TBPN | Tuesday, June 17th (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:03) [Music] (00:00:14) [Music] (00:00:19) Come on. Come on. Come on. Come on. Come (00:00:19) on. Come on. Come on. (00:00:22) [Music] (00:00:32) [Music] (00:00:39) Hey. (00:00:46) Hey. Hey. (00:00:57) [Music] (00:01:08) [Music] (00:01:17) [Music] (00:01:29) [Music] (00:01:40) [Music] (00:01:43) Heat. Heat. (00:01:49) [Music] (00:02:01) [Music] (00:02:10) [Music] (00:02:22) [Music] (00:02:31) [Music] (00:02:33) feeling. (00:02:34) [Music] (00:02:44) [Music] (00:02:53) [Music] (00:03:17) I see multiple journalists on the (00:03:19) horizon. (00:03:29) You came to this world (00:03:33) [Music] (00:03:35) to reach the stars (00:03:40) to shape our future. (00:03:44) [Music] (00:03:47) to feel the new (00:03:58) [Music] (00:04:07) surrounded by journalists. Hold your (00:04:08) position. (00:04:20) [Music] (00:04:30) Let me (00:04:36) [Music] (00:04:46) Team Deathmatch (00:04:49) 5 (00:04:53) You're watching TBPN. Today is Tuesday, (00:04:56) June 17th, 2025. We are live from the (00:05:00) TBPN Ultra Dome, the temple of (00:05:02) technology, the fortress of finance, the (00:05:05) capital of capital. Welcome to the show. (00:05:09) Massive day. Ramp has announced a new (00:05:11) valuation. $16 billion. Let the robots (00:05:14) chase your receipts and close your books (00:05:16) so you can use your brain to build (00:05:17) things, says Eric Lyman, CEO of (00:05:19) ramp.com. Save time is money. Time is (00:05:22) money. Save both. Go to ramp.com. Switch (00:05:24) your business to ramp.com. We have a (00:05:26) great lineup for you today. Uh let's run (00:05:29) through some timeline just to give you (00:05:31) an idea of what's going on in the news. (00:05:33) Obviously uh the because one of the big (00:05:35) issues war is still is still on the (00:05:36) front page of the Financial Times. Um (00:05:39) the Wall Street Journal is taking a (00:05:40) little bit of a more positive view uh (00:05:43) highlighting uh peace talks at the G7 (00:05:47) and this idea that tan signals readiness (00:05:50) to renew diplomacy. Iran says it wants (00:05:53) nuclear talks as long as US stays out of (00:05:56) the conflict with Israel. And so that's (00:05:58) where the major front page news is. But (00:06:00) there's a ton of other stuff that's more (00:06:02) important that's happening in tech. And (00:06:04) so we got to talk about tech. Big issue (00:06:05) is if you want to use X, the everything (00:06:07) app for news today, good luck because I (00:06:10) personally can't go on there without the (00:06:12) first 10 posts being about ramp. I love (00:06:15) it. And certainly we are contributing to (00:06:17) that. But uh we'll try to cover. It's a (00:06:19) timeline takeover. It's a timeline (00:06:20) takeover, folks. But there is other (00:06:22) news. Um, on the front page of the Wall (00:06:25) Street Journal, Berber Jinn has a scoop (00:06:27) about OpenAI. Mo OpenAI Microsoft (00:06:30) tensions are reaching a boiling point. (00:06:33) Startup frustrated with its partner has (00:06:35) discussed making antitrust complaints. (00:06:37) Tensions between OpenAI and Microsoft (00:06:38) over the future of their famed AI (00:06:42) partnership are flaring up. OpenAI wants (00:06:43) to loosen Microsoft's grip on its AI (00:06:46) products and computing resources and (00:06:48) secure the tech giant's blessing for its (00:06:51) conversion into a for-profit company. (00:06:52) Microsoft's approval of the conversion (00:06:54) is key to OpenAI's ability to raise more (00:06:56) money and go public. But the (00:06:58) negotiations have been so difficult that (00:06:59) in recent weeks, OpenAI's executives (00:07:02) have discussed what they view as a (00:07:03) nuclear option, accusing Microsoft of (00:07:05) anti-competitive behavior during their (00:07:08) partnership. people familiar with the (00:07:10) matter said. And so there's a whole (00:07:11) bunch more analysis on this that we'll (00:07:13) go into today and I'm sure we'll talk to (00:07:15) some folks on the show about it. Don't (00:07:18) use the Mword monopoly. Yes, it's (00:07:21) banned. It's banned. It's banned. No, (00:07:24) that really is a nuclear option. Uh they (00:07:26) were very happy partners for seemingly (00:07:29) about a year and a half. Yeah. And (00:07:32) uh clearly Satia and Sam have different (00:07:36) visions for their partnership going (00:07:37) forward. I think at the time when they (00:07:39) did their original $10 billion (00:07:42) investment, it felt like everybody was (00:07:44) getting a good deal. I think with the (00:07:46) growth of Chat GPT, (00:07:48) um, in hindsight, maybe that wasn't the (00:07:50) best structure, the best, you know, way (00:07:52) to to do a partnership. And, um, (00:07:54) certainly there was some (00:07:57) some fine print that they're now trying (00:07:59) to walk back. Um and I and I there had (00:08:01) been some other chatter around um one of (00:08:04) the issues and the reason for the (00:08:07) windsurf acquisition to not be formally (00:08:10) closed and and announced was that uh if (00:08:13) they just sort of went forward with it (00:08:15) with the existing structure of the (00:08:16) OpenAI Microsoft agreement. uh Microsoft (00:08:19) might have some claim over over (00:08:22) Windserf's IP and um again it's all very (00:08:26) complicated because there's 20 different (00:08:27) entities uh and ultimately you know it's (00:08:31) hard to know what fits in where but um (00:08:34) again MASA seems I wonder how real that (00:08:37) is like I I I haven't used windsurf (00:08:38) enough maybe we should ask Tyler have (00:08:40) you used windsurf at all cursor what do (00:08:42) you use I've used cursor I I haven't (00:08:44) tried windurf though give give windsurf (00:08:46) a try today I want I want your a little (00:08:48) review. I want to know how uh how it (00:08:50) differs because specifically I want to (00:08:52) know is the intellectual property are (00:08:56) there like you know specific designs or (00:08:58) specific algorithms in Windsurf that (00:09:00) that if copied by Microsoft would (00:09:02) improve GitHub copilot because my my my (00:09:05) thinking is that the real value with (00:09:07) windsurf is the aggregation being the (00:09:09) front door to AI codegen and then (00:09:12) generating data and feedback and then (00:09:14) feeding that back into a reinforcement (00:09:17) learning (00:09:18) system and doing another training run. (00:09:20) And so if if it's just it like are you (00:09:24) buying windsurf is the value of windsurf (00:09:26) the fact that they have a lot of people (00:09:27) using it or is it that they've designed (00:09:29) something you know unique that if copied (00:09:32) would immediately give you the same (00:09:34) product like would everyone switch to (00:09:35) GitHub copilot if if they copied Windsor (00:09:38) because Gemini has copied a lot of the (00:09:41) chatbt features I'm pretty sure chatbt (00:09:45) has like 99% penetration right well part (00:09:47) of it too was opening knows that codegen (00:09:51) is going to be very important to their (00:09:52) business long term and they'll spend $3 (00:09:54) billion to get a really talented team (00:09:57) that's that's figured out some uh you (00:10:00) know key they have they have real (00:10:02) traction they're they're they're growing (00:10:04) quickly and uh it can just accelerate (00:10:07) what they're already doing on the cogen (00:10:08) side. I think um copying I don't know is (00:10:12) how how much of a concern that is (00:10:15) because big tech again will copy (00:10:17) everything. If something works they'll (00:10:18) copy it. um the the stories and and and (00:10:21) pull to refresh all of these different (00:10:23) things. Yeah, some of these were even (00:10:24) patented, but tech has this like weird (00:10:27) thesis around patents where they (00:10:29) shouldn't patent UI elements. And so I (00:10:31) think all the tech companies own a lot (00:10:33) of patents, but they never really (00:10:34) enforce them. You never and you never (00:10:35) really hear about like, oh, this company (00:10:37) has this tech and like, yeah, no one (00:10:39) else has an alarm clock app because (00:10:42) Apple patented the alarm clock on the (00:10:44) phone. Like that doesn't happen. the the (00:10:46) dynamic that I think is fascinating is (00:10:47) OpenAI has a lot of traditional venture (00:10:49) capitalists on the cap table or on one (00:10:53) of the uh different entities. Yes. Um (00:10:56) but Microsoft and Microsoft was making (00:10:58) sort of gross stage very high-risisk (00:11:00) investments into OpenAI, but they don't (00:11:03) have to be founder friendly. They don't (00:11:05) have to be, you know, they don't have to (00:11:07) like play by the same rules as the VCs (00:11:09) where normally I I can I can almost bet (00:11:12) that any investor that is on the that (00:11:15) that you know, traditional VC in OpenAI (00:11:17) if Sam went to them and said, "Hey, I (00:11:20) know this is going to be, you know, I (00:11:21) know the structure is not, you know, (00:11:23) exactly what you would have liked, but (00:11:25) you know, let's like you're going to go (00:11:28) forward with it." Yes. Yes. Yes. Sam (00:11:29) can't do that to Microsoft three (00:11:32) trillion dollar company when Satia has (00:11:35) has representing all of Microsoft (00:11:38) shareholders, right? Yeah. So, so I I (00:11:39) and it's not like Microsoft is trying to (00:11:42) lead a bunch of these deals a year and (00:11:44) and they're worried about their (00:11:46) reputation of being founder friendly. It (00:11:48) also sounds like OpenAI leadership just (00:11:50) doesn't like the original deal that they (00:11:52) did and they agreed to and they want to (00:11:55) try to unwind some of that now. So yeah, (00:11:57) I I do wonder how much of this is just (00:11:59) like like the we we keep hearing about (00:12:01) like the nuclear option, the most (00:12:02) aggressive option, like taking it to the (00:12:04) courts, taking it to antitrust, taking (00:12:06) it to whatever. And uh you know, we've (00:12:08) certainly seen that with the with the (00:12:09) XAI, open AI battles that have played (00:12:12) out because Elon Musk was a donor to the (00:12:15) nonprofit. And I was always just (00:12:18) thinking, yeah, it it like all of these (00:12:20) deals. I think that the stress is so (00:12:23) much higher. Not just because like super (00:12:26) intelligence could potentially there's a (00:12:28) chance that it's like God in a box and (00:12:29) you become like the most powerful person (00:12:30) in the world, but but aside from that, (00:12:32) it's like we are clearly looking at a (00:12:35) category that will have a power law (00:12:37) winner that will be worth over a (00:12:39) trillion dollars. (00:12:41) I mean, I I don't I don't think that's (00:12:43) controversial to say. May maybe you (00:12:44) could say, "Oh, actually like the (00:12:46) market's only 100 billion or something, (00:12:47) but like it seems like consumer (00:12:50) artificial intelligence is going to be (00:12:52) as big as the search engine market or as (00:12:54) big as the iPhone market or as big of as (00:12:57) uh you know any of these other uh GPU (00:12:59) accelerators market. So like we're (00:13:02) looking at the the next hyperscaler, the (00:13:03) next the the eighth company to join the (00:13:06) Mag 7 will probably be an AI company. (00:13:09) OpenAI looks like a leading candidate (00:13:11) there. And so the like what is at stake (00:13:14) is so much higher that it's potentially (00:13:17) worth it to go through every possible (00:13:20) option. PR, leaking to the journal, (00:13:24) going to the courts, lobbying, getting (00:13:26) Trump to do a press conference with you (00:13:29) or getting Trump to put pressure. Like (00:13:31) there's no amount of money that you can (00:13:33) spend. there's no amount of political (00:13:35) capital that you can expend that that (00:13:37) looks ROI negative when even getting a (00:13:40) small slice could be hundred billion (00:13:42) dollars like very very material and so I (00:13:44) think that's actually driving more of (00:13:45) the dynamic than like when you see like (00:13:47) a billionaire who takes a flyer on some (00:13:50) sort of early stage company and they're (00:13:53) and they're in it for a million bucks (00:13:54) and then a deal happens and it's like oh (00:13:56) you're going to get some and we need to (00:13:58) we need to cram you down or do something (00:14:00) like the company's not doing that well (00:14:02) like it like you're shuffling it around (00:14:04) it. It's not that much of a lever on (00:14:06) people's net worth, but like we're at a (00:14:08) scale where owning artificial (00:14:11) intelligence would be a lever on even (00:14:12) Elon Musk's net worth. Well, yeah. And (00:14:15) the other thing here is the way that (00:14:16) OpenAI's or sorry, Microsoft's (00:14:18) investment in OpenAI were structured and (00:14:20) that it was more of a profit share, (00:14:23) right? It was this 49% profit share (00:14:25) capped at a 10x. Yeah. And it's also (00:14:28) possible that Satia doesn't actually (00:14:32) feel like maybe that that's the right (00:14:34) structure, right? When he's like, "Hey, (00:14:35) I actually want to own a piece of of the (00:14:38) front door to consumer AI." Yeah. I (00:14:40) mean, on both sides, you know, they they (00:14:43) did this deal at a at, you know, in many (00:14:46) ways 20 years have happened in the last (00:14:48) few years and so it's very possible that (00:14:50) both sides are kind of unhappy with the (00:14:52) deal and need to rework it. So, but it (00:14:56) really would be nuclear for OpenAI to (00:14:59) start complaining to about, you know, (00:15:02) antitrust, anti-competitive. Would it be (00:15:04) though? I mean, Microsoft's dealt with (00:15:06) like the worst antitrust ever. Like, (00:15:08) they went through the whole browser wars (00:15:10) and it destroyed Bill Gates's life for a (00:15:13) couple years and he had to step down. (00:15:14) Balmer came in. Yeah. That's pretty (00:15:15) nuclear if samples. We're talking about (00:15:18) a company that's, you know, this is uh (00:15:21) this is uh the training test site over (00:15:23) there. They've been they they've they've (00:15:25) seen nuclear bombs go off. The whole (00:15:28) sand is glass. They're they're immune (00:15:29) maybe. I don't know. Maybe. Yeah, we'll (00:15:31) have to dig into it. Anyway, uh we we'll (00:15:33) bring you more analysis on that in a (00:15:34) little bit. Um let's run through some (00:15:36) more timeline. Uh open AI or a speaking (00:15:39) of AI, Sam Alman spoke at YC's AI demo (00:15:43) day. Uh, and so did uh the Perplexity (00:15:47) CEO. Uh, somebody said, "How do you stay (00:15:50) motivated when you're down?" And Arvin (00:15:51) from uh, Perplexity says, "I just (00:15:53) watched the Elon Musk YouTube videos." (00:15:56) And I love this comment from Hard Deep. (00:15:57) He says, "You imagine the CEO of a 14 (00:15:59) billion valued company to say something (00:16:01) philosophical." And he ends up being one (00:16:04) of us. I just watched the Elon YouTube (00:16:07) videos. I love it. Just searching Elon (00:16:09) Musk in YouTube. Hopefully searching it (00:16:11) in perplexity. Well, Arvin, we we if (00:16:14) you're if you're just getting into (00:16:15) motivational YouTube videos, Elon Musk (00:16:17) is just the front door. You got to get (00:16:18) into David Gogggins. You got to get into (00:16:20) Sam. Sam Sulac, we we need you to watch (00:16:23) some uh uh Oh, what who's who's David (00:16:27) Senra's uh favorite person? Uh Jaco (00:16:29) Willink. Jaco. Joo. Good. Get deep down (00:16:33) the rabbit hole. We'll send you the Jaco (00:16:35) Willink good video and you'll be Get Joo (00:16:37) to actually go come give a talk at your (00:16:40) office. You can do that. That's an (00:16:41) option. (00:16:42) Anyway, uh, massive news from Superbase. (00:16:45) Uh, what's funny is that if you'd asked (00:16:47) me about Superbase's growth in December (00:16:49) 2024, I would have told you it was (00:16:50) excellent, says Jared Freeman from Y (00:16:52) Combinator. And in 2025, they are just (00:16:56) an absolute rocket ship. And I think a (00:16:58) lot of this is due to the fact that (00:16:59) they've become one of the preferred uh, (00:17:02) one of the preferred database vendors (00:17:04) for uh, a lot of those AI scaffolding (00:17:07) companies, vibe coding tools. And so (00:17:09) people are setting up these databases (00:17:10) very quickly and the growth is (00:17:12) incredible. (00:17:14) Um very very exciting for them. So (00:17:15) congrats to them. Also some good news in (00:17:17) OpenAI world. Even though they're (00:17:19) they're going through the some battles (00:17:21) with Microsoft. Uh they scored a $200 (00:17:23) million US defense contract. So they're (00:17:25) working with the DoD. The most unhinged (00:17:28) this photo. I know. How did this photo (00:17:30) shoot happen? Like this doesn't this (00:17:32) looks like he's about to go on stage and (00:17:35) he has a lav mic and it's from a low (00:17:37) angle but like what lighting scenario (00:17:40) created this photo. It's amazing but (00:17:43) it's very impressive. Anyway, great (00:17:45) selection by not beating the uh you know (00:17:48) evil arms dealer allegations with this (00:17:52) picture. Yeah, if you're if you're (00:17:53) founder, you got to be careful how the (00:17:55) angles people photograph you. But at a (00:17:56) certain point if you're on stage they're (00:17:57) going to take photos in any direction. (00:17:59) But uh yeah, and if they take enough (00:18:01) photos, they'll get every single (00:18:02) expression. So get ready. Uh other news, (00:18:05) Meta finally put ads into WhatsApp. Uh (00:18:08) in January 2012, they said we don't sell (00:18:11) ads, which is still live. Um but this is (00:18:15) why we love them and this is why uh (00:18:18) WhatsApp should have ads. The backbone (00:18:20) of the internet. It's the backbone of (00:18:21) the internet. Uh and uh um Signal says (00:18:25) that's why he's an Apple fanboy because (00:18:27) they don't do ads. (00:18:29) have a huge ads business. I bet you this (00:18:31) is just them listening to their users. I (00:18:32) bet you know the users, you know, all (00:18:34) over the globe said, you know, the only (00:18:36) thing that could make WhatsApp better is (00:18:38) just putting some ads in this bad boy. (00:18:40) It's the one place I go that I don't (00:18:42) felt like this was already happening. I (00:18:43) felt like this leaked like 5 years ago. (00:18:45) How much it was a $20 billion (00:18:47) acquisition? Of course, they were going (00:18:48) to put some ads in some point. Of (00:18:49) course, Senator, we sell ads. Uh anyway, (00:18:52) we have uh Eric Lyman from RAMP in the (00:18:54) studio. Welcome to the show, Eric. (00:18:56) Congratulations, Jordy. Get that mallet (00:18:59) ready. I got a gong head coming up. The (00:19:01) news. What happened today? Tell us. (00:19:05) $2,283 (00:19:08) and we're at $16 billion valuation. (00:19:15) I've been waiting to do that one. Yeah. (00:19:17) Is that the number of days since you (00:19:19) started the company? That is. That's how (00:19:21) long we've been at it. Um we have a long (00:19:23) way to go and guys couldn't be happier (00:19:25) to be here. (00:19:26) So is the job finished? (00:19:29) Job's not finished. There you go. I (00:19:31) don't think so. That's great. Uh what (00:19:34) what is coming up? What can we expect? (00:19:36) Uh are are there is there a clear public (00:19:39) roadmap? Are I mean the the the number (00:19:42) of features launched last year and this (00:19:43) year is a ton. What can you tell us (00:19:45) about the product? (00:19:47) Look, I I I think the the world is (00:19:49) moving faster than ever. for your the (00:19:51) whole introduction of this segment was (00:19:53) uh you about folks working on AGI, ASI, (00:19:56) whatever it is. Yeah. Um you know I I (00:19:59) don't know how far we are along that (00:20:01) path but I can tell you AI is definitely (00:20:03) smart enough to do your expense reports. (00:20:05) It can definitely do bill payment runs. (00:20:07) It can definitely speed up accounting. (00:20:09) And those are the kind of uh I I think (00:20:12) real innovations and support. We're (00:20:13) actually bringing customers today over (00:20:15) 40,000 businesses, you know, and so (00:20:17) we're we're putting it deeply into the (00:20:18) product and we can go through all that. (00:20:20) Uh we're also using it to ship a lot (00:20:22) faster. All of last year, we thought we (00:20:24) moved fast. We were pretty proud that we (00:20:26) shipped 207 features. Uh so far in just (00:20:29) the first five months, uh that number is (00:20:31) over 270. So we're moving a lot faster (00:20:34) thanks to the help. How big is the team (00:20:36) now? (00:20:38) Um we are uh over,00 people currently. (00:20:43) It's remarkable. And how much has that (00:20:45) grown over the last few years? Is that (00:20:47) accelerating? Because it seems like (00:20:48) there's a narrative where like you can (00:20:50) do more with less, but obviously you're (00:20:53) trying to grow as fast as possible. So (00:20:54) are you growing headcount and the amount (00:20:57) of features you're shipping and the size (00:20:59) of and and the footprint of the (00:21:00) businesses that you serve as well? It (00:21:03) it's all growing, but certainly at (00:21:05) different rates. Um so first I mean the (00:21:08) business itself um it's really unusual (00:21:10) usually just it's like law of physics (00:21:11) the bigger companies get gravity weighs (00:21:14) you down you just go uh more slowly and (00:21:16) I think the most unusual thing and you (00:21:18) know Trey at foundersson pointed this (00:21:20) out um in an interview with Bloomberg uh (00:21:22) revenue growth has actually been faster (00:21:24) uh so far in 2025 than even 2024 at much (00:21:27) larger scale uh and so revenue growth is (00:21:30) is picking up as well as purchase volume (00:21:33) um the team size is growing too. Um, but (00:21:36) to your point, we're seeing leverage. (00:21:38) Uh, I think around this time last year, (00:21:41) uh, I I want to say we were somewhere, (00:21:44) you know, 700 to 800ish, uh, folks. Uh, (00:21:47) and so what we've grown, um, the pace of (00:21:49) the the top line and the bottom line has (00:21:51) grown much faster. Uh, and and when I (00:21:54) look at, um, you know, last couple (00:21:55) things, developer productivity is way (00:21:57) up. uh over the last four months the (00:21:59) average engineer at ramp uh is um (00:22:02) shipping uh 50% more commits um you know (00:22:06) in a given day uh than they were just (00:22:08) months ago and so more productivity from (00:22:10) developers and from you know I think the (00:22:12) best go to market uh organization uh in (00:22:15) the business you know look that team is (00:22:16) growing you have teams that are (00:22:17) extraordinarily productive and so you (00:22:19) know at at the core we're hiring across (00:22:21) all of them um but we are seeing these (00:22:23) gains you're in a unique position where (00:22:25) you interface with a huge swath of the (00:22:29) American business community. Uh h how do (00:22:32) you track kind of the the general vibe (00:22:35) or the outlook of American business (00:22:38) leaders right now? There's a lot of (00:22:40) uncertainty. There's geopolitical news (00:22:42) every day on the cover of the Wall (00:22:44) Street Journal. Um how are just everyday (00:22:47) American businesses feeling? (00:22:51) I I think right now as far as the you (00:22:55) know data shows I would say like fairly (00:22:58) hopeful and optimistic. Um I mean back (00:23:00) in 2022 when interest rates really (00:23:03) spiked um I mean you would see a sudden (00:23:06) pull back uh in terms of spending um we (00:23:09) put these out regularly. It's it's now (00:23:11) monthly reports. If you go to (00:23:12) ramp.com/data you can see our benchmarks (00:23:15) across industries and verticals. I think (00:23:17) the biggest warning sign is spend on (00:23:20) advertising is starting to slow down as (00:23:22) well as some spend on recruitment. Um (00:23:25) computers, things that are usually some (00:23:27) leading indicators about how bullish (00:23:29) companies are. Um but if you look month (00:23:31) by month, companies are tending to (00:23:32) actually spend more and so still (00:23:34) optimistic. But I I I think that the (00:23:36) other big thing that's really top of (00:23:38) mind for folks and and and we take (00:23:39) really seriously is almost where you (00:23:42) started um uh where I think every (00:23:45) certainly the valley companies are being (00:23:46) asked by their boards what are you doing (00:23:48) about AI but you know this is true of of (00:23:50) of small businesses of of old school (00:23:52) enterprises you name it and I I think (00:23:54) that for most of the 40,000 businesses (00:23:56) we serve um they don't have a single (00:23:58) software engineer uh let alone an (00:24:01) engineer working just for their finance (00:24:04) teams. And so, you know, it's a big part (00:24:05) of why we spend over half of our our (00:24:07) payroll ultimately, uh, just on research (00:24:10) and development, which is an extremely (00:24:11) high ratio. Um, we want to be these (00:24:14) organizations finance teams, these these (00:24:17) engineers for these finance teams. So, (00:24:18) um, uh, those are the main things that (00:24:20) we hear about that we see. Yeah, it's (00:24:22) interesting like there's there's like (00:24:24) new headlines every day and yet the (00:24:26) market has been kind of just kangarooing (00:24:29) around, jumping up and down. And I think (00:24:31) it reflects that kind of like there's (00:24:32) some uncertainty but still enough (00:24:34) optimism in the system that we're not (00:24:36) seeing like a full full pullback. Yeah. (00:24:38) I mean there's there's so much broader (00:24:40) geopolitical uncertainty it's hard to (00:24:42) make super big long-term strategic (00:24:45) decisions of should we grow headcount (00:24:47) 10%. Should we should we've been there's (00:24:50) been a number of like of like almost (00:24:52) false starts where it's been like oh (00:24:54) we're going into a radically different (00:24:56) tariff regime and then it's like a (00:24:57) tariffs actually kind of rolled back. (00:24:58) It's like, oh, there's like this (00:24:59) narrative. And then, so if you're (00:25:01) actually planning, like by the time you (00:25:04) have the all the staff meetings to plan (00:25:06) for the thing, it's like kind of over (00:25:07) and then you're kind of just back to (00:25:09) business as normal. So, you got to chop (00:25:10) wood. Um, how how do you how do you (00:25:13) continue to build uh build in this sort (00:25:16) of um general like how do you how do you (00:25:19) continue to have this mindset of being (00:25:21) day one, right? The growth is just (00:25:23) absolutely insane. I think uh you're in (00:25:26) a position right now that that every CEO (00:25:28) when they go raise their first dollar of (00:25:29) venture capital, they they want they (00:25:31) they they imagine this, you know, (00:25:33) playbook where a couple thousand days (00:25:34) in, they're a multi-billion dollar (00:25:36) company, but it it rarely happens. (00:25:39) What's you know, what what kind of (00:25:41) wisdom do you do you look to in terms (00:25:43) of, you know, continuing to just (00:25:45) motivate the team day in and day out? (00:25:46) We're having Saquon on the show uh (00:25:49) later, which we're excited to talk to (00:25:50) him about kind of that that championship (00:25:52) mindset. But um what what what do you (00:25:55) kind of draw on to to keep that drive (00:25:57) high? (00:25:59) You know, I I uh you know, John, I'm I'm (00:26:02) happy that you you posted this earlier. (00:26:04) You know, for me, this this creates a (00:26:05) lot of emotion. You you you mentioned it (00:26:07) straight away. Look, um we did hit this (00:26:09) $16 billion valuation. Uh it really (00:26:12) should be a day of celebration, but you (00:26:14) know, I'm I'm sad, too. I'm I'm thinking (00:26:16) about the 98% of businesses that, you (00:26:19) know, are not on ramp um that are uh (00:26:23) losing time, that are losing money, um (00:26:25) are still experiencing that worst hour (00:26:27) of their month doing expenses the old (00:26:29) way. And you know, there there's just no (00:26:31) need for it. And so I I think um you (00:26:34) know, there's an element of of just uh (00:26:36) you know, trying to solve a problem and (00:26:38) make the world a better place is uh what (00:26:40) gets us up and out of bed in the (00:26:41) morning. I I'd also say, look, I'm I'm (00:26:43) happy you're you're interviewing Saquon. (00:26:44) He is he is an inspiration and a (00:26:46) motivation I think to to a lot of us. (00:26:47) And it's not just that he's he's great (00:26:49) at what he does. I think there's lots of (00:26:51) uh folks who who are extremely talented. (00:26:53) Um I I think you know time and time (00:26:56) again his focus on uh you know whether (00:26:58) it's letting the young guys eat um you (00:27:00) know uh uplifting others, bringing folks (00:27:02) who should have been in the parade (00:27:04) lifting them over the gates to bring (00:27:05) them in. Uh I think the the drive that (00:27:08) he has of uh seeing others on the team (00:27:10) win is really motivating. I mean, (00:27:11) there's a lot of folks who I think RAM (00:27:13) famously hires um you know, freshman (00:27:16) interns, dropouts, uh folks early in (00:27:19) their career and take a bet on them. And (00:27:20) for me, um seeing folks come in uh (00:27:23) succeed, um get great at their craft and (00:27:26) build teams in the space to do that is (00:27:29) is uh on a more serious note, very (00:27:30) genuinely motivating. And so those are (00:27:32) the things. Yeah. Keith Boy talks about (00:27:34) that idea of like if you're not if you (00:27:36) don't have something burning in your (00:27:37) belly to go build and like become a (00:27:39) founder like you should go work for a (00:27:41) company uh that's that's scaling has (00:27:43) product market fit is going to be high (00:27:46) growth but set you up for success if (00:27:48) you're yeah learn what excellent looks (00:27:50) like what greatness really is and what (00:27:52) good management looks like make the (00:27:54) transition from individual contributor (00:27:55) to manager or back or something like (00:27:57) that and like you can tell that he's (00:27:59) talking about ramp obviously he's like (00:28:01) basically like go work at ramp Um yeah (00:28:03) talk I mean I I think it'd be the the (00:28:05) thing that I think would be most (00:28:06) interesting is like talk about the scale (00:28:09) the scale of your ambition right I think (00:28:11) it's helpful to think about it okay 2% (00:28:13) of the businesses in in the US are (00:28:15) running on ramp that's great we (00:28:16) obviously want to get the other 98% (00:28:18) we're we we work on it here every we're (00:28:21) going to do that we're going to get 100% (00:28:23) of the business on ramp but then the (00:28:24) real challenge starts because then we (00:28:26) have to encourage entrepreneurship to (00:28:28) create new businesses to create new ramp (00:28:31) customers exactly That's the real (00:28:33) long-term goal. The second I mean the (00:28:35) second lover. But on a serious my (00:28:37) question about the 98% is like how many (00:28:40) of them just aren't doing any expense (00:28:42) reporting at at all and are just like (00:28:44) yeah I have a credit card and I don't (00:28:46) know how much people spend versus