holiday break.

Speaker 11:
[46:45] Yes. No, that's my understanding. Obviously, Nick first did the big expose about the daycares in Minnesota, which were getting funded by Medicaid or something of that sort. And so that also helps drive a lot of awareness and interest in our company as well, because it's just sort of visceral. I guess first there was Doge, where people just got really mad about how their money was being spent. And then the daycare sort of was the next step. And I think there's just a big anti-fraud moment in the United States these days, where people are just upset about their dollars aren't going far enough, right? I mean, like it's $600 billion a year. That's like seven, eight percent of the federal budget. So that means like a good chunk of your hard-earned tax dollars is being stolen by people who are contracting with the government in the wrong way. And I mean, we need to restore trust in our institutions. We need to have good governments, have people have faith in our institutions. I think that, again, aligning the incentives, having sort of private actors come in and do this and rewarding them when they're able to claw back money for taxpayers, the most of the money that does claw back will go back to the Treasury and to the taxpayers. I think this is kind of a no-brainer.

Speaker 2:
[48:00] How do you think about the actual application of AI tools and models in this particular case? I understand that you're sort of creating a database or mirror a data lake of all the different sources. But then, do you need a bunch of examples and to fine-tune a model? Do you just need to load examples of red flags into the context window? How far down are you on the AI research side of understanding the problem? Because I can imagine, with a frontier model, you can go and if you give it a lot of examples and data points, you could potentially just one shot a detection and then it's just applying it at scale is the problem. But how are you thinking about applying the actual technology?

Speaker 11:
[48:46] So there are potentially two steps here. The first step is that a lot of this data is unstructured. A lot of this is documents. LLMs really speed that up. It's sort of getting structured entities out of the unstructured data. And then we ontologize this. We have like this company has this contract, has this relationship with this employee, etc. We build sort of this ontology knowledge graph like thing. And then we have the second layer, the rules model essentially. My co-founders are both lawyers, have a legal background. They've been working on sort of the fraud issue in the legal sphere. And so they know what that looks like. And with their input and with their expertise, they were able to essentially develop this rules model, this rules layer, compare that to what we're seeing in the real world and flag things that look like violations in real time or at least stuff that could be violations and we need to get more information on.

Speaker 2:
[49:34] Did you drop out? Because if you have two lawyer co-founders, did they drop out? Like what's the story? Who's the dropout?

Speaker 11:
[49:42] I'm the dropout. My co-founders have JDs, unfortunately. They're also a bit older, so they're not eligible for the Teal Fellowship, unfortunately. But I did drop out of Brown in 2025 to work at Palantir and then start this company.

Speaker 2:
[49:58] That's great. That's great. Well, good luck and thank you for everything that you do.

Speaker 3:
[50:03] How big is the team now?

Speaker 11:
[50:05] We have 11 full-time employees, but we're hiring engineers. If you're an AI engineer that wants to work on this problem, visit antifraudcompany.com and we'd love to work with you.

Speaker 2:
[50:15] That's a great name. Well, good luck out there.

Speaker 3:
[50:18] Great to see you, Alex.

Speaker 2:
[50:19] Thank you for the great fight and let us know when you catch a big culprit so you can come on and tell the story because I'm sure it'll be riveting.

Speaker 11:
[50:25] Yes, indeed. Thanks, fellas.

Speaker 2:
[50:26] We'll talk to you soon.

Speaker 3:
[50:27] Great stuff.

Speaker 2:
[50:27] Goodbye.

Speaker 3:
[50:28] It would be funny if he was investigating a company to have the founder on. We're interviewing them and then Alex joins the call.

Speaker 2:
[50:38] Sort of, yeah, the catcher-pride or whatever you want to do with fraud. This is a new media thing.

Speaker 3:
[50:45] This is the new media opportunity.

Speaker 2:
[50:47] New media for pumping dumps and schemes and all sorts of stuff and rug pulls. That inadvertently happened during the NFT and crypto boom because there were a number of founders who went on shows and then it was revealed. I mean, the famous one is Joe Weisenthal and some Bloomberg reporters talking to SBF, I believe, on odd lots and they ask him, so this is like a black box that you put money in and you just get more money out? And he was like, yeah, exactly. Exactly.

Speaker 3:
[51:18] Wait, SBF was on odd lots?

Speaker 2:
[51:20] I'm pretty sure it was odd lots. I know Joe was on that podcast with SBF. And he basically describes a Ponzi scheme.

Speaker 3:
[51:34] This is crazy. SBF and Matt Levine on odd lots.

Speaker 2:
[51:38] And Matt Levine asks, like, so he describes a Ponzi scheme and SBF basically just says, like, exactly. Like, that's why it's good. It's like you put more money in and then it grows and then it's this magical system. He was like describing like crazy DeFi schemes. It was a rough time. It was a crazy time. But we lived through it and we became stronger in the process. We do have our next guest.

Speaker 3:
[52:00] Yes, April 27th, 2022. And so just six months later, everything would go.

Speaker 2:
[52:08] It was a wild, wild time. Well, we have Nick from Every Ticker in the waiting room. Let's bring him in to the TBPN Altruim. Nick, how are you doing?

Speaker 9:
[52:16] Pretty good. How are you?

Speaker 3:
[52:17] We're great.

Speaker 2:
[52:17] Great to meet you. Thanks so much for taking the time. Great to meet you.

Speaker 3:
[52:20] We hear you are the youngest ever Teal Fellow. Is that true?

Speaker 2:
[52:24] That is true, yeah.

Speaker 3:
[52:26] Wow.

Speaker 2:
[52:26] That's amazing. So are you dropping out of college?

Speaker 3:
[52:28] You can retire now. You should drop out of the Teal Fellowship now.

Speaker 9:
[52:33] Yeah, I'm dropping out of high school. It's a pretty crazy experience.

Speaker 2:
[52:36] Wow. Congratulations.

Speaker 3:
[52:41] What have you done to date where you get identified and chosen for an opportunity like that?

Speaker 2:
[52:48] Yeah.

Speaker 9:
[52:48] So the main thing I've been working on for the past year or so is Every Ticker. I'll start with the problem that we're solving. It's that the vast majority of the US equity market is on the small, mid and micro cap stocks. But Wall Street only covers the mega and large cap stocks. So what we do is we use LLMs to generate high quality research on every single US stock, including the ones that Wall Street doesn't cover.

Speaker 2:
[53:13] Interesting. Okay. So like the big stocks would have an equity research report from Goldman Sachs or Morgan Stanley, these are the 20 page PDFs that you see come out every quarter or so with a buy sell hold. Are you offering financial advice? Like who is the end consumer of this if it's some small or micro cap?

Speaker 9:
[53:34] So our philosophy is that we explicitly do not offer any sort of financial advice. We don't give a buy, sell or hold recommendation. Since from talking to my users, I find that they want to use their experience and evaluate the stocks themselves. So what we do is we aggregate and synthesize all of this research. We come up with a thesis and then we let the end consumer decide for themselves whether it's a stock that they're interested in.

Speaker 2:
[54:00] Yeah, that makes sense. And then what about the business model? Is this a subscription? I know if you want those Goldman Sachs equity research reports, you got to pay a pretty penny. How much does it cost to get an Every Ticker report on a smaller microcap?

Speaker 9:
[54:14] So right now, all over 5,000 reports are completely free. Eventually, we want to transition into a subscription model. But right now, we just want to grow as fast as possible.

Speaker 2:
[54:24] That makes sense.

Speaker 3:
[54:25] Love it. How do you think about competing with LLMs over time? I'm sure people are doing this where they say, acts like a Goldman equity research analyst, analyze this company, blah, blah, blah. You can prompt your way to it right now. It'll probably get easier over time. So how are you thinking about that dynamic over time? And I guess, how do you expand the product? Because I'm assuming that's where you'd go.

Speaker 9:
[54:53] So for our competitive advantage, if you ask ChatGPT to write you an equity report on, say, Apple, it will generate you something that's decently surface level, but I would not say that it's as high quality as an analyst. So our competitive mode is that we have a agentic system that is fine tuned specifically for this task. It acts like an actual analyst. And then we have all sorts of different data feeds that are proprietary. And as for the product, this is just the starting point. We want to get really good at the specific niche of high quality research reports. And then we want to expand to adjacent parts of the research workflow in the future.

Speaker 2:
[55:31] Yeah, so is one way to think about what you've built is sort of like an agentic harness on top of the frontier LLMs so that you can deliver...

Speaker 3:
[55:41] Yeah, but it's running in the background and then just producing the report.

Speaker 2:
[55:43] Oh, and then the report's just instantly available at the click of a button, correct?

Speaker 9:
[55:48] Not even at the click of a button. We pre-generate all of the reports so you can just search a stock and it's there.

Speaker 2:
[55:53] It's already there, got it. And then you're probably putting those on some sort of cron job that runs every quarter or every month or something.

Speaker 3:
[56:00] How did you kind of stumble into this opportunity where you were you investing a bunch yourself and just kind of struggling to find great research on some of these smaller names?

Speaker 9:
[56:10] So in seventh grade I convinced my parents to open a Fidelity account for me and I just instantly fell in love with the stock market. I really liked the idea of it and I quickly got into fundamentals focused investing and learned from Warren Buffett and that type of philosophy. And over the summer of eighth grade I was looking into smaller cap stocks because I think that's where a lot of the opportunities are. There's more mispricing is there, right? And I realized that you can use LLMs to generate decently high quality research on all these smaller stocks. And so I just built a quick MVP in three days, published it, and then I found that so many people found value in it. So I just started scaling it from there.