paper (00:28:48) receipts versus other solutions. Do you (00:28:50) have an idea of like the pie chart of (00:28:52) what American like there's millions of (00:28:54) businesses like what is the most common (00:28:57) like you know um alternative that's (00:28:59) currently in place. Yeah. No, and I'm (00:29:02) happy you pointed this out. I mean, the (00:29:04) the one and a half% we think is by (00:29:06) volume of corporate and small business (00:29:08) card transactions. Um, by businesses, (00:29:10) there's it's probably closer to 99%. Um, (00:29:13) so there's about 3 and a half million uh (00:29:16) to four million B businesses that have (00:29:18) five or more employees. And so typically (00:29:21) at this point you're you're employing (00:29:22) others. You're you're keeping some book (00:29:24) and books and records hopefully paying (00:29:26) taxes um and um you know keeping those (00:29:29) records and and so you know when we look (00:29:31) at that 40,000 is a lot but it's it's (00:29:33) tiny. It's it's not yet one and a half%. (00:29:35) There's about 30 million businesses. Um (00:29:37) a lot of those are LLC's maybe with a (00:29:39) single employee or no no employees. And (00:29:41) so the number is much starker. And I and (00:29:44) just to drive this home, you know, guys, (00:29:46) like it is 1 and a half% if we just look (00:29:49) at card volume, but it it turns out (00:29:51) there's other ways to pay for things. Um (00:29:53) you know, there's a um check wires. Um (00:29:56) we we also shared for the first time uh (00:29:58) the volume uh on uh non-card purchases (00:30:02) has exceeded card volumes ultimately on (00:30:05) ramp. And so we're just getting going. I (00:30:07) think our market share is closer to 0% (00:30:09) and one on that. uh and I've you know (00:30:11) it's also been brought to our attention (00:30:12) there are there are more countries than (00:30:14) just the US too and so you know there (00:30:16) there's there there's just a huge amount (00:30:17) of of scale this opportunity and and I (00:30:20) think um (00:30:22) honestly like uh you know I it may be a (00:30:26) bit injust but I think I think you're (00:30:28) actually right John like most businesses (00:30:31) uh never get off the ground uh if they (00:30:33) do I think one in eight fail in a given (00:30:35) year um and uh most businesses are (00:30:38) operating on really slim margins uh and (00:30:40) if you actually go and you know uh you (00:30:43) can increase uh when you look at the (00:30:45) average American business um the eight (00:30:48) and a half% profit margin uh (00:30:50) mathematically uh if you can either um (00:30:53) you know a penny save is not a penny (00:30:55) earned um you know a dollar of savings (00:30:58) is actually worth 12 a.5 earned uh on (00:31:00) the top line and so if you can actually (00:31:03) go and you know add percentage points to (00:31:06) the bottom line I think not only will a (00:31:08) lot more businesses is make it and live (00:31:10) to fight another day. But I I think that (00:31:11) a lot of folks who have huge talents but (00:31:14) you know don't have specialization in (00:31:16) finance um don't have expertise in this (00:31:18) field if you can go and actually just (00:31:20) get this out of a box through software (00:31:21) or just build a business that you're (00:31:22) passionate about I think the world gets (00:31:24) a lot more interesting and so I actually (00:31:26) think it's a very real point that you're (00:31:27) you're bringing up. Yeah, it's crazy how (00:31:30) how like I don't know like culturally (00:31:32) you can change the mindset of every (00:31:35) employee. Like I mean our producer like (00:31:38) will send requests to us on ramp for (00:31:41) like individual pieces of equipment he's (00:31:43) buying and and and that is like (00:31:46) incredibly fine grain but it makes sure (00:31:48) that we're not like spending too much on (00:31:50) on random stuff. Uh and and and that (00:31:53) that type of like you know every penny (00:31:55) matters culture is something that like (00:31:58) ramp makes it easy to actually like (00:32:00) enforce that in a way that's not very (00:32:03) cumbersome but you can still instill (00:32:05) that that that uh that like behavior (00:32:09) which I think is really really valuable. (00:32:10) Well, I have some breaking news here. Uh (00:32:12) a Jeb Bush post just in the timeline. (00:32:16) The silence this morning was deafening, (00:32:18) but he came out and said, "Ranch Ramp (00:32:19) reached a$16 billion valuation. Ramp (00:32:22) isn't just a tech story. Businesses (00:32:23) across the US are saving billions of (00:32:25) dollars because of their technology. (00:32:27) Eric and Kareem are making our economy (00:32:28) stronger and I thank them for it. This (00:32:31) is American innovation at work. So, (00:32:33) let's give it up for Bush and the whole (00:32:35) RAM team. Um, it is he's been he's been (00:32:38) a backer for like years now, right? (00:32:40) Yeah. Yeah. He got in early early and (00:32:42) right, you know. So, fantastic. I I have (00:32:45) to give a a shout out to the governor, (00:32:47) too. I mean, unironically, like I have (00:32:49) to say there's a lot of uh investors, (00:32:52) venture capitalists, angels listening to (00:32:55) this podcast. And by the way, you're (00:32:57) you're totally right. This is the best (00:32:58) place to come to. I I I learn a lot. (00:33:00) But, uh look, I I have to say that that (00:33:03) the governor actually exhibits, I think, (00:33:05) everything you would hope for an (00:33:06) investor. um you reach out to him. He's (00:33:09) uh he's responsive. He helps close new (00:33:11) business. Um great. He helps encourage (00:33:13) more people to join. When you have great (00:33:15) news, he shares it. Uh he's there. I (00:33:18) wish everyone, you know, approach (00:33:20) investing the way the the governor did. (00:33:22) Everybody can learn a little about being (00:33:24) a helpful VC from Jebush Bush. We love (00:33:27) it. So, it's amazing. Well, thank you so (00:33:29) much for stopping by. I'm sure you got a (00:33:30) busy Congratulations on on the (00:33:32) milestone. It is it's an honor to to (00:33:35) watch you and the team work. Uh, and we (00:33:38) uh, you know, couldn't couldn't be more (00:33:39) excited to cover the next decade of ramp (00:33:42) domination. So, hit it again. Hit it (00:33:44) again, John. And then we'll let There we (00:33:46) go. (00:33:48) Congratulations. Thank you for coming (00:33:49) on, Eric. Uh, enjoy the day. Um, I know (00:33:53) it is just day one. Cheers. We'll talk (00:33:56) to you soon. Cheers. Bye. Later. Bye. (00:33:59) Another air horn on an absolute tear. (00:34:03) Another absolute tear. Uh, this was an (00:34:05) interesting post I saw from modem (00:34:08) introducing the dream recorder, the (00:34:10) magical bedside open open- source device (00:34:14) that plays your dreams back as cinematic (00:34:16) reels. So, it will I I don't know how it (00:34:20) does this if you wear a headband and it (00:34:23) tries to read your brain waves or if (00:34:24) it's just like listening to you talk in (00:34:26) your sleep. I I don't know. We were (00:34:28) discussing with the with some friends (00:34:30) about like whether or not people dream. (00:34:33) There's like a new It's like the new (00:34:34) internal monologue. Are you like a no (00:34:36) monologue, no internal monologue person (00:34:39) there? Apparently, there are people that (00:34:40) just don't have dreams. I can't imagine (00:34:42) it. I often wind up talking out loud in (00:34:45) my sleep. I talk about business (00:34:46) exclusively though, which is very very (00:34:48) funny. Um, but I imagine that if I if I (00:34:53) if I use this device, put it by my (00:34:55) bedside, it were able to read my dreams, (00:34:56) it would actually just be (00:34:59) I'm on the website. Yeah, please. I (00:35:01) don't see any information on how it (00:35:03) actually works, (00:35:05) which is my one major question before I (00:35:09) set this device up next to my bed. It (00:35:11) would be funny, you know, you're you're (00:35:12) watching a dream, you know, you're (00:35:14) playing back a dream you had or you're (00:35:16) kind of experiencing it in real time and (00:35:18) it's like, okay, I'm riding a cow and I (00:35:20) jumped off a cliff and I landed on an F1 (00:35:22) track and now I'm That's like perfect (00:35:24) for AI generation, honestly. It's just (00:35:27) like your trippy dreams can like come to (00:35:29) life. Yeah, it's pretty cool. I wonder (00:35:31) if completely they open sourced it. I (00:35:35) don't know how I don't know. It's hard (00:35:36) to say how real this is. It's a cool (00:35:40) cool 3D render that they're showing and (00:35:42) seems like a cool use case for AI. So (00:35:44) anyway, good good luck to them. Um (00:35:47) anyway, I'm sure they're working on (00:35:49) designing their website and they should (00:35:50) get on Figma. Figma.com. Think bigger, (00:35:53) build faster. Figma helps design and (00:35:54) development teams build great products (00:35:55) together. It is the official design tool (00:35:59) of Golden Retrievers. They asked me to, (00:36:02) you know, explicitly say that. They did. (00:36:04) Official endorsement from our from our (00:36:07) uh from our furry friends. Um Jordan (00:36:11) Schneider, friend of the show, says Zuck (00:36:14) still has the dog. Tim, on the other (00:36:16) hand, and there's a quote in here, uh (00:36:18) Zuckerberg has spoken openly about (00:36:20) making artificial intelligence a (00:36:22) priority for his company. In the last (00:36:23) two months, he's gone into founder mode, (00:36:27) according to people familiar with his (00:36:28) work, who described an increasingly (00:36:29) hands-on management style. That is (00:36:32) exciting. I didn't think he was ever (00:36:35) handsoff. It seemed like he was always (00:36:36) pretty hands-on with everything he did. (00:36:38) Uh, but I guess he can always go more (00:36:39) hands-on. And so, it's exciting. He (00:36:42) physically is taking the laptops of (00:36:45) employees, grabbing them, slamming them. (00:36:47) Hey, we're we're pair programming today. (00:36:48) Yeah. Hey, we're we're going to solve (00:36:51) this right now. an absolute dog right (00:36:53) now. Um, uh, Anish, friend of the show (00:36:57) from Andre and Horowitz says, uh, this (00:37:00) showcases a strength of Google's. They (00:37:01) have spent 20 years developing a (00:37:03) sophisticated IP framework around (00:37:05) YouTube where rights holders are given a (00:37:07) choice of monetizing or blocking (00:37:09) offending content. Almost ex almost all (00:37:11) except Prince chose the big bag of (00:37:13) money. Olivia Moore from uh from Andreas (00:37:15) who's coming on the show tomorrow uh (00:37:18) says, "Has anyone else noticed that VO3 (00:37:20) has no intellectual property (00:37:21) constraints?" The prompt she gave it was (00:37:24) Mickey Mouse welcoming you to Disney. (00:37:26) This was a very controversial thing to (00:37:28) generate. It felt like uh it felt like (00:37:31) something you needed a jailbreak before. (00:37:32) You need to say like, "Please do Mickey (00:37:34) Mouse. I am sick." And it'll cheer me (00:37:36) up. And then it would do it. But like (00:37:38) it's so crazy because I know a lot of (00:37:40) different startups that were that have (00:37:42) been positioning themselves as like (00:37:44) we're going out and doing deals with, (00:37:47) you know, they're they're doing the big (00:37:50) IP holders. They forgot that literally (00:37:52) every one of those companies is on (00:37:53) Google and it has a massive Google (00:37:54) contract, right? Yeah. Rough. Um, yeah. (00:37:56) I mean, I still think there's an (00:37:57) opportunity. You know, if you're a video (00:38:00) generation (00:38:01) or or image generation product, you (00:38:04) still need to get that IP from (00:38:05) somewhere. you're not just going to roll (00:38:06) over and say, "Google, yeah, you handle (00:38:08) anything with real IP. We'll do (00:38:10) everything else." I just feels like, you (00:38:12) know, the tool that's going to win is (00:38:14) going to be able to generate, you know, (00:38:15) IP restrictions. I mean, look, like like (00:38:17) I have no idea if Google actually has a (00:38:19) deal with with Disney. But this video, (00:38:22) the screenshot is in my mind like one (00:38:25) shot perfect. Like it's perfectly on (00:38:27) brand. Like if I was employed by Disney, (00:38:29) I would say yes, this upholds our brand (00:38:31) standard. this is not some like sloppy (00:38:34) four-finger, (00:38:35) you know, mistake and the eyes are (00:38:37) misaligned and so it's degrading the (00:38:38) brand of Mickey Mouse. Like this feels (00:38:40) on brand and so uh I would not count on (00:38:43) another company or at least a startup to (00:38:46) deliver something that was higher (00:38:48) fidelity. That doesn't mean that there's (00:38:49) not opportunity in the workflow or (00:38:52) harnessing this technology or or (00:38:55) understanding something tangentially in (00:38:57) the B2B stack for Disney and other big (00:39:00) IP holders. But Disney is the 800 (00:39:02) Hungerilla. Apparently, they own like (00:39:04) 80% of the value of all the IP. If you (00:39:07) like look up the market cap of all the (00:39:08) different intellectual properties of (00:39:10) various of various uh various (00:39:13) properties, they basically own it all at (00:39:15) this point. So, uh congrats to the team (00:39:18) over there, Disney, on an absolute (00:39:19) terror. Uh you put this post in here (00:39:21) from Jerick Isaacman. This was (00:39:23) interesting. I I thought this was just a (00:39:24) part of the tied into the nothing ever (00:39:27) happens meta which is dominant right (00:39:30) now. Supposed to be NASA administrator (00:39:33) and is an accomplished entrepreneur and (00:39:35) the first commercial or the first (00:39:37) independent uh citizen uh civilian (00:39:39) astronaut. Right. So uh story career um (00:39:43) and uh I don't know if you want to give (00:39:45) this a read or you Yeah, I'll read (00:39:47) through it. So he says, "Apologies for (00:39:48) the TLDDR, but when you step back, it is (00:39:51) kind of wild what we've all lived (00:39:52) through over the last 5 years. No wonder (00:39:54) so many young people are anxious about (00:39:55) the future. The quote unquote (00:39:57) disturbance in the force feels stronger (00:39:59) by the day. I don't have any grand (00:40:01) takeaways other than this. The world (00:40:03) could use an immediate course correction (00:40:05) in the direction of boring or we may (00:40:07) really need those Mars rockets sooner (00:40:08) than expected. One thing is for sure, (00:40:11) Israel is making a compelling case for (00:40:13) Golden Dome. and he kind of lists out a (00:40:17) few different kind of uh broader events. (00:40:19) So he says, "An once in a century (00:40:20) pandemic shuts the world down, no matter (00:40:22) how you view it in hindsight, both (00:40:23) allies and adversaries were nearly (00:40:25) unified in halting the global economy (00:40:27) and banishing society to lockdowns and (00:40:29) high pressure mask and vaccination (00:40:31) campaigns. We tried to print our way out (00:40:34) of the system shock, triggering the most (00:40:35) euphoric markets since the dot bubble. (00:40:38) Let's give it up for euphoric markets." (00:40:40) Um, pre-revenue IPOs uh reappeared for (00:40:43) some reason and people forgot that good (00:40:45) companies generally don't spec. The (00:40:47) digital revolution kicked into (00:40:48) overdrive, work from home, virtual (00:40:50) education, traumatized parents, Zoom (00:40:52) cocktail parties, Pelaton, Door Dash, (00:40:54) and Microsoft Teams. Probably the most (00:40:56) painful development taking shots. Um (00:41:00) civil unrest emerged alongside deepening (00:41:03) social and political divides and um (00:41:08) a disheartening end to the war in (00:41:09) Afghanistan. Trillions spent, thousands (00:41:11) of lives lost and the Taliban is still (00:41:13) running the show. Market euphoria gave (00:41:16) way to historic inflation. Interest (00:41:18) rates shot up to cool things down. The (00:41:19) tide went out. The uh SH (00:41:23) the bad companies failed. Centralized (00:41:25) crypto exchanges gambled customer (00:41:26) deposits. Hedge funds weren't hedged. (00:41:29) BCheavy banks like SVB collapsed, (00:41:31) triggering a temporary panic in the (00:41:33) regional banking systems. The big banks (00:41:35) got even bigger this. For the first time (00:41:38) since the Soviet invasion of (00:41:39) Afghanistan, a nuclear superpower (00:41:41) launched a full-scale invasion of a (00:41:43) neighboring country. The West isolates (00:41:45) Russia and we witness a asymmetric (00:41:47) dynamic in warfare. Cheap drones, (00:41:49) missile swarms, all playing out in real (00:41:51) time on social media. (00:41:53) the metaverse and web 3 died quickly as (00:41:55) the mag 7 let a market rebound on the (00:41:58) promise of AI China closes gaps and (00:42:00) anyways it goes on and on and on and on (00:42:02) and on uh and he says all in just 5 (00:42:04) years so the the reason I thought this (00:42:06) was interesting is you know the the (00:42:09) anything almost happens and then doesn't (00:42:12) happen and people just say nothing ever (00:42:13) happens but it's like if you actually (00:42:15) look back a ton of history a lot has (00:42:18) happened things are happening um I think (00:42:22) It it almost feels like depends on the (00:42:24) frame. Nothing ever happens is now just (00:42:26) like it would have to be like a a 90% (00:42:29) stock market. Yes. Correction. Yes. A (00:42:32) nuke going off. You know, there's a (00:42:34) number of things that people would be (00:42:35) like, "Okay, something happened." Yes. (00:42:37) Yes. The definition of happened just got (00:42:40) really really degraded. We're in this (00:42:42) new normal. Any really significant event (00:42:43) as long as it stretches out over, you (00:42:45) know, 6 to 12 months and then everybody (00:42:48) moves on, it just doesn't count as as (00:42:50) happening. Yeah. Hopefully, our defense (00:42:52) and policy leaders are paying attention (00:42:54) and making some course corrections. (00:42:55) Congressional leadership is mostly (00:42:56) well-intentioned, but often fights for (00:42:58) expensive job programs. Exactly the kind (00:43:00) of thing an overconolidated defense (00:43:02) industry encourages, even as we stare (00:43:04) down an unsustainable 36 trillion (00:43:06) dollars of national debt. That's how you (00:43:08) end up holding a fleet of battleships (00:43:10) during the advent of the aircraft (00:43:11) carrier. Only this time, the analogy (00:43:13) breaks down because as a nation, we have (00:43:16) forgotten how to build ships. So (00:43:18) instead, we will have $300 million (00:43:20) fighter jets. We can't afford arriving a (00:43:22) decade too late in quantities that may (00:43:23) not even matter. Disrupted by (00:43:25) million-dollar hypersonic laser equipped (00:43:26) drones that our adversaries will likely (00:43:28) produce at scale until perhaps a dark (00:43:30) nor the dark horseynet T1000 shows up. (00:43:35) This is the time, especially in such (00:43:37) politically charged environment when we (00:43:39) need to be finding more ways to come (00:43:40) together instead of moving apart. A time (00:43:42) to be rooting for America and our (00:43:44) leadership, not betting on the next (00:43:45) catastrophe. because if uh if the next 5 (00:43:50) years look anything like the last (00:43:51) military parades and trade imbalances (00:43:53) will be the least of our problems. And (00:43:55) Tom Mueller uh says, "Great summary of (00:43:57) our current situation. We need to stop (00:43:59) with the division and build our way out (00:44:00) of it." And Casey Hanmer says, "Thanks (00:44:03) Jared. I can't believe how many healthy, (00:44:04) smart, well educated, well-resourced (00:44:06) people aren't building out factories (00:44:08) right now." Yep. Very interesting. If (00:44:11) you are healthy, smart, well educated, (00:44:13) well resourced for the factory. I hope I (00:44:15) hope Isaac Jerry Eisenman like lands a (00:44:18) role somewhere in the government at some (00:44:19) point. This was like very detailed and (00:44:21) very inspirational. I feel like he (00:44:23) should he should do I don't know that be (00:44:26) empowered somewhere and do do something (00:44:27) cool. Uh because he was I was excited (00:44:29) about him as as uh head of NASA, but now (00:44:32) that he's a free agent, hopefully he (00:44:33) gets traded to something uh bigger. It' (00:44:36) be great. Uh anyway, um let's tell you (00:44:40) about Vanta. Automated compliance, (00:44:42) manage risk, prove trust continuously. (00:44:44) Vanta trust management platform takes (00:44:46) the manual work out of your security and (00:44:47) compliance process and replaces it with (00:44:49) continuous automation. Whether you're (00:44:50) pursuing your first framework or (00:44:51) managing a complex program, get on the (00:44:53) platform trusted by Duolingo Intercom (00:44:55) Rampingo. Oh yeah, nice. Uh Nick Carter (00:44:58) has funny post. I feel like we almost (00:45:00) touched on this yesterday, but he says, (00:45:02) "Faking an email from a VC saying they (00:45:04) are sending you a term sheet while (00:45:06) raising is securities fraud, (00:45:07) misrepresenting material facts in (00:45:09) connection with a securities offering." (00:45:11) Uh Gary Tan had some notes for us. Uh at (00:45:14) YCW, don't commit securities fraud. (00:45:16) Yeah, we asked him about the uh (00:45:18) inflation and and the you know, (00:45:21) different interpretations of what ARR (00:45:23) is. Oh, yeah. Yeah. Yeah. Contracted AR. (00:45:25) You know, be uh be very very explicit (00:45:28) about your revenue. Y um VCs are very (00:45:32) fine, very okay with risk. You just have (00:45:34) to give them full insight into the (00:45:37) business. So yeah, very very silly. (00:45:40) Yeah. Uh we are we are we're definitely (00:45:42) in the new stage of a new bubble, a new (00:45:45) market, a new bull market and so there's (00:45:47) a lot of froth and there's a lot of (00:45:48) people uh testing the barriers, testing (00:45:51) the testing the limits of these things. (00:45:54) Anyway, this next post is wild from Joe (00:45:56) Cohen. A Cal Trains employee built (00:45:59) himself an illegal apartment in a train (00:46:02) station and he has unfortunately been (00:46:05) sentenced to 6 months in prison. Uh (00:46:08) because it was an illegal apartment in a (00:46:10) train station. Um and when I first read (00:46:13) this, I thought it was built in a train (00:46:15) and I was like that would be so sweet if (00:46:17) he like somehow took secretly took over (00:46:19) a train. It's got a extra link in the (00:46:21) train, you know. (00:46:24) Yeah. And and then is just building (00:46:25) there. But and then apparently (00:46:27) everyone's response to this is it's (00:46:29) impressive what he was able to do with (00:46:30) $42,000. Maybe he should be in charge of (00:46:33) California housing. And Sheil has some (00:46:35) photos. Uh you know, it's not it's not (00:46:38) luxurious luxurious living, but uh it's (00:46:41) not bad either. I mean, this is this is (00:46:43) pretty comfortable. How did he actually (00:46:45) build this though? He went and got all (00:46:46) the furniture and installed the the (00:46:48) panels. I'm I'm interested like how do (00:46:50) you do plumbing in an environment like (00:46:51) this? Like that is a shower with a (00:46:54) drain. Uh it seems to be working. It's (00:46:56) you know it's pretty shoddy but it's (00:46:57) not. It it definitely seems like it (00:47:00) works. A weird story. And to think you (00:47:03) know this is one one story like this (00:47:06) around the world. There's probably (00:47:08) thousands and thousands of people that (00:47:10) have built little secret apartments in (00:47:12) places they shouldn't be and they're (00:47:14) just scurrying around, you know, like a (00:47:17) like a like a mouse setting up setting (00:47:19) up shop where (00:47:21) setting up shop in the walls like like (00:47:24) uh like sand through marbles. It finds (00:47:26) all the cracks. It's it's a you know an (00:47:28) efficient we like resource utilization. (00:47:30) I'm I'm I'm I'm for this guy. you can (00:47:32) go. Yeah, there maybe there should be a (00:47:35) rule where it's like if you set up a (00:47:36) secret apartment and you and you and you (00:47:39) sleep there for 10,000 nights straight (00:47:41) and you're not caught, you then are (00:47:44) you get Yeah. you you get the deed. You (00:47:46) you actually own the place. Yeah. Well, (00:47:49) I mean, I I used uh I rented a house at (00:47:51) one point uh with a roommate like a (00:47:53) decade ago or something, and we had (00:47:55) these two bedrooms that were right next (00:47:57) to each other, and they'd built uh they (00:47:59) built built-in closets between them, (00:48:01) between the two uh rooms, and as I was (00:48:05) moving out, I was looking and there was (00:48:06) a window and we looked in the window and (00:48:08) the window like went to a room that (00:48:11) didn't exist in the house. So, but it (00:48:14) wasn't it wasn't like someone was living (00:48:15) there. It was more like they when they (00:48:18) built the built-ins, they built two (00:48:20) walls and there was just a window there (00:48:21) that they didn't like patch up. And so (00:48:24) there was just like empty space where (00:48:26) the closets didn't really make sense to (00:48:27) fit. And so there was just like dead (00:48:30) space. It was probably like five square (00:48:31) feet, not much, but it was still like (00:48:33) this like room and you could look (00:48:35) because the window is cracked open (00:48:37) inwards and so you could look in and see (00:48:38) this like kind of empty tiny room and it (00:48:41) was very creepy. Weird. Yeah. Weird. not (00:48:44) not not not the best. But uh you know I (00:48:46) think these I think these like like (00:48:48) empty rooms kind of happen from uh like (00:48:51) shoddy remodel after shoddy remodel and (00:48:53) like moving really quickly and being (00:48:55) like oh yeah like we don't really care (00:48:56) about that space whatever. It's you know (00:48:58) we lost some square footage but it (00:48:59) doesn't matter. Like we just need to (00:49:00) slot in these closets in the right (00:49:02) place. Anyway, if you're planning a (00:49:04) remodel uh make sure you use linear. (00:49:06) Linear is a purpose-built tool for (00:49:07) planning and building products. And why (00:49:09) not uh $42,000 apartments? Maybe you (00:49:12) should have used linear. wouldn't have (00:49:13) gotten caught. Depends. You're (00:49:15) stretching stretching a little bit now. (00:49:17) But if you are building software (00:49:20) development because it is ramp day ramp (00:49:22) uses linear. Open AAI uses linear (00:49:25) streamline complexity projects and (00:49:27) product. It says streamline projects. (00:49:29) Boom. Aerospace. You're telling me I (00:49:31) could not build an apartment using (00:49:33) linear. Me use the Use the tool. You're (00:49:36) challenging me to build an illegal. (00:49:37) You're twisting my arm making me build (00:49:39) an illegal apartment using this. A (00:49:40) little challenge for you. Little (00:49:41) challenge. You mean just give it to (00:49:43) Tyler. Tyler, build build a get on (00:49:46) linear a a cubiclesized apartment (00:49:49) somewhere in the studio. But don't let (00:49:51) us know where somewhere on this lot. So (00:49:53) he's in the ramp office now. That that (00:49:56) really looks like the ramp office. Is (00:49:57) that a photo or is that AI? I can't even (00:49:59) tell. That really does. What? It's a (00:50:01) real photo. Oh, it's a real photo. There (00:50:03) we go. Yeah, that is that is exactly the (00:50:05) layout. Wow. Um yeah, Jordy, look over (00:50:08) there. You see behind me there's a (00:50:10) there's a door. Who's living in there? (00:50:11) Where does it go to? Do we know? We (00:50:13) don't know. That could be where Tyler's (00:50:14) sleeping. We don't know. Anybody could (00:50:15) be in there. We haven't seen his house. (00:50:17) We don't know. Um Edward Mayor uh has a (00:50:20) good post here about what he likes in (00:50:22) tech companies. He says, "I've been (00:50:24) thinking about what kind of tech (00:50:25) companies actually excite me to work for (00:50:27) or help build and why. I came up with a (00:50:29) list. I call it sacred engineering. What (00:50:33) companies do you know that fit the (00:50:34) bill?" One, real time consequence. (00:50:36) There's no undo. what's built or (00:50:38) launched happens and it either works or (00:50:40) fails visibly. Thinking of SpaceX (00:50:42) obviously, but there's lots of other (00:50:43) examples. Even Rainaker, you spend all (00:50:46) this time, you know, launching uh, you (00:50:48) know, some some drone that's supposed to (00:50:50) control the weather. Augustus gotten a (00:50:52) little got a little dust up on the (00:50:54) timeline with Trey Stevens because he (00:50:55) was caught building a tent outside and (00:50:58) Trey said, "Hey, if you can control the (00:50:59) weather, why do you need a tent?" (00:51:03) But to Augustus's credit, he said, "We (00:51:05) can only make it rain. We can't make it (00:51:08) not rain." There we go. He's the rain (00:51:11) maker, not the rain pretor. Figure out (00:51:13) how to do both, I guess. Yeah. Yeah. (00:51:14) Yeah. You got to you got to figure out (00:51:15) both. There's obviously applications to (00:51:17) both. Um, mastery over extremes, pushing (00:51:20) the limits of heat, speed, pressure, (00:51:21) complexity. Nature isn't a backdrop. (00:51:23) It's an adversary or gatekeeper. (00:51:25) Precision with soul. Every bolt, weld, (00:51:27) line of code must be perfect, yet it (00:51:28) serves a larger purpose or dream. Like a (00:51:31) temple built to reach the divine (00:51:33) collective will. Sacred engineering is (00:51:35) never solo. It's many minds and hands (00:51:37) moving as one. Symbolic payload. It (00:51:39) means something. A launch, a cure, a (00:51:41) mission. Everyone knows it's more than (00:51:43) just tech. It's a statement. This this (00:51:45) is good writing. This is This sounds (00:51:46) like something I would hear like a (00:51:47) voiced over Super Bowl ad. Everyone (00:51:50) knows it's more than just tech. It's a (00:51:52) statement by a Dodge Ram. (00:51:56) Uh transformation. something (00:51:58) irreversible, changes, terrain, orbit, (00:52:00) political landscape, or human (00:52:02) perception. So, making the world a (00:52:04) different place. I like it. Very good. (00:52:06) We have a post here from Chris at Pace. (00:52:09) He says, "To those of you watching at (00:52:11) home, this is your canary in the coal (00:52:12) mine to let you know that the next wave (00:52:15) of consumer apps is incoming. No (00:52:17) surprise that iPad is the first wedge (00:52:19) given the intersection of Mchip and App (00:52:21) Store distribution." So, somebody at who (00:52:24) was at WWDC says, "I turned my iPad into (00:52:28) Tom Riddle's diary with Apple's new (00:52:30) foundation model, an LLM built into iOS. (00:52:32) It's lightning fast." Yeah. And so, I (00:52:34) mean, this this is this is such an (00:52:37) exciting time. Like I remember the early (00:52:39) app store days and we talk about the (00:52:41) beer app, but there were so many other (00:52:44) cool apps that people were just ripping (00:52:46) and some of them turned into massive (00:52:48) massive companies like people forget (00:52:50) that at one time like Uber was just