Speaker 2:
[56:50] Have you thought about what you're going to do with the Teal Fellowship money? I know some previous Teal Fellows that have made angel investments very successfully. Do you play, is there something in your mind that wants to potentially dip your toe into the hedge fund world or actually managing a portfolio?

Speaker 3:
[57:06] YOLO into Neo clouds. Maybe.

Speaker 9:
[57:10] Yeah. I don't think I would do that. In the short term, I think it would probably just go to paying my server costs and things like that. I'll be moving to San Francisco. So also things like rent and things like that. But if there's any opportunities that I see, I would definitely be interested in that.

Speaker 2:
[57:27] Do you have a team yet? Have you built out an executive team of grizzled 60 year olds yet? I imagine that that's the next step for you.

Speaker 9:
[57:34] No, right now I think that just being a solo founder is pretty good. It allows me to move super fast.

Speaker 2:
[57:40] Yeah, that's awesome.

Speaker 3:
[57:41] That's very cool. How did you make your first dollar on the internet? Was it through Every Ticker?

Speaker 9:
[57:47] Actually, no. In fifth grade, from fifth grade to seventh grade, I built a Roblox game. It was tens of thousands of lines of code. I was spending five hours every day. It ended up being a complete flop, but I did make a bit of money from it.

Speaker 2:
[58:00] There you go.

Speaker 3:
[58:00] Amazing.

Speaker 2:
[58:01] That's great. What advice would you give to young people who want to invest in the stock market? Where should they start?

Speaker 9:
[58:09] Well, every ticker, but yeah, I think that I personally would advise fundamentals focused long-term investing. I think a lot of people get carried away by these super volatile stocks and all of these moonshot opportunities. I think it's important to stay grounded in where the real value is.

Speaker 2:
[58:30] Yeah, that makes sense. Yeah, that's interesting. What goes into a great equity research report? Can you share a little bit about differentiation? What I always find interesting personally is the research reports that can contextualize a company in its broader competitive set in the market. But what do you think makes for a good flavor, like the qualitative elements that go into something that's readable, digestible, informational, effective, and ultimately satisfactory to the user?

Speaker 9:
[59:08] Yeah, the qualitative element is one of the most important things, and it's one of my competitive advantages because I know exactly what the user wants. I'm one of my users and I talk to a lot of my users to see what they want. So the research report is, I take that into account. So as you said, you have to conceptualize it in its broader industry. And the goal of the research report is that after reading this 10-minute report, you'll be able to understand exactly where the company is, its competitive advantages, its modes, its trajectory, and you should be able to determine if it's something that you want to invest in or not.

Speaker 2:
[59:42] How do you think about the towels of LLM writing? You're absolutely right. It's not this, it's that. Those stylistic flourishes, that contrastive parallelism, that annoys tech insiders because it reads as LLM-generated and that reads as maybe subpar, but what are you actually seeing from users? Do they like that quality of writing? Are they looking for a different stylistic tone in these research reports?

Speaker 9:
[60:15] So I personally am quite annoyed by that. So I have a method of making things sound very human. So from talking to some of my users, one of their first questions is, how do you have so many research reports? And they don't even realize that it's LLMs or AI. So yeah, the language in our research reports is much more human than some of the other AI stuff you see out there.

Speaker 2:
[60:39] I love it. I love it. I won't ask you to divulge the secret, but good luck. It seems like it's going very well and congratulations on the Teal Fellowship.

Speaker 5:
[60:46] It's great to meet you.

Speaker 2:
[60:46] Great to meet you and excited to follow the journey, come back on whenever there's a milestone. We'd love to talk to you again.

Speaker 9:
[60:52] Yeah, thank you so much for your time.

Speaker 2:
[60:53] Have a great day.

Speaker 5:
[60:54] Great to have you Nick.

Speaker 2:
[60:54] We'll talk to you soon.

Speaker 5:
[60:55] Cheers.

Speaker 2:
[60:57] Here's a micro cap stock that might be making a breakout move. Spirit Airlines. They potentially are getting a pay line from the US government to the tune of $500 million in rescue funding. The money would come with equity warrants that could make government majority owner of struggling discount carrier. I know you're not a fan of Spirit Airlines, Jordi.

Speaker 10:
[61:21] Well, I can't say I'm not a...

Speaker 2:
[61:23] You've never got it?

Speaker 3:
[61:24] I've never flown Spirit.

Speaker 2:
[61:25] Oh, it's bad. It's real bad. It's real, real bad. You have to pay for everything. You have to pay for where you board and if you get a water bottle, if you can go to the bathroom, they charge you for everything. But it allows them to...

Speaker 10:
[61:38] They charge for the bathroom? Maybe.

Speaker 3:
[61:40] You get charged for water, too.

Speaker 2:
[61:41] You get charged for water.

Speaker 3:
[61:42] Unless you tell them that you need it for an emergency, then they'll give it to you for free.

Speaker 2:
[61:47] But it allows them to advertise very, very low headline prices. So if you're looking for LA to Las Vegas, you'll see something like $50. And you're like, that's impossible. And it's because you actually cannot just get through the plane. Because even for a carry-on, sometimes they'll charge, I believe. I don't know. A lot of this is just my memory from a decade ago. But the Trump administration is looking to invest as much as $500 million in Spirit Airlines to fund the discount carrier's exit from bankruptcy. The government money would take form of a senior loan with equity warrants that would come with it. Eventually, owning a majority stake in Spirit, which has struggled with high operating costs, stiff competition and surging jet fuel costs amid the Iran War. Well, we will follow that story more in the future. But we have our next guest returning to the show, Ishan Gupta from Juicebox. Welcome to the show. How are you doing?

Speaker 7:
[62:41] Hello.

Speaker 3:
[62:41] I'm doing awesome.

Speaker 2:
[62:42] Thank you so much for taking the time.

Speaker 3:
[62:46] Great to have you.

Speaker 2:
[62:47] Introduce yourself in the company.

Speaker 7:
[62:49] Yeah.

Speaker 8:
[62:50] So I'm Ishan. I'm one of the co-founders of Juicebox. We are an AI recruiting platform and we help companies find and engage talent using LLMs. We build agents that will go out there, find the right people for every role you're trying to fill and get them into your process.

Speaker 2:
[63:02] Okay. We have a friend who's an executive at a big tech company. He's been getting absolutely spammed with new outreach from what appears to be a new Gmail account that has been created. And what's so remarkable about this cold outreach is that it even nails the sort of like cringe footer that says like, you know, do not print this for the environment's sake and like, you know, a little quote and like, it doesn't feel just like a prompt box shot in there, there's a huge variety. But it hasn't been successful. He's annoyed by it. So how do you think about actually targeting the outreach so that you're not annoying potential new hires?

Speaker 8:
[63:50] Of course, yeah. I think the main problem is that the exact side is almost like the extreme side of it because they're getting like so much outreach anytime an exec is available and everyone's kind of spamming them. The thing is that for the average role, the best way to guarantee that you're reaching out to the right people is to spend more time on the search. What ends up happening in organizations today is that there's a hiring manager who really understands what they're looking for. And then there's sort of these pattern matching that happens afterwards where talent teams are basically relying on these really hard filters like, oh, I'm going to reach out to every software engineer at Google. That's sort of smart strategy, right? What LLMs are able to do is to truly understand.

Speaker 3:
[64:25] Are you interesting?

Speaker 8:
[64:29] It's like LLMs are able to truly understand what makes someone successful in a role and try to actually find people who are good at that. And instead of just pattern matching based on what company you're at or what job title you're at, we'll actually go in there. We'll analyze your real work product. We'll see what you're doing on GitHub, what you're doing on different platforms, what companies you've been at, and develop a more deeper understanding of who's going to be a fit and then reach out to the right person. So that in fact reduces spam.

Speaker 2:
[64:54] Yeah. Let's flip it around to the person that's looking for a job. They want to get spammed with great offers. What should they be doing? Because there's a lot of software engineers where they can't contribute publicly as they work for an important company. There's a lot of folks who their work happens behind the scenes. And maybe they are sharing it on LinkedIn. But can your scrapers reach into LinkedIn? Should they have a personal website? And then put that in some sort of robots.txt and web crawler filter so that they get indexed properly. Like, how can job seekers show up in results more aggressively at a time when more people are looking for jobs more aggressively than ever?

Speaker 8:
[65:36] Essentially, put your work product out there. Talk about what you're working on. Putting out blogs, putting out open source contributions, putting out any projects that you have worked on. Those things help. Because then that information gets...

Speaker 3:
[65:45] Would three hours of live daily content help?

Speaker 8:
[65:50] I'm sure. I mean, it helps with an acquisition.

Speaker 3:
[65:54] That's true.

Speaker 2:
[65:55] You just talk about AI comms for three hours a day, and yeah, you turn some, raise some eyebrows. Anyway, yeah, what about... So are you purely focused on technical roles because you keep coming back to GitHub contributions, breaking down example work products, and I'm just wondering for someone who's in more of a knowledge role, or they work in PR or marketing or biz dev or finance, should they be educating people? Should they be explaining their strategies? Like if they're not a natural writer or they don't just have a piece of like an obvious side project that they can open source, how can they show up to the AI recruiters of the future, your company included?

Speaker 8:
[66:40] Yeah, of course. So we started out more tech focused because that's the industry we understood the most, and we started out working with tech startups, and we started working with companies like Ramp, Scale, all of these kinds of companies. Then eventually, what has happened in the last year is we've grown way, way past that. So now tech does not represent the majority of our customer base. We have more than 5,000 customers, and a lot of them are larger enterprises, and they're looking for all sorts of people. So what we are able to do is we're able to join profile data and your experiences on all the company data that we have. So essentially, if you put a little bit of information about yourself on different platforms, we're able to take that information, enrich that, and do a deeper research on every single company we're in at, every single skill you've had, everything you've worked on, and build a better understanding of exactly what your area of expertise is, and then make better matches using that. And it applies across industries. So for example, we have people looking for traveling nurses in the Midwest. Very different type of role. But what we can do is we can look up profiles. There are registries available for nurses on the Internet as well. We'll look up those profiles, we'll understand the different places you've worked at, we'll try to build a good understanding of what these different environments are, what kind of people do these other companies hire, where you've been at in the past and then make a good inference on whether or not they are fit for the role you're looking for.

Speaker 2:
[67:56] Okay.

Speaker 3:
[67:57] Why do you think there hasn't been a big, like a dekakorn scale outcome in recruiting tech today?

Speaker 2:
[68:06] What is indeed?

Speaker 8:
[68:07] I would actually argue against that. LinkedIn is a pretty-

Speaker 2:
[68:10] LinkedIn is- Oh, got it. Mog.

Speaker 8:
[68:12] Indeed is a pretty massive outcome. Yeah, recruiting holdings?

Speaker 2:
[68:15] 70 billion enterprise value, bro. Come on.

Speaker 3:
[68:20] But that's like kind of where I'm going. Isn't that like a holding company of like staffing and recruiting firms?

Speaker 2:
[68:26] Yeah.

Speaker 3:
[68:27] So more like labor-intensive, but I can imagine-

Speaker 2:
[68:31] It's not the database market. Let's just be-

Speaker 3:
[68:32] Yeah.

Speaker 2:
[68:33] It's not like, oh, there's a snowflake and a-

Speaker 3:
[68:35] It feels like you could have, you could have, like if you execute properly, there could be an outcome closer to that holding company that you just mentioned than some of these other kind of recruiting tech platform. Yeah.

Speaker 2:
[68:48] I guess like where's the source of economic power and lock-in and scale that would allow for a really, really broad outcome here as opposed to, okay, there's a whole bunch of AI recruiting firms and companies sort of bid them all down and there's not insanely high margins good businesses, but not the $100 billion outcome. Have you been thinking through that? I know it's early, but have you been thinking through like, what does this look like at mega scale?

Speaker 8:
[69:15] Of course, yeah. We've thought a lot about that. The thing is that the main value in the recruiting industry always accrues in the services layer. So far, it has never accrued as much in the software layer. Because it's really actually the work that goes into finding people. It's the ability to search for the right person and get them into any role. That is what people pay for. And so far, there has been no way of actually automating that because you cannot do that. It requires judgment. It requires actually understanding what someone's working on or what their capabilities are. That is exactly what agents are able to do. So this is really the first time when you have a massive opportunity in a market that's incredibly important, and you're able to build that with LLMs, because it's just not been possible until now. And most companies have traditionally in recruiting tried to target this like software layer, which is like, hey, we'll build another CRM, we'll build another ATS and all of these different platforms. But the main value always accrues in being able to find the right person. That is what someone wants to pay for. And that is exactly what we're able to do with LLMs.

Speaker 2:
[70:12] Very cool.

Speaker 3:
[70:13] Yeah, I remember first discovering that just everyday recruiting services companies could be big. Because, hey, there's a public company called Hays, hays.com.

Speaker 2:
[70:24] No way. Really?

Speaker 3:
[70:25] It's just a public recruiting, like, staffing firm. Wow.

Speaker 2:
[70:29] Yeah, good business.

Speaker 3:
[70:30] That's a very, very big opportunity for you guys.

Speaker 2:
[70:32] Congratulations on the Thiel Fellowship. Thank you so much for coming on and breaking it down for us. And hope to see you soon. I'm sure there's a lot of milestones in the near future.

Speaker 3:
[70:40] Yeah, great stuff.

Speaker 2:
[70:40] Have a great rest of your day.

Speaker 8:
[70:42] Thank you so much for having me. It was awesome.

Speaker 2:
[70:43] Goodbye.

Speaker 3:
[70:44] Cheers.

Speaker 2:
[70:45] Up next, we are running one minute behind, but we're going to catch up because we have, we have Derpetual next, building infrastructure to enable derivatives trading on any asset. Interesting. Okay, Derpetual, Antoni Kiszka.

Speaker 3:
[71:02] Before we go there, let's pull up this post from none other.

Speaker 2:
[71:08] Then who?

Speaker 3:
[71:09] Then Ben Horowitz, not Ben Horowitz.

Speaker 2:
[71:12] Yes.

Speaker 3:
[71:13] Ben Horowitz.

Speaker 2:
[71:14] With the launch of the day.

Speaker 3:
[71:15] The anti-grammarly.

Speaker 2:
[71:17] The anti-grammarly.

Speaker 3:
[71:17] Mess up your emails with AI.

Speaker 2:
[71:19] I love this.

Speaker 3:
[71:19] If you're worried about people accusing you of using AI to write your emails, use sincerely.

Speaker 2:
[71:26] Yes, misspelled.

Speaker 3:
[71:28] To use more AI to add misspelling.

Speaker 2:
[71:31] Yes. It condenses down that long, boring email that you were about to send that says, to whom it may concern, I wanted to reach out to express my interest in connecting regarding potential synergies between organizations, and it just dumbs it down to wanted to reach out. Let's talk. This is a newsflash. This is how people actually email and communicate when it's person-to-person, and I think everyone would be better off with a little bit tighter communication methods, especially in the age of AI. So this seems like a joke or a drop, but I would imagine this being a good product, and I think people might actually pay for this. I think there are so many times.

Speaker 3:
[72:13] I think it's actually a real product.

Speaker 2:
[72:15] I hope so.

Speaker 3:
[72:16] They're going to charge five bucks a month.

Speaker 2:
[72:17] Alex Lieberman said such a good viral drop idea, but I would not expect this to go away. I think this is a business, and I think this will be successful, and I think this is a valuable tool. It's something that would be hard for Google to justify baking into Gmail. It is a little bit counter-positioned against.

Speaker 3:
[72:40] Yeah, it's a tough pitch.

Speaker 2:
[72:41] It's a tough pitch to be.

Speaker 3:
[72:42] We want to help our users get more typos.

Speaker 2:
[72:44] It might be a good April Fool's Day joke that they could bring in and then leave it around.

Speaker 3:
[72:48] Not anymore, because Ben did it first.

Speaker 2:
[72:49] Ben did it first now, so good luck to Ben, and sincerely, we'll check in with him later. But we have our next guest from Derpetual in the waiting room that's bringing him in to the TBPN UltraDrop. How are you?

Speaker 6:
[73:01] Welcome to the show. Hey guys, it's great to be here.

Speaker 2:
[73:02] Thank you so much for helping on. Please introduce yourself and the company.

Speaker 6:
[73:07] I'm Antoni, and I'm building a company called Derpetual, and we want to create the derivatives exchange for everything. So basically, we created a new kind of derivative, and we want to use this type of derivative to bring derivatives to assets beyond the top 1,000 that exist now.

Speaker 2:
[73:25] Okay. Is there any crypto involved?

Speaker 6:
[73:29] Currently, yes. Our protocol is working on meme coins, so you can trade meme coins with longs and shorts. But it's honestly a great way for us to test everything. We have the craziest markets in the world and the craziest people in the world doing stuff on our tech, so we get to improve and check if everything is fine.

Speaker 2:
[73:55] Yes. Problem. Meme coins weren't crazy enough, but we made them crazier. You're welcome.

Speaker 6:
[73:59] Yes.

Speaker 2:
[74:02] Is there a plan to bring this to real world assets, prediction markets? We've seen the different financial approaches to so many different categories. How do you think this actually crosses the chasm because we went through the meme coin boom and they did go mainstream, but I think a lot of people just didn't really find lasting value there necessarily, and they moved on and it's become a bifurcated market. I'm wondering, is there a business to be built just in the meme coin community, or do you want this to go broader and wind up on Wall Street or wind up with retail investors? How do you think about the long-term plan?

Speaker 6:
[74:44] We definitely want to go to traditional violence and end up on Wall Street. In the end, the problem with meme coins and with most things in crypto is that they are not useful. So, of course, there is a market for them, but that market is going to be a thousand, ten thousand times smaller than something that actually creates value instead of just being a casino.

Speaker 2:
[75:05] Yeah, that makes a lot of sense. How do you think about the current strategy that we've seen Kalshi roll through around CFTC and futures contracts? It feels like there's been a major acceleration in at least the, I don't even know, it's like the approval process for these new futures contracts. I know that for every new game, if the Super Bowl is happening, they have to get that contract approved. And it feels like with their technology...

Speaker 3:
[75:42] Yeah, I think there's like a 24-hour cycle.

Speaker 2:
[75:44] Yeah, there's like a 24-hour cycle. And so that restricts some of the things they're doing, but it still feels like the fastest that the financial sector has ever moved. Is that an interesting pathway for you, or do you think that there's an entirely different way to attack the problem?

Speaker 6:
[76:01] Definitely. I think Kalshi has been doing it right. And in the end, their strategy was pretty unique. You probably know they started their company, and for a couple of years, they didn't do anything and just wrote letters to the CFTC about how they please want to be regulated. And in the end, that worked out, and no one else is even close. So they are definitely something that we want to be learning from.

Speaker 2:
[76:35] Is there a North Star asset that needs a derivative that would resonate with a broader everyday audience? Because the derivatives on Mean Coins, that seems like there's a community for that, but that's probably not something that's super aspirational for the everyday American. How are you thinking about the actual application? The one that I always go to with derivatives like weather futures for farmers, that's a very practical or hedging the price of fertilizer. These things are financialization, but they're clearly beneficial if you're running a farm. So that's somewhat relatable to people. Do you have other derivatives that you think need to be unlocked in the near future in the real world?

Speaker 6:
[77:26] Definitely. I think the inspiration is something that we probably want to have as one of our first assets in traditional finance is hydrogen. Because there is no hydrogen derivatives anywhere whatsoever. And it's a very popular asset for the industry, for energy, for things like that. But it's just to spread out. The production is to be centralized, so you can't just create one hydrogen index and have it be traded. So I think this decentralization makes it impossible for traditional markets to bring liquidity to that.

Speaker 3:
[78:06] What were you doing before this?

Speaker 6:
[78:09] A lot of things. I was doing trading terminal for crypto. I was doing a software house, and that was my first company, and I was also making money selling Minecraft items.

Speaker 3:
[78:26] There you go. A lot of in-game item sellers so far in this fellowship.

Speaker 2:
[78:32] Any chance that we will see Onion futures traded in the future?

Speaker 6:
[78:41] My fellowship money might be deployed to Lobbing for finally getting Onion futures.

Speaker 2:
[78:48] Onion futures. There's many single issue voters out there for bringing back Onion futures. They were banned famously, very obscure piece of financial regulation.

Speaker 6:
[78:59] Yeah. I think in general, farmers in the US have a lot of political power, and especially Onion farmers, they really want their futures.

Speaker 2:
[79:07] They do. Well, hopefully, it happens. Thank you so much for coming on the show.

Speaker 5:
[79:11] Yeah.

Speaker 6:
[79:11] Great to meet you.

Speaker 2:
[79:12] Congratulations. We'll talk to you soon.

Speaker 5:
[79:13] Cheers.

Speaker 2:
[79:14] Goodbye.

Speaker 5:
[79:14] Cheers.

Speaker 2:
[79:15] While we bring in our next guest, we were in Cultured Magazine, culturedmag.com. What happens when OpenAI buys your tech podcast? Ask the boys at TBPN. We did an interview with them and we had some fun with them.

Speaker 3:
[79:28] Yeah, we had Sarah on the show.

Speaker 2:
[79:29] Yeah, we have. So the first question that they asked us was, what's one work of art that got you through an important moment in your life? I said Godfrey Reggio's 1982 masterpiece, Coyonis Cotsi. Jordi, what did you say?

Speaker 3:
[79:45] Borat. Borat.

Speaker 2:
[79:47] I think that's good. I think that's good. They asked us, what keeps us up at night? We said data centers being even slightly delayed. That would be terrible. I lose sleep thinking about it. A single supercomputer delayed by even minutes. Even minutes.

Speaker 3:
[80:04] We missed opportunity to say supercomputer.

Speaker 2:
[80:07] They did ask us. We had some fun with some questions. We told some serious answers. We mixed it up. They asked us, what do you think your biggest contribution to culture has been? This is the question of culture. What do we say?

Speaker 3:
[80:19] Making tech people put on a suit.

Speaker 2:
[80:21] We've seen a few suits on this TALE Fellowship gigastream. We've seen some suits creep into the tech podcasting world. Job's not finished, but we're making progress. We'll go back to our TALE Fellowship gigastream with Milan from Opt32, developing compute infrastructure from compilers to chips for physical autonomy and local intelligence. Welcome to the show. How are you?

Speaker 7:
[80:46] Good. How are you guys doing? Great to be here.

Speaker 2:
[80:48] Thank you so much for hopping on. Please introduce yourself and the company.

Speaker 7:
[80:52] Yeah, totally. So I'm Milan. I'm one of the co-founders of Opt32. And at Opt32, what we're trying to do is essentially build modern full-stack compute infrastructure for physical autonomy. So more concretely, like you said, that's everything from software, so compilers, down to chips, custom accelerators, to run on-device machine learning in things like robots, drones, cars, autonomous defense systems, other autonomous vehicles.

Speaker 2:
[81:17] That seems extremely, extremely, extremely broad. Are you going to narrow it down? Is there a beachhead that will happen where you will find a particular niche, or is going broad a piece of the...

Speaker 3:
[81:30] He's like, is that the strategy? The idea to pick a small niche market and dominate it, I think it's wrong.

Speaker 2:
[81:36] Maybe, maybe.

Speaker 3:
[81:36] We're gonna dominate everything all at once.

Speaker 2:
[81:38] I'm super curious.

Speaker 7:
[81:39] Yeah, so I guess a couple things to touch on there. So the first is that, you know, the nice thing about building infrastructure and compute infrastructure in particular, is that it does generalize fairly well to many different applications. So you can use the same chip and the same software to run ML in a robot as you could in an autonomous vehicle. I do think we are, you know, taking a sort of an entry approach into the market, particularly from the robotic side. And from, you know, our side, we are starting at the software layer and then gradually working our way down the stack towards hardware.

Speaker 2:
[82:10] Okay. And in terms of, so if you're starting with the software layer, walk me through the infrastructure stack that you might be buying off the shelf. Are you buying, like, Nvidia GPUs? I know that, like, when I think of, like, previous autonomy stacks, I think of, like, Tesla being very vertically integrated, and then Nvidia and a few other partners sort of being, like, Mobileye, sort of going around to the rest of the automakers and the OEMs and sort of plugging in with a little bit more flexibility. But those companies that provide that stack aren't fully integrated. So I see the opportunity, but I'm curious about how you're solving the short term where, even if you wanted to design custom silicon today, it's probably not going to ship to you in a year, right?

Speaker 7:
[82:56] Yeah, certainly. So what we do right now is we essentially build a fully automated model optimization platform, right? So say you're a robotics company, you're deploying some perception model on your robots. You very often need extremely low latency. You have compute constraints, and you can't just toss a massive server scale GPU on a robot that runs off, say, a battery. So what we do is we work with these companies to get their models running faster, or for them to be able to fit more intelligent models on cheaper hardware, stuff like that. And right now, our software layer is pretty much hardware agnostic, so it could, in theory, run an NVIDIA GPU or even like a, I don't know, Qualcomm accelerator or something like that, but primarily we do target NVIDIA GPUs as our backend. Okay.

Speaker 2:
[83:42] Yeah, where, what, are you excited about like robots that are too small to house an NVIDIA GPU? I'm thinking of like the Matic robotic sweeper or the Roomba. Like, and I've been super excited about the potential, like the Roomba had such wide deployment. It really did break through across the chasm in terms of like robotics in the home. And I'm wondering if you're excited about, or optimistic about sort of a shorter timeline to those more incremental steps to robotics, robotic deployment versus, you know, we've heard a lot of pitches for like the straight shot to humanoid AGI and I think that's coming, but is there a wall crawl run that, you know, technology sector will do here?

Speaker 7:
[84:34] Yeah, I definitely agree with you there. I think we're very excited about sort of the gradual deployment of robots and autonomy into our everyday lives over the next few years. I do very strongly agree that as we get maybe, you know, jumping three years into the future, we'll see a lot more autonomy in consumer life. So, you know, that could be something like a Roomba, could be something like a cooking robot, could be civic duties like street cleaning, as well as another area I'm particularly excited about is autonomy and manufacturing. I think it can do a great job at sort of augmenting human workers and helping to sort of bridge this gap we're seeing where there's not necessarily enough skilled, say, like welders or something to fulfill all the manufacturing wants.

Speaker 2:
[85:19] Is there enough data in manufacturing to do anything at scale with machine learning or is there enough, you know, transfer learning from the big models?

Speaker 7:
[85:30] Today, maybe not. Over the next few years, I would hope so. We don't concern ourselves with the model layer. We try to be everything after that. So we don't build our own models. We don't train models. We work on running models, essentially. Sure. Sure.

Speaker 2:
[85:43] Jordi?

Speaker 3:
[85:43] Very cool. What were you doing before this?

Speaker 7:
[85:46] I was a freshman at Harvard studying CS and philosophy.

Speaker 2:
[85:50] That's cool.

Speaker 3:
[85:51] Why did you get into Harvard other than you seem like a smart guy?

Speaker 5:
[85:58] I have no idea.

Speaker 7:
[85:59] I don't know. I spent most of high school doing CS research. I started writing compilers back when I was a freshman. I worked in a bunch of different university research labs, all on compilers, computer architecture, programming languages, apply to and have that acceleration. Yeah.

Speaker 2:
[86:15] What do you make of the decline in computer science as a major across universities? It feels like there's a... I've seen some stats that show a pretty steep drop-off, and yet...

Speaker 3:
[86:27] He's contributing to it, John.

Speaker 2:
[86:29] No, no, no. I mean, you're still CS major, effectively. But it feels like the route that... The fear around don't major in CS is very much like don't major in CS and then try and go get a front-end engineering job that's just writing code. But if you're majoring in CS, that still might be the best path into working in robotics, working on infrastructure, doing a lot of different things. So how have you processed the value? Like do you feel like the CS that you've learned is less relevant today?

Speaker 7:
[87:04] Yeah, that's what I was going to say. I think there's sort of going to need to be a reallocation, so to speak, of talent and focus within the field of CS, where as we see artificial intelligence capabilities advance, we see a move up layers of abstraction to where skills like system architecting, sort of higher level theory, are going to become more and more important. And kind of these very low level implementation details, like you said, being like a front-end developer, doesn't provide as much value. So I don't necessarily think CS as a whole is dropping off, so much as it's reallocating towards those higher levels of abstraction.

Speaker 2:
[87:36] Yeah. How big is the team? How are you thinking about the capital intensivity of this business? When I hear custom silicon, I'm hearing hundreds of millions of dollars to do really anything interesting. We've had some previous Teal Fellows from Etched on the show, or I've talked to them and done podcasts with them. It feels like this can be extremely capital intensive, or you can find partnerships and do something that's much more on the actual software and infrastructure side that might be less capital intensive. But how are you thinking about it?

Speaker 7:
[88:12] Yeah. Right now, the team is just the three of us, co-founders, me and my two high school best friends, all technical. We are piring and hoping to grow pretty quickly. In terms of capital intensivity, the really great thing about building full-stack compute infrastructure is that we can actually sell our software in isolation of the hardware. We are already doing that and working with some new design partners. We can get revenue, we can do that quickly. Software development costs are relatively cheap. The cost of operating our software is essentially zero. Any revenue from that is pretty much pure profit. In terms of hardware development, I think there's a great path towards gradually moving towards a full production run of a custom ASIC. In particular, what we are going to do is build single board computers, so PCVs, around some existing accelerator chips and build the entire software layer there. Then we will move on to maybe implementing some chip architectures and FPGAs. Then from there, you could do a smaller, not super advanced process node, kind of prototype tape out run. Then say one, two years down the line, once we've raised a lot more capital, that's when you go for the big advanced process node, full production tape out.

Speaker 2:
[89:17] Yeah.

Speaker 3:
[89:17] Are you working with Victor and Cavalla yet? The chat is asking.

Speaker 2:
[89:22] Or are you competing in bitter rivals?

Speaker 3:
[89:25] No, no, no. I think we will. They'd be natural partners in some degree.

Speaker 7:
[89:29] Yes. I think we'll be definitely working with Cavalla at some point in the future. Right now, we're mostly focused on internal technical work, but shortly.

Speaker 2:
[89:42] What is the, especially in the manufacturing sector, is there not a lot of energy being devoted to not tell operation, but just putting compute, like, I guess like a thin client for robotics would be the term, like you have a server room with a NVL-72 rack on site, and then you're doing inference in the IT cabinet or closet, or the local data center, effectively, and then your robot can just be much lighter and has like barely, it just has a camera on that's just feeding, and like, yes, there's maybe a little bit of latency, but you're talking about, you know, the speed of light across high bandwidth Wi-Fi. It seems like that might be an interesting solution. Like, is that happening? Is that one of many strategies, or is that a dead end?

Speaker 7:
[90:33] Yeah, I think it's a split, right? I think there are definitely use cases where you can do that. And there's also other use cases where you might have some sort of network constraint or some sort of extreme latency constraint where you can't. And, you know, some things that we work on are actually like splitting the workload across both a cloud server call and maybe something that's more latency sensitive can run on device.

Speaker 2:
[90:53] Yeah, that makes a ton of sense.

Speaker 3:
[90:55] Have you guys raised money already outside of obviously, you know, fellowship is, fellowship is for you. But, what about the company?

Speaker 7:
[91:02] Yeah, we raised our seed round about two months ago. We raised five million co-led by Box Group and by Venture.

Speaker 3:
[91:09] Box, petitionator, petitionator would get into this one. We love, we love him, great, great pickup. Man, great to meet you. I'm sure we'll be back on soon.

Speaker 7:
[91:21] Yeah, great to meet you guys as well.

Speaker 2:
[91:22] Well, thanks so much for coming on the show. We'll talk to you soon. Have a good day.

Speaker 7:
[91:25] Thank you.

Speaker 2:
[91:26] Goodbye. Up next, we have Galen Mead from Standard Intelligence, focused on building a line.

Speaker 3:
[91:33] Not to be confused with non-standard super intelligence.

Speaker 2:
[91:37] Non-standard unintelligence. You don't want to be unintelligent. Building Aligned General Learners, Galen Mead joins us on the show. And I believe Galen is in the waiting room, so we will bring him in to the TBPN Ultradome as soon as he's ready. Galen, how are you doing?

Speaker 5:
[91:58] Hey, doing well.

Speaker 2:
[92:00] Reintroduce yourself and the company.

Speaker 5:
[92:04] Yeah, we build computer use models, the general research company towards Aligned AGI. I dropped out around three years ago. Didn't spend much time in university. And I just wanted to do research full time.

Speaker 2:
[92:21] Did you jump straight to computer use or was there something in the intervening years?

Speaker 5:
[92:27] We, at the time that we were started, there wasn't really this notion of a neolab. So we, from the start, wanted to be a general research company. But initially we started with some work on audio models that I had continued from SF compute.

Speaker 2:
[92:46] Interesting.

Speaker 5:
[92:47] Just because you train some state of the art model and people see that you can train models and you can go from there. I think this is maybe one of the worst decisions. Not worst, but worse.

Speaker 2:
[93:00] Yeah, sure.

Speaker 5:
[93:02] Strategically. But computer use models are nice because it's a really general form factor.

Speaker 2:
[93:07] Why?

Speaker 5:
[93:08] It always wanted to have a space for taking actions.

Speaker 2:
[93:12] Yeah, why is computer use important? I think there was maybe like a one week period where everyone was super everything will be a CLI, and there's maybe like a resurgence in computer use. But what uniquely captivates you about computer use as a goal here?

Speaker 5:
[93:32] So it's a training recipe that I had always wanted to do. You can see in some of the early work on games from DeepMind, there's this notion of training a policy on tons and tons of supervised data from diverse environments and getting something that is like a base model for actions. And it occurs to us that this exists for real world work in the form of like screen recordings with a computer being the universal actuator that humans use to interact in very diverse environments. So this real opportunity to pre-train base models for being agents and this is a very appealing concept from a research perspective if you want to make very competent agents.

Speaker 2:
[94:26] Do you think, how far out do you think we are from using like a truly complex piece of software like Premiere Pro or Cinema 4D or AutoCAD, any of these tools? It feels like we're on the cusp. Is that this year? Are you seeing glimmers of, OK, the average Premiere Pro user will probably be interacting with it through a prompt pretty quickly in the same way that software engineers migrated from the IDE to the CLI and the text box pretty quickly?

Speaker 5:
[95:04] Yeah, I think it'll be this year. I, back in middle school, did freelance animation. And I had to bring that out recently for our model launch a bit over a month ago to put together Dharma videos. And it honestly felt like Stone Ages compared to working with the AI-assisted code editors for my day-to-day job. So I think we'll get there by the end of this year. And I think that will be a pretty big step change for all of these industries that aren't used to working with Copilot.

Speaker 2:
[95:42] Yeah. Is there any difficulty with the lack of open-source tools in other computer use areas? Because it feels like part of the reason potentially for the acceleration in coding is that the IDEs were open-source and also the programming language is open-source and GitHub is this rich repository. But will there be this battle fighting between some proprietary piece of software that doesn't want to be used by an agent and then you're trying to figure out how you can do it? Have you grappled with any of that?

Speaker 5:
[96:23] So I think we're somewhat unique in that we are rather obstinate about hitting the input and output of a human exactly. So we take video inputs. We output mouse state deltas and character level keystrokes. And nobody's copyrighted the form factor of a screen and a keyboard. So if we interact on that level, it's fully general.

Speaker 2:
[96:48] Yeah. How do you want to instantiate this with a customer?

Speaker 5:
[96:56] We're pretty agnostic on that right now. There's a lot of ways to go on the model capabilities. We can get pretty crisp signal. I like there's a lot of things where we should be able to tell the model to do something and it should do it.

Speaker 2:
[97:08] Yeah. And once we're at that point, it's just download the app and start paying probably.

Speaker 3:
[97:13] A little bit of a tangent, but how have you processed the SaaSpocalypse? There are still some SaaS bowls and they'll tell you, we might even have more seats because we're going to have agents that are using computers and using existing software just like a human would. But how have you processed it overall?

Speaker 5:
[97:33] I mean, I have one friend's doing a search product to print it nicely, that there's an opportunity for a lot more usage if you have tons and tons more agents actively using these products. I haven't been following this stuff very closely. I'm pretty off the Internet.

Speaker 3:
[97:52] Locked in. Love it.

Speaker 2:
[97:54] Did we get how you made your first dollar? That's always-

Speaker 3:
[97:57] Yeah, was that doing Blender?

Speaker 5:
[97:58] Freelance animation.

Speaker 2:
[97:59] Freelance animation.

Speaker 5:
[98:00] Basically, artist.

Speaker 2:
[98:03] What was the actual project? Is it like marketing videos or?

Speaker 5:
[98:08] It was a bunch of like forum graphics stuff. It was a very niche space, yeah.

Speaker 2:
[98:14] That's cool. Well, thank you so much for taking the time to come share with us.

Speaker 3:
[98:17] Great to get the update. Come back on for your next big launch.

Speaker 2:
[98:19] Yeah, congrats on the progress. We'll talk to you soon.

Speaker 3:
[98:22] Cheers.

Speaker 2:
[98:23] Back to the Culture Magazine article. They asked us what grounds you and what invigorates you. We said the daily schedule grounds us. The show goes live every weekday at 11. So there's no room for projects to drift or spiral. Booking a fascinating guest on the same day. Everyone is discovering them is incredibly energizing. That's the Soham Perique moment in a nutshell.

Speaker 3:
[98:46] They asked what would you wear to meet your greatest enemy?

Speaker 2:
[98:49] What did we say?

Speaker 3:
[98:49] A TBPN racing jacket covered in logos. So at least we're getting ad impressions for our sponsors.

Speaker 2:
[98:56] That's great. Yeah, we had a lot of fun. What are we looking forward to this year? Self-improving AI systems. That's on brand. Anyway, we have our next guest from Swoop, building a super app for Africa, starting with food delivery in the waiting room. Let's bring in Aubrey from Swoop. Aubrey, how are you doing? Welcome to the show.

Speaker 11:
[99:16] Hey, I'm doing well. Yeah.

Speaker 2:
[99:19] Please introduce yourself and the company.

Speaker 11:
[99:22] So yeah, we're building a super app for Africa. So I live in Nigeria now. I started my first company in Africa when I was 15. So I got really into geography growing up. I had this first company, it was a recruitment company in Southern Africa. And I was spending all my summer and winter breaks in high school with the Africa business company. And I realized that all the biggest opportunities basically, you had to live in Africa. All these export industries were way more competitive. Anything where you had to live in Africa was a lot bigger opportunity. So I was figuring out, what is the biggest opportunity in African tech? I knew I wanted to move to Africa, build a tech company, landed on this idea of building a super app. Biggest opportunity has been tech, but I felt like doing a super app is going to be the way that you can get there, how you can build a multi-billion dollar consumer business in Africa.

Speaker 2:
[100:18] Let's start with the state of food delivery in Africa. Are the major players there in a minor capacity? Are they failing or is it a wide open blue ocean market?

Speaker 11:
[100:29] Yeah, it definitely depends on the country. For instance, in Africa, Uber Eats is doing quite well, but Nigeria is our big market right now. Nigeria, where we just launched, we have two major competitors. We have Glovo, it's an international competitor, and you have Chowdeck, which is actually the market leader. It's a local company, but Nigeria is very early stage. If you look at the ratio of GDP to how much GMB food delivery is doing, Nigeria is five or seven times behind a lot of comparable countries, and it's still growing more than 150% of a year. So it's still very early days. What you see is, in food delivery in particular, once that kind of service becomes available, the culture adapts and it can continue to grow for a really long time.

Speaker 3:
[101:19] What do you think the other players are doing poorly? Where is the opportunity to differentiate?

Speaker 11:
[101:25] Yeah, I mean, the biggest thing is price. Prices are really expensive in the market. We're a lot cheaper. And then, I think, obviously, to have a lower price, you have to have a vision for how you can keep those prices low. So that's when we get to this super app idea. If we can convert a significant percentage of our food delivery customers over to being payments customers for us, that's something that's really exciting because payments are extremely large. There's 18 trillion in digital payments being sent in Africa every year, 2.6 trillion in just Nigeria. If you can capture a significant percentage of that and you can be converting each food delivery customer, maybe one in five of them, starts using your peer-to-peer payments product, sending money to other users, that becomes not only a really effective way to acquire customers for something that's very high margin, but also it allows you to keep the food delivery service cheaper because you don't necessarily need to make your revenue from it. Whereas a company like Guvo, they're international, they only run a food delivery service in Europe. It's too late to build a super app in Europe, so they're not set up to build a super app. They'll never launch a payment service in Nigeria because that's not their business, but it is ours, so we do have an advantage there.

Speaker 2:
[102:46] What does the North Star super app look like these days? What is the canonical example of success?

Speaker 3:
[102:52] It's still WeChat.

Speaker 11:
[102:53] I'd say WeChat is definitely big. I think one company that I get really excited about in terms of comparison is Caspi. They're not only the biggest e-commerce company in Kazakhstan, but they're also one of the biggest payments companies, one of the biggest banks in Kazakhstan. They do e-commerce, they ride a show, they do a bunch of verticals. I think it's a really exciting business, and Kazakhstan's not that much of a bigger economy than Nigeria. It's about the same size, but Caspi is doing 2.4 billion in profit every year in just Kazakhstan. I think that you could do that in a number of African countries with the state of what the competition is. I think it really is early days, these verticals are growing very fast. The competition is not established to the point that it's impossible to win. In a lot of these verticals and in a smaller market, you have this phenomenon where the gains to scaling become greater than the gains to specialization. Whereas in a bigger market, you might need to specialize because the fixed costs are relevant relative to the potential size of the business.

Speaker 2:
[103:58] How big of a deal is the Belt and Road Initiative in Africa these days? We were reading about some potential bailouts for some loans from the IMF, being discussed at a very high level. But is the Belt and Road Initiative a big deal from being actually there? What's your perception of it?

Speaker 11:
[104:20] I wouldn't say it's maybe top two or three things that we're thinking about in terms of what's going to shape a country. But obviously, it's very real. There's a lot of infrastructure that's going on that's being built. Actually, one of the countries that we work in, our first country, actually, Aswatini, you might know it as Swaziland, that was the old name. In Swaziland, that's the only country that hasn't taken Belt and Road money. And it's interesting to see because they actually recognize Taiwan, so China will not do Belt and Road. There's 53 African countries that do get Belt and Road funding. Swaziland is the one that doesn't. But there is a difference there in who's really roads, who's building different infrastructure. But I don't think it's one of the top three things that I'm looking at as something that's going to change development in Africa.

Speaker 2:
[105:10] Yeah. How do you think about geographical expansion? How local do you want to be focused from a staffing perspective? Do you imagine having satellite offices internationally? Or do you want to centralize everything? What are the trade-offs there?

Speaker 11:
[105:26] In terms of recruitment, first of all, our team is primarily across Nigeria and India. So I decided that we would go international with the software development function in particular to be able to capitalize on just all the markets. My first was a recruiting company. That's what I felt was the right thing to do. And I think we've been able to bring on a great team that way. That being said, the majority of our team is in Nigeria. And going forward, I expect to expand primarily in Nigeria let's say the next year. And then from there, we'll be looking at primarily launching other African countries and hiring there. So the vast majority of staff is in Africa. There's amazing talent in African countries, particularly in Nigeria. And if you are able to identify that, if you have a good process, I think there's amazing talent there. In terms of geographical expansion for the business itself, definitely excited about a couple of other African markets and we just need to be able to get to a point where launching another country would not hinder our business in Nigeria, which that's not where we are today. Obviously, we just started in Nigeria, started marketing actually last week. So we're not at that stage yet, but I would love to be at that stage where we could start launching other African countries.

Speaker 2:
[106:51] What did you learn from the recruiting company? What was the story of that business?

Speaker 11:
[106:56] Yeah, I think the most important thing I learned is how to recruit. The second most important thing I learned is how to operate a business in Africa, what it's like to live in Africa, what are the challenges, what are the biggest opportunities. I learned a lot about what the opportunities were. But I think number one most important thing is how to recruit. I feel that most companies in the world, but maybe particularly in Africa are not meritocratic with hiring. If you are, if you could go out into a new market, if you can figure out how to source thousand candidates, if you can reach out proactively to people that you know are good, if you can do more than reliable referrals, if you could build an effective way to actually assess people that you don't know before. If you can actually build that, there's incredible talent on cloud. We've been able to bring some amazing people onto our team. I think that's the biggest thing I learned. Basically, how can you get an advantage in recruiting?

Speaker 2:
[107:57] Yeah. Well, congratulations. Thank you so much, Jordi. You have anything else?

Speaker 3:
[108:01] Yeah, very cool.

Speaker 2:
[108:02] Have a great rest of your day. We'll talk to you soon.

Speaker 3:
[108:04] Congrats on the lunch.

Speaker 2:
[108:05] Congrats on the fellow show. Yeah.

Speaker 11:
[108:06] Thank you so much.

Speaker 3:
[108:07] Cheers.

Speaker 2:
[108:07] Good night. People having more fun with the OpenAI image model, a hydrologically accurate cutaway of the straight-up Hormuz drawn by Richard Scarry. Are you a Richard Scarry guy? I'm a huge fan of Richard Scarry. With the cats, this is a good way to learn. It's a good way to learn. Riley Walls was doing some crazy stuff where, or actually it's Riley Goodside, the prompt engineer. I got them mixed up. So Riley Goodside worked at DeepMind and ScaleAI, and he asked, Chat, GPT Images 2.0 Pro, generate a photo of a cake decorated with an SVG that when transcribed to a file renders the cake, another cake. And so it's absolutely wild.

Speaker 3:
[108:55] People have been putting the model through its paces.

Speaker 2:
[108:58] For sure, for sure.

Speaker 3:
[108:59] I like this other one from Tender.

Speaker 2:
[109:01] Yes.

Speaker 3:
[109:01] That got to say GPT Images 2 is pretty clutch for interior decorating ideas.

Speaker 2:
[109:07] Yes. If you have an empty wall, put a fridge of Sobe, Monster and Red Bull in your blank wall.

Speaker 3:
[109:17] Sobe has been discontinued, John.

Speaker 2:
[109:19] Really?

Speaker 3:
[109:20] Insane opportunity for certain people to remember who they are and bring this back.

Speaker 2:
[109:25] Yes. I was never a Sobe drinker, but I always respected the brand and what it stood for, the funny ads and what a wild time. Was it highly caffeinated or was it not caffeinated enough? It seemed like it just got steamrolled by Red Bull and Monster eventually. It would be interesting to dig into the story of Sobe and what happened there, but we will have to do that another day because we have Samuel from Prazo creating an AI-powered infrastructure for wholesale commerce, covering procurement, credit and workflows for SMBs. Welcome to the show. How are you doing?

Speaker 4:
[110:00] Good. How are you?

Speaker 2:
[110:01] We're good.

Speaker 3:
[110:01] Good to meet you.

Speaker 2:
[110:02] Please introduce yourself and the company.

Speaker 4:
[110:04] Yeah, of course. Well, I'm originally from Northeast Brazil, a city called Recife. I ended up going to Stanford for undergrad, but dropped out and came back to Brazil and started Prazo. Okay. We're tackling a major problem in Brazil and trying to build the new infrastructure for wholesale.

Speaker 2:
[110:23] When you say infrastructure, that could mean everything from warehouses to warehouse management system, to e-commerce software. Help me understand what infrastructure means in this context.

Speaker 4:
[110:33] Yeah, for sure. We're trying to build it end-to-end. So the idea here behind business is that while retail has seen a lot of digitization in the last 10, 20 years in Brazil or in other countries across the world, wholesale in our view has been left behind. And these small and medium sized businesses are still procuring the same way that they were 20, 30 years ago. Distribution hasn't changed. We're basically building a tech-enabled wholesale commerce player where we deal end to end. So we deal with manufacturers and we help them reach small and medium sized businesses in a more efficient data driven way.

Speaker 2:
[111:12] Sure. So is Alibaba a reasonable comp here? I feel like that's where a lot of people meet wholesalers and suppliers if they're operating in China at least.

Speaker 4:
[111:24] Yeah, but we're only tackling Brazil. We deal with Brazilian suppliers and Brazilian merchants. We started out with food and beverage, which is the highest frequency category with the most number of merchants. So today we serve a little bit over 10,000 restaurants, and just in Northeast Brazil so far. But only restaurants, if you look at restaurant procurement is an over $50 billion a year market. But if you go to other verticals in Brazil, if you go to Latam, it's double the size, so about $100 billion a year. And if you go to other verticals than restaurants, the market just multiplies by five or six. So it's a massive market we're starting out with restaurants.

Speaker 2:
[112:03] Yeah. What is the secret to getting 10,000 restaurants on board? Is this cold email, phone calls, sales reps, advertisements? What's the funnel look like to actually grow? Because you're operating this two-sided marketplace, I imagine that demand generation is top of mind at all times.

Speaker 4:
[112:21] Yeah. I think the trick with serving small and medium businesses is that they have high natural attrition. So the way to grow in an efficient way is having very low CAC. And the only way you can grow at a very low CAC is if merchants love your product, then they will refer you to other merchants. So basically today about 40% of our customer acquisition is sales led. So we have sales reps just reaching out to these merchants and onboarding them into Prazo. But 60% of our merchant acquisition comes through either organic or merchant referrals or channels which are more product led.

Speaker 2:
[113:02] How do you, is disintermediation a problem for you? Like how do you stay, I imagine that there's transaction fees for finding and working with a supplier, if you're a company that's on the platform. Is there a fear of someone meets a great supplier for their food or beverage product and then they start dealing with them directly, is that an issue?

Speaker 4:
[113:24] Really, and the reason for that is that we are serving these very small merchants. So the biggest challenge for a manufacturer to serve them is not, it's one of them might be reaching them, but they couldn't do the logistics by themselves, which is why we do the logistics as well and we also do the credit. So oftentimes, a manufacturer, they don't want to get into these small merchants because the drop size for delivery is very low and they can't do it in an economically feasible way.

Speaker 3:
[113:51] I want 500 Coca Colas, and I want credit.

Speaker 4:
[113:56] Exactly, and they want credit as well, and manufacturers are, in general, one, they're not very good at assessing credit risk of a small merchant, but second, they're not senior in the merchant's payment stack, right? At Prazo, since we're serving them as a one-stop shop and we are getting more and more share of their procurement, we are becoming more and more senior in their payment stack, so as we grow and we become more relevant to these merchants, our delinquency rates just go much lower.

Speaker 2:
[114:24] Do you think you'll have to raise a lot of money to underwrite more credit products, or is there a banking system that you can plug into and sort of use another fintech ally to sort of service that?

Speaker 4:
[114:39] Yeah, today we already use a partner for financing, but I don't think we would have to raise a lot of money for that. And the good thing about our business is that we don't believe in offering very long credit to small merchants because they're very volatile, right? So we have to offer short-term loans. So our average loan is about 10 days. So we turn that three times a month. And that helps a lot in keeping a low outstanding balance but compounding at a very high rate. The company overall is already profitable. So we've raised funding to date, but we've become profitable. Thank you.

Speaker 2:
[115:18] Congrats.

Speaker 4:
[115:19] How big is the team? About 100 people.

Speaker 2:
[115:21] Whoa. When did you start this then?

Speaker 4:
[115:26] I started the company about four years ago and moved back to the northeast of Brazil. Today, we only operate in two cities in northeast of Brazil. So there's major opportunity for growth from there. These two cities combined are only 4% of Brazilian GDP. So we have an opportunity to tap like a 25 times bigger market just in Brazil without even expanding to other countries at that time.

Speaker 3:
[115:53] Do you feel a little bit like an unk being a part of this? Teal Fellowship class, we had the youngest ever Teal Fellow. You seem like maybe a 10 year age gap with him. Is that correct?

Speaker 4:
[116:06] Yeah, well, not that much because I got it to Stanford a little bit earlier. So I had skipped grades in school. So I got it to Stanford. I was a little bit younger. But I think I might be the oldest in this class.

Speaker 3:
[116:21] So we're going to call you the elder, the wise elder.

Speaker 4:
[116:24] I'm probably the youngest wherever I go to somewhere. But in this group, I'm definitely the oldest.

Speaker 2:
[116:31] Well, we'll keep you aggressive. Can you zoom out for us? Oh, sorry.

Speaker 3:
[116:35] Yes, I had one more question around, what do you think the path for this business is? Is this eventually come in IPO in America? How deep are Brazilian capital markets? How do you think about scaling?

Speaker 4:
[116:52] Yeah, for sure. Yeah, most of our investors are US based. So it definitely helps a lot to raise venture funding here. And I think ultimately we would want to IPO in the US. But if you look at this, there's just a major opportunity. I think the market potential is even bigger than food delivery. So for example, iFood, which is the largest food delivery player in Brazil is controlled by Prazo's. It's an over $10 billion market cap company. And we think there's an opportunity to build something even bigger. But while iFood goes in the consumer space, we go B2B. And we think there's an opportunity to build a multi-billion dollar profit company. So such a massive market and there's a long way to building it.

Speaker 2:
[117:41] Last question for me was, zooming out, can you get us up to speed on the Brazilian economy broadly? You know, China's been growing very quickly. America is at the risk of some stagflation. Gas prices are high. There's some economic growth. It's heavy in data centers. And we've been tracking the American market. But how are things going in Brazil broadly? Like, what is the economic outlook?

Speaker 4:
[118:07] Yeah. I would say it's not great. However, the way we generally put this is that, compared to the other emerging markets, we are better. So they're all in a very bad position. So, relatively, we are better. So that's why, like, if you look at the Brazilian stock bar, it has been going up.

Speaker 2:
[118:23] Sure.

Speaker 4:
[118:24] But it's mostly because I think investors want to invest in emerging markets have relatively worse places to put their money.

Speaker 2:
[118:32] Yeah.

Speaker 4:
[118:33] But I think Brazil tends to grow a lot in the next, maybe, decades. And the reason for that is, too, I think, especially energy. So, like, the demand for energy will probably, like, go up a lot in the next few years. Brazil has, sort of, a very big renewable matrix. And we have a lot of opportunity for energy. So I think that helps a lot. Over 50% of our energy is hydro. I think over 70% is renewable.

Speaker 2:
[119:02] That's crazy.

Speaker 4:
[119:02] I had no idea. Yeah. And the second thing is, we have one of the largest reserves of rare earth metals in the world, which I think is very strategic as well. So I think Brazil is a very well-positioned country. It just needs great leadership to take it to the next phase.

Speaker 2:
[119:19] Well, we're glad you're taking the helm and they're lucky to have you.

Speaker 3:
[119:24] Yeah. Super impressive.

Speaker 2:
[119:25] Well, thank you so much for coming.

Speaker 3:
[119:26] Come back on the show next time you guys have a big milestone.

Speaker 2:
[119:29] Yeah.

Speaker 3:
[119:29] Maybe a profitability milestone like your first eight figure, even a quarter or something like that. Yeah.

Speaker 2:
[119:37] That would be fantastic.

Speaker 3:
[119:38] I can see it in your future.

Speaker 2:
[119:39] Well, have a great rest of your day.

Speaker 4:
[119:40] We'll talk to you soon. Thank you very much.

Speaker 3:
[119:42] Great to meet.

Speaker 2:
[119:43] Goodbye. Someone went to the new ChatGT images 2.0 model and said, make me the most AI slop image that ever AI slop, the pinnacle of slop, a seminal work on AI slop.

Speaker 1:
[120:00] And people are saying, this is unironically a work of art. This is a crazy, crazy image that's all over the place. It's really just like every style, sort of. There's some Fortnite in there. There's a jazz cup, a car with an alien, and clippies in there somehow. I wonder if this is the real prompt. You never know when people post online. But certainly, a...

Speaker 2:
[120:24] Certainly rough to look at.

Speaker 1:
[120:26] It is very weird. It didn't nail the beauty of many of these things. Anyway, we have our last guest of the show, Claire Wang, building biologically accurate simulations of nervous systems.

Speaker 2:
[120:40] No company, pure research.

Speaker 1:
[120:42] Welcome to the show, Claire. How are you doing?

Speaker 2:
[120:44] Great to meet you.

Speaker 3:
[120:45] Hi, nice to meet you.

Speaker 1:
[120:46] Thank you so much for taking the time. Please introduce yourself.

Speaker 2:
[120:49] Age of research.

Speaker 1:
[120:50] Age of research. Yeah, tell us a little bit about what you're working on.

Speaker 3:
[120:53] Yeah, of course. I'm Claire. I actually just struck out, but I was a junior at MIT studying electrical engineering, computer science. A lot of my interests in the past bit has been in the field of neurotech which includes a focus in whole brain emulation with the focus on doing this with worms first, but also how useful this could be for brain-computer interfaces and understanding consciousness maybe, in that sort of direction.