one (00:52:52) of a many of a bunch of apps that you (00:52:54) tried. It was a hot app of the month. (00:52:56) every uh every Apple could have Apple (00:52:59) could have messaged this so much better (00:53:00) and just said like introducing free (00:53:03) inference free inference on device like (00:53:05) go is going to be a a gold rush for (00:53:09) developers that are building cool things (00:53:11) all of these apps there's going to be so (00:53:12) many applications for for kids to you (00:53:15) know story time and all these different (00:53:17) playbooks and stuff like yes it's going (00:53:19) to be hard to go and disrupt like (00:53:21) Salesforce on mobile because like (00:53:22) Salesforce is going to want to build (00:53:24) that or whatever like some of big or re (00:53:26) rebuilding uh you know Google Docs and (00:53:29) all the crazy enterprise stuff is (00:53:31) probably going to be a little bit more (00:53:32) entrenched but just having fun and (00:53:35) building something that's really unique (00:53:36) that's completely enabled and AI native (00:53:38) is like it's total game on and obviously (00:53:41) it's getting even easier to actually (00:53:42) build these apps uh they're partnering (00:53:44) with Anthropic to speed up Xcode and so (00:53:46) there's going to be a ton of really (00:53:47) really cool things coming out of this uh (00:53:49) what a time to be a an app developer if (00:53:52) you're developing a cool app you're (00:53:54) going viral hit us up come on show, (00:53:55) break it down for us. Um, but I I I (00:53:58) think this is going to be a really (00:53:58) really fun time. I'm definitely looking (00:54:00) forward to the next iPad Pro. Um, in (00:54:03) just a couple months, I think I'll (00:54:05) probably pick one up. That'd be great. (00:54:06) And when I do, I'll probably have to (00:54:08) play sales tax. Hopefully, the sales tax (00:54:10) will be processed with numeral, (00:54:11) numeralhq.com, (00:54:13) sales tax on autopilot. Spend less than (00:54:15) five minutes per month on sales tax (00:54:16) compliance. I prefer to buy products (00:54:19) that use numeral for sales tax. It's now (00:54:22) it's now like a key decision factor. (00:54:24) Totally. Choosing between two products, (00:54:26) go with the one that's using sales tax. (00:54:28) You're so right about that, John. Get (00:54:32) get on there right now. Um, this uh post (00:54:34) from Jacob was funny. My wander vacation (00:54:37) morning routine, having my eight sleep (00:54:39) wake me up, checking ramp first thing in (00:54:41) the morning to ensure I'm saving time (00:54:42) and money, putting my Nautilus on from (00:54:44) bezel, and turning on TBPN to monitor (00:54:47) the situation. So, Jacob is absolutely (00:54:50) locked in monitoring the situation from (00:54:52) his happy place. I love it. You love to (00:54:54) see it. I love it. Um, Lulu came out (00:54:58) very funny. Uh, a little bit uh some (00:55:00) people might say this this could be a (00:55:02) little bit spicy, but consider launching (00:55:04) on June 19th. Big companies usually (00:55:06) avoid that day for announcements. They (00:55:07) don't work on Junth and mainstream media (00:55:09) have the day off, too. But tech people (00:55:11) are all still online. Good opportunity (00:55:14) for attention arbitrage. What does (00:55:15) mainstream media has the day off? Does (00:55:17) that mean like there's no CNBC that day? (00:55:20) like you just turn on the TV. It's maybe (00:55:22) it's more like runs or something. Wall (00:55:23) Street Journal is the market. (00:55:26) I don't know. Is it a is it like a bank (00:55:28) holiday now officially? I have heard (00:55:30) that some companies are are are off. (00:55:32) Yes. Uh Thursday the US stock and bonds (00:55:34) markets will be closed for Junth. Um so (00:55:38) well we'll be on and hopefully some (00:55:40) companies will will launch. We already (00:55:42) have a couple companies coming on the (00:55:43) show that day. So that'll be fun. But (00:55:46) yes, I like this idea of of figuring out (00:55:48) creative days to launch. It does seem (00:55:50) like the there's like a glut of startup (00:55:52) launches and then they come and go and (00:55:55) they go back and forth because there's (00:55:57) some days when it just kind of lines up (00:55:59) that everyone's launching the same day (00:56:01) because of like the various travel (00:56:03) schedules and stuff. Um, but finding (00:56:05) those little pockets of alpha where (00:56:07) where the timeline's easier to break (00:56:08) through is good. You don't want to get (00:56:10) steamrolled. It's always rough. Yeah. It (00:56:12) was interesting. Christian yesterday (00:56:13) said things are still very busy. Yeah. (00:56:15) Um, and that tracks with what I'm (00:56:18) seeing, but um, I think a lot of the (00:56:21) rounds that get down get done over the (00:56:23) next month won't even announce until the (00:56:27) fall. But, I mean, a lot of these (00:56:28) announcements get pushed because of uh, (00:56:32) scoops. Like, if a journalist gets a (00:56:33) hold of it, usually that moves up the (00:56:35) timeline. And so, you say, "Hey, like, (00:56:37) you know, it's leaking. We got to go (00:56:39) live with this today or tomorrow. We (00:56:40) don't have a time to to schedule (00:56:43) everything, but we can do the best we (00:56:44) can. That's why that's why we exist. If (00:56:47) your if your stuff's leaking, give us a (00:56:48) call. Come on the show the same day. Why (00:56:50) not? Uh Tane from Wing, he's been on the (00:56:53) show. He says, "For all the slack Masa (00:56:56) gets, I give him slack." Is that is that (00:56:59) the right flack? Flack. Flack. For all (00:57:01) the flack he gets, I guess you got to (00:57:03) give him slack. Give him some slack. Um (00:57:05) for all the flack he gets, respect, he (00:57:07) might be the only person ever to make a (00:57:09) hundred billion dollars on two different (00:57:11) investments. Alibaba, a $20 million (00:57:14) investment in 2000, turned into a (00:57:16) hundred billion realized. ARM, a 32 (00:57:19) billion acquisition in 2016. Soft Bank (00:57:21) Stake is worth 135 billion now, mostly (00:57:24) unrealized. But, you know, there's (00:57:25) there's probably liquidity there. And (00:57:27) so, uh, he's done it twice. Insane the (00:57:31) scale it took to do the second one. Uh, (00:57:33) really really a testament to being a (00:57:34) size lord. But you only get to be a size (00:57:36) lord if you get the first one right. So (00:57:39) if advice for emerging managers, just go (00:57:42) find a $20 million investment that you (00:57:44) can make that will realize you hundred (00:57:45) billion dollars. Yeah. That's step one. (00:57:48) Start there. And then you'll still get a (00:57:50) lot of hate and need to do it again. (00:57:52) You'll you'll still need to kind of (00:57:53) justify your existence. Exactly. People (00:57:56) will hate on you, write negative (00:57:57) articles, talk down to Exactly. Even (00:58:00) though you did they'll meme you. They (00:58:02) will meme you. Uh but yeah, what a (00:58:04) legend. uh he he participated in the (00:58:07) CNBC deep dive on uh on on the uh uh on (00:58:12) the Abolene Center for uh with with with (00:58:14) Crusoe and OpenAI and and SoftBank and (00:58:17) Stargate. Um and so we got to get him on (00:58:19) the show. We'd love to chat with him. Be (00:58:21) fantastic. Uh anyway, Delian's uh (00:58:24) bringing some spice to the timeline. He (00:58:26) says, "It's been 49 days since the (00:58:28) Discord founder stepped down. I assumed (00:58:30) a reporter would manage to get this (00:58:31) story, but since no one figured it out, (00:58:33) I've had it confirmed by multiple (00:58:34) insiders that Benchmark again pushed out (00:58:36) this founder since they were unhappy (00:58:38) with the IPO path. Tricky. I feel like (00:58:41) Jason Citroen's (00:58:43) great entrepreneur. I'm surprised that (00:58:45) this that this happened this way. Um, I (00:58:48) don't know exactly what Discord needs to (00:58:50) do to get on the IPO path, but it seemed (00:58:53) like it was maybe highly valued in the (00:58:56) private markets because when I think (00:58:58) about Discord going out at like 10, that (00:59:01) seems easy, but I think that they I (00:59:03) think that the the later funding rounds (00:59:04) were much much higher. But again, like (00:59:06) how much can you even put on the CEO of (00:59:09) this company that started it, you know, (00:59:11) what, a decade ago, who's been building (00:59:14) this for so long, and it's like all of a (00:59:16) sudden like, oh, the IPO windows closed (00:59:18) for a couple months, and you you're (00:59:19) upset about this guy. I don't know. I (00:59:21) would I would batten down the hatches (00:59:23) and and ride it out. But we'll see. (00:59:25) Yeah, I don't understand. Uh, they were (00:59:28) unhappy with the IPO path. Was that just (00:59:32) who knows? Citroen maybe wanted to delay (00:59:33) it more. Sounds like more CFO. Well, so (00:59:37) they brought in Blizzard Activision (00:59:39) Activision former CFO Humam (00:59:43) Sakini. Um maybe mispronouncing that, (00:59:46) but um he's a new CEO still seemingly (00:59:49) taking them down the IPO path. But but (00:59:52) um I don't know. Yeah, I mean like how (00:59:54) much of that is on the CEO? I mean, (00:59:56) maybe the road shows, like he's not (00:59:57) telling the right story or something to (00:59:59) these to these uh like institutional (01:00:02) investors who would be the anchors, (01:00:04) anchor buyers in a road show, but I (01:00:06) don't know. I think Jason's scouted. Get (01:00:08) him back in there. Bring him back. Yeah. (01:00:11) You don't hear it first time the founder (01:00:14) took a step back. Super fascinating (01:00:15) story. Came back in. Yeah. No, no, he (01:00:18) has a super fascinating story. He he he (01:00:19) he (01:00:21) didn't follow like the traditional uh (01:00:24) Silicon Valley. actually has like an un (01:00:26) uh what do they call it like (01:00:28) untraditional path untraditional (01:00:29) non-traditional background like he went (01:00:31) to a a small college focused on like (01:00:34) game development built a couple games (01:00:36) wound up selling one to another company (01:00:39) um and then started Discord and kind of (01:00:41) grew from there's some old footage of (01:00:43) him at like TechCrunch Disrupt like (01:00:44) pitching his like mobile League of (01:00:47) Legends competitor is very fascinating (01:00:50) um but yeah he's been on absolute tear (01:00:51) let the guy let him cook he's been at it (01:00:53) for 12 years doing his leg work. Let him (01:00:55) let him give him more more work. Anyway, (01:00:58) Delian continues to be on uh talking (01:01:01) absolute tear. Says Bill Gurley has now (01:01:04) commented on three different podcasts (01:01:05) that the only reason Hillen Valley is (01:01:07) hawkish on China is because we're (01:01:09) talking our book. Or perhaps it's (01:01:10) because the organizers are proud (01:01:12) patriots and recognize that China is (01:01:13) playing a 50-year game looking to unseat (01:01:15) the United States. Uh yeah, the tension (01:01:18) between these two palpable and I think (01:01:21) we should get them. both on having (01:01:23) debate for a pay-per-view. Yeah, (01:01:24) pay-per-view. (01:01:26) Delian, (01:01:28) Mr. Bill Gurley, you're welcome to come (01:01:31) on and settle settle the score. Settle (01:01:33) the score. Just a couple venture (01:01:37) capitalists talking it out. Yeah, I (01:01:41) think they could find common ground. I I (01:01:43) I would certainly hope so. Um anyway, if (01:01:46) you're looking to get in the action, (01:01:47) head over to public.com. Investing for (01:01:49) those who take it seriously. They got (01:01:50) multiasset investing, industryleading (01:01:52) yields, and they're trusted by millions, (01:01:54) millions. Um, Semi analysis has a fun uh (01:01:57) little anecdote here about why chips are (01:02:01) rectangular when r when wafers are (01:02:04) round. Silicon wafers are round because (01:02:06) the silicon is grown in a cylindrical (01:02:08) ingot using the Salski method and then (01:02:12) this ingot is sliced into circular (01:02:14) wafers. Circular wafers pose a (01:02:16) challenge. Rectangular dyes cannot fit (01:02:19) cannot perfectly tile a circle. So any (01:02:21) die that overlaps the curved edge is (01:02:23) incomplete and must be discarded. By (01:02:26) contrast, the in panel packaging we use (01:02:28) organic uh in in in panel level (01:02:32) packaging. We use organic laminate (01:02:34) materials, not grown crystals. So they (01:02:36) don't have to be circular. To illustrate (01:02:37) the amount of area wasted, we used the (01:02:39) semi-analysis die yield calculator, (01:02:42) which they provide for free uh to (01:02:45) compare how much waste area there is (01:02:47) using a 300 millimeter wafer v versus a (01:02:50) uh 510x 515 mm panel wafer using an (01:02:56) interposer that is 61x 68 millimeters. (01:02:59) Our results show that panel wafers offer (01:03:01) significantly better area utilization (01:03:02) than circular wafers. Uh very (01:03:05) interesting that Dylan Patel is going (01:03:07) in. This is why companies like TSMC are (01:03:09) actively developing panel level (01:03:10) technology with many of the industries (01:03:13) framing. I just want to say if you're (01:03:14) analyzing semiconductors, this is who (01:03:17) you're analyzing against. It really he (01:03:19) really is incredible. Anyway, uh we we (01:03:22) have our next guest in the studio, uh (01:03:24) George Hots. How you doing, George? Good (01:03:26) to hear from you. What's going on? (01:03:27) Welcome. (01:03:29) Can we hear you? (01:03:31) Oh, can you hear me? Yes, we can hear (01:03:33) you. Gotcha. Uh uh let's kick it off (01:03:36) with something uh simple. I want to take (01:03:39) your temperature on uh AGI timelines, (01:03:42) pdoom, the the easy and fun stuff. I (01:03:46) don't know what AGI means and I don't (01:03:47) know what you mean by doom. No. (01:03:50) Is are are these terms just like (01:03:52) entirely irrelevant? I mean now we've (01:03:54) shifted to like super intelligence. (01:03:55) They're all buzzwords, but at the same (01:03:57) time like there is there is an idea of (01:03:59) like the like I don't know that that the (01:04:02) conversation is maybe shifting to like (01:04:04) the (01:04:06) AI generating more economic value than (01:04:08) humans. Is that a relevant metric to (01:04:10) track? Machines have been generating (01:04:12) more economic value than humans since (01:04:14) the industrial revolution. (01:04:16) Is there some is there some other metric (01:04:18) that that we should be tracking (01:04:21) or is it just like irrelevant? You're (01:04:23) just talking about like hype. Like I (01:04:26) don't know. I mean I I like I I don't (01:04:29) know what you mean. Like you can talk (01:04:31) about concrete things. The term like AGI (01:04:33) means nothing, right? Like computers (01:04:36) everything that's a touring machine is a (01:04:37) general purpose computer. Is that what (01:04:39) you call intelligence? I don't know what (01:04:40) you mean. Is a linear regression (01:04:42) intelligent? What if it's big enough? (01:04:43) The Chinese does know Chinese. Yeah. Um, (01:04:46) what what I mean what about uh uh your (01:04:50) your decision to get on a spaceship (01:04:54) traveling at 0.9 C away from the from (01:04:57) the Earth? Like how close are we to (01:04:59) that? Are we closer than the last time (01:05:01) we talked which was like a couple years (01:05:02) ago and it seemed like it was maybe (01:05:04) going to happen within your lifetime. (01:05:07) Has it moved at all? Yeah. I don't know. (01:05:08) I don't know if I'm actually going to (01:05:10) get that spaceship but it's kind of like (01:05:12) in an ideal world what I would want to (01:05:13) do, you know? Yep. Just just just back (01:05:16) away and chill and don't look back. (01:05:18) Actually, you can't look back. (01:05:21) They're all there. You need the you need (01:05:23) the blast shield, right? (01:05:26) You need the information shield. (01:05:28) Information. What do you mean? (01:05:31) Oh, that's how they're going to get you. (01:05:32) Okay. Right. I mean, okay. So, like (01:05:35) here's a way you can think about AI, (01:05:37) right? Yeah. Um, imagine there were 10 (01:05:39) CIA agents assigned to you and they're (01:05:43) running at a thousandx real time. So, (01:05:44) they're like hyperfast CIA agents that (01:05:46) devote their entire lifespan to your (01:05:49) day. Mhm. And they're trying to (01:05:51) manipulate you. Maybe to get you to buy (01:05:54) things, maybe to get you to vote for a (01:05:56) certain guy, whatever. But like that's (01:05:58) what you're going to be up against with (01:05:59) AI. What we're currently building, what (01:06:01) what what if you think about the biggest (01:06:03) companies in AI, what they do is (01:06:04) advertising. What advertising is is just (01:06:06) manipulation of humans. Um, so you're (01:06:09) going to have a team of CIA agents (01:06:12) thinking about you and trying to (01:06:13) manipulate you at all times. And now you (01:06:15) see why you want to head away at the (01:06:16) speed of light, right? Mhm. Even CIA (01:06:19) agents can't do that. Is there is there (01:06:21) some world where there's like a capital (01:06:23) war and I'm paying for a more powerful (01:06:25) ad blocker? (01:06:28) Yeah, I mean that sounds good. Like (01:06:30) another question is kind of to say like (01:06:32) okay if you think that you either think (01:06:35) that current like capital uh (01:06:37) accumulation dynamics are going to (01:06:39) continue and that uh the rich are going (01:06:41) to continue to get richer and if you (01:06:43) believe that the question is kind of (01:06:44) well how many people are going to (01:06:45) survive in the future? How many people (01:06:46) are going to have any modum of (01:06:48) independence? Mhm. (01:06:50) Um, right. Like you have some far AI (01:06:53) people who think that there's going to (01:06:54) be a singleton, right? Think that (01:06:56) there's going to be literally one, (01:06:58) right? Um, you know, some people maybe (01:06:59) think it's 10, some people a thousand, (01:07:01) 10,000. Uh, some people think that all (01:07:03) the humans will get to continue to exist (01:07:05) as independent entities. Are they (01:07:07) already independent entities? That's a (01:07:09) question, right? I don't know. Good (01:07:10) question. Uh I mean if you were try to (01:07:12) put it in like the form of a bet human (01:07:15) population above or below 8 billion in (01:07:18) 2030 (01:07:21) above I think I would just give you a (01:07:22) normal trend. (01:07:25) Oh is that what the trend says? Yeah. (01:07:26) Just go the trend says I don't think (01:07:28) there's going to be any discontinuities (01:07:29) to any trends really. Well (01:07:32) yeah I mean I mean at some point uh but (01:07:34) but the question is like how far out do (01:07:36) you have to go until you start seeing (01:07:37) these effects? What do you mean by (01:07:38) human? Right. What about someone who (01:07:40) lies in bed all day and watches TikTok? (01:07:41) Are they human? Yeah, that is odd. They (01:07:44) kind of drop out of society. I think I (01:07:46) think uh question that that popped up (01:07:48) for me is is this uh all this debate (01:07:52) about AI safety and what should labs be (01:07:54) doing? What should labs not be doing? It (01:07:56) feels like your angle is it should be (01:07:59) each individual's responsibility look to (01:08:01) look after their own safety in the (01:08:03) context of of AI. Is that at all? I I (01:08:07) just I I just like this whole like (01:08:08) should shouldn't like what I don't know. (01:08:11) I'm not a sadistic [ __ ] who wants to (01:08:13) manipulate other people like the people (01:08:14) in power. Like I don't know. Yeah. But I (01:08:17) mean people still look to you as like an (01:08:19) example of like uh someone who might (01:08:22) have uh answers. No, I don't have any (01:08:25) answers. Not not necessarily answers. (01:08:27) Just like uh but you can buy my shitcoin (01:08:30) here. Did I shield a shitcoin? Here you (01:08:31) go. Just click this QR code and you can (01:08:34) buy a George Hots coin and that will (01:08:36) give you answers. You will find (01:08:38) satisfaction and fulfillment in your (01:08:40) life after purchasing a George Hots (01:08:42) coin. (01:08:44) Is that Is that the end state? We all (01:08:46) have our own coins. I guess. No, no, no. (01:08:49) I I don't mean it like that. I mean, (01:08:50) like I think that a lot of people are (01:08:53) like they don't really know what they're (01:08:54) looking for and that uh vacuum is is a (01:08:59) very uh you know it's very dangerous and (01:09:01) it's going to be filled by dumb [ __ ] and (01:09:03) don't have that vacuum, right? You got (01:09:05) to you got to stand for something, you (01:09:06) know, or something. I don't know. Yeah. (01:09:08) I mean, do do you think that there's a (01:09:10) chance that someone is able to take a (01:09:12) stand and and actually uh bend the arc (01:09:15) of of AI progress in the way that uh I (01:09:19) mean it happened with nuclear, right? (01:09:21) Like like nuclear development did stall. (01:09:24) There was a stagnation in real world (01:09:27) buildout of nuclear capability on the (01:09:30) energy side. Yeah. I mean there's a few (01:09:32) things about nuclear that make it (01:09:34) different. Uh so nuclear uh even as a (01:09:37) weapon is incredibly hard to deploy (01:09:38) tactically, right? So so if a country (01:09:41) has has nuclear uh weapons, they're (01:09:43) aside from like a mutually assured (01:09:45) destruction idea, they're not all that (01:09:46) useful. It's not like you can use a (01:09:48) nuclear weapon to accomplish tactical (01:09:50) objectives, you know, if you could I (01:09:51) think Russia would have already done it. (01:09:52) Yeah. Right. Right. Russia has some (01:09:54) tactical objectives they might want to (01:09:55) accomplish, but uh nukes aren't really (01:09:56) going to do it, right? I mean, from a (01:09:58) pure real politique perspective, not (01:10:00) even from a uh like uh oh, like a taboo (01:10:04) moral perspective, like what do you want (01:10:06) in a radiated pile of rubble? Like (01:10:08) that's what you're going to get. No, (01:10:09) what you want is drones that are hyper (01:10:11) specific and can take out exactly who (01:10:13) you want, can control areas, right? So (01:10:15) like as a military technology, nukes are (01:10:17) not that good. AI is way better. Yeah, (01:10:19) but what about as an energy technology? (01:10:21) It feels like the it feels like the fear (01:10:24) like the mimemetic fear of nuclear war (01:10:26) and total destruction caused a whole (01:10:29) bunch of regulation to pour into a (01:10:30) sector and essentially a stalling of (01:10:32) nuclear energy buildout. And if if if (01:10:35) the AI doom scenario, whether it's real (01:10:38) or not, becomes so mimetically powerful (01:10:40) that someone's able to harness that and (01:10:42) actually say if you try and build a big (01:10:44) data center, we will shoot you. Then (01:10:46) maybe it stagnates. No, I don't really (01:10:48) think that's the reason for nuclear. I (01:10:50) think it has more to do with why we (01:10:51) can't do other big infrastructure (01:10:53) projects in this country, right? Like it (01:10:55) doesn't have to do with the new we also (01:10:56) can't build dams, right? Yeah. And if (01:10:59) you look like that's the thing, if (01:11:00) people think that there's some weird (01:11:01) taboo around nuclear, right? But then (01:11:03) okay, look at hydroelectric, right? (01:11:05) There's no taboo around hydroelectric, (01:11:07) but China leads in installation of both (01:11:09) nuclear and hydroelectric and coal and (01:11:12) everything. It's almost like they're (01:11:13) correlated, right? So the thing is not (01:11:16) there's a specific fear around nuclear. (01:11:18) It's like, you know, the US decided that (01:11:19) they're a developed country and we're (01:11:20) not going to develop anymore because (01:11:22) we're already developed. You see the D (01:11:23) on the end, right? Like, interesting (01:11:25) stuff. Is that So, so is that just (01:11:27) cultural then when you are like the (01:11:29) Malaysia sets in, would you expect that (01:11:31) to happen to China when they catch up? (01:11:34) I don't know. I I Yeah, I mean, maybe (01:11:36) it's just like this normal story arc uh (01:11:39) of like (01:11:41) uh you know, it's it's (01:11:45) I don't know. I don't know. I I think (01:11:47) that like you have a real problem when (01:11:49) the kids can't live better than their (01:11:50) parents. Yeah. Um (01:11:53) so, but no, I don't have anything more (01:11:55) to speculate on that. Do do you have (01:11:57) more context on on China and (01:12:00) specifically in like the AI context? Um (01:12:03) like US electricity looks like this and (01:12:05) China electricity looks like this. Is (01:12:06) that all that matters? Pretty much. (01:12:09) Yeah. I mean, that's a pretty good proxy (01:12:10) for everything, right? Yeah. Um, like (01:12:13) there's two things. There's two things. (01:12:14) You know, people are like, "George, how (01:12:16) do you feel about the Trump (01:12:16) administration?" I'm looking at two (01:12:18) things. Yeah. With any administration, (01:12:19) I'm looking at two things. Did you (01:12:21) decrease government spending and did you (01:12:23) increase total electricity production of (01:12:25) America? Those are the only two numbers (01:12:27) I care about. Those will capture (01:12:28) everything. Why does uh why does (01:12:31) government spending matter? We were (01:12:33) joking that, you know, Trump must be (01:12:35) extremely AGI pilled if he's running up (01:12:38) a massive budget deficit. What the hell (01:12:39) is AGI? I don't know what this is. Uh (01:12:43) like in in this never seen it. Never (01:12:45) seen it in this formulation. It's that (01:12:47) it's numbers. Yes. Yes. Yes. But but it (01:12:51) but is an extra lever on on labor and (01:12:54) capital and it creates more GDP that (01:12:56) then can be taxed to pay down the (01:12:58) increasing amount of debt. Super super (01:13:00) Excel. Yeah. Super Excel. I get it. What (01:13:02) is What is What does Super Excel do that (01:13:04) normal Excel doesn't? Let's give it up (01:13:05) for Super Excel. Yeah. Yes. We need that (01:13:09) Excel 2.0. Right. Yes. The thing is (01:13:12) Excel was the the final piece of (01:13:15) software and then but in order to add (01:13:17) another, you know, hundred trillion (01:13:19) dollars to to global GDP, we needed to (01:13:21) like kind of rebrand it. And so now we (01:13:24) get AGI. GDP is the the complete it's (01:13:27) biggest [ __ ] thing ever, right? Like (01:13:29) I always joke with my friend and I that (01:13:30) we're going to start companies and be (01:13:31) billionaires. And I'll tell you how (01:13:32) we're going to do it. So, okay. All (01:13:33) right. I start a company, he starts a (01:13:35) company. Uh we both write contracts to (01:13:37) each other. Yep. Right. like I'll buy (01:13:39) something from him for a million dollars (01:13:40) and he'll buy something from me for a (01:13:42) million dollars. We'll just do this real (01:13:43) fast. We'll keep passing the money back (01:13:44) and forth. Whoa, look at our revenue. (01:13:47) Wow, that all contributes to GDP. Wow, (01:13:49) we made we're billionaires overnight, (01:13:51) right? Yep. Like like and that's my (01:13:54) argument is the economy is just that (01:13:56) with a lot of extra steps, right? You (01:13:57) can't use services is not part of GDP. (01:14:00) This is complete nonsense, right? You (01:14:02) can't you can't have services. No, like (01:14:03) literally literally you take the steel (01:14:05) out of the ground, you grow the corn. (01:14:06) Okay, that's GDP. But is it I mean if if (01:14:09) that GDP is fake, is not the de is the (01:14:11) deficit not fake? Like is is government (01:14:13) spending less? We owe that [ __ ] to (01:14:14) people. It's not fake. (01:14:17) But can't you just tax the fake the fake (01:14:19) money? Like if you tax your scenario (01:14:20) where you're generating a billion (01:14:22) dollars in fake money. You can't tax the (01:14:24) fake money because we're passing the (01:14:25) same dollars back and forth. The minute (01:14:27) you tax it, that falls off so fast. (01:14:29) Yeah. Yeah. Yeah. You can only tax (01:14:31) productive work. (01:14:35) Uh, is uh is AMD doing productive work (01:14:38) right now? AMD is doing all right. Yeah, (01:14:41) either Nvidia's really overvalued or AMD (01:14:43) is really undervalued. It has to be one (01:14:44) or the other. How does it all play out? (01:14:46) Like what what does AMD actually need to (01:14:48) do to get back on track or realize their (01:14:51) potential? Nvidia needs to stumble. I (01:14:53) mean, it worked for AMD and Intel, (01:14:55) right? Like so AMD ended up beating (01:14:56) Intel in the entire like no one would (01:14:58) buy a data center Intel CPU anymore. (01:15:01) Yeah. And it's just because well, you (01:15:03) know, they stumbled and now Intel owns (01:15:05) that market. Yeah. So, you know, AMD (01:15:08) just sits there in second place. They're (01:15:10) pretty they'd be in a better second (01:15:11) place than they were a few years ago. (01:15:13) Yeah. And then when Nvidia stumbles, AMD (01:15:15) is like, "Oh, hey, we're here." Is is (01:15:18) DGX Leptton like their their cloud (01:15:20) offering a potential stumbling block or (01:15:24) is it uh or is it the right move for (01:15:26) them? I don't know what that is. What's (01:15:27) an Nvidia cloud [ __ ] Yeah, exactly. (01:15:29) Cloud's dumb. cloud though you can you (01:15:31) can break AI down basically into like (01:15:33) there's like five five tiers right like (01:15:35) at the base level you have like (01:15:36) electricity and data centers and land (01:15:38) and like things like that tier two are (01:15:40) like TSMC ASML Samsung Intel right fabs (01:15:43) uh Nvidia AMD open AI anthropic and then (01:15:47) on top you have like completely (01:15:49) worthless things like cursor and windurf (01:15:51) um you know these character AI all these (01:15:53) people who think oh we're the app we're (01:15:54) going to we're going to get the ARR no (01:15:56) that worked the web it won't work for AI (01:15:57) and I can go into why but it's kind of (01:15:59) boring I want I want No, keep going. (01:16:00) Keep going. Keep going. Basically, okay, (01:16:04) so like here's the difference between AI (01:16:05) and and web. Um when you want to run a (01:16:08) service like Gmail, one server can serve (01:16:10) 10,000 people easily, right? And there's (01:16:13) no demand for like better Gmail, right? (01:16:15) It's not like it's not like I can click (01:16:16) and get like, yeah, you can buy Gmail (01:16:18) Pro and it'll have a few things, but (01:16:19) most people don't really care, right? (01:16:20) There's no limit to the ceiling of how (01:16:22) good you want your AI to be, right? Or (01:16:24) how fast you want your AI to be. Maybe (01:16:26) there's a limit to the speed, but like (01:16:27) when you're at like a thousand tokens (01:16:29) per second, I want the biggest model in (01:16:30) the world, right? Like so there there's (01:16:32) very little limit on on that. Uh but (01:16:35) suddenly you can't serve one 10,000 (01:16:37) users from one server anymore. Mhm. (01:16:39) Right. And the whole dynamics of the (01:16:41) web, the whole reason some of the value (01:16:43) aggregated to these end players and they (01:16:45) still didn't aggregate to the cursor and (01:16:46) the winds, they aggregated to the open (01:16:48) eye and the entropics. Right. Nobody (01:16:50) nobody nobody who built like an email (01:16:52) client survived. They all got eaten up (01:16:54) by the the tier fours of the web, right? (01:16:55) the Googles, the Facebooks, um, all of (01:16:58) these like app providers, right? Where's (01:16:59) a Zinga today? You know, like this (01:17:01) already happened, right? People just (01:17:02) don't, where's Zinga? Oh, Zinga is going (01:17:04) to be the next thing, man. Like, no, (01:17:06) it's not. Facebook ate all of that (01:17:08) value, right? Google ate all of the (01:17:10) value from all the people building on (01:17:11) top of Google. So, the tier fours ate (01:17:13) all that value. Yeah. So, OpenAI, (01:17:15) Anthropic will eat all the value from (01:17:17) the cursors and the wind surfs of the (01:17:18) world. They'll acquire some of them. (01:17:19) They'll compete with some of them, (01:17:20) right? Same as you saw on the web. Uh, (01:17:22) but I argue that the tier fours aren't (01:17:24) even going to have value because the (01:17:25) tier fours, (01:17:27) this ain't the web. This ain't where you (01:17:29) can have one server serve lots and lots (01:17:31) and lots of people. You know, I'm (01:17:33) running 03. I'm running, you know how (01:17:34) much I cost OpenAI every month. (01:17:38) I pay the $200 a month and I cost him a (01:17:40) lot more than that. Codeex, you can now (01:17:42) click on Codeex. Yeah. Spin up four (01:17:44) nodes. Yeah. Why would I not click four? (01:17:46) It's not my computer. (01:17:50) You gave me the button. Hey, I'm just (01:17:51) using I'm just using George Hots (01:17:55) single-handedly bankrupts. So, $300 (01:17:57) million. Is there no value in just being (01:17:59) the the the front end to AI applications (01:18:02) to be like the the the the front door, (01:18:04) just the the the default button? because (01:18:07) we see these we see these these these uh (01:18:10) these models kind of go back and forth (01:18:11) in terms of benchmarks or what's hot and (01:18:13) there isn't as much customer churn as (01:18:15) you would expect because people are are (01:18:17) just kind of like defaulted into the app (01:18:19) that they installed whenever and so even (01:18:21) if Gemini gets better in terms of the (01:18:24) actual performance metrics people don't (01:18:26) switch from OpenAI to Google because (01:18:29) it's so it's so negligible you got to (01:18:31) make something 10x better right you got (01:18:33) to make something 10x better so like (01:18:35) this whole game is open AI eyes unless (01:18:37) they stumble. Sure. Um I'm not switching (01:18:39) to Gemini because it's 20% better and I (01:18:41) download some new app and think about (01:18:43) all the thing, right? No one's going to (01:18:44) switch. Is there is there a chance for a (01:18:46) company to kind of come out with (01:18:48) something that's 10x better with an (01:18:50) algorithmic improvement or is it just a (01:18:52) race for scale? Like what could actually (01:18:54) be that next? It felt like GPT 3.5 when (01:18:58) they really broke through with Da Vinci (01:19:00) and then 40 or and then four. like it (01:19:03) felt like this kind of like binary (01:19:05) moment when a lot of people realize that (01:19:07) this was usable for their daily life, (01:19:10) even if it's just a Google search (01:19:11) replacement or whatever, write a poem or (01:19:13) whatever. Uh like a a 10x what you're (01:19:16) describing in like a 10x improvement (01:19:18) feels like that kind of like qualitative (01:19:20) binary shift. Is that possible with just (01:19:23) scale or is this something that we need (01:19:25) a different model for? I don't know. Um (01:19:29) I don't know. I would bet majority still (01:19:33) on like these big labs are also (01:19:35) attracting the talent. Um, but it is (01:19:38) also like it's uh pretty commoditized a (01:19:41) lot more so than like Google search, (01:19:43) right? Like you can look at people track (01:19:44) how far open source is behind. It's not (01:19:46) that far behind. Y um so no, I don't (01:19:50) know. I think this game is mostly going (01:19:52) to be uh chat GPTs. I think Elon's aware (01:19:56) of this too. That's why he's trying to (01:19:58) go 10x bigger with the data center. Yep. (01:20:00) We'll see. Maybe it'll work. You know, (01:20:02) there's there's someone to bet on. (01:20:04) anthropic I'm not that bullish on but uh (01:20:08) maybe (01:20:09) you kind of predicted the uh the (01:20:11) pre-training wall uh but that's not a (01:20:14) reputation of the bitter lesson and (01:20:16) we're going to see similar scale play (01:20:18) out in reinforcement learning or is (01:20:19) there going to be something else that (01:20:20) we're building the big data centers for (01:20:22) there's something that we don't (01:20:24) understand uh in terms of data (01:20:27) efficiency um so like when you think of (01:20:30) how long it takes a GPT to learn to talk (01:20:31) like how much data it takes it takes (01:20:33) like terab terabytes of data. In order (01:20:35) to make a GPT talk like a normal person, (01:20:37) it takes terabytes of data. Okay? (01:20:38) Whereas a human trains on megabytes. (01:20:41) Yeah. Right. How is it that if you take (01:20:43) all the text that you've ever heard in (01:20:44) your life and you you put it to Whisper (01:20:46) and you you uh you transcribe it, it's (01:20:48) going to be a couple megabytes, 10 (01:20:49) megabytes, maybe 100 megabytes. Yeah. (01:20:51) So, humans have this thousandx data (01:20:54) efficiency (01:20:56) uh advantage. And we're going to have to (01:20:59) fix that if we want like reinforcement (01:21:00) learning to work. especially like (01:21:02) reinforcement learning that you want to (01:21:03) do in the real world. A humans could do (01:21:05) humans can learn from very few samples. (01:21:08) Yep. Um, and yeah, I think that like it (01:21:10) might be okay if these foundation models (01:21:12) train unsupervised on lots and lots of (01:21:14) stuff, but uh, yeah. (01:21:17) Is that a (01:21:20) is that something that (01:21:22) somebody's working on just like a a new (01:21:24) more data efficient algorithm to drop (01:21:28) into the pipeline? Or do we have any (01:21:31) like leads there? because it feels like (01:21:33) right now we're going down the path of (01:21:35) like reinforcement learning with (01:21:36) verifiable rewards and we're going after (01:21:38) like individual business use cases that (01:21:40) are increasingly long tail and that (01:21:43) could be kind of like valuable but it (01:21:45) doesn't feel like the breakthrough that (01:21:46) you're talking about (01:21:48) like has there ever been a breakthrough (01:21:50) right like people think GBTs were a (01:21:51) breakthrough no they weren't like you (01:21:53) just if you watch the the world it was (01:21:55) just it was all just smooth but but but (01:21:58) what I will say about AI scaling laws oh (01:22:00) man you see People get excited about AI (01:22:03) scaling laws, but here's a pitch that'll (01:22:04) kill your excitement immediately. Ready? (01:22:07) AI scaling laws. You can put in (01:22:09) exponentially more money to get linear (01:22:12) returns. Exactly. (01:22:15) Uh do do you believe that uh the real (01:22:18) value is investing in in humanoid (01:22:20) robotics then? (01:22:24) Uh have you heard this theory? So, so it (01:22:26) it So, I mean, if you if you put (01:22:29) exponential more money into humanoid (01:22:31) robotics, assuming that they work and (01:22:33) assuming you can uh you can like you (01:22:36) make 10 times as many robots, you get 10 (01:22:38) times as much output. Anyone Anyone who (01:22:40) wants a humanoid robot has never worked (01:22:42) in a factory in their life. Okay. Right. (01:22:45) Bring it down. Anyone wants a human Oh, (01:22:46) yo. Yo, it's going to walk around. Oh, (01:22:48) good thing it has legs, right? No. (01:22:51) Here's what I want. (01:22:53) Yeah. Yeah. We got a laugh track. Okay. (01:22:56) Can you show me a robot arm that's (01:22:58) capable of putting a screw in something? (01:23:00) Probably. Put the screw in the thing. (01:23:03) Yeah. No, no, no, no. Not like you (01:23:06) carefully jigged up the screw and have a (01:23:07) screw dispenser like the way a normal (01:23:09) human does it where the screw sitting (01:23:11) there in a little bucket on the thing (01:23:12) and it picks up one screw and it puts it (01:23:14) in. It takes a screwdriver. No, we're (01:23:15) No, no, we're we're we're not close, but (01:23:17) also it feels like we're not far. It (01:23:19) feels like that's what I feel like I (01:23:21) feel like humanoids are this interesting (01:23:23) sort of like space because a lot of (01:23:26) smart people just say like here's the 20 (01:23:28) reasons why they won't work and and like (01:23:30) why we shouldn't build them but then so (01:23:33) much capital and so many different teams (01:23:34) are trying to make them work that they (01:23:37) they very well might work for some (01:23:39) things and like they just humanity might (01:23:41) brute force it because we saw it in a (01:23:43) sci-fi movie you know 30 years ago. Why? (01:23:46) Why? Why? Why are we cooked on? This is (01:23:48) as dumb as self-driving cars was, right? (01:23:51) And nobody learns their lesson. And (01:23:52) people like Kyle Boat should be ashamed (01:23:53) of themselves. Like they really should. (01:23:55) These people who go and raise large (01:23:57) amounts of money for another thing that (01:23:59) like they should know, they should know (01:24:01) better, right? Here's basically like (01:24:04) remember in 2012 when Google said that, (01:24:06) you know, my my my 12-year-old daughter (01:24:08) would never have to get her driver's (01:24:09) license. Yeah. Come on. That's nonsense, (01:24:12) right? And like now, okay, they shipped (01:24:15) Whimo. It's in a few cities. They're (01:24:18) teley opt (01:24:21) how how tea operated are they in your (01:24:24) opinion? Is it is it effectively one to (01:24:26) one? It's more than one to one. There's (01:24:28) probably about I would say there's 1.2 (01:24:30) operators per car. Um but it's not they (01:24:33) don't have a steering wheel and pedals. (01:24:34) Yeah, it is it is an autonomous system (01:24:37) that they're probably doing some higher (01:24:39) level inputs on. They're definitely like (01:24:41) saying when you can be aggressive, when (01:24:42) you should slow down, uh, you know, (01:24:44) whether you can turn to the stop sign or (01:24:45) not. Yeah. Um, you know, again, like (01:24:49) here's here's the simple reason to know (01:24:50) that it's like that, right? There's (01:24:52) definitely some teleopaths, right? Yeah. (01:24:55) Have you ever seen a picture of that (01:24:56) room? No. Yeah. Why not? (01:24:59) I've always thought it was like an ace (01:25:01) up their sleeve because like if if (01:25:02) there's a lot of pressure on them to say (01:25:04) these Whimos aren't safe, they can pull (01:25:06) off pull the the sheet off of the ghost (01:25:08) and say there's actually a human in the (01:25:10) loop. Don't worry, it's safer than you (01:25:12) thought. Yeah. Yeah. Like you the fact (01:25:16) that you've never seen that room tells (01:25:18) you that it's way worse than you think (01:25:19) it is, right? Tells you that there's way (01:25:21) more telly op than you think it is. If (01:25:22) it was really one person supervising 10 (01:25:24) cars, Google would post those pictures (01:25:26) all over the place. You don't see any (01:25:27) pictures. There's so cruise that (01:25:29) actually came out in the lawsuit. I (01:25:30) think it was like 1.5 or 1.7 humans per (01:25:33) car, right? Or or vice versa. Like, (01:25:35) right. 1.5 cars per person, right? No. (01:25:38) No. Wait. More people than cars. Yeah, (01:25:41) that's what he's saying. It's still that (01:25:43) it's still that one. An Uber only (01:25:45) requires one person. Yes. Yeah. But but (01:25:47) so so maybe the real innovation is just (01:25:49) allowing somebody to get in a car with (01:25:51) and not have to talk about the weather (01:25:53) or or you know. Exactly. Exact. I'll pay (01:25:56) more for that. I'll pay more for that. (01:25:57) Yeah. But I mean, is there any hope that (01:25:59) we drive this down and we get to two (01:26:02) cars per person, then four cars per (01:26:03) person, it starts doubling exponentially (01:26:05) and eventually like we are there. I (01:26:07) mean, yeah, like it's obviously going to (01:26:08) happen, right? It's obviously eventually (01:26:10) going to happen. If you want to see (01:26:11) where the real state-of-the-art of (01:26:12) unsupervised self-driving is today, (01:26:14) right? There's no person with FSD. When (01:26:16) you get your Tesla, that's not tell. You (01:26:18) can go press FSD and that's real AI. Uh, (01:26:21) and well, you can see how good it is, (01:26:24) right? Would I uh take a nap in there (01:26:27) even for five minutes? No way in hell. (01:26:29) Yeah. Um you'd be stupid, right? How are (01:26:32) things going on the comma side? Uh give (01:26:34) us the update there. Pretty good. You (01:26:36) know, we're we're we're on track to um (01:26:40) we're on track to be two years behind (01:26:42) Tesla. So there you go. So So two years (01:26:44) behind Tesla. Uh but you know, here's (01:26:46) why we win, right? Like because like (01:26:51) it's cheap. Uh okay. So, when you think (01:26:55) about self-reming cars, uh it doesn't (01:26:57) look anything like the roll out of Uber, (01:26:59) right? Or Airbnb. When you roll out (01:27:00) something like that, you're trying to (01:27:01) roll out a two-sided marketplace. Mhm. (01:27:03) Uh you got to spend tons of money on (01:27:05) customer acquisition costs. You got to (01:27:06) make sure that you've perfectly matched (01:27:07) that marketplace right away because if (01:27:09) drivers aren't getting rides, they're (01:27:10) going to leave the platform. If riders (01:27:12) have to wait too long for drivers, (01:27:13) they're gonna leave the platform. So, (01:27:14) it's this careful balancing act. But (01:27:16) once you get this marketplace, you got (01:27:18) you got a moat, right? Switching costs (01:27:20) are real high. Try to get everybody to (01:27:21) switch at the same time. It's a chilling (01:27:22) point. ever doing. Self-driving cars (01:27:25) don't look anything like that. (01:27:26) Self-driving cars look like scooters. (01:27:30) The only thing that it's going to take (01:27:31) to roll out big fleets of self-driving (01:27:32) cars is capital, right? It's just (01:27:34) strictly a capital market. You could (01:27:36) just I can if I look at a city, I can (01:27:38) calculate how many Whimo there are. If I (01:27:40) want to build my own network and deploy (01:27:42) that network and run at a lower cost, (01:27:43) it's straight up capital. Easiest thing (01:27:45) for investors to calculate, very little (01:27:47) risk. So, self-driving cars are going to (01:27:50) be this awesome race to the bottom, (01:27:52) right? It's going to be like scooters (01:27:53) where there's going to be like 10 (01:27:54) providers of these things for a while (01:27:55) and then they're going to consolidate (01:27:56) and like one is going to do it. But um (01:27:59) yeah, people are really going to win. (01:28:01) What is what's most valuable in terms of (01:28:04) developing the next like the next better (01:28:08) version of full self-driving? Is it (01:28:10) having a lot of data, building a big (01:28:11) data center, having a great team to (01:28:14) actually design the system? What's most (01:28:16) important? Are they all equal? Yeah, all (01:28:19) those things matter, right? I think the (01:28:21) main thing that matters more than (01:28:22) anything else is just time. Like we're (01:28:25) figuring things out with research. (01:28:27) Infrastructure is getting better. I (01:28:29) think that a lot of it's just (01:28:30) infrastructure. My new company's AI (01:28:31) infrastructure, right? Like the (01:28:33) infrastructure gets better. Um my uh my (01:28:37) uh coworker has a saying is like uh what (01:28:40) we do is that we make the uh hard things (01:28:44) easy and the impossible things hard. And (01:28:48) that's like the goal of infrastructure, (01:28:49) right? You build infrastructure, your (01:28:50) infrastructure gets better. And then (01:28:52) what was what you couldn't even dream of (01:28:53) doing 10 years ago is now one command (01:28:55) today. And today, you know what what you (01:28:58) you you (01:29:00) uh Yeah. What's the current use case for (01:29:03) most people with tiny boxes? (01:29:07) Is that that's by design, right? You're (01:29:09) not supposed to know. But I mean, so I (01:29:11) sell the computer. It has specs, right? (01:29:12) Like so many people want to tell you, (01:29:14) and I hate this. I hate this. They're (01:29:15) telling you like how the product is (01:29:16) going to impact your life or what you (01:29:18) can use the product for. Oh my god, who (01:29:21) cares? Yeah, here's what it is. I'm (01:29:24) going to tell you what it is. That's (01:29:25) your job, right? I'm not an advertiser, (01:29:29) but I mean I mean our intern wants to (01:29:32) build something with a tiny box. I want (01:29:33) to give him some ideas. Go buy one. I (01:29:35) don't know. Why do you want to build (01:29:37) something with a tiny box? I mean, is it (01:29:38) good? Yeah, it's just a It's a bunch of (01:29:40) GPUs in a box, you know? It's nice to (01:29:41) know boxes. GPUs in a box weight to it. (01:29:44) What um it what robotic form factors are (01:29:48) you most bullish on? We've touched (01:29:49) humanoids. You gave a a great review (01:29:52) there. We've touched autonomous (01:29:54) vehicles. Sounds like generally bullish, (01:29:56) but Capital Wars, Race to the Bottom, uh (01:29:59) all that stuff. Are are there any other (01:30:01) kind of form factors that you're (01:30:02) thinking about that you are generally (01:30:05) optimistic or excited about? ARM. Arm (01:30:09) the two arm, right? Just two arm, right? (01:30:12) Because I look I run a factory. I run a (01:30:14) factory in San Diego. We make all the (01:30:15) commas right here. And I can't wait to (01:30:17) get a whole lot of robots in there. But (01:30:19) uh I don't need humanoids. I'm just (01:30:21) going to stick two arms to the table. (01:30:24) And then it's going to grab a comma. (01:30:25) It's going to put the screen on the (01:30:27) front. It's going to flip it over. It's (01:30:28) going to put the four screws in it. And (01:30:30) then it's going to pass it on. Yep. (01:30:31) Right. Show me anything that's anywhere (01:30:33) near that level today. Yeah. What what (01:30:36) uh what would you do if you were trying (01:30:37) to build like a truly multi-purpose (01:30:40) robotic arm? The arms are already good (01:30:42) enough. It's the AR offtheshelf arms are (01:30:44) fine. It's all software. Again, it's (01:30:46) always all software. Autonomous vehicles (01:30:48) are all software. Robotics is all (01:30:49) software. But everybody loves to bike (01:30:52) shed. Yeah. What color are we going to (01:30:54) paint the humanoids, you know, like like (01:30:56) let's have a great conversation about (01:30:58) that. Well, you know, we don't want to (01:30:59) paint them red because that might scare (01:31:01) people in a terminator. This is actually (01:31:03) the level of stupidity that I see in (01:31:05) most discussions about human robots. (01:31:06) Yeah. Would you uh would you trade in (01:31:09) your legs for wheels if you could? I got (01:31:12) that. I trade in my legs for wheels. (01:31:14) Yeah, this is a question from Aaron (01:31:16) Frank, friend of the show. He's asking (01:31:18) this in real time. Weird wheel guy then. (01:31:20) Like people had asked me if the wheels (01:31:21) are like making a statement. I just (01:31:23) don't want to have to have this (01:31:24) conversation. What What What about the (01:31:27) sim tore gap in robotics? Like how how (01:31:29) is how is simulated data? You know, you (01:31:32) you build a bunch of data in Unreal (01:31:34) Engine, then you try and transfer learn (01:31:35) it back. Uh obviously there's been a (01:31:38) bunch of experiments of that with (01:31:39) self-driving cars. Is that a path that (01:31:41) we should be going down for the for the (01:31:44) robotic arm development? Yeah. So I (01:31:47) think with a lot of sim to real stuff, (01:31:49) the reason people are excited about it (01:31:50) is because of that data efficiency gap (01:31:52) we spoke about, right? Like current (01:31:54) machine learning algorithms like a (01:31:55) thousandx less data efficient than (01:31:57) humans. Uh so yeah, you're going to need (01:31:59) a thousandx more data, right? If a human (01:32:01) can learn something in one example or 10 (01:32:03) examples, the computer is going to need (01:32:05) a th000 or 10,000. Now, do you really (01:32:07) want to reset the stupid state of the (01:32:09) physical world 10,000 times? You might (01:32:11) do it 10, but you're not going to do it (01:32:12) 10,000, right? So, that's where you want (01:32:14) to simulator where you can just click (01:32:15) reset and everything's back to exactly (01:32:17) how it was. Um, so I think this stuff's (01:32:19) going to play a role, but I think more (01:32:21) fundamentally that data efficiency gap (01:32:24) has to be understood. (01:32:26) Uh, we talked a little bit about coding (01:32:29) agents. We talked about how you're (01:32:30) bankrupting OpenAI by spinning up a lot (01:32:33) of different uh codec agents. Um what uh (01:32:36) what other a sort of agentic software (01:32:40) are you excited about? Do you expect to (01:32:42) uh you know what's a gentic mean? Yeah, (01:32:45) basically bots. It's it's like what (01:32:48) we're calling bots now. Um but but but (01:32:50) anyways like I just you know from Star (01:32:53) Trek. Yeah. Yeah. Maybe. No but but I (01:32:57) think about a world in the future you Do (01:32:59) do you expect to be I don't know if (01:33:00) you're a a Slack guy, an iMessage guy, (01:33:03) Discord, maybe no messaging at all, just (01:33:05) you know um you know telepathy (01:33:08) telepathy. Uh but but do you expect a (01:33:10) world in the future where you're just (01:33:11) you know a perfect interaction between (01:33:14) you know human employees and agents or (01:33:18) is it you know going to be more like you (01:33:20) know you'll do the odd deep research or (01:33:22) maybe you send some automated outbound (01:33:24) emails or have some codeex bots running? (01:33:27) I don't even like follow this. When do (01:33:29) you think you'll be able to book a (01:33:30) flight just by saying I'm trying to get (01:33:32) to New York tomorrow? Oh, see the worst (01:33:34) part about this is like like and that's (01:33:37) going to come pretty soon actually. (01:33:39) Okay. Right. We're going to pretty soon (01:33:40) have computer use models that are (01:33:42) actually capable of going to delta.com (01:33:43) and booking a flight. Yeah. But then (01:33:44) what's actually going to happen is (01:33:46) Delta's going to partner with whatever (01:33:47) company does that and they're going to (01:33:49) put it behind the stupid uh thing and (01:33:51) like Yeah. So that's gonna Yeah. That's (01:33:53) going to be here in a few years, right? (01:33:54) Not with agentic [ __ ] but just with (01:33:56) normal hooking the APIs together, right? (01:33:58) Yeah. Wait, so yeah, what is the bullish (01:34:01) on APIs? What is the mistake about like (01:34:03) the agentic buzzword? Like what what are (01:34:06) people like even describing? Again, it's (01:34:07) another thing that I I really have no (01:34:09) idea what it means. You know, I I was (01:34:10) hanging out with some friends last night (01:34:12) and like like uh my friend in this VR (01:34:14) company and the you know, the the CEO is (01:34:17) really interested in things being open (01:34:18) source, but he's also really interested (01:34:20) in making sure that things are (01:34:22) protecting our intellectual property and (01:34:24) proprietary. And the truth is he has no (01:34:27) idea what the word open source means. He (01:34:29) has no idea what it means that they can (01:34:31) copy his [ __ ] right? Like that someone (01:34:33) else could use it. He just he just heard (01:34:34) the word open source in some like (01:34:36) buzzword thing and he's like, "Do we (01:34:38) have the open source? Do we have the (01:34:40) open source in the thing? Okay, so check (01:34:42) the box. Check the open source box. (01:34:44) Let's protect our IP. Last question (01:34:47) about the a last question about the (01:34:49) agentic buzzword. I think that there is (01:34:51) something that people are picking up on, (01:34:53) which is that these models seem to be (01:34:55) very smart for short amount of time, but (01:34:57) if you run them for a long time, they (01:34:59) start hallucinating and kind of going (01:35:00) off the rails. And so you you have like (01:35:02) 10-minute AGI. feels incredible, but as (01:35:06) you let it run and do more work, you (01:35:08) can't just say, "Hey, go do a week's (01:35:10) worth of work. Come back to me when (01:35:11) you're" But it's superhuman in one (01:35:13) minute. And so, is that kind of (01:35:15) trade-off curve real? And then is it (01:35:17) just a matter of like better harnessing (01:35:19) to actually get to two hours of work, (01:35:21) which is kind of what the agentic people (01:35:23) are like advocating for. No. So, I don't (01:35:26) think it's better hard, but this is (01:35:27) definitely a real phenomenon. This is (01:35:28) definitely a real phenomenon. Uh, you (01:35:30) can experience this. There's papers (01:35:32) exploring it which show that if in 10 (01:35:35) seconds there's absolutely no way I'll (01:35:37) come even close to a modern LM totally (01:35:41) um because the first shot from the LLM (01:35:43) is great. Yeah. And then it kind of (01:35:45) degrades and it degrades pretty quickly (01:35:48) whereas humans look a lot more like (01:35:51) this. Humans can stay coherent (01:35:53) internally for much longer. (01:35:56) Um so yeah I I think that that's a real (01:36:00) thing. I think that that's mostly going (01:36:02) to be fixed by like long context, just (01:36:06) more energy. Long context RL. Yeah, just (01:36:10) like you just got to do it. We'll figure (01:36:11) out new ways to make the context better. (01:36:14) We'll combine diffusion and uh and auto (01:36:18) reggression in some clever ways. Yeah, I (01:36:21) think that this is just going to be uh (01:36:22) like there's not going to be a (01:36:23) breakthrough here. There's not like one (01:36:25) magical thing that we're missing. Yeah, (01:36:27) I think it will be a continued plot. The (01:36:28) same thing with data efficiency. I think (01:36:30) people will start to care about it. Some (01:36:32) new tricks will come out. Some of them (01:36:33) will work, some of them won't work. (01:36:35) We'll continue to do graduate student (01:36:37) descent until we find a (01:36:41) anything that's last question for me. (01:36:44) Anything that you're particularly (01:36:45) optimistic about? Anything? You check (01:36:47) the timeline and you think, "This is (01:36:48) awesome. I love this. I love this. I (01:36:51) want to see more of this." A little (01:36:53) maybe a little white pill to kind of cap (01:36:54) it off. Yeah. So, here's something I'm (01:36:57) optimistic about. That fact that the one (01:37:00) server can't run 10,000 users, that is (01:37:04) most of the reason that the modern (01:37:05) internet that that is one of the reasons (01:37:07) that the modern internet sucks. (01:37:09) That that that so much of the stuff is (01:37:12) in non-recurring expense and then it (01:37:14) becomes really really hard to compete (01:37:16) with these people, right? Like you could (01:37:18) run Twitter on one computer. Yeah. (01:37:20) Right. and 20 people could do it too. (01:37:22) But like they don't because again these (01:37:25) companies have moes and they invest in (01:37:27) making sure that their modes can't be (01:37:28) broken. Um with AI I think there's going (01:37:31) to be a much less of a mo especially (01:37:34) when you look at the move from auto (01:37:35) reggression to diffusion. So auto (01:37:37) reggression can run in large batch (01:37:38) sizes. When you run chat GBT you're (01:37:40) running with a whole bunch of other (01:37:41) people on that same computer. Yeah, it's (01:37:44) only 100. It's not 10,000 but still it's (01:37:46) 100. Diffusion is run in the cloud at (01:37:48) batch size one. And once you're in batch (01:37:51) size one land, running it locally starts (01:37:53) to make sense. (01:37:55) Actually running the models locally or (01:37:57) at least having your own computer in the (01:38:00) cloud. Yeah. Not being some shared (01:38:02) resource that's really controlled by (01:38:04) some (01:38:06) else. (01:38:09) Um so yeah, this was never a thing (01:38:11) because you can't put lots of people on (01:38:12) a GPU. That makes sense. They tried some (01:38:15) weird stuff with the licensing, but (01:38:16) yeah. Fantastic. Well, thank you so much (01:38:19) for stopping by. This is a great (01:38:20) conversation. Yeah, I wish we had a full (01:38:21) hour. This is great. We'll talk to you (01:38:22) soon, George. George. Cool. Bye. See you (01:38:24) later. Cheers. Bye. Next up, we have uh (01:38:27) Joseph Troian, (01:38:29) author of The Party's Interests Come (01:38:31) First. He was recommended by Jordan (01:38:33) Schneider of China Talk. We're very (01:38:35) excited to talk to him about uh the life (01:38:37) of Cinping's father, Ciong. Welcome to (01:38:41) the show, Joseph. Good to have you. (01:38:43) Thank you for Thanks so much for coming. (01:38:46) It's great to have you. (01:38:48) I would love to get a little bit of your (01:38:49) background how you landed on this topic. (01:38:51) It's incredibly difficult to uh to to (01:38:55) research. I was uh digging just just (01:38:57) into the life of uh Cinping and uh there (01:39:00) were no English biographies for a long (01:39:03) time. One of the most uh deepest dives (01:39:06) on Cinping was from the economist (01:39:07) actually in the form of this this (01:39:09) podcast and and this reporting about the (01:39:11) prince uh and and I was interested how (01:39:14) you got into this what your background (01:39:15) is and then we could go through some of (01:39:16) the story. So, I had just finished a (01:39:19) book about elite power struggles in the (01:39:21) Soviet Union and China after the deaths (01:39:23) of Stalin and Mao. And someone asked me (01:39:26) to write about party history and (01:39:28) Xiinping. And I thought I would do a (01:39:29) short article about Xiinping, a little (01:39:32) bit about his father. But what I did (01:39:34) find out was the more research I did, (01:39:37) the more I could learn and that I could (01:39:39) tell a really interesting story through (01:39:40) Xi Jun that wasn't just about Xiinping (01:39:43) but could be a sort of microcosm of (01:39:45) Chinese Communist Party history in the (01:39:47) 20th century. Got it. Um, one of the (01:39:50) themes that I've been kind of wrestling (01:39:51) with and my takeaway from studying (01:39:53) Cinping was that uh there is this (01:39:56) constant narrative of being both a a a (01:39:59) victim and a perpetrator of party (01:40:03) aggression essentially. And I was (01:40:05) wondering if that narrative feels right (01:40:07) to you, where that comes from, why (01:40:10) there's so why why these these leaders (01:40:13) are so reluctant to reject the system (01:40:17) that's sending them to jail or putting (01:40:19) their family in hardship. Like it feels (01:40:22) unique and it doesn't feel like (01:40:24) something