Speaker 1:
[121:18] So, I mean, we've talked to a number of companies that are working on brain-computer interfaces, bumping into this general scientific discipline. What was the, take me through the decision to not join one of the existing efforts or stay in academia and actually go out on your own?

Speaker 3:
[121:39] Yeah, that's a good question. So I would say like, there's still a lot of current efforts that I think I really believe in. And that like, I'm good friends with them, would help out with them. I think there's a lot of bets that people have to make. And these bets are unfortunately kind of fundamental science bets of like, oh, is this physics going to work? Is this like biological, like truth actually true? So I think like, while these people are, I think all these bets are really interesting, there's no 100%. And so I think it kind of makes sense to instead make your own bet on what you think makes the most logical sense. That's kind of my reasoning here.

Speaker 1:
[122:11] Yeah. Do you also think it's a better time than ever to be working more independently on biotech generally because of the advance of AI? And we've talked to a couple of companies that are sort of working on, you know, the AWS of lab equipment, more rental projects, so that maybe you don't need to raise, you know, a billion dollars out of the gate and build a biosafety level two lab on day one. Do you feel like more empowered to validate those hypotheses that you were articulating earlier?

Speaker 3:
[122:47] Yeah, no, I think this is probably the right time. I think it's a mix, like, definitely AI, definitely insurgents of alternate science funding, a willingness for academics and, for example, people in startups to work together. I also think that people are just much more open to new ideas and research risk in startups, which I think like 10 years ago, you'd be told they never ever invest in research risk versus now, like everything is a researcher. I think that's a really good time for you to do it right now.

Speaker 1:
[123:23] Yeah, Jordi, you had a question there?

Speaker 2:
[123:25] Yeah, I mean, my main question was like figuring out when that right moment is, because especially after the announcement of the fellowship this week, I'm sure you've been offered. I'm sure people would offer you millions of dollars over e-mail. Let's just say like it's fine if you want to just keep doing research, but I guess you'll know it.

Speaker 1:
[123:46] Well, help me bridge the gap between what we've seen in BCI with mostly reading from the brain, getting an X and Y output so you can control a mouse on a computer. Incredibly impactful technology and then simulating nervous systems. How are these two things related in your mind? Why are they important to overlap? What is the overlap?

Speaker 3:
[124:12] Yeah, I am. So I think right now BCI technology is very powerful in the sense that it is on a great path to its clinical applications. Soon people who are paraplegics can maybe walk again, people who are blind can see again. But a lot of this comes from almost like post-hoc problem solving. You throw enough data at a model of someone's brain, and maybe you can help them move their arm in the X, Y direction or move a mouse in a specific way. But when you are able to actually decode information from the brain and read and understand exact signals coming from every region of the brain, being able to truly control the brain, for example, if you know which regions to activate, then you have a more naturalistic control method. Instead of only being able to move my arm left to right, you can move your arm very naturalistically in any degree of freedom. I think that's the power of being able to simulate the brain. You could do a lot of research and understanding of one of the most complex things in the universe. Obviously, we're not starting with the human brain, but any level tells you the sort of data that you need and tells you the sort of imaging techniques that work. I think through that, you get a lot more information and progress in BCI.

Speaker 1:
[125:25] Why C elegans and not mice or monkeys or something else? Is it cost or do you have a firm belief that if it works in C elegans, it will scale? Or is there prior art, like what excites you about that particular target?

Speaker 3:
[125:41] Yeah. So, C elegans is 300 neurons, it's really dumb, it's like really just not, I mean, it's not close to humans at all. But I think the argument is like we can't even estimate the C elegans. There's a lot of benefits C elegans gives us. It is translucent, it's much easier to do gene therapy, so like fluorescence therapy, so easier to image as well. And also they're just like very simple. They're like the same graded potentials instead of action potentials. So I think the idea is we can't even do C elegans, so we have to start with that. And there's a lot of like what is the sort of data that you need? Like do I need voltage data or is calcium data enough? Is light she or electron microscopy enough? So like there's all these questions that C elegans can answer. And once we answer that question, obviously the goal is to move on to like zebrafish and mice and fly and so on. But right now, like we are not close or like we're not close to mouse, for example, because like there's no way to image the mouse brain while the mouse is still alive. Yeah.

Speaker 1:
[126:38] But you can't see all of them.

Speaker 2:
[126:39] Fascinating.

Speaker 1:
[126:41] Are you thinking about, do you think you'll wind up with co-founders? Like how early are you in the process of like turning something into a company? Do you want to just sort of like be on your own or do you want to build a some sort of team even if it's loose? How do you think about like research collaboration?

Speaker 3:
[126:58] Yeah, so I mean, I've been able to, I've been lucky to work with some of the most amazing researchers and I think these are people I want to continue to be around and learn from. But I think in terms of specifics like finding co-founders, finding the right bet, for example, the scientific bet I want to make. That's still up in the air. I'm still like learning. I'm still meeting a lot of people and seeing like, what do I believe in and what do I think makes the most sense? So it's still quite early stage for me.

Speaker 1:
[127:20] Are you going to move to San Francisco?

Speaker 3:
[127:22] Yeah. Yeah, probably.

Speaker 2:
[127:25] Reluctantly, maybe.

Speaker 3:
[127:27] I mean, I like SF. I lived in SF in the past, but I don't know. Yeah. It's kind of a principles thing.

Speaker 1:
[127:34] Yeah. Have you always been pure researcher or do you have entrepreneurship background as well? What else have you done?

Speaker 3:
[127:39] Oh, yeah. I mean, I would say almost most of my background is a very clean mix between the startup space and research. I've done a lot of work in example, working at various startups, helping out with startups, some small scale investing stuff. So I think I have a good mix up of both sides.

Speaker 1:
[127:57] That's very fun. Well, good luck.

Speaker 2:
[127:58] Well, come back on whenever you have news and congratulations.

Speaker 1:
[128:02] Yeah, we'd love to catch up soon.

Speaker 2:
[128:03] Great to meet you.

Speaker 1:
[128:04] Have a great rest of your day.

Speaker 2:
[128:05] We'll talk to you soon.

Speaker 1:
[128:06] Goodbye.

Speaker 2:
[128:08] Intel is up.

Speaker 1:
[128:10] Intel's up.

Speaker 2:
[128:11] Massively.

Speaker 1:
[128:12] The market is broadly down today.

Speaker 2:
[128:15] Intel is up 15% after hours. They reported earnings.

Speaker 1:
[128:20] That's good news.

Speaker 2:
[128:20] Let us see. Intel reported.

Speaker 1:
[128:23] Leopold Aschenbrenner continues to cook with his Intel bet.

Speaker 2:
[128:26] He needed another win. He needed another win. It had been a couple of days since he had a massive win.

Speaker 1:
[128:30] Yep. I think there's actually another historian here from today about a memory company he invested in that's doing very well.

Speaker 2:
[128:37] Well, more breaking news in the journal. Bob Iger is returning to where?

Speaker 1:
[128:43] Disney.

Speaker 2:
[128:43] Thrive.

Speaker 1:
[128:44] Thrive. No way. That's amazing.

Speaker 2:
[128:47] Back to Thrive.

Speaker 1:
[128:48] Love it.

Speaker 2:
[128:49] Yeah, I think he's been an LP in Thrive. He also bought a piece of Thrive.

Speaker 1:
[128:54] Yeah, that's right. Okay. Well, that will be a good next act for him. I'm very interested to see where he goes. There's a whole alumni class coming together. Reed Hastings is out and on to the next thing. We'll see where they go, hopefully.

Speaker 2:
[129:07] Back to Intel. Intel announced its first quarter earnings after the bell on Thursday, beating analysts expectations on the top and bottom line and providing better than anticipated Q2 guidance and strong data center sales. Intel said it expects revenue of 13.8 and 14.8 billion for the second quarter. Wall Street was anticipating 13 billion. And as of this morning, they were at around 100 times PE. And so it's only, I guess, only up for Intel.

Speaker 1:
[129:44] Climb in the ranks, climb in the ranks. Tesla also released Q1 earnings. Revenue of 22.4 billion versus 21.4 billion estimated. So they beat on top line. They also beat on net income. 1.45 billion versus 1.17 billion estimate. The interesting article in the journal was that Elon was being more cautious about Tesla talking, saying, I think we need to get realistic about some timelines. So he is certainly not pumping everyone up and he's trying to sort of reset around the fundamentals. And so we will see where that goes. It is a wild timeline that SpaceX, if it goes out at 1.75 trillion will be bigger than Tesla, which is sitting around 1.1, 1.2 trillion these days. So still both huge companies.

Speaker 2:
[130:38] Credit to Bubble Boy over on X. Two hours ago he says, everyone asked me about how I'm playing earnings. He says, doubling down 25% of my portfolio is in Intel calls.

Speaker 1:
[130:49] Wow. There we go, Bubble Boy. Congrats.

Speaker 2:
[130:52] Well done. Well played.

Speaker 1:
[130:54] Stuff.

Speaker 2:
[130:54] Well played.

Speaker 1:
[130:55] Thank you for tuning in to our Teal Fellowship Giga Stream. We will be back tomorrow at 11 a.m. Pacific. Leave us five stars on Apple Podcasts and Spotify Center for a newsletter at tbpn.com. Throw that flashbang, Jordi, because we are out of here.

Speaker 2:
[131:09] It's been an honor.

Speaker 1:
[131:10] It's been an honor.

Speaker 2:
[131:11] Thanks for hanging out with us.

Speaker 1:
[131:12] We will see you tomorrow.

Speaker 2:
[131:14] We'll see you soon.

Speaker 1:
[131:15] Goodbye.