that happens in America. Or (01:40:25) maybe I'm just brainwashed by American (01:40:28) politics or something and it is (01:40:29) happening over here and I'm just not (01:40:30) picking up on it. But is that a theme (01:40:32) that you picked up on and and and you (01:40:34) think is like worth digging into? Is it (01:40:35) is it what's unique about the story? It (01:40:38) sounds like you read my book quite (01:40:40) closely. That is certainly one of the (01:40:41) themes. There's a puzzle there which is (01:40:44) Xi Jangun, the father of Xiinping, was (01:40:46) persecuted by his own party on several (01:40:49) occasions and the party asked him to do (01:40:52) things that he thought were wrong. (01:40:54) Nevertheless, he remained devoted. And I (01:40:56) think to appreciate that we need to look (01:40:58) at this bolevik communist political (01:41:00) culture. And it's because these people (01:41:04) see the party as a source of meaning and (01:41:06) purpose in their lives. So in that sense (01:41:09) to reject the party would have meant (01:41:10) rejecting themselves. So perhaps (01:41:13) counterintuitively when the party hurt (01:41:15) them the motivation was to redouble (01:41:17) their efforts to win back the party's (01:41:19) trust in them. (01:41:21) Do you think that there's I'm I'm I'm (01:41:23) I'm still struggling to understand the (01:41:25) dynamic of like how dominant the single (01:41:30) party is in China. And I'm wondering if (01:41:32) if we compare it with American politics, (01:41:34) we it feels like we're always in this (01:41:36) 50-50 stalemate where every four years (01:41:39) it's like it's neck andneck and then and (01:41:42) then yeah, maybe one party wins by some (01:41:44) sort of margin, but I'm I've always (01:41:46) wondered if that's a if you talk to (01:41:48) somebody who's like a socialist, they're (01:41:50) always really upset because they're (01:41:51) basically like, well, they're both (01:41:53) capitalists who are running. uh and so (01:41:55) they see it as as a false choice, but (01:41:57) the vast majority of the American (01:41:59) electorate is incredibly animated by the (01:42:01) by the differences even though they (01:42:03) might be somewhat minor if you zoom out (01:42:05) a little bit. Um and I'm wondering like (01:42:08) the the the more single party system (01:42:11) versus the dual party system. How does (01:42:13) that emerge? Can you give me some some (01:42:15) history of the party and and why is it (01:42:18) so stable? (01:42:20) So Lenin said that what we are doing is (01:42:22) building a party of a new type and he (01:42:26) thought that western political parties (01:42:27) as you said were essentially (01:42:29) representative of dominant class (01:42:31) interests. What Lenin wanted to do was (01:42:33) create an organizational weapon and the (01:42:36) reason for that was he was running a (01:42:38) conspiracy and he wanted to take over (01:42:40) the country and then use violence to (01:42:42) transform it into something that was (01:42:44) completely different. And to do that, (01:42:46) you need to create an institution that (01:42:48) can force people to do things that they (01:42:51) might not want to do. It's by design (01:42:53) constructed so that the top leader can (01:42:56) make a choice and then everyone has to (01:42:58) follow along with that choice. And (01:42:59) sometimes that choice is wrong uh and (01:43:02) has devastating impact for the party and (01:43:04) the nation. But you can see that if you (01:43:06) have such an ambitious agenda, what you (01:43:09) want to do is create a system where the (01:43:11) top leader is sort of firewalled from (01:43:13) any kind of political consideration. (01:43:15) Mhm. Can you I mean there's so many (01:43:18) there's so many anecdotes that could (01:43:20) kind of tell the story. Would you mind (01:43:22) telling me which story from Xi Jean (01:43:25) Chung's life really stuck out to you as (01:43:27) as emblematic or or maybe uh even just (01:43:31) give me the high level because we kind (01:43:32) of just jumped into it. give me like the (01:43:34) highlevel story arc that you decided to (01:43:36) uh to weave the entire narrative (01:43:38) through. Yeah. So he was born in 1913. (01:43:42) That was only two years after the (01:43:43) collapse of theQing dynasty. He was born (01:43:46) into a rather dramatic place. It he was (01:43:50) born near where the Turkata soldiers (01:43:52) are, which probably many of your viewers (01:43:54) know about. That was where Chin Shaang (01:43:56) forged the first unified Chinese state (01:43:58) thousands of years before. And it was (01:44:00) the city where emperors had ruled for (01:44:02) for for millennia. Uh but by the time Xi (01:44:06) Jang was born, it had fallen into years (01:44:08) of banditry and war and famine. And he (01:44:12) was attracted to radicalism, trying to (01:44:14) figure out a way to uh help China escape (01:44:18) from these uh imperialist encroachments (01:44:20) and uh political infighting at home. And (01:44:24) his first uh political act was an (01:44:26) attempt to assassinate an academic (01:44:28) administrator. and it failed. He got a (01:44:30) bunch of teachers sick uh and was thrown (01:44:33) into prison and joined the party there. (01:44:34) Was he again? He wasn't he like 15 when (01:44:36) he was sent out. That's right. He was (01:44:38) very young. And what's interesting too (01:44:39) is he didn't really have an intellectual (01:44:41) attraction to communism. Yeah. And when (01:44:44) he was released from incarceration, he (01:44:48) read this novel uh that's a terrible (01:44:50) novel. uh it's basically just a (01:44:52) protagonist who goes from one disaster (01:44:54) to another, but it fetishizes this idea (01:44:56) of resistance and and struggle. And so (01:45:00) it speaks to the fact that uh you know (01:45:01) for a lot of these people their (01:45:04) intuition was that something needed to (01:45:05) be changed, but they weren't exactly (01:45:07) sure what and the party was a source of (01:45:08) meaning for them. H um I I I know you (01:45:12) mentioned that he had a lot of (01:45:14) interaction with the party around uh (01:45:17) relationships with other religions and I (01:45:19) was wondering if you could if you could (01:45:20) kind of like when I grew up I remember (01:45:22) like the free Tibet movement and it's (01:45:24) kind of like fallen by the wayside. (01:45:26) Tibet's not really in the in the in the (01:45:29) news much anymore. Could you take me (01:45:31) through the story of his interaction (01:45:33) with some of these uh kind of tangential (01:45:36) groups in China and uh and how how the (01:45:39) history of those relationships played (01:45:40) out? Yeah, I'm glad you asked that. So, (01:45:43) within the Chinese Communist Party, you (01:45:45) would have people that were experts in (01:45:47) the military on economics. Xi Jun was (01:45:49) the leading United Front figure. So, (01:45:52) what's the United Front? Well, Mazudon (01:45:54) called it one of the CCP's magic (01:45:57) weapons, and it's basically a political (01:46:00) influence campaigns. And so, what you do (01:46:02) is you figure out who's on your side, (01:46:03) who's in the middle, who's against you, (01:46:05) and you empower the ones uh that like (01:46:07) you, you win over the ones that aren't (01:46:09) sure, and then you hurt the ones that (01:46:11) are that are deadenders. And this was (01:46:13) something that Xi Jang continuously (01:46:15) applied to uh China's ethnic minorities. (01:46:18) And so he was in the last years of the (01:46:21) civil war and the early years of the (01:46:23) people's republic the leader of the (01:46:24) so-called northwest bureau which (01:46:28) included a huge expanse of China (01:46:30) including Shing Jang but there were also (01:46:32) lots of Tibetans in Shing Hai and Gansu (01:46:34) which were other uh massive provinces uh (01:46:37) in China and so he was trying to figure (01:46:38) out how to incorporate them and it was (01:46:40) definitely bloody but he also wasn't (01:46:42) blind to the advantages of finding local (01:46:44) power brokers and winning them over. Um (01:46:47) but over the 1950s the party decided (01:46:49) that that approach was dated because (01:46:51) people weren't coming to socialism on (01:46:53) their own. And so essentially um uh the (01:46:56) party declared war on them with tanks (01:46:58) and planes and um during the cultural (01:47:01) revolution any sign of an ethnic (01:47:02) difference was seen as class struggle. (01:47:04) And then in the 1980s the party (01:47:06) understood that they had really screwed (01:47:08) up and that they needed to have a new (01:47:10) approach. And Xiang ran uh ethnic (01:47:13) politics for the secretariat which is (01:47:15) sort of the party's brain. and he tried (01:47:17) to use economic development, bringing (01:47:19) religion out into the open so it could (01:47:20) be better controlled, (01:47:22) uh, allying with local power brokers, (01:47:25) but by the end of that decade, uh, many (01:47:29) people within the party concluded that (01:47:30) when you do that and you give people (01:47:31) space, they just use it to hurt you. Uh (01:47:34) and so the party decided that growing (01:47:36) protests weren't a sign of uh uh (01:47:40) reaching a new equilibrium but uh you (01:47:42) know hitting some road bumps on the way (01:47:44) but that uh that fundamental approach (01:47:46) had been a mistake. (01:47:48) Yeah. Yeah, there's this there's this (01:47:49) narrative that keeps popping up whenever (01:47:51) you hear about turmoil in the party uh (01:47:54) around purges, which is a term that I (01:47:59) don't I'm not really I can't really map (01:48:01) to as an American um because we don't (01:48:04) really have those, I guess. Um and and a (01:48:07) lot of this is always tied to uh there's (01:48:11) uh there's corruption and we're going (01:48:12) after corruption and every every kind of (01:48:15) major defenestration feels like it's (01:48:17) tied to corruption and I'm always (01:48:20) wondering how we are evaluating those (01:48:24) claims because at the same time in a (01:48:28) growing developing country there (01:48:30) probably is a lot of corruption and (01:48:32) there's probably a lot of people that (01:48:33) are uh banditry for example that you (01:48:35) mentioned. (01:48:36) um there are people that are taking (01:48:38) advantage of different parts of the (01:48:39) government. So the so the corruption (01:48:41) might be real, but then also corruption (01:48:43) is used as a weapon to to amass power or (01:48:47) or remove certain people from power. And (01:48:49) so can you walk me through um how (01:48:52) corruption is used as a tool? How real (01:48:56) do you think the various corruption (01:48:57) narratives have been? Maybe if there's (01:48:58) any examples of of corruption being (01:49:01) wielded as a tool for uh for party (01:49:03) power. (01:49:05) So as soon as the party was able to take (01:49:09) over the country, they immediately faced (01:49:12) a question which was how they were going (01:49:14) to ensure that they didn't take such (01:49:18) pride and arrogance in their victory (01:49:20) that they allowed bourgeoa uh (01:49:23) liberalization meaning individualism to (01:49:25) seep in and that they would be divorced (01:49:27) from the masses and that they would care (01:49:29) about uh their own materialist needs. (01:49:32) And so from the very beginning the party (01:49:34) struggled to figure out a way to (01:49:36) eliminate this problem. And so (01:49:38) especially during the Mao era you had (01:49:40) constant constant rolling campaigns that (01:49:42) inevitably went too far. Uh and now (01:49:45) under Xiinping you see that he believes (01:49:47) that corruption for him is is about (01:49:51) whether the party can survive because (01:49:53) for Xiinping ideals and conviction and a (01:49:55) sense of motivation are necessary for (01:49:58) the party to exist. And the biggest (01:50:00) danger to that is corruption because (01:50:02) it's uh it basically means putting (01:50:05) yourself first but also corruption is a (01:50:07) problem because it's a vector for (01:50:09) western influence because there's this (01:50:11) idea that the west is materialistic and (01:50:14) uh oriented towards consumption and once (01:50:16) that gets into China then uh the west (01:50:20) can use it as a weapon and in fact many (01:50:22) people in China thinks that what (01:50:24) happened to the Soviet Union uh that (01:50:26) essentially uh the leaders of the regime (01:50:28) became oriented towards the West and um (01:50:31) the uh the west won what they call a war (01:50:33) without gunpowder. Mean meaning you (01:50:35) don't use a you don't use violence to (01:50:37) destroy the regimes you don't like. You (01:50:38) do it by winning over certain people uh (01:50:40) within it. And the one other thing uh to (01:50:43) say too here is that you're right like (01:50:44) corruption is partly about regime (01:50:46) security, but it's also about getting (01:50:48) rid of people that you don't like. And (01:50:49) so what's interesting here is, you know, (01:50:52) why do people keep uh doing things that (01:50:55) make them vulnerable to accusations of (01:50:57) corruption? And here I think speaks to (01:50:59) something else with um about the system (01:51:01) which is nobody's quite quite sure where (01:51:02) the red lines are. Uh and so sometimes (01:51:06) you just get it wrong and sometimes the (01:51:07) top leader changes where the red lines (01:51:09) are and you can see that um in a system (01:51:11) where the top leader is the one who gets (01:51:12) to make those decisions, it's a very (01:51:14) powerful weapon. Yeah. What is (01:51:18) what is the current sentiment (01:51:21) uh from the party about the state of the (01:51:24) world? Right. I I feel like here in (01:51:26) America right now, you have a (01:51:27) frustration. China's become the factory (01:51:29) of the world. We're we're deeply reliant (01:51:33) on them. Uh the at the same time, the (01:51:36) world feels very uncertain consumerrist (01:51:39) because they're certainly buying Xiaomi (01:51:40) phones and and uh you know. Yeah. So, so (01:51:43) I'd be curious like any type of insight (01:51:46) around the party's sentiment around (01:51:47) what's happening in China culturally and (01:51:49) then sort of geopolitically on a on a (01:51:52) world stage. do they feel like they're (01:51:54) making progress? You know, we we study, (01:51:56) you know, their goals around individual (01:51:58) industries like semiconductors, AI, (01:52:00) defense, etc. Um, but I'm curious what (01:52:03) what you know sentiment in in and out of (01:52:05) the country. So, it's interesting to see (01:52:08) how ambitious Xiinping is in terms of (01:52:11) how he characterizes his goals in a (01:52:14) holistic sense. He says that for (01:52:16) millennia you would have one dynasty (01:52:18) collapse after another and that the (01:52:21) Chinese Communist Party needs to break (01:52:23) that. And his solution is this idea of (01:52:25) self-revolution. And basically what he (01:52:29) wants to do is figure out a way of how (01:52:31) you win over the third and fourth (01:52:32) generation (01:52:34) uh of young Chinese uh to the (01:52:36) revolutionary cause. And your questions (01:52:38) speak to a dilemma there, right? like (01:52:40) you can see how a message of sacrifice (01:52:43) and rejuvenation is meaningful maybe for (01:52:46) many people but that he also recognizes (01:52:48) that economic development uh is (01:52:51) essential as well. So, you know, how you (01:52:53) balance those two things at the same (01:52:55) time is not clear and it's something (01:52:57) that I think uh the party is has (01:53:00) constantly wrestled with from the very (01:53:01) beginning. How you balance ideals and (01:53:04) conviction and motivation with being (01:53:06) practical and thinking about economics (01:53:08) and that kind of thing. You know, in (01:53:10) terms of how China thinks about uh the (01:53:12) world, uh they see real inherent (01:53:15) strengths in their system. And even (01:53:17) though they're communists and therefore (01:53:19) should be m uh people who only think (01:53:21) about the world in terms of um uh (01:53:24) concrete uh objective conditions, they (01:53:27) talk about spiritual civilization all (01:53:28) the time. And in fact, they think that (01:53:30) China has one, but the west does not (01:53:32) because in capitalist societies, all (01:53:34) they do is care about money and (01:53:35) consumption. So in Xiinping's mind, he (01:53:38) sees those problems only getting worse (01:53:40) and worse in the west as opposed to (01:53:41) China where in his mind they can tell a (01:53:44) good story, a motivating story, but also (01:53:45) the party can organize interests in a (01:53:47) way that doesn't allow one class or one (01:53:50) group to dominate. How does that (01:53:52) actually play out? Because I feel like (01:53:54) we there's incredible amount of wealth (01:53:57) still in China like like there's still a (01:53:59) wealth inequality. Um there are many (01:54:02) tech billionaires and massively (01:54:04) successful folks over there. Of course, (01:54:07) when they get really really big, they (01:54:08) tend to disappear for a little bit and (01:54:10) it doesn't seem like a great place to be (01:54:12) a Jeff Bezos type. But at the same time, (01:54:15) it doesn't feel like oh yes, like they (01:54:17) are truly communist. Everyone has the (01:54:19) exact same standard of living. And so is (01:54:21) that even the goal? Are they okay with (01:54:23) some level of wealth inequality? And or (01:54:25) or do they see it as a failure? and and (01:54:27) and even that digging a level deeper if (01:54:30) if she and and party leaders can speak (01:54:33) to the sort of civilization, you know, (01:54:35) the spirit of the the Chinese (01:54:37) civilization and and how great it is and (01:54:40) how the the system of the west is (01:54:41) failing. What do three, you know, five (01:54:45) levels below that in in in the Chinese (01:54:47) system in terms of, you know, how much (01:54:49) do you try to dig in and, you know, what (01:54:51) is the average factory worker? Do they (01:54:53) feel that same sense of of pride in in (01:54:56) the in the spirit of China or is there (01:54:58) kind of a disconnect? (01:55:01) Yeah, that's a really those are some (01:55:02) really good questions. So I think for (01:55:05) Xiinping (01:55:06) he (01:55:08) can do course corrections to a certain (01:55:10) extent, right? So this is someone who a (01:55:14) few years ago was really doing a very (01:55:16) severe crackdown uh in the tech sector (01:55:19) and people who made a lot of money were (01:55:22) suddenly very worried. Uh Xiinping was (01:55:24) talking about common prosperity. (01:55:26) uh but over the last few months uh they (01:55:30) have been talking more about the economy (01:55:32) and it seems that Xiinping has empowered (01:55:33) his premier uh to focus on development (01:55:37) and that doesn't mean that they've (01:55:39) suddenly stopped caring about security (01:55:41) or ideology. It just means that (01:55:44) they can move around a little bit uh (01:55:47) without admitting that they were wrong. (01:55:49) And I think the reason for that is that (01:55:51) the party when they talk about ideology, (01:55:53) they're more careful I think than people (01:55:55) uh give them credit sometimes, right? So (01:55:57) for example, if you look at the latest (01:56:00) uh party congress report, it begins with (01:56:01) ideology and it talks about how we need (01:56:03) to believe in this stuff and we still (01:56:05) are, you know, socialist and communist. (01:56:06) But then it says other things too. It (01:56:08) says the reason communism works in our (01:56:10) country but failed in other others is (01:56:11) that we cynicize it. We made it Chinese, (01:56:13) which basically means we made it say (01:56:15) whatever uh we want it to be, right? And (01:56:17) so then you also see other language (01:56:19) about markets still being decisive and (01:56:22) China still being in the stage of (01:56:24) primary accumulation which means that (01:56:26) development not restructuring (01:56:28) um uh society is the top priority. So (01:56:31) you know they talk about common (01:56:32) prosperity because they know that uh (01:56:35) they need to tell a good story about (01:56:37) getting rid of inequality because they (01:56:38) subscribe to socialism. Then you have (01:56:40) all these buts and ants, right? So they (01:56:42) say, you know, we're not going to become (01:56:43) populist like Latin American countries (01:56:45) by allowing people to have such a social (01:56:48) safety net that they feel that they (01:56:49) don't have to work hard. And so for all (01:56:52) of these reasons, I think that uh uh he (01:56:55) he's trying to manage dilemmas because (01:56:58) he he can't solve problems, if that (01:57:00) makes sense. Yeah. Yeah. Have have you (01:57:02) uh um Yeah. I mean, we've been talking (01:57:05) to people about how uh some of China's (01:57:08) investments in the semiconductor (01:57:10) industry are in some ways more (01:57:12) capitalist than the way America (01:57:13) subsidizes our our like the chips act (01:57:16) compared to the the series of five-year (01:57:18) plans that have been somewhat cutthroat (01:57:20) over in China between a whole bunch of (01:57:22) different companies competing for (01:57:24) grants. And yes, there's a lot of (01:57:25) government money pump pumping into the (01:57:26) sector and they've been able to stay at (01:57:28) the lagging edge for years, but it's (01:57:30) been uh designed to be kind of doggy dog (01:57:34) and hasn't just been a situation where (01:57:35) China's just picked one winner and said, (01:57:37) "Here's, you know, $50 billion and go (01:57:40) run with it." Um, I'm I'm interested to (01:57:42) hear the story about uh the development (01:57:44) of Hong Kong relative to some of the (01:57:46) mainland uh uh cities. Uh, can you can (01:57:50) you uh walk me through what that taught (01:57:53) about China and Sei Jong Shun because I (01:57:56) believe he saw kind of that happened. (01:57:58) I've seen the time lapses uh and I've (01:58:00) I've been to Guanghou once very briefly (01:58:03) uh and and and the balancing act there (01:58:05) between those those various cities and (01:58:07) what that kind of taught uh taught (01:58:09) China. I'm really glad you've asked (01:58:11) about Hong Kong. I've done a handful of (01:58:13) interviews and you're the first person (01:58:14) to raise it and it's it's a quite an (01:58:16) interesting story and it's revealing in (01:58:18) many ways. So, Xi Jong Fin, he was (01:58:20) persecuted for 16 years, right? Uh and (01:58:23) then he's released and the first job (01:58:26) that he gets is the party boss of (01:58:28) GuangDong, which of course is the (01:58:29) province that borders Hong Kong. and he (01:58:32) saw physically just how far behind China (01:58:35) had become. And his first (01:58:39) task that he had to deal with were the (01:58:43) thousands and thousands and thousands of (01:58:46) Chinese people who were fleeing to Hong (01:58:47) Kong, often losing their lives in the (01:58:49) process. So, it was such a striking (01:58:51) physical manifestation of how far the (01:58:53) socialist motherland uh had had had (01:58:57) fallen behind. And so for him uh that (01:59:00) helps explains the special economic (01:59:02) zones which uh he played a role in (01:59:04) establishing and you know he has this (01:59:06) idea that one of the problems is that is (01:59:08) ideology. We need to make people you (01:59:10) know believe in the cause but he wasn't (01:59:11) so foolish as to not recognize that you (01:59:13) also had an economic angle here that you (01:59:15) needed to figure out. And then ensuing (01:59:18) years uh in Beijing when he was running (01:59:20) the United Front, he had to figure out (01:59:22) how the people in Hong Kong were going (01:59:26) to feel about returning to the mainland. (01:59:28) So he kept meeting with these Hong Kong (01:59:29) people uh during the handover (01:59:31) negotiations and he would say things (01:59:33) like, you know, we're all Chinese. Uh go (01:59:35) overseas and look around and you'll (01:59:37) realize that actually uh Western (01:59:38) countries aren't as great as you think. (01:59:40) You put your money in a bank and the (01:59:42) fees are greater than any interest that (01:59:43) you would make. you know, these kind of (01:59:45) like sort of very conventional communist (01:59:48) views of of of of boom and bust and that (01:59:50) uh uh he uh but you know what's (01:59:54) interesting is for them like in in Hong (01:59:56) Kong uh one of them during these (01:59:58) meetings a lawyer from Hong Kong said (02:00:01) you know you in the mainland you feel (02:00:02) like you're coming out of a tunnel (02:00:03) because the cultural revolution is over (02:00:05) and you're reforming but we in Hong Kong (02:00:06) we feel like we're going into a tunnel (02:00:08) um because we're uh we're looking at (02:00:11) unification with you but we're not sure (02:00:13) about what direction you're going and (02:00:14) and were frightened. And it was it was (02:00:16) Xi Junction's job to win their trust and (02:00:19) faith to this uh um to to the handover. (02:00:22) Interesting. Um Xi is up for reelection (02:00:25) in 2028. There's no term limits. what (02:00:29) what should be people be looking out for (02:00:32) in terms of the leadup and and what he's (02:00:36) is it the right um kind of outlook that (02:00:39) that he should be you know really trying (02:00:41) to prove like does this next few years (02:00:44) really matter in terms of proving that (02:00:46) he's the right leader for the next five (02:00:48) years beyond that or or how do you how (02:00:50) do you look at the election? So he is in (02:00:53) a situation where whatever he decides is (02:00:56) what will go. Nevertheless, he is facing (02:01:00) really really powerful challenges when (02:01:03) he looks at the succession because as (02:01:06) his father's story shows if you read my (02:01:09) book there's it's hard to think of (02:01:11) anything more explosive than succession (02:01:13) politics in an alenist regime and it (02:01:14) gets back to what we were talking about (02:01:15) earlier which is that these are very (02:01:17) lead leader friendly systems. The whole (02:01:19) purpose is to have a really really (02:01:21) powerful person at the center who can (02:01:22) make choices and then everybody has to (02:01:24) listen. So Xiinping if he if he picks a (02:01:27) successor (02:01:28) uh it raises the question of well who's (02:01:31) actually really boss and then this the (02:01:33) name successor will have to figure out (02:01:35) how much space he has and whether or not (02:01:37) he's actually getting Xiinping right and (02:01:39) he want he'll want to impress Xiinping (02:01:41) but he also won't want Xiinping to think (02:01:43) that he is getting too big for his (02:01:45) bridges but if Xiinping doesn't pick a (02:01:47) successor then uh he might be afraid (02:01:49) about what will happen to the party (02:01:51) after he dies uh and he will likely want (02:01:55) to make sure that the person that he (02:01:57) thinks is best is the person that uh (02:01:59) becomes the next leader. So, he's kind (02:02:00) of stuck, right? Because neither of (02:02:02) those uh options are very good, but um (02:02:06) and he's probably also thinking in his (02:02:07) own mind and changing his mind and (02:02:09) testing certain people and uh seeing how (02:02:11) well they read him and how well they (02:02:12) handle situations. So, it really is (02:02:14) something to watch and a lot of it will (02:02:16) be about personal chemistry. Yeah, I'd (02:02:18) love to know how hereditary dynasties (02:02:22) play into that dynamic and if you could (02:02:24) ground it by explaining the concept and (02:02:26) history of the princelings and then kind (02:02:29) of the current status of the idea of a (02:02:31) princling. I don't know if that's kind (02:02:33) of gone out of fashion or if that's (02:02:34) still uh if if if uh uh if that matters (02:02:38) today. Yeah. So in the 1980s, the party (02:02:44) as it was trying to figure out how to (02:02:46) survive after the founding generation (02:02:47) died, they looked at young people and (02:02:49) they were a little afraid because during (02:02:51) the cultural revolution, young people (02:02:53) had beaten them up, brought them to (02:02:55) struggle sessions, incarcerated them uh (02:02:58) while they were still alive, which you (02:02:59) know raised questions about what would (02:03:00) happen after they died. And a lot of (02:03:03) them had been betrayed by their own (02:03:04) secretaries. And so in the 1980s a lot (02:03:07) of them were picking princlings uh to (02:03:09) work in their offices uh because they (02:03:11) thought that they would be more (02:03:12) trustworthy. The problem with princlings (02:03:14) is that they were not wellliked more (02:03:17) broadly within the population and in (02:03:19) many circles of the party because they (02:03:20) were seen as entitled. They were seen as (02:03:22) arrogant. They were seen as benefiting (02:03:24) from their parents' status. And so they (02:03:27) had this weird sense of both entitlement (02:03:28) and vulnerability uh at the same time. (02:03:31) And Xiinping was a witness to that. And (02:03:33) that was one of the reasons I think why (02:03:35) Xiinping decided to go work in the (02:03:38) grassroots as opposed to pursue a career (02:03:40) that was purely in Beijing so that he (02:03:41) could avoid those kinds of charges. So (02:03:44) uh you know Xiinping's career was hurt (02:03:46) in many ways because he was a princling (02:03:48) although he also did benefit his father. (02:03:50) So it's a little bit of a complicated (02:03:52) story. you know, now in um uh China (02:03:54) today, I think that uh Xiinping probably (02:03:58) doesn't have a great relationship with (02:03:59) these princlings because he doesn't want (02:04:00) them to think they have special purchase (02:04:02) on him because of who their family is (02:04:04) and Xiinping thinks everyone should put (02:04:06) the party's interest first. So, uh in (02:04:08) that sense, I think that princlings (02:04:11) matter because Xiinping is a princling (02:04:12) and the fact that his background, you (02:04:14) know, tells us something about that. (02:04:16) But, uh I think many other princlings (02:04:17) feel that they don't have much of a say (02:04:19) in the direction China is going. Well, (02:04:21) that's the name of the book, The Party's (02:04:22) Interests Come First. And thank you so (02:04:25) much for stopping by. This is a (02:04:26) fantastic conversation. Yeah, we'll have (02:04:28) Thanks so much for having me back on (02:04:29) when there's when there's Yeah. Yeah. I (02:04:31) mean, we'd love to have you back and and (02:04:32) talk when there's uh anything that you (02:04:34) haven't announced, but also uh anything (02:04:36) in the news that you could comment on (02:04:37) would be fantastic. Uh but (02:04:39) congratulations on the book. Uh and I (02:04:41) highly recommend everyone go pick it up. (02:04:42) The party's interests come first. Uh (02:04:44) thanks so much for stopping by, Joseph. (02:04:45) Great questions. Thank you. Bye. Next (02:04:48) up, we have Paul from Browserbase coming (02:04:51) into the studio. Uh, he's not here yet. (02:04:54) And that and you know what that means. (02:04:55) We get to do some ads, baby. Adio (02:04:57) customer relationship magic. Adio is the (02:05:00) AI native CRM that builds, scales, and (02:05:02) grows your company to the next level. Go (02:05:05) to addio.com. (02:05:08) Also, how did you sleep? (02:05:11) Jordy, Coca-Cola, modal USV. I think I (02:05:15) got you. Got rain. Think I beat you. (02:05:17) How'd you sleep, John? 93. Only 7 hours (02:05:21) and 19 minutes, but great on quality. (02:05:23) Great. Back on my back. 97. Oh, it's a (02:05:27) disaster. A clean eight hours. Oh, (02:05:29) brutal. Okay. Well, congratulations. (02:05:31) You're you're back. You've been You had (02:05:34) a rough rough week. It was bit about a (02:05:36) 10day period that perfectly coincided (02:05:38) with being You beat me 10 weeks in a (02:05:40) row, but I beat you for one week and it (02:05:43) felt like 20 weeks of victory for me. I (02:05:46) was pumped. That's right. Anyway, I will (02:05:47) let you take the intro with Paul, get (02:05:50) the details from him on what's going on (02:05:53) in his world, in the browser based (02:05:54) world. There he is, Mr. Paul Klein IVth. (02:05:58) Welcome. Hey guys. Big uh big day for (02:06:02) you. Break it down. What's going on? (02:06:04) Good. Uh you know, series B announcement (02:06:07) today. Of course, you know, Sequin has (02:06:10) his touchdown chain. I brought the (02:06:12) browserbased chain in to the big You got (02:06:15) the big B. All right. Well, I'm gonna (02:06:17) hit the gong for the big B if if the (02:06:19) team can go to the wide. Get ready here. (02:06:24) There we go. Solid connection for you (02:06:27) and the browserbased team. Um, love to (02:06:30) see it. Love to see it. So, the chain, (02:06:32) what how do you have any sort of like (02:06:34) milestones? The chain stays on until 100 (02:06:37) million of of ARR, you know, the chain (02:06:40) is currently taped on. Uh, it falls off. (02:06:42) It's from Amazon. So, I I think that the (02:06:44) series D will upgrade to a better chain. (02:06:47) Real. Okay. Okay. So, a couple couple (02:06:49) rounds away. Um, awesome. Who who (02:06:52) participated in the round? Um, I'm I'm (02:06:55) sure you had a pretty fun process. Uh, (02:06:57) but but break it down for us. Yeah, you (02:06:59) know, we're super lucky at Browser Base. (02:07:01) Every round that we've done has been (02:07:03) preemptive. So, it it was a quick one (02:07:05) and that's because we work with people (02:07:06) we've known for a long time. you know, (02:07:07) Glenn Solomon from Notable Capital. He's (02:07:09) someone who met with us in the earliest (02:07:11) days and uh the Notable team, formerly (02:07:13) known GGV Capital, they're just killing (02:07:15) it right now. They just did features and (02:07:17) labels series B as well. And I think (02:07:18) that really started the conversation for (02:07:20) us. We're like, "Hey, sounds like you (02:07:21) guys are doing B's. We'd love to kind of (02:07:23) have a conversation." Of course, CRV (02:07:25) doubling down as well. Re rechristian on (02:07:27) our board. Uh, you know, Pler Perkins as (02:07:30) well, one of our earliest investors at (02:07:31) the seed, participation. (02:07:35) I feel like this board is a good (02:07:36) starting lineup and, uh, we can go (02:07:38) pretty far with it. So, we're excited. (02:07:40) Amazing. You guys also had a big launch (02:07:42) today. Is that right? Um, director, (02:07:45) what's going on there? Our whole thing (02:07:47) is that you know when we think about how (02:07:49) people automate the web and browser base (02:07:51) for those who aren't familiar we're (02:07:53) basically a web browser that can be (02:07:55) controlled by your AI really an (02:07:56) infrastructure company so we sell to (02:07:57) developers who are building features (02:07:59) that want to go automate the web and we (02:08:01) realized that one of the places that (02:08:03) people are starting when they're (02:08:04) building their applications is often in (02:08:05) these kind of lovable vzero or bolt (02:08:08) experiences they're vibe coding and we (02:08:11) kind of looked at the way vibe coding (02:08:12) applications were set up to help (02:08:13) developers who build like you know (02:08:15) automation scripts like submitting your (02:08:16) Delaware franchise tax or grabbing a (02:08:19) quote from a website of a supplier you (02:08:20) want to work with. And for us, we wanted (02:08:22) to build something built for the vibe (02:08:24) coders who actually want to build (02:08:25) software that will do work on their (02:08:26) behalf. And that's what director is. (02:08:28) Director.ai allows you to go to the (02:08:30) website, prompt, and it's going to (02:08:32) output this repeatable script that will (02:08:34) go to a website, click buttons, fill in (02:08:36) forms, download files all on your (02:08:38) behalf. We have a few cool prompts on (02:08:39) there as demos. One is like going to (02:08:41) Kshi and checking the poll for uh will (02:08:44) Trump unfollow Elon on Twitter. Another (02:08:46) one's like hey go to the NASDAQ website (02:08:48) and find me all the earnings calls for (02:08:50) this week. And then it outputs not only (02:08:52) it does it just does that for you and (02:08:53) then also outputs code so you can (02:08:55) integrate that into your application and (02:08:57) run that every single week in a (02:08:58) repeatable way. Very cool. George Hotz (02:09:01) was actually surprisingly bullish on (02:09:04) computer use. He's he he usually pours a (02:09:06) lot of cold water on a bunch of (02:09:08) different stuff, but he was pretty (02:09:09) pretty optimistic about uh me being able (02:09:11) to book a flight on delta.com with an (02:09:14) agent coming soon. Uh what I didn't get (02:09:16) from him and what I want to get from you (02:09:17) is where does that interaction live? He (02:09:20) was saying that yes, it's coming. You'll (02:09:22) be able to just say I'm trying to fly to (02:09:24) New York tomorrow. Book me a flight that (02:09:27) gets me there in the evening and it will (02:09:29) do it for you and it'll interact with (02:09:30) the web page or the API. The question is (02:09:33) where does that interaction live? Is (02:09:35) that in a consumer app that I uh (02:09:37) subscribe to or pay for is ad supported (02:09:40) or is that a delta.com feature or back (02:09:43) and forth like where does the where (02:09:45) where does more of the agentic (02:09:48) interaction (02:09:49) live? Yeah, I think for for consumer use (02:09:52) cases like booking a flight, it has to (02:09:54) live where the distribution is, right? (02:09:56) And that's probably going to be Google, (02:09:58) Apple um on device. your AI will control (02:10:00) the browser on your device. But if you (02:10:02) think of companies like RAMP, uh if Ramp (02:10:04) wants to go automate collecting an (02:10:06) invoice from a website where they don't (02:10:07) have a first-party connection, the way (02:10:09) they can agentically do that is with a (02:10:10) web browser, going to that website, you (02:10:12) know, getting your invoice for you and (02:10:13) putting it through their, you know, bill (02:10:14) pay. I'm a ramp customer, so you know, (02:10:16) it's something I've been asking for for (02:10:18) as a feature for a while. So I think (02:10:19) that like you know for consumer (02:10:20) automation use cases it certainly will (02:10:22) live on device or in some sort of (02:10:24) consumer app maybe chatbt but for the (02:10:26) B2B use cases where you're actually (02:10:28) using some software and the automation (02:10:29) is an extension of the existing software (02:10:32) that's where I think the the browser and (02:10:34) more generalistic computer use that (02:10:35) lives within a company like browser base (02:10:37) it's really going to stand out. Yeah, le (02:10:39) let's dig into that that ramp example uh (02:10:41) more because I' i've had to do this (02:10:42) where I've wanted to classify like an (02:10:44) Uber receipt. For some reason, I wasn't (02:10:46) subscribed to the emails and so I go (02:10:48) into the Uber website and I pull the (02:10:50) receipt down and then I upload that to (02:10:52) ramp, right? Um with browserbased, they (02:10:55) could potentially do that if they have (02:10:56) my login. um long term there's probably (02:10:58) some sort of API interaction and so I (02:11:01) almost feel like there's this weird (02:11:04) world where like you see a lot of people (02:11:06) using computer use and then once (02:11:08) something really takes off then they (02:11:09) then they automate with an integration (02:11:11) but then there's like a new wave so is (02:11:13) your growth kind of like a series of (02:11:15) scurves is that the way to think about (02:11:17) it or do you think that people will just (02:11:19) be so happy with the results from (02:11:21) computer use that they will just say why (02:11:23) would I even bother building the direct (02:11:24) API connection well I think There will (02:11:26) always be direct API connections and (02:11:28) there should be. You know, if someone (02:11:29) has appetite to build that I I highly (02:11:31) encourage it, but I think about like the (02:11:32) rest of the internet, the internet that (02:11:34) already exists. Are we going to rebuild (02:11:35) that internet? The analogy I like to use (02:11:37) and I use in this piece today with Alex (02:11:38) Conrad is that, you know, when we think (02:11:39) about Whimo, you know, Whimo drives on (02:11:41) the roads we've already built. And Whimo (02:11:43) is not good enough to navigate those (02:11:44) roads. It might be more efficient for us (02:11:46) to build highways just for Whimo or just (02:11:48) for your self-driving cars to go go out (02:11:50) and browse, but in in the end like if AI (02:11:54) can use the internet the same way we can (02:11:56) just as well as we can, why not just let (02:11:58) it browse the web like we do? Short (02:12:00) highspeed rail I'm hearing (02:12:03) because when you think dedicated roads (02:12:07) for something that's autonomous that (02:12:09) doesn't need to steer, that's a (02:12:10) high-speed rail. But yeah, uh why not (02:12:12) just be in a in a self-driving car? How (02:12:14) much are you thinking about bigger (02:12:16) partnerships long term? There will be (02:12:18) some, you know, uh, companies that will (02:12:21) not be excited about, you know, computer (02:12:23) use agents kind of, you know, (02:12:25) interacting with their web properties (02:12:27) and and at some point you might be, you (02:12:29) know, constantly going back and forth (02:12:31) with them, you know, trying to figure (02:12:32) out a way to I I don't know exactly what (02:12:34) it would look like, get get around (02:12:36) captures and and things of that nature, (02:12:38) but is that something that over time, I (02:12:40) mean, we've had Zach from Plaid on the (02:12:42) show uh many times and, you know, early (02:12:44) on they they um, you know, were sort of (02:12:47) doing things maybe manually and then and (02:12:49) then um you know with software but then (02:12:52) they ended up developing out you know (02:12:53) meaningful partnerships to enable these (02:12:56) sort of interactions to be more (02:12:58) seamless. Is that something on your road (02:12:59) map at all? I can imagine just after you (02:13:02) guys have raised you know over $50 (02:13:04) million in a very short period of time. (02:13:05) I imagine people are kind of knocking on (02:13:07) your door now at this point to have (02:13:10) conversations as well. Yeah, we're (02:13:12) certainly excited to partner with the (02:13:14) antibbot detection companies. In our (02:13:16) mind, like browser has an opportunity to (02:13:17) be an arbiter of good bots. You know, (02:13:19) for example, we may be able to help a (02:13:21) customer of ours say, "Hey, we're this (02:13:24) small startup. Here's our use case. (02:13:26) We've done KYC with browserbased. We (02:13:27) have, you know, restricted domains that (02:13:29) we can only browse to." And we could (02:13:31) help kind of represent them to the (02:13:33) broader, you know, antibbot community (02:13:35) and say, "This is a customer that we've (02:13:36) certified." And I think in the for the (02:13:38) longest time, antibiot was built because (02:13:40) there's only bad bots online. But now (02:13:42) there's good and bad bots. And on (02:13:43) browse, we hope that we can empower, (02:13:45) show, and and prove that these are good (02:13:47) bots and use the brand that we built as (02:13:49) a center part of our trust. But (02:13:51) regardless, like, you know, Plaid is an (02:13:52) amazing company. I've really looked up (02:13:54) to Plaid. It is possible to build (02:13:55) integrations for every single bank out (02:13:57) there. There's there is like a finite (02:13:59) number of banks or at least like a (02:14:00) workable or tractable number of banks. U (02:14:02) there is billions of websites out there. (02:14:04) So regardless, you still are going to (02:14:06) have to have some AI for when an agent (02:14:09) or a customer prompts your agent, hey, (02:14:11) can you work with this specific website? (02:14:13) Uh, you might not have an integration (02:14:14) built. And browser base is kind of that (02:14:16) primitive that you know the integration (02:14:18) of last resort. And we hope that if (02:14:20) you're building an agent, you include a (02:14:22) browser tool more for like the what if (02:14:23) it happens, not if for the when it (02:14:25) happens. How do you think about (02:14:26) competition coming from the hyperscalers (02:14:29) and the big cloud platforms? uh (02:14:32) companies like Stripe and Plaid haven't (02:14:34) seen much competition from GCP, Azure, (02:14:37) AWS, but on the flip side, some of the (02:14:40) other parts of the AI stack have been (02:14:43) either cloned or offered or vended in (02:14:46) through some of the bigger cloud (02:14:48) platforms. Is that a risk or or do you (02:14:51) have an idea for uh a moat that is is (02:14:55) too high for for the king of moes, Jeff (02:14:58) Bezos, to clear? (02:15:00) Well, I think it starts with a few (02:15:01) things. One, I think you have to build a (02:15:03) great developer product and that comes (02:15:05) down to actually having a dashboard of (02:15:07) features that people love to use and (02:15:08) work really well. Um, yeah, Jeff Bezos (02:15:11) is the king, but I don't think any (02:15:12) developer really raves about the AWS (02:15:13) console, right? And secondly, you have (02:15:16) to have more than just infrastructure. (02:15:17) Infrastructure always kind of becomes a (02:15:18) commodity, right? Sure. So, moving up (02:15:20) the stack one layer, that's what we did (02:15:21) with Stage hand. Stage hand is our (02:15:23) browser automation framework. When you (02:15:25) build it, you kind of write code and it (02:15:27) generates the the integration. Sorry. (02:15:29) When you write stage hand code, it (02:15:32) actually generates the browser (02:15:33) automation functionality in browser (02:15:34) base. So it's like one layer up. The (02:15:36) analogy here is like burcell and x.js y (02:15:38) browser base and stage hand. So when you (02:15:40) own the framework, you're really able to (02:15:41) offer more stickiness and just build (02:15:43) more firstparty integrations into the (02:15:45) framework that run better on (02:15:46) browserbased. And then finally, director (02:15:48) is at top of that stack, right? Well, if (02:15:50) you have a framework, you probably want (02:15:51) to have a way to generate that framework (02:15:53) code. So we moved one layer higher to (02:15:54) the application level. And I haven't (02:15:56) seen any incumbents really do well at (02:15:58) adding, you know, building developer (02:16:00) love around new frameworks. Um, and then (02:16:02) when they hire like building (02:16:03) applications to generate them yet. So (02:16:05) for me, I'm kind of betting on (02:16:06) innovation here. I'm betting on us (02:16:07) expanding our portfolio. And there's (02:16:09) this Parker Conrad tweet which was like, (02:16:11) hey, this is going to be one on the (02:16:12) field. Who can build it better? Who can (02:16:14) sell it? Who can market more? Who can (02:16:15) innovate more? And uh, we're ready to go (02:16:17) battle on the field with our new fundra (02:16:20) in the field, in the bathrooms, in the (02:16:22) slack. It's fought out everywhere. Uh, (02:16:25) we love it. Um, talk to me about the (02:16:28) good bots versus bad bots dynamic uh (02:16:30) trends in the robots.txt. (02:16:33) Is there a need for some sort of like I (02:16:37) don't know like almost universal (02:16:39) uh like trust rating or something that (02:16:43) identifies you as a good bot as you go (02:16:46) around because there's going to be lots (02:16:47) of anti-bot protections and yet I (02:16:50) imagine that having some sort of like tr (02:16:53) coming from a trusted IP address is (02:16:56) something that gets you out of a lot of (02:16:58) Cloudflare captas. Um, and and I imagine (02:17:02) that that there's probably work that you (02:17:04) can do almost on like the business (02:17:06) development side to to kind of grease (02:17:10) the wheels as you try and automate more (02:17:12) and more of the web. Is that something (02:17:14) that you're thinking about? What what (02:17:15) are the different approaches that you (02:17:17) can take to make sure that as the (02:17:20) anti-bot protections get more and more (02:17:22) robust, you don't get bogged down in (02:17:23) that and you're still offering a good (02:17:25) experience to developers? Yeah, it's a (02:17:26) great question. All um one of our (02:17:28) investors Jeff Lawson he he always (02:17:30) frames great infrastruure companies as (02:17:31) kind of being one of three things. Yeah. (02:17:33) One is like APIs representing capital. (02:17:36) So it's like hey if I want to deploy a (02:17:38) thousand servers I don't have the (02:17:39) capital to stand that up. AWS I can just (02:17:40) deploy a thousand servers. The second (02:17:42) one is algorithmic complexity. So you (02:17:44) know your inference as a service right (02:17:45) you have a bunch of inference you know (02:17:47) how can you run that more efficiently (02:17:48) than anyone else. How can you serve it (02:17:49) in a innovative way? And then the third (02:17:51) one is really bisdev as an API and I (02:17:53) think browserbased long-term (02:17:54) partnerships are extremely important to (02:17:55) us and that's why we're really proud to (02:17:57) work with companies like work OS clerk (02:17:59) stitch and octa which all have major (02:18:01) capture presences on the w online (02:18:03) because that's what protects (02:18:04) authentication. Secondly, I think the (02:18:06) credentiing actually happens when you (02:18:08) log in. Like that's how you show you are (02:18:10) who you are. And I think solving agent (02:18:12) off, which I think these authentication (02:18:14) companies I just mentioned will solve, (02:18:16) really allows our customers to say, (02:18:17) "Hey, I'm Paul. I'm logging into my (02:18:19) United account and this is my agent, but (02:18:21) it's acting on behalf of me because I've (02:18:23) authenticated it." I think that's pretty (02:18:25) compelling. You know, a lot of the (02:18:26) reason Antib exists is to prevent, you (02:18:27) know, account takeovers or spam. But if (02:18:29) it can authenticate and say, "Hey, I'm (02:18:31) acting on behalf of Paul." I really (02:18:33) think that's a good way to show proof of (02:18:34) ownership or proof of humanood of a bot (02:18:37) and that's going to be solved very soon. (02:18:39) It feels like there's a bunch of stuff (02:18:40) happening at Agentic O that is very very (02:18:43) promising and that's going to be great (02:18:44) for browser base and we're happy to (02:18:45) partner with all those companies to make (02:18:46) it happen. That's fantastic. Well, (02:18:48) congrats on the big round uh and thanks (02:18:51) for stopping by. We'll talk to you soon. (02:18:52) Always welcome. Congratulations to you (02:18:54) and the team. Excited to watch you guys (02:18:55) work. Have a good rest of the day. Talk (02:18:56) to you soon. See you around. Appreciate (02:18:58) it. And really quickly, let me tell you (02:18:59) about adquick.com. Out of home (02:19:00) advertising made easy and measurable. (02:19:02) Say goodbye to the headaches of out of (02:19:03) home advertising. Only adqu combines (02:19:05) technology, out ofome expertise and data (02:19:07) to enable efficient, seamless ad buying (02:19:09) across the globe. Our next guest is Alex (02:19:12) Canerwitz from big technology. We love (02:19:15) big technology and we got the big (02:19:18) technology in the studio. Welcome to the (02:19:21) show. Thanks so much for joining us. How (02:19:23) are you doing? Thanks guys. I'm doing (02:19:25) great. Great to see you. Thanks for (02:19:26) having me on. uh would you mind uh doing (02:19:28) a quick introduction on kind of how you (02:19:30) frame yourself in the in the giant pool (02:19:33) of big technologists and big technology (02:19:35) folks? Definitely. I think it's fairly (02:19:38) straightforward. I'm an independent (02:19:39) journalist. So I used to be working (02:19:41) within newsrooms in 2020. I quit (02:19:43) BuzzFeed News, started my own thing. (02:19:45) I've been writing the newsletter for 5 (02:19:47) years and doing big technology podcast (02:19:50) for just underneath that. Amazing. Uh, (02:19:53) and I mean we this is just we can go (02:19:56) into a bunch of other stuff, but I want (02:19:57) to talk like the the meta level, the (02:19:59) inside baseball for a little bit. Uh, do (02:20:01) you see uh do you see kind of a a (02:20:04) meaningful difference between being an (02:20:06) analyst and being a a reporter, an (02:20:09) investigative journalist, someone who (02:20:11) gets scoops versus someone who provides (02:20:14) op eds? Do you uh do you think that's a (02:20:17) meaningful distinction in kind of the (02:20:19) modern independent journalism era or the (02:20:22) going direct era or anything like that? (02:20:24) We see ourselves not as breaking news (02:20:27) here. We see ourselves more as like a (02:20:29) reaction and opinion show and and talk (02:20:31) show or anything like that. But I'm (02:20:32) interested to see how you think it's (02:20:34) evolving and then how you see yourself (02:20:35) in the ecosystem. Well, it's so (02:20:37) interesting because the definitions, (02:20:39) they sort of blur. And I think if you do (02:20:42) this for long enough, you're going to (02:20:43) realize that you're a little bit of (02:20:44) each, right? You're going to want to uh (02:20:46) get interested in a subject and then (02:20:48) just make a bunch of calls and, you (02:20:50) know, maybe you'll find something out (02:20:51) and now next thing you know, you're an (02:20:52) investigative journalist or you write a (02:20:54) piece of analysis, now you're an (02:20:56) analyst. Or I had a situation where I (02:20:58) interviewed Blake Lemony because I was (02:21:00) interested in this guy who said that (02:21:03) Google's AI was sentient. And Blake (02:21:06) Blake is late to the podcast interview. (02:21:08) And I'm like, what's going on? And it (02:21:10) turns out that while I was waiting for (02:21:11) him, Google was firing. Fired. Yeah, I (02:21:13) knew. Correct. So I said mid- interview, (02:21:16) "Hey Blake, this is news. Uh, can I like (02:21:18) get a comment from Google and publish (02:21:20) it?" And he goes, "Yeah, sure. Go for (02:21:21) it." So then I write to Google mid-in. (02:21:24) We publish the story on big technology (02:21:26) because I have a Substack. Um, I wake up (02:21:28) the next morning and it's a story in the (02:21:30) Journal and Bloomberg and the New York (02:21:32) Times and a bunch of other places. most (02:21:34) of them said as first reported in big (02:21:36) technology. So then I'm a reporter with (02:21:38) scoops. So I think that you know I think (02:21:40) the key to this stuff is are you (02:21:42) president? Are you curious? Are you (02:21:44) going to try to serve your audience? I (02:21:45) know you guys are doing this so you're (02:21:47) probably you know in all these buckets (02:21:48) as well. And I think it's one of the (02:21:50) cool things about doing you know (02:21:52) reporting or news or information on the (02:21:54) internet is you get to play um you know (02:21:56) in all these different buckets. That's (02:21:58) fascinating. Uh yeah I had I had not (02:22:02) thought about it that way at all. Uh (02:22:04) that's fascinating. Uh anyway, let's (02:22:06) move on to the big news. I'd love kind (02:22:08) of your reaction and breakdown from uh (02:22:11) WWDC. Uh we talked to a lot of folks (02:22:14) about it and um I I mean obviously a lot (02:22:17) of mixed reactions, but uh I came away (02:22:20) kind of optimistic that we could be (02:22:21) going into this era of like a new (02:22:23) explosion in AI apps with leveraging (02:22:27) ondevice inference. And I was extremely (02:22:29) optimistic about it, but uh there's a (02:22:31) lot of other narratives going on. What (02:22:33) was your takeaway? Yeah, it was really (02:22:35) interesting. So, I was there. I was on (02:22:37) the broadcast riser because I was doing (02:22:38) some stuff with CNBC. Cool. And I have (02:22:41) never seen a more different uh event one (02:22:45) year to the next. Last year, super (02:22:48) enthusiastic, a little bit more open, (02:22:50) big vision setting with Apple (02:22:52) Intelligence. this year. I think they (02:22:54) spent, this is a conversation I from a (02:22:56) conversation I had with MG Seagler uh (02:22:59) last week. They spent more time talking (02:23:01) about the phone app than they did AI. (02:23:04) So, it was clear that the expectations (02:23:07) have changed for Apple and they ran into (02:23:09) a lot of difficulty building AI in the (02:23:11) same way a lot of companies have run (02:23:13) into difficulty building AI because u I (02:23:16) think anyone who builds with this (02:23:17) technology knows. I'd be curious to hear (02:23:18) from your listeners and viewers if this (02:23:20) is the right uh perspective, but it's (02:23:22) probabilistic. It doesn't act the way (02:23:23) that you want to all the time. Uh if (02:23:25) you're a company like Apple with a very (02:23:27) high quality bar, it's not very easy to (02:23:29) wrangle uh this technology. This is why (02:23:31) I think that they should buy Perplexity (02:23:33) because they can just integrate an AI (02:23:35) search engine into their products and (02:23:37) have that control that Apple wants. Um, (02:23:39) so we definitely saw a a very big change (02:23:42) last year uh to this year and the focus (02:23:45) was really on this new design uh liquid (02:23:48) glass that they've brought into iOS (02:23:50) which is starting to roll out in like (02:23:52) the developer release and with a bunch (02:23:54) of fixes we'll all get access to it at (02:23:56) some point soon. Um but it was (02:23:58) definitely you know a very subdued event (02:24:00) for this company and you're it's (02:24:03) interesting that that I'd actually be (02:24:04) curious to hear why you think that this (02:24:05) ondevice development uh is the thing (02:24:07) that is going to lead to an explosion in (02:24:10) AI apps. The conventional wisdom from (02:24:12) the people that I speak with about this (02:24:14) is that you can already build fairly (02:24:16) fast applications using the stuff that's (02:24:19) available today and you can build much (02:24:21) smarter applications with um these (02:24:24) larger parameter models than this very (02:24:26) small model that Apple is making (02:24:28) available on device. So I didn't really (02:24:30) find that to be a needle mover but uh (02:24:33) I'm curious to hear why you think it is (02:24:35) and what your audience is telling you (02:24:36) about it. Yeah, I mean I think there's (02:24:38) still like I think there there's this (02:24:40) hangover from war stories where (02:24:41) developers like created a cool (02:24:43) experience, released it, and then (02:24:45) incurred massive cost from it going (02:24:47) viral because a lot of these generative (02:24:49) AI features are naturally so exciting (02:24:52) and novel that people just have this (02:24:54) excitement to use them. And that was (02:24:56) leading to some meaningful costs that (02:24:58) independent developers historically you (02:25:01) could just spin up a simple mobile app (02:25:03) or SAS tool and not be thinking about (02:25:05) what your cloud bill was going to look (02:25:06) like. Maybe you were even just had (02:25:08) credits and it and it you weren't going (02:25:09) to incur it at all. It's the it's the (02:25:11) kid without a credit card that is this (02:25:15) super long tale of developers. They're (02:25:17) not super mature. They don't they're not (02:25:18) venturebacked, but you get a ton of (02:25:21) those random solo developers that don't (02:25:24) even want to think about incurring costs (02:25:26) and they want to build, you know, the (02:25:28) the the next flashlight app, the the (02:25:31) beer app or Flappy Bird. These types of (02:25:34) all these different apps that kind of (02:25:36) start as little demos and then maybe (02:25:38) they grow into something with proper (02:25:39) infrastructure and and a backend. But (02:25:42) the more that they can leverage the free (02:25:44) out of the box tools um the faster they (02:25:47) can get going and yes eventually they'll (02:25:49) raise money eventually they might switch (02:25:51) to you know cloud models that are better (02:25:54) and more expensive and they need to (02:25:55) think about the business model there but (02:25:57) being able to use AI in an app that (02:26:00) lives in the app store uses app store (02:26:02) viral distribution truly zero cost seems (02:26:07) like it unlocks a different tier of (02:26:09) entrepreneurship or or app development. (02:26:11) that that could be a really really (02:26:13) interesting canvas to explore. It's kind (02:26:16) of like when the when the when the (02:26:19) original AI video came out, there was (02:26:21) that one viral video Harry Potter (02:26:23) Balenciaga and it's like that's (02:26:26) something that that could have been done (02:26:28) with a traditional CGI pipeline. Like (02:26:31) you could have gotten ILM or Pixar team (02:26:34) to go and make that like it was doable (02:26:37) with the previous technology. But when (02:26:39) you lowered the cost from a million (02:26:41) dollars to do a you know a 3D face scan (02:26:44) of Dumbledore and and put him in (02:26:46) Balenciaga or hire the actors and shoot (02:26:49) it with cinema cameras. When you lower (02:26:50) that to just a couple creative prompts, (02:26:52) like that human creativity part comes (02:26:54) out and you get a viral video, Harry (02:26:57) Potter, Balenciaga, right? And and the (02:26:59) same thing I think could potentially be (02:27:01) on the verge of happening with viral (02:27:03) apps as the as the the the the overhead (02:27:06) and the cost to develop software drops (02:27:08) with products like Cursor and GitHub (02:27:10) Copilot and Windinsurf and Codeex and (02:27:13) Devon and as that drops and then also as (02:27:15) the ability to call AI inference on the (02:27:18) device. Uh we're already seeing it. (02:27:21) There was a cool app that went um went (02:27:23) out on X earlier today or yesterday. Uh (02:27:26) somebody made the Tom Riddle (02:27:28) uh the Tom Riddle iPad app. So you can (02:27:31) you can write hello and it kind of (02:27:33) dissolves into the UI and then and then (02:27:36) a script writes back to you. Uh and (02:27:38) that's all generative text. And so it's (02:27:40) just a simple fine-tuned prompt. It (02:27:43) doesn't need to be, you know, oh imo (02:27:45) gold medalist like amazing math like PhD (02:27:48) level, you know, it's not leading (02:27:49) benchmarks. It doesn't need to lead (02:27:50) benchmarks. It just needs to be (02:27:52) conversational and with an interesting (02:27:54) UI. And it's something that this (02:27:55) developer was able to build pretty (02:27:57) quickly. And for like a young Harry (02:27:59) Potter fan, that's going to be a magical (02:28:01) experience for, you know, maybe a couple (02:28:03) minutes, maybe an hour, maybe may maybe (02:28:04) it doesn't turn into the next big thing, (02:28:06) but you get a million of those in the (02:28:08) app store and then one of them, you (02:28:11) know, develops and develops and becomes (02:28:12) something really really cool. So that's (02:28:13) why I've I've been very optimistic. Let (02:28:15) let me give the counterpoint which is (02:28:17) that with the biggest models that we (02:28:19) have and the such powerful models that (02:28:21) we have we haven't yet seen a you know (02:28:23) critical mass of hit apps uh that you (02:28:26) would say oh if we could just bring the (02:28:27) cost down we would get an explosion but (02:28:29) you have opened my mind I I'm willing to (02:28:31) to say that this is definitely something (02:28:33) that I hadn't considered where you could (02:28:34) have the sort of low lift fun apps that (02:28:37) don't you know make the developer go (02:28:39) broke that maybe were cost prohibitive (02:28:42) to build beforehand that we might see. (02:28:43) So let's let's talk about this in a (02:28:45) couple months and then we'll see we'll (02:28:47) see what happened. Yeah. So I mean I I I (02:28:49) do agree with you. Uh and uh the the (02:28:52) example that I think would be worth (02:28:53) digging into is do you remember that app (02:28:55) Lensza? It was this magic avatar app. (02:28:58) You download the app and you take and (02:29:00) you take and you take like 10 photos of (02:29:02) yourself and then it can generate uh a (02:29:06) like an AI image of you. It was kind of (02:29:08) a precursor to the Studio Gibli moment. (02:29:11) Yeah. and they were extremely good at (02:29:13) monetization because they would be like, (02:29:15) "Yeah, we're we're making the image. Do (02:29:16) you want it? Buy this $20 one inapp (02:29:19) purchase." They did very well. Uh it was (02:29:21) a little bit of a flash in the pan. I (02:29:22) think it went it went really big. And (02:29:24) then they haven't shipped an update (02:29:25) since December 20, 2021. Yeah, it it (02:29:28) didn't really go anywhere, but that felt (02:29:30) like an important shelling point for uh (02:29:33) for for generative AI imagery. And then (02:29:36) we saw it again with Studio Gibli really (02:29:37) taking over the internet. I feel like we (02:29:39) haven't had that in text yet. There (02:29:42) hasn't been that like where's the Harry (02:29:43) Potter Balenciaga for something that was (02:29:45) just purely text generated. Um but uh (02:29:49) but but I think that it might be a cost (02:29:50) issue. It might be something that people (02:29:52) just haven't had a chance to to to play (02:29:54) with at low enough cost. And I think (02:29:55) that anytime you you reduce the cost (02:29:57) like you know GPS, you drop the cost of (02:29:59) that really really low and you get Uber. (02:30:01) And I I I'm just still optimistic that (02:30:03) that there's a there's a there is a (02:30:05) binary difference in in 0001 cent and (02:30:09) zero cents. Like zero is actually Yeah. (02:30:12) I'm smiling because I'm remembering this (02:30:14) Lensa moment and there's been this like (02:30:16) it was the start of people being like (02:30:18) don't upload your face to AI because (02:30:20) then they'll have your, you know, your (02:30:22) biometric data. And I remember in the (02:30:24) middle of the Studio Ghibli thing, there (02:30:26) was a tweet that came out like that (02:30:27) where someone was like, you know, be (02:30:29) careful about uploading your face to (02:30:31) chat GPT and then someone took their (02:30:33) Twitter avatar and jiblied it underneath (02:30:35) course. And I was like, all right, well, (02:30:37) no good deed goes unpunished, I guess. (02:30:39) It's so brutal cuz the company that did (02:30:41) Lensa pivoted to something called (02:30:43) Prisma, which was a photo editor with (02:30:45) like cool artist. This is a It's a (02:30:47) Russian company, I think. It's cool (02:30:50) artistic filters and they basically (02:30:52) built the Gibli type experience and then (02:30:56) I think they were actually Prisma before (02:30:58) cuz I I think I've used that. That's the (02:31:00) name Prisma Labs is is the name of the (02:31:01) company and uh and it was never quite (02:31:03) there and then OpenAI just kind of (02:31:05) leaprogged and I think that will be a (02:31:07) dynamic that happens a ton. I think that (02:31:09) there will be uh kind of solo (02:31:12) developers, kids that build a magical (02:31:14) experience like the flashlight app and (02:31:16) then yes, Apple will steamroll them, but (02:31:19) that solo developer, that kid will (02:31:22) probably make a ton of money with like a (02:31:24) $2 inapp purchase just for upgrades or (02:31:26) throw some ads in it or something, make (02:31:27) some initial money. But even more (02:31:29) important, the experience of having like (02:31:30) a hit app sets you up for, okay, the (02:31:33) next goround, I'm ready to raise money. (02:31:35) I'm ready to think about the the the the (02:31:37) seven powers or the moes that I should (02:31:39) build around my business. Maybe I wind (02:31:41) up going into B2B software or or (02:31:43) infrastructure or something else, but (02:31:45) but they've caught the itch of of (02:31:47) entrepreneurship because they've had (02:31:48) that taste of like I built I built a (02:31:50) product that had an impact and that's (02:31:52) something that uh I I love when those (02:31:54) things get unlocked at earlier earlier (02:31:56) stages. It's a it's a very cool idea. I (02:31:58) mean, I've had this idea to build this (02:32:00) generative AI app where it's like a (02:32:01) choose your own history where you can (02:32:03) either play uh through these scenarios (02:32:05) cuz you can do this. You can like prompt (02:32:06) Claude today and you can do it today uh (02:32:09) where you could like be a historical (02:32:11) character like let's say you wanted to (02:32:12) be like Alexander Hamilton during the (02:32:14) American Revolution and sort of like (02:32:16) experience uh that as he did quote (02:32:18) unquote right but doing it through a (02:32:20) genai experience you can do that. I (02:32:22) think the problem has been that the (02:32:24) models haven't been powerful enough for (02:32:25) the thing that I want to build. But I I (02:32:27) think that you guys are on to something (02:32:29) like if this is a moment where Apple, (02:32:32) you know, Apple might start with the (02:32:33) smaller model. Let's see what happens. (02:32:35) Maybe you get the equivalent of that (02:32:36) beer drinking app. Uh if that works out, (02:32:39) then there's a lot of incentive for them (02:32:41) to put much bigger models uh within the (02:32:43) operating system. And then you could (02:32:45) really unlock some very cool things. And (02:32:47) and it it does make sense that if you (02:32:49) end up uh not going broke by building a (02:32:52) Genai app that becomes popular, you (02:32:54) might want to build another one. Yeah, I (02:32:56) actually had this exact experience. The (02:32:57) historical character on uh character AI, (02:33:01) the uh the the the the app where you can (02:33:03) chat to a fine-tuned LLM on a particular (02:33:07) person. Uh I picked uh Stalin. (02:33:12) And what does that say about you? No, (02:33:14) I'm kidding. Well, I I'll tell you. So, (02:33:16) I was like, I want to go and argue with (02:33:19) Stalin about communism, and I want him (02:33:21) to play the steelman, the literal (02:33:24) steelman in his case. Uh, and and and I (02:33:27) and I want to and I want to have a real (02:33:29) hardcore debate about what Stalin did, (02:33:31) my perception of him. I I feel like I (02:33:34) have a lot of good ammunition to to (02:33:36) fight back against his ideology. Uh, but (02:33:38) let's go have this debate. And so I (02:33:40) started debating with this character AI, (02:33:41) the Stalin, but it had been so RLHFD and (02:33:45) so fine-tuned on like American values (02:33:47) that Stalin would come to me with (02:33:49) something like, yes, like uh you know, I (02:33:51) did a lot of bad things and I'm like, (02:33:52) sorry about it. I was like, I don't (02:33:54) think Stalin would talk like that to me. (02:33:55) I think Stalin would be very proud of (02:33:57) what he did and and not admit fault. And (02:33:59) I was like, "Okay, uh, character, you (02:34:01) clearly didn't fine-tune this enough (02:34:03) because it's still it's still rejecting, (02:34:05) uh, you know, the the the real the the (02:34:07) real communism. Real communism in in (02:34:09) characterized (02:34:10) with it and realized that, yeah, you (02:34:13) know, maybe I did get a few things (02:34:14) wrong. That's not the stalin I wanted to (02:34:17) debate." (02:34:17) The genesis of my idea was just dumping (02:34:20) Wikipedia pages into uh I think it was (02:34:23) Claude and basically playing these (02:34:25) scenarios as like uh maybe somebody (02:34:28) who's not like in the power position in (02:34:30) history. So I always find it (02:34:31) interesting. So my wife is European so (02:34:33) we've like visited a lot of like um (02:34:35) historical sites in Europe and I've (02:34:37) always thought like well what happens if (02:34:39) you're not like the lord or the lady (02:34:40) that they feature or the king that they (02:34:42) feature in these museums? like what if (02:34:44) you're just like somebody on the line (02:34:46) and what would your life be like? So for (02:34:49) me, cuz I'm a fun guy, I dump Wikipedia (02:34:51) pages into these bots and say, "Well, (02:34:53) what would that be like?" And then you (02:34:55) can role play through it. And I think (02:34:56) that um if you want to if you find a way (02:34:59) to like prompt these bots to give you (02:35:01) like the um the non-sanitized version, (02:35:04) it can be really interesting and and I (02:35:06) don't know, fun. I don't know if fun is (02:35:07) the right way uh to put it, but uh worth (02:35:09) worth going through these scenarios and (02:35:11) playing the games. Yeah. Yeah. I mean, (02:35:12) totally. it's a great idea. Somebody (02:35:14) will probably build it. Uh but even (02:35:16) right now, it's like even even if they (02:35:18) would get you to pay, that's an extra (02:35:21) hurdle that causes conversion rates and (02:35:23) there's a million different things that (02:35:24) that that kind of downstream from that. (02:35:26) Even if it's just writing extra code to (02:35:28) make sure that you're processing a (02:35:29) payment in order to pay for your cloud (02:35:31) bill and your cloud bill. Uh so I'm (02:35:33) excited to see what happens. I I don't (02:35:34) know. We we could see how it plays out. (02:35:36) Uh we some other Yeah. Can you give us (02:35:38) an update on the app store? There's just (02:35:41) been a lot of back and forth between uh (02:35:44) it's Tim Sweeney, right, over at Epic (02:35:47) and Apple uh chirping at each other. Uh (02:35:51) what what is the state of the app store? (02:35:53) Are alternative, you know, payment rails (02:35:56) getting, you know, real adoption yet? (02:35:59) Can you give us kind of a broad (02:36:00) overview? Well, we're super early and (02:36:03) there might be some court cases to still (02:36:05) work through uh in terms of this (02:36:07) alternative payment on the app store. (02:36:08) But basically what Tim Sweeney and Epic (02:36:10) Games are trying to push through is (02:36:13) Apple if you want to buy stuff within (02:36:15) apps you have to use Apple's inapp (02:36:17) payments and when you use Apple's inapp (02:36:19) payments Apple gets a cut a sizable cut (02:36:21) uh of the money that you spend and that (02:36:23) is very important for their services (02:36:25) business. And just to take a step back, (02:36:27) so basically Epic is trying to say uh (02:36:30) that that should be illegal uh whatever (02:36:32) it is, monopolistic or anti-competitive, (02:36:34) and we want to be able to uh process the (02:36:37) payments on our own or via the web and (02:36:39) not pay Apple the 30% or whatever it is (02:36:42) from every dollar that goes through our (02:36:44) services. Um it is so I was going to (02:36:47) take a step back because it's just (02:36:49) happening in this very interesting (02:36:50) moment. So, if you think about Apple's (02:36:52) financial position right now, um almost (02:36:55) everything is either flat or shrinking. (02:36:57) Uh the iPhone revenue 2024 versus 2023 (02:37:01) flat. Maybe it grew a tiny percent, but (02:37:03) like you look in there, uh 10K or (02:37:05) whatever it is, and you see the the uh (02:37:08) you know, the growth or the loss is just (02:37:09) the dash flat iPhone revenue. And by the (02:37:12) way, that's happening as iPhone sales (02:37:14) decline. And the only reason revenue is (02:37:17) flat is because they're getting a higher (02:37:19) average selling price per iPhone. So (02:37:21) Apple is in sales decline with the (02:37:23) iPhone right now. And that's happening (02:37:25) as people spend more time with their (02:37:26) phones and as people uh in China are (02:37:30) less interested in buying iPhones (02:37:31) because Huawei has become this object of (02:37:33) national pride and it's almost as good. (02:37:35) So they're starting to buy Huawei. And (02:37:37) by the way, they're not locked into the (02:37:38) Apple ecosystem because they use WeChat. (02:37:41) Mhm. So the company is is seeing flat (02:37:44) sales revenue on the iPhone. They're (02:37:46) seeing uh declines in things like (02:37:48) wearable and the iPad. I think MacBook (02:37:50) grew about 2% last year. That's the uh (02:37:53) only like traditional Apple business (02:37:55) category that grew. Uh but there is the (02:37:58) services category uh that grew 13% last (02:38:01) year. And what does that include? It (02:38:03) includes a number of things under (02:38:04) threat. Uh that includes those inapp (02:38:06) payments that you just brought up and it (02:38:08) also includes this 20 plus billion (02:38:10) dollar payment a year that Apple gets (02:38:12) from Google which is also under threat (02:38:14) because there's this federal case going (02:38:16) on in DC that Google has already lost. (02:38:19) Whoa, whoa, let me stop you. It also (02:38:20) includes Apple TV. Let's talk about F1. (02:38:22) Let's talk about some of the movies (02:38:23) they're making. Let's talk about all the (02:38:25) stuff that they highlight in the I'm not (02:38:27) here to be super negative. Not here to (02:38:28) be super negative, but let me be (02:38:30) realistic. If that Google payment in (02:38:32) 2024 went away and like let's say that (02:38:35) was 20 billion it was probably more but (02:38:36) let's say it went away in 2020 24 any (02:38:39) any thoughts about what happened? All (02:38:40) right Apple as a as a company shrinks uh (02:38:44) services the one the one uh division (02:38:47) growing double digits growing 13% (02:38:50) actually contracts. Now it's a one-time (02:38:52) hit. So you would go down and then you (02:38:53) would grow from there. But the reason (02:38:55) why Apple is in the $3 trillion range is (02:38:57) because it has this services uh division (02:39:00) which investors will value as uh more of (02:39:04) like a software company. So there you (02:39:05) could get into like the the 30 or the (02:39:07) 35x multiples versus a hardware company (02:39:09) which is the other the other thing you (02:39:10) didn't mention is margin iMessage has (02:39:13) also opened up pretty dramatically too (02:39:15) where you have red red receipts you know (02:39:18) between you guys have seen the Android (02:39:21) uh folks typing and been like what's (02:39:22) that or seen the red receipts like why (02:39:24) is the green bubble doing the typing (02:39:26) thing starting to happen so you're right (02:39:28) I I think gold bubbles are coming the (02:39:31) open AI phone is coming that he's going (02:39:33) to pick a different (02:39:35) Trump phone. Yeah, that Trump phones (02:39:36) will have go bubbles for sure. No doubt (02:39:38) about that. The go bubbles are coming. (02:39:40) Uh talk to us about uh about Google. Uh (02:39:43) you interviewed Sergey Brin recently. Um (02:39:46) uh what what what was your read on his (02:39:49) involvement because I feel like there's (02:39:51) a press cycle every few months about (02:39:53) like he's coming back. He's he's he's (02:39:54) he's back in founder mode. He's deeper (02:39:57) than ever. he's not fully stepping into (02:39:59) the CEO seat, but they're clearly uh (02:40:02) stepping up to the AI moment. At the (02:40:04) same time, they're dropping a ton of (02:40:06) product updates. Uh Google IO was a (02:40:08) massive event with like a ton of (02:40:10) different features and then they're also (02:40:12) at the paro frontier of so many AI (02:40:15) models on ter in terms of performance (02:40:17) and cost and yet a lot of stuff's not (02:40:19) breaking through. What's your read on (02:40:20) Google right now? I think they are they (02:40:23) knew they were behind on AI. I mean (02:40:26) basically they had they developed this (02:40:28) is going to be old history for a lot of (02:40:30) people but for those who don't know they (02:40:31) developed this transformer model within (02:40:33) Google uh they were very safety (02:40:35) concerned we we talked about Blake Leo a (02:40:37) little bit ago um he was using a version (02:40:40) of their LLM chatbot called Lambda that (02:40:43) he believed was a person. So they had (02:40:45) this advanced technology working within (02:40:47) the company but OpenAI of course was (02:40:49) first to market with something that (02:40:51) exploded with chat GPT. So once I think (02:40:55) Sundar Pachai realized what was going (02:40:57) on, he said, "Okay, we're going to turn (02:40:59) all of our focus toward uh you know this (02:41:02) AI moment building large language models (02:41:04) and he was lucky or fortunate or he (02:41:07) planned uh whatever you want to say. I (02:41:08) mean the company has Demisabis the head (02:41:10) of DeepMind in there and he consolidated (02:41:13) these two divisions, DeepMind and Google (02:41:15) Brain, which had traditionally kind of (02:41:17) struggled over resources and realized (02:41:19) they needed to coordinate. So, Demis (02:41:21) takes over both of those divisions and (02:41:23) now you see that Google is shipping like (02:41:25) crazy uh with AI and they have some (02:41:27) great I mean they obviously previewed (02:41:29) some very cool things at Google IO some (02:41:32) great features working already. I mean (02:41:33) notebook LM is really interesting. I (02:41:36) think Gemini could work a little better (02:41:37) in places like Gmail, but we'll probably (02:41:39) see it like being able to search your (02:41:41) email is a pretty big deal. uh and and (02:41:43) do like things like I sometimes will say (02:41:45) like pull out the emails um from all of (02:41:48) Big Technologies paid subscribers (02:41:49) recently to um so I can invite them you (02:41:52) know to our to our private discord and (02:41:54) Gemini gets that right about 50% of the (02:41:57) time. So I think it's clear that they're (02:41:59) expect Yeah. What do you expect to see (02:42:01) out of the Google Open AAI relationship? (02:42:04) Right now we're seeing tension between (02:42:06) Microsoft and OpenAI, their partners. (02:42:08) They they maybe both are unhappy with (02:42:11) with how the deal is worked out. I think (02:42:13) all this stuff is evolving, but in my (02:42:16) view over the next 5 to 10 years, I just (02:42:18) see this massive war and battle for the (02:42:22) consumer between OpenAI and Google. Is (02:42:24) that how you're looking at it as well? (02:42:27) And do you expect to see more attention (02:42:28) there? It probably will happen. So just (02:42:30) to I'm going to answer that question. (02:42:31) I'll just finish up what I was saying (02:42:33) that Google's in this moment of (02:42:34) reinvention. So they're going to move (02:42:36) from search to AI. And also I think (02:42:38) Sergey is really involved on the (02:42:39) technical side of things to answer that (02:42:40) part. Uh so the partnership that Google (02:42:43) and and OpenAI did is is for basically (02:42:46) um server space. Uh I think OpenAI is (02:42:50) really running out of GPUs. Uh, and of (02:42:52) course this they have this hosting deal (02:42:54) with Microsoft uh, that's kind of like (02:42:56) up in the air. We're not quite sure (02:42:58) where we're going to go on that front. (02:43:00) And you see Sam Alman every couple days (02:43:02) saying, "All right, we're out of GPUs. (02:43:03) Stop burning our servers." He's joking, (02:43:05) but it's a real issue for the company. (02:43:07) So, of course, you look to Google, which (02:43:09) has the TPUs uh, to sort of offload some (02:43:12) of that um, some of that sort of compute (02:43:15) need. Now, we don't see OpenAI being (02:43:18) made, unless I'm wrong here, we don't (02:43:20) see Open AI as models being made (02:43:22) available within Google Cloud. That's (02:43:24) the other side of the equation. I spoke (02:43:26) with Google Cloud CEO Thomas Curry in a (02:43:28) couple of months ago and he said we'd (02:43:30) welcome OpenAI on the platform. So, (02:43:32) that's another step I think that we (02:43:34) should uh anticipate. And then on the (02:43:36) consumer side, I mean again like and you (02:43:39) guys know this better than anyone else, (02:43:40) like Silicon Valley is filled with (02:43:42) frenemies. Um, and we have another (02:43:44) situation here where you're probably (02:43:46) going to have Google, you'll have (02:43:47) OpenAI, we already have it, OpenAI (02:43:49) relying on Google infrastructure, but (02:43:51) competing with them on, let's say, the (02:43:53) assistant front or the chatbot front. (02:43:55) And I think that's one of the cool (02:43:56) things about watching tech and for those (02:43:58) who are in tech being in tech is there (02:44:01) is almost uh you know a reliance on one (02:44:03) another's innovations and products but (02:44:05) also a recognition that everyone is (02:44:08) working to build the best products and (02:44:10) may the best product wins win and you (02:44:12) know one product may be the king one day (02:44:14) and then it could get leaprogged the (02:44:16) next and build on top of each other. So (02:44:18) I do see this being a heated battle um (02:44:21) and we're going to see a lot more drama. (02:44:23) I anticipate from all these companies (02:44:26) coming up uh pretty soon. Uh but it is (02:44:28) notable that now OpenAI and Google have (02:44:31) opened up the line of communications (02:44:33) line of communication and maybe more is (02:44:35) coming. (02:44:37) Last question. No, I'm excited to I'm (02:44:40) excited to see all this play out. My (02:44:42) last question when when can we place (02:44:44) OpenAI in the category of big tech? Are (02:44:48) they there yet? Does 300 mill $300 (02:44:51) billion valuation does that do it? What (02:44:53) what's your framework? (02:44:55) Uh it's so interesting because yesterday (02:44:57) I was seeing um people on X talk about (02:44:59) like whatever happened to Fang, right? (02:45:01) It was Facebook, Amazon, Netflix, uh (02:45:04) Apple, and Google, and people are (02:45:06) calling it Mango now. You guys saw that (02:45:08) like Meta Amazon jokes. I just like Mag (02:45:13) 7. Mag 7 works. But we're hoping uh (02:45:16) we're hoping that, you know, YC can make (02:45:18) enough startups so we get the Mag 70. We (02:45:21) see you like to expand it. Yeah. So I I (02:45:24) would say look I'm not I'm not a mango (02:45:26) head like the Ow and Mango is open AI. (02:45:28) I'm not a Mango head yet but it (02:45:30) certainly seems like they're on the (02:45:32) trajectory. So I'm I'm uh I'll put my (02:45:34) order in and and we'll wait on it for a (02:45:36) little bit. Now how much have you dug (02:45:38) into the innovators dilemma with regard (02:45:40) to Google? Uh there's this whole story (02:45:43) about like Sundar Pachai hasn't read the (02:45:45) book and does it matter and Ben Thompson (02:45:47) says well it doesn't matter because it's (02:45:48) structural. But the thing that I keep (02:45:50) replaying is like is like what if Google (02:45:52) had actually just released that Blake (02:45:54) Lemony chatbot? Like what if they were (02:45:56) the first mover in chat bots before chat (02:46:00) GPT? It feels it feels like that chat (02:46:03) GPT moment would have been theirs and (02:46:05) then they could have compounded off off (02:46:07) of that and it would have been fine, but (02:46:09) maybe they'd be disrupting themselves. (02:46:11) Do you have any thoughts on on that? I (02:46:13) mean it gives more credit to Chris (02:46:15) Pike's thesis around top down versus (02:46:17) bottoms up innovation but anyways Alex. (02:46:21) Yeah. So I mean I wrote this book that (02:46:23) came out in April 2020. Don't recommend (02:46:25) releasing books at the first wave of a (02:46:27) pandemic. Um but it is sort of what (02:46:29) launched big technology for me. Uh and (02:46:31) it's called always day one. And the idea (02:46:33) is comes from Bezos. uh that the reason (02:46:36) why the big tech companies have been (02:46:38) able to stay competitive and dominant (02:46:40) for longer than the average right the (02:46:42) average company on the S&P 500 last 15 (02:46:45) years uh today as opposed to 67 years (02:46:48) last century and the reason why these (02:46:50) companies have been able to uh sustain (02:46:52) their dominance is because they reinvent (02:46:55) and they don't really uh care too much (02:46:57) about their legacy business they're able (02:46:59) to say what is going to take me into the (02:47:01) next generation uh and sometimes they're (02:47:04) late I I mean Microsoft is a perfect (02:47:05) example. They held on to Windows (02:47:07) forever. They had the lost decade and (02:47:10) then Satya who ran the server and tools (02:47:12) business which was all about onrem (02:47:14) servers said you know what let's do this (02:47:16) cloud thing and now they're where they (02:47:18) are today. Um Google had maybe a lost (02:47:22) couple years but I think they've (02:47:23) definitely gotten with the program. Like (02:47:25) I mentioned, they're in reinvention (02:47:27) territory right now. And uh I think you (02:47:30) could really tell the difference between (02:47:32) what they were doing and what Apple was (02:47:33) doing at uh WWDC, although some might (02:47:36) disagree. Um but they're all in on on AI (02:47:38) right now. And I think that um you don't (02:47:41) have to be first, you have to be all in (02:47:42) and you have to nail it. And uh the book (02:47:45) the the book isn't yet written on where (02:47:47) Google ends up after this because it (02:47:49) does threaten search. Uh but I think (02:47:51) they have as good a chance of as any (02:47:53) given the like we talked about the (02:47:55) compute, the talent and the data uh to (02:47:58) be one of the leaders if not the leader (02:47:59) in generative AI. Well, hopefully you're (02:48:01) the one to write the book. Um that would (02:48:03) be cool. Thank you so much for stopping (02:48:05) by. Yeah. Uh we'd love you guys. This is (02:48:08) fantastic conversation. Definitely (02:48:09) great. Fan of your show. Thank you for (02:48:11) having me on. Have a great rest of your (02:48:12) day. Looking forward to the next talk (02:48:13) soon. Cheers. Bye. Take care, guys. Uh (02:48:15) next up we have Saquon Barkley from (02:48:18) technology investor. uh coming in from (02:48:20) the the Philadelphia Eagles. Uh let's (02:48:22) bring him in the studio and we will talk (02:48:25) to him about everything going on in his (02:48:28) world. Welcome to the show. Ramp day. It (02:48:31) is ramp day. Wouldn't be ramp day (02:48:32) without Squan. Hold on. We are going to (02:48:35) do one ad first. Uh if you're looking (02:48:38) for a luxury watch, go to getbzzle.com. (02:48:40) Your bezel concier is now available to (02:48:43) source you any watch on the planet. (02:48:44) Seriously, any watch. Um, we are working (02:48:48) to bring in Seaquan. Um, and do we have (02:48:52) more news to talk about in the meantime? (02:48:54) We can also talk about Wander. Find your (02:48:56) happy place. Book a wander with (02:48:57) inspiring views, hotel grade amenities, (02:48:59) dreamy beds, top tier cleaning, and 247 (02:49:02) concier service. It's a vacation home (02:49:05) but better. Um, (02:49:10) um, we are working on that. Oh, I think (02:49:12) he's here. Welcome to the show. How are (02:49:14) you doing? What's up, guys? Thanks for (02:49:16) having me. Thanks so much for jumping (02:49:18) on. Um, been looking forward to this. (02:49:20) Congratulations on a on a massive day in (02:49:23) Ramp World. We'd love to get your story. (02:49:26) How did you meet Eric Glimman? Uh, (02:49:27) what's it been like to work with him? (02:49:30) Um, it's been it's been great. Uh, it's (02:49:33) been great. The conversation and (02:49:35) introduction kind of came from uh my (02:49:38) manager cuz um who's been super helpful (02:49:41) uh with me and learning and exploring (02:49:44) more in the tech world. Um but getting (02:49:47) to know getting to know Eric and being (02:49:48) involved Eric has been amazing. Uh he's (02:49:51) been become a really good friend. My my (02:49:54) favorite thing I like to tell people (02:49:55) about Eric is he he has this contagious (02:49:58) smile. He always has this big smile on (02:50:00) his face every single time you're around (02:50:02) him. Um he makes you feel welcome and uh (02:50:05) just him and his team like they've been (02:50:07) doing an amazing job and excited for the (02:50:09) the news that they announced today. (02:50:11) Yeah, it's huge. Uh talk to me about the (02:50:12) Super Bowl ad. We have your picture from (02:50:14) the Super Bowl ad up on the stream right (02:50:17) now. Um how did that come together? Um (02:50:20) how fast was that process? How was the (02:50:22) shoot? (02:50:24) Yeah, we've been talking to RAM for for (02:50:26) a while, the Ram team for a while. Um, I (02:50:29) had dinner with Sam and Max in PA. Oh, (02:50:32) yeah. Um, (02:50:35) a couple of them, they were able to come (02:50:36) to a game. Um, and I forgot it was maybe (02:50:40) like 10 days, a week or two before uh (02:50:44) the Super Bowl. Uh, my manager Cuz asked (02:50:47) me about this idea of of running an ad. (02:50:51) usually, you know, I probably would say (02:50:53) no to Super Bowl commercial uh 10 days (02:50:56) before, but the the ramp team uh they're (02:50:59) they're super efficient and you know, (02:51:01) they they made it so easy and they were (02:51:03) so helpful for me. Um and you know, they (02:51:06) they obviously knew I had a lot at stake (02:51:08) and wanted to make sure that was the (02:51:09) main focus, but they did a great job. I (02:51:12) think the ad came out amazing. They also (02:51:14) showed me that, you know, in past these (02:51:16) brands that I've worked with, they (02:51:18) probably took too much of my time. um (02:51:20) when you're able to do this and under (02:51:22) power and create something so special (02:51:24) and that fans love. Um but the whole (02:51:26) team, the whole ramp team is incredible. (02:51:27) Uh Kareem, Sam, Max, all of them, (02:51:29) they're all special people and I'm (02:51:31) excited to be able to work with them. (02:51:33) What was the reaction from the rest of (02:51:34) the team to see you in an enterprise (02:51:37) software ad in the Super Bowl? Uh was (02:51:40) that surprising or was everyone just (02:51:41) kind of excited to to to see you take (02:51:43) that step? Um yeah, I think it was more (02:51:46) just exciting. More excited. Uh, I've (02:51:48) always, you know, always talking about (02:51:50) how can I do, how can I be creative in (02:51:53) new ways and, you know, we always talk (02:51:54) about financial freedom and involving (02:51:56) yourself with with the best companies (02:51:58) and the best founders and um, you know, (02:52:01) it it's it's been pretty cool just so (02:52:04) far in my career, all the great things (02:52:06) I've been able to do, but the platform (02:52:08) that has been established because of (02:52:10) football, the things football's able to (02:52:14) bring me into and the world is is able (02:52:15) to bring me into and um, I don't know (02:52:18) all the stuff yet, but continue to learn (02:52:20) in this space and continue to meet, you (02:52:22) know, extremely incredible people and (02:52:25) incredible teams. Yeah, it's an exciting (02:52:27) time. Uh, I want to talk about, uh, (02:52:29) lessons from popular books. A lot of (02:52:31) people in tech have read The Score Takes (02:52:34) Care of Itself by the 49ers coach and (02:52:36) borrowed ideas from football and taken (02:52:38) them back to business. Uh, you you read (02:52:41) 0ero to one. uh what is a concept that (02:52:43) you took from 0ero to one or you think (02:52:45) other football players could take back (02:52:47) into the world of football from that (02:52:49) book? (02:52:50) Um that's a great question. Uh Peter (02:52:53) talks about, you know, when you're (02:52:55) starting a company, um you never want to (02:52:58) have two people doing the same job. And (02:53:01) an interesting thing that Nick Serani (02:53:03) did this year, uh, well, my first year, (02:53:07) I'm pretty sure he did in years prior (02:53:08) with the Eagles, my first year with the (02:53:10) Eagles, he he had all of us in a meeting (02:53:13) and let all of us know what our role was (02:53:15) and what was expected of us. Not saying (02:53:17) that our role can expand. Um, not saying (02:53:20) it can't increase, but what's needed for (02:53:23) us to be able to go out there and (02:53:24) compete and accomplish what we (02:53:28) ultimately want to accomplish the Super (02:53:30) Bowl. And I I think the things you could (02:53:34) take is just from (02:53:42) might be losing you. (02:53:49) is it's all right to buy in knowing what (02:53:53) you got to do, knowing what you got to (02:53:55) accomplish. Um, sometimes your role is (02:53:58) not doing what the team asks you to do (02:53:59) to be successful and it takes a great (02:54:01) team. And you guys hear me? Yeah. Yeah, (02:54:03) you're back. (02:54:09) Yeah, it kind of dropped out for a (02:54:10) second, but we're we're we got most of (02:54:12) that. Yeah. Can you hear me? Yep. Yep. (02:54:16) there. Oh, sorry about that. No worries. (02:54:20) Yeah, you hear me? Yeah. Yeah. Uh, we're (02:54:23) good on this end, I think. Yep, I can (02:54:24) hear you guys. Okay, cool. Uh, let's (02:54:26) move on to another question. I want to (02:54:28) talk about uh similarities between the (02:54:30) offense defense dynamic on the football (02:54:32) field to uh you know, these two units, (02:54:35) they aren't necessarily on the field at (02:54:36) the same time, but they both need to (02:54:38) perform. There's some similarities to (02:54:40) business where maybe the sales team (02:54:42) wants to go and sell a product that the (02:54:45) engineering team needs to go and deliver (02:54:46) and make. There are two two teams that (02:54:49) need to be working somewhat in concert. (02:54:51) Um but they're doing very different (02:54:53) roles. Uh what can you tell us about the (02:54:56) team building that goes on between the (02:54:59) offense and the defense to keep the (02:55:00) whole group working in concert? (02:55:06) We lose. Yeah. Super important. Um, team (02:55:09) is the the most important thing when it (02:55:10) comes to Can you hear me? Yep. Yep. (02:55:13) Hello. Sorry, I think I'm in a dead (02:55:15) spot. Um, I apologize. It's okay. Uh, (02:55:18) but I I think team I think team is the (02:55:20) most important thing uh and especially (02:55:23) in in my sport or just kind of any (02:55:25) profession to be honest. I think the (02:55:27) parallels um from business, from (02:55:31) science, from sports, uh all those (02:55:33) things, you know, have similarities and (02:55:36) it all comes down to building a great (02:55:38) team. So, we know how important it is (02:55:40) for, you know, on the offensive side for (02:55:42) us to accomplish what we need to do in (02:55:45) the line of scrimmage. But the same (02:55:46) thing on the defensive side, establish (02:55:48) the line of scrimmage. A lot of games in (02:55:50) football is one up front and it takes (02:55:52) all phases, offense, defense, and (02:55:53) special teams for us to go out there and (02:55:55) accomplish what we want to do. And when (02:55:57) you look at it in any profession, that's (02:56:00) why I say the parallels are so simpy (02:56:05) into the roles. You need people doing (02:56:08) all the little things it takes to be (02:56:09) successful. And that's why you have, you (02:56:12) know, successful teams and you have like (02:56:14) I don't think I can name any individual. (02:56:16) whoever you think is successful, whether (02:56:17) it's a billionaire, whether it's an MVP, (02:56:20) it takes everybody. It takes a family, (02:56:21) it takes it takes a culture, it takes a (02:56:23) village to be able to have that um have (02:56:26) that success. (02:56:27) Uh I I I want to let you go because (02:56:29) obviously you're very busy. Uh la last (02:56:31) question for me and if Jordy has (02:56:32) anything he can ask too. Um tell me (02:56:35) about Big Dom, the chief security (02:56:37) officer. What's his role in the (02:56:38) organization? (02:56:40) What is Big Dom's role in the (02:56:41) organization? (02:56:43) I don't even know if I have an answer (02:56:45) for that to be completely honest. Like (02:56:47) big dumb I know I know he gets a lot of (02:56:50) love and you know ever since his uh the (02:56:54) situation with the 49ers on the sideline (02:56:56) you know he's I go to events and you (02:56:59) know people asking for my autograph and (02:57:01) Big Dom is probably getting more (02:57:03) requests for autographs and pictures (02:57:04) than than me to be honest. But to be (02:57:07) like the the best way I can put it, um (02:57:10) he's just the glue to the team and it's (02:57:14) I'm not even exaggerating when I say (02:57:15) that. Like he is the glue to the team. (02:57:18) He he keeps everything in line and he he (02:57:21) makes sure he has everybody's back and (02:57:23) that's one person that I know outside of (02:57:25) my family member and friends that if I (02:57:27) need anything and I need to make one (02:57:30) phone call, I'm picking up the phone and (02:57:31) I'm calling Big Dom. I love it. That's (02:57:34) amazing. Jordan, you have anything else (02:57:35) you want to run through? Yeah, I wanted (02:57:36) to ask, you know, general, you know, how (02:57:39) how you how you you and and and friends (02:57:41) and maybe other players are thinking (02:57:42) about investments today. You've invested (02:57:44) in RAM. You've invested in and these are (02:57:46) some of you know the top companies in (02:57:48) Silicon Valley. Um in in Silicon Valley, (02:57:51) a lot of people, you know, uh will start (02:57:53) a big company, sell it, and then make a (02:57:55) series of of silly investments. How how (02:57:57) do you avoid maybe making how do you (02:58:00) avoid the bad investments? What's your (02:58:01) kind of framework for that? Um, yeah. (02:58:04) How do you avoid making the bad (02:58:06) investments? That's a that's a great (02:58:08) question. You you get u as a as an (02:58:10) athlete um you get a lot of things (02:58:13) thrown at you to say the least and you (02:58:17) don't know what's the right thing to get (02:58:19) involved in and um you just got to try (02:58:22) your try to educate yourself as best as (02:58:25) you can. And to be honest, (02:58:28) when you come from the NFL, a lot of us (02:58:31) go to college, go to college for three (02:58:33) years, maybe four, and you know, our (02:58:36) financial educ (02:58:38) how how educated we are on the financial (02:58:40) side of things, it's not that great. And (02:58:42) I'm open and honest and I'm able to (02:58:44) admit that to myself um that coming out (02:58:47) the NFL, it it wasn't that great. And it (02:58:49) still has so much room to improve. But (02:58:51) the way I handle is by surrounding (02:58:53) myself with the right people. And you (02:58:56) have to trust people. And one, I have an (02:58:57) amazing manager cuz um and he's pretty (02:59:01) good at saying no to be honest. Uh he he (02:59:03) he will say no to pretty much everything (02:59:06) unless Zack Franklin calls. And when (02:59:08) Zach calls us, uh anything anything he (02:59:10) say is kind of like, uh yeah, let's (02:59:12) let's just do it right now. That's (02:59:13) right. That's right. (02:59:16) Let's just go right to it. Um but it's (02:59:18) God's like that. like you you just (02:59:20) surround yourself with people like that (02:59:21) who who are super smart and super (02:59:22) successful and done it the right way and (02:59:24) you're able to build relationships with (02:59:26) them. Um and you able to earn you able (02:59:29) to earn their trust and you're able to (02:59:31) trust them and uh make those decisions. (02:59:33) So it it's tough because you get a lot (02:59:34) of things thrown at you and I don't know (02:59:36) if I have a the exact perfect answer but (02:59:39) I that's a pretty good answer if you ask (02:59:41) me. That's a pretty good answer. Yeah. (02:59:42) Right. Yeah, that's a great answer. I (02:59:45) think a lot of people in in the tech (02:59:47) world, you know, have spent decades (02:59:50) investing, professional investors, but (02:59:52) they should follow the same framework. (02:59:53) Say no to those things, and then if a (02:59:55) certain person calls you, you just say, (02:59:56) "Yeah, you back up the truck." You back (02:59:58) up the truck. That's amazing. Well, (03:00:01) thank you so much for stopping by. Uh (03:00:03) we'd love to have you back. Uh hope to (03:00:04) see you soon in person. Uh yeah, (03:00:06) congrats on the markup, too. First of (03:00:08) first of many, I imagine. Thank you guys (03:00:10) so much. And I apologize for the NFL. (03:00:13) I'm leaving. I was coming back from uh (03:00:16) an event at a hospital. So, I might I (03:00:18) might be a little shaky. I apologize, (03:00:20) but thank you guys for having me. Yeah, (03:00:22) we'd love to have you in the studio (03:00:23) soon. Yeah. Uh we'll we'll we'll get you (03:00:25) under the lights here for uh for a (03:00:27) couple minutes. We can grab some of your (03:00:28) time because this is fantastic. This is (03:00:30) great. We'll talk to you soon. Cheers, (03:00:32) Saquon. Have a good one. (03:00:35) That is an amazing answer. That's the (03:00:36) best investment framework. I I know (03:00:38) multiple smart very smart people that (03:00:40) have basically the same. That's amazing. (03:00:43) If it works, you know, it works. Wait, (03:00:45) wait, wait for that one person to call. (03:00:49) That was incredible. Uh, anyway, we have (03:00:51) our uh we have our next guest coming up (03:00:53) in three minutes. In the meantime, let's (03:00:55) run through some timeline. Yeah, we got (03:00:57) we want to talk about Yeah, I mean the (03:00:58) big thing (03:01:00) for for the people monitoring the (03:01:03) situation at home. Uh apparently uh (03:01:08) President Trump is warming to the idea (03:01:09) of using US military assets to strike. I (03:01:12) thought you were gonna make a joke about (03:01:13) monitor the real monitoring the (03:01:15) situation is the Microsoft open AI (03:01:16) situation because that's the one that (03:01:18) most people in tech are monitoring on a (03:01:20) daily basis. But we do need to cover the (03:01:22) geopolitics. What's the update there? Uh (03:01:25) yeah, apparently Trump uh Trump has been (03:01:27) warming up to the idea of military (03:01:30) action against Iran. Um so much so that (03:01:34) the poly market is spiking up to 75% (03:01:38) chance of military action against Iran (03:01:40) before it's July. So couple weeks left. (03:01:43) Now there would be a difference between (03:01:44) military action and boots on the ground. (03:01:46) And I think that there's probably a a a (03:01:49) bit of a Rubicon to cross there mentally (03:01:52) for the American people. Um obviously (03:01:55) people don't like the idea or there are (03:01:56) many people in America that don't like (03:01:58) the idea of sending money to fight a (03:02:00) foreign war and they think we should rep (03:02:02) prioritize America and American paying (03:02:04) down the debt for example all those (03:02:05) different things. Uh the Ukraine war has (03:02:07) gotten some push back from a lot of (03:02:09) folks in tech and beyond um for that (03:02:11) reason. But the real red line for a lot (03:02:13) of people is boots on the ground. Don't (03:02:15) send our soldiers over there. It's a (03:02:17) quagmire. We live through the global war (03:02:19) on terror. Let's never do that again. Uh (03:02:21) and so uh it'll be interesting to track (03:02:23) which way this leans because there's (03:02:25) obviously multiple tiers to engaging. (03:02:28) One is just you you know you sell the (03:02:30) weapons or technology to another comp to (03:02:32) another uh government that uses them. (03:02:34) Then you're deploying American military (03:02:36) assets but no boots on the ground and (03:02:37) then the final is of course uh boots on (03:02:39) the ground. Uh but hopefully this whole (03:02:41) thing resolves quickly and we can get to (03:02:43) a uh a peace negotiation and a trade (03:02:45) deal. Um because uh we want we want (03:02:48) freedom and we want capitalism, (03:02:49) stability, we want business. We want we (03:02:52) want we want 10 more whizzes out of (03:02:55) Iran. Give me the Iranian whiz. That's (03:02:57) what I want. Yeah. Give me some cyber (03:02:59) security companies that we can flip to (03:03:01) Google for 30 million. (03:03:04) That's what I want. Crazy. Anyway, um we (03:03:06) have some more. Apparently, there's some (03:03:07) news leaking about uh OpenAI's uh or (03:03:11) sorry, XAI's fundraising. Oh, okay. (03:03:13) What's going on there? They are uh (03:03:16) spending 4.7 billion in the next uh (03:03:21) three months on a new out of their new (03:03:23) $9.3 billion raise. Uh they're (03:03:26) projecting 18 billion for new data (03:03:28) center capex through 2027 (03:03:31) and uh they are the XAI employee payroll (03:03:36) has almost quadrupled since the series (03:03:38) C. It's interesting to think about. (03:03:40) Should we hit the G? Yeah, do it John. (03:03:42) We'll give you one of these air horns (03:03:43) too. (03:03:44) There we go. Gong. It's interesting that (03:03:47) uh X was going through this like, you (03:03:49) know, intense austerity, limiting (03:03:51) spending, you know, keeping employee (03:03:53) counts low. Boom. Now it's just scaling. (03:03:57) Scaling. I love it. We love scale. It's (03:03:59) great. Um so, uh what else is in the (03:04:02) news? Oracle uh has a new initiative to (03:04:04) help companies sell technology to the (03:04:05) Pentagon. Uh Larry Ellison, absolute dog (03:04:10) working to help bridge the gap between (03:04:13) startups, companies, and the DoD. We've (03:04:15) seen this from a couple companies. (03:04:17) Palunteer has the AIP program. Uh (03:04:20) Ander's been acquiring some companies (03:04:22) and and leveraging their connections in (03:04:24) the DoD to actually get distribution on (03:04:25) these products once they're built. uh (03:04:27) every company has a different kind of (03:04:29) angle on the problem, but it feels like (03:04:31) the big narrative is like the we heard (03:04:33) this from the secretary of the army like (03:04:35) the DoD is ready to buy new capability (03:04:38) from the private sector, but it's still (03:04:40) really hard to actually get a program (03:04:41) record. It's actually really hard to get (03:04:43) the products in, educate people, build (03:04:45) the right consensus, figure out what the (03:04:48) budgets are, and get the products in the (03:04:50) hands of the war fighter. And so, uh (03:04:51) Oracle is stepping up with a program (03:04:53) called the Oracle Defense Ecosystem. (03:04:56) It's structured to help smaller (03:04:57) companies break through the challenges (03:04:58) they typically face in selling tech to (03:05:00) the defense department. Uh it's far too (03:05:02) hard to serve the American defense (03:05:04) enterprise. We can provide an easy path (03:05:06) for these companies to get better access (03:05:08) to the defense market. So shout out (03:05:10) Oracle consistently underrated. I say (03:05:13) put them in the MAG 7 right now. The uh (03:05:16) if if you just saw the chart of Oracle, (03:05:18) you would think it was a Bitcoin chart. (03:05:20) Like it's just basically up and to the (03:05:22) right like crazy. It's an absolute (03:05:24) hockey stick. Um, in other news, Morgan (03:05:27) Barrett is posting uh just 30 minutes (03:05:30) ago. He said, "When I heard that there (03:05:32) was rampant payments corruption in Iran (03:05:34) among the military, which is causing the (03:05:36) regime to collapse, I initially thought (03:05:37) that was a TVPN bit advertising ramp." (03:05:41) So, yeah, we want to, you know, there's (03:05:43) uh once once, you know, Iran is is uh (03:05:46) we'll get them on ramp, free and (03:05:48) capitalist, we will are happy to go over (03:05:50) there and and spread the good word. Get (03:05:51) everyone on ramp. Anyway, we have our (03:05:53) next guest, Ryan from Crosby coming in (03:05:56) the studio. Ryan, how you doing? There (03:05:57) he is. Welcome. (03:06:00) Big, congratulations. Yeah. Kick us off (03:06:03) with an introduction on the company, (03:06:04) yourself, the news, everything. Amazing, (03:06:06) guys. I'm so happy to be here. Thanks (03:06:08) for getting me on. Are we ringing the (03:06:09) gong? I mean, you got to tell us. Is (03:06:11) there something wrong worthy to ring the (03:06:13) bell? We're uh we're announcing Crosby (03:06:16) today. We've been in Stealth since uh (03:06:18) about the fall, and we're announcing (03:06:20) that we raised our $5.8 $88 million seed (03:06:22) for congratulations. (03:06:26) There we go. (03:06:29) Incredible. Massive. Congrats. (03:06:31) Incredible. Congrats. Uh, so break it (03:06:33) down. What What are you building? And (03:06:34) then give us some differentiation. How (03:06:36) do you see the market playing out? We'll (03:06:37) we'll dig into it all, but give us the (03:06:39) Yeah, 100%. I'll give maybe like a few (03:06:41) seconds of my background and how I got (03:06:42) here. Yeah, this is a live tracker, by (03:06:44) the way. I'll get to that in a second. (03:06:45) Yeah, I want to hear about that for (03:06:46) sure. Janky way under these boxes for (03:06:49) today. So uh my background is have been (03:06:53) between tech and law for like a decade (03:06:54) early at a couple of startups that did (03:06:56) well and also for various reasons ended (03:06:58) up going to law school practicing. My (03:07:00) last company we grew really quickly from (03:07:02) like 10 to 100 people in about a year (03:07:04) and the only thing slowing us down was (03:07:06) contracts and because I was the only (03:07:08) person there who knew anything about (03:07:09) legal. It was like my problem and (03:07:12) obviously I tried out every legal AI (03:07:14) thing I could get my hands on um over (03:07:16) the last few years. such an interesting (03:07:17) space, but I kept feeling like something (03:07:19) was missing. A lot of the legal AI (03:07:22) tools, plugins, word add-ons on the (03:07:24) market made me a little faster as a (03:07:26) lawyer, but I was still doing so much (03:07:28) manual work. And so I started working (03:07:30) with my co-founder John Sarah really (03:07:32) early at RAMP. So we're celebrating (03:07:33) twice today. Let's go. Really interested (03:07:36) in like the RAMP alumni day. (03:07:40) Yeah. John's big post of the day is that (03:07:43) finance agents could walk so legal (03:07:44) agents could run. Oh, that's great. (03:07:46) That's as good as we did. Yeah. Yeah. (03:07:48) So, we got, you know, John like he was (03:07:50) obsessed with helping um, you know, fast (03:07:52) growing businesses grow faster and save (03:07:53) money and save time. Like that was his (03:07:54) thing. And I was obsessed with this (03:07:56) contract problem. We started jamming (03:07:57) idea last summer and really quickly we (03:07:59) came up with the idea of just starting (03:08:01) our own law firm like what would that (03:08:03) look like full you know starting from (03:08:05) scratch um hire lawyers try to replace (03:08:08) their work agentically like what work (03:08:10) can actually be done with agents what (03:08:11) can't how do we get the speed of AI with (03:08:13) the you know expertise of lawyers and (03:08:15) you know we we had this feeling like (03:08:18) contracts are the API of business like (03:08:20) anytime there's any economic growth (03:08:22) transaction between two parties there is (03:08:24) a contract and we should make that (03:08:25) faster they haven't changed in 50 years (03:08:27) the way we review them. So that's what (03:08:28) we're doing. That's Crosby. Okay. Yeah. (03:08:31) I have been (03:08:33) very upset with uh with No, no, no. (03:08:36) Well, well, that uh that uh I love our (03:08:40) lawyer, but uh you know, it's it's you (03:08:42) know, it's definitely a cost center. Um (03:08:45) no, but I've been I've been kind of (03:08:46) interested to follow, you know, the fact (03:08:48) that that uh we haven't adopted more AI. (03:08:52) We we're an SMB, right? Um, we use AI (03:08:55) all over the place. We use AI across the (03:08:57) entire business from content generation (03:08:59) to management, transcription, sorts, (03:09:01) transcription, research, all this stuff. (03:09:03) And yet, uh, you know, when it comes to (03:09:05) actually generating and, you know, going (03:09:08) back and forth on contracts, it's still (03:09:10) very manual and it just feels like a (03:09:14) place that uh, that we already should be (03:09:17) have have picked things up. So, yeah. (03:09:19) Can you talk about um the different (03:09:22) tiers of legal work? Like my my mental (03:09:25) model for this is like there's like form (03:09:27) filling which did actually get somewhat (03:09:31) commoditized by just CRUD apps in the (03:09:35) web 2.0 era. You have companies like (03:09:37) Legal Zoom. If you need to just fill out (03:09:38) a form to form an LLC, they'll do that. (03:09:41) uh Stripe uh uh Stripe Atlas for example (03:09:45) doesn't use didn't use AI but was able (03:09:47) to advance a lot of things. Then on the (03:09:50) far other end you have someone who looks (03:09:52) more like a legal artist like the top (03:09:55) tier of lawyers who are thinking really (03:09:57) creatively about how to construct all (03:10:00) sort I mean whoever came up with the (03:10:01) open AI or chart clearly that person's (03:10:04) not getting displaced by AI anytime soon (03:10:06) because that was a work of art. It's (03:10:08) like a It's a monument to big law. (03:10:12) Exactly. Yes. I mean, I'm sure the legal (03:10:13) bills there is probably in the hundreds (03:10:16) of millions of dollars. Um but but but (03:10:18) it seems like every year as we chip away (03:10:20) at better foundation models, more (03:10:22) agentic reasoning, longer longer test (03:10:24) time inference, longer uh reasoning (03:10:26) agents, we can move things from one (03:10:29) bucket to the other. and and and so so (03:10:33) talk to me about the flow of of like the (03:10:36) the the road map of moving more and more (03:10:40) uh into the AI world. What's what's you (03:10:44) know today? What's in six months? What (03:10:46) are you tracking on the foundation model (03:10:49) side looking forward to an unlock? (03:10:52) Yeah, I mean these are like I mean these (03:10:54) are like the fundamental questions that (03:10:56) John and I were playing with back last (03:10:58) fall when it was like me sitting there (03:11:00) doing legal work and him looking over my (03:11:01) shoulder trying to convince me he could (03:11:03) automate the things I was doing and me (03:11:04) being really skeptical and like we did (03:11:06) this for weeks and um I think I mean you (03:11:09) I think you nailed it. There's sort of (03:11:10) like we we kind of mapped out and (03:11:12) there's probably thousands of SKs of (03:11:14) legal work from like sure reformatting a (03:11:16) word document to you know writing an (03:11:18) appellet brief for for a court. I think (03:11:20) as you go from more least complicated to (03:11:23) more complicated, the work gets slower. (03:11:25) It gets more expensive and it looks more (03:11:27) like an art and like there's a real (03:11:28) craft to it. It gets more sophisticated. (03:11:30) And I think part of what's going on (03:11:31) there is that part of legal work is (03:11:33) fundamentally like a dialogue between (03:11:35) humans, right? It's a lawyer to a judge (03:11:38) um or to a jury or two contract, you (03:11:40) know, like parties negotiating with each (03:11:42) other. And there's like nuance and (03:11:44) subtlety that we we had a really hard (03:11:46) time like like the difference between (03:11:48) the words reasonable and commercially (03:11:49) reasonable are like and a betting sort (03:11:52) of concept almost the same but but (03:11:53) actually quite different. So like so (03:11:55) this is what we're struggling with. I (03:11:57) think look we obviously have a long bet (03:11:58) on where the models will get to in terms (03:12:00) of like getting better and better so (03:12:02) that we can have a fully agentic law (03:12:04) firm. Like that is where surely this (03:12:05) will go. And so the idea that we started (03:12:08) with like a word add-on. Yeah. As like (03:12:10) the packaging for this is kind of str (03:12:11) like no lawyer has ever or nobody's ever (03:12:13) gone and said, "I want a word add-on." (03:12:14) They just they want a lawyer. They want (03:12:15) to throw it over the fence and not think (03:12:17) about it. So that's what we're doing (03:12:18) here today. And right at like we call (03:12:20) like right at the sort of cutting edge (03:12:23) like of where the machines sort of stop (03:12:25) and the human intuition or judgment or (03:12:27) just accuracy has to come in. Um that's (03:12:31) kind of what we've been trying to model (03:12:32) like from a workflow perspective where (03:12:34) lawyers will jump in. That's where we (03:12:35) are today. Yeah. Um to be more precise (03:12:38) on the like where what kind of what SKS (03:12:40) like where do things land. I think right (03:12:42) now we're focused on contracts. It's (03:12:43) crazy. It's like there's $40 billion or (03:12:45) so that we map spent on contract review (03:12:47) in the US. These are like contracts that (03:12:49) are like a lease to like a merger (03:12:50) agreement that's 80 pages. And so we (03:12:52) just focus there. And so our our goal is (03:12:54) to move up the complexity scale. We're (03:12:55) with these commercial agreements now (03:12:56) that are like let's call it you know (03:12:59) first quartile you know not the easiest (03:13:01) things but but still needs some human in (03:13:02) the loop. talk about um uh I want to (03:13:05) give you an opportunity to talk about (03:13:07) maybe how so we we recently moved into (03:13:10) this new studio the typical process you (03:13:13) wait a long time to get a lease when you (03:13:15) finally get the lease you know I I take (03:13:17) a a quick skim throw it over to our (03:13:20) lawyer they're like cool it's going to (03:13:22) take us a while to review this right (03:13:24) it's variable maybe they're busy maybe (03:13:25) they're not they turn it around you get (03:13:27) through these issues um how much uh do (03:13:30) you already feel like you're an order of (03:13:32) magnitude better than the traditional (03:13:34) process? Like, can I basically (03:13:35) immediately send a contract to Crosby (03:13:37) and you guys can just like basically go (03:13:40) line by line almost immediately and and (03:13:42) you know cuz cuz the typical process is (03:13:44) like, okay, you have a lease, maybe (03:13:46) there's 15 or so places that you might (03:13:48) want to push back or change or or things (03:13:50) you want to add. Uh, and you're (03:13:52) basically just going through and and (03:13:53) talking through those things. And it and (03:13:55) it feels like that could be done. I I (03:13:57) imagine I I don't know how far along you (03:13:59) guys are. It feels like you could (03:14:00) already be at a point where the speed (03:14:02) the the the sort it's almost like it's (03:14:04) almost like I want a an agent on my side (03:14:07) and I want them to have an agent on (03:14:08) their side and I want the two agents to (03:14:10) fight it out and do two weeks of (03:14:12) fighting in two minutes because like (03:14:15) really I want representation and they (03:14:17) deserve representation but I just want (03:14:19) things to go faster because what kills (03:14:21) me is not that they're pushing back on (03:14:23) is it $2,000 or $5,000. What I hate is (03:14:26) that it's two weeks in between a turn of (03:14:28) documents. Okay. Like I like honestly (03:14:31) like the the first thing about the (03:14:32) mapping out legal complexity was like (03:14:34) the first slide in a pitch deck and then (03:14:35) like the last slide was this this is (03:14:37) this elucory but crazy idea but it's (03:14:39) like it's getting more real that there's (03:14:41) a professor that was doing research I (03:14:42) think at Stanford Law School that like (03:14:44) if you have two law students doing mock (03:14:46) negotiations for divorce negotiations. (03:14:48) So it's very like stylized. It's like (03:14:50) who gets the kids on these days and that (03:14:52) date and then you give their same bottom (03:14:54) lines to agents. the repeated agentic (03:14:56) negotiations, they simulate that that (03:14:58) that process. They always get to better (03:15:00) outcomes for both parties in minutes, (03:15:02) right? So, it's it's simple. There's not (03:15:03) that many things to negotiate. So, John (03:15:05) and I were thinking like, okay, there's (03:15:07) like about 50 main sticking points in a (03:15:09) commercial agreement. This is like if (03:15:11) you're any startup, you were selling (03:15:12) software, you were negotiating MSAs (03:15:14) constantly. 50 sticking points. What if (03:15:16) we could map like Adobe's preferences (03:15:18) and like Microsoft preferences and (03:15:20) rather than having months of back and (03:15:21) forth just agentically simulate it and (03:15:24) in minutes say guys this is what you're (03:15:25) going to get to. And so that's like (03:15:26) that's like maybe a few years out but (03:15:28) that's the goal of this company. We want (03:15:29) to be the API for businesses to connect (03:15:32) with each other from a legal and risking (03:15:34) standpoint. Um where we are today how it (03:15:36) works like in terms of because we got (03:15:38) money now we're signing a new office and (03:15:40) we're doing that same thing to the (03:15:42) commercial. It's been three months. (03:15:44) Exactly. we need office, but we we like (03:15:47) these commercial agreements, right? Like (03:15:48) you do a few of them every week. We're (03:15:51) we're actually I would say and like I (03:15:53) think for the fastest growing companies, (03:15:55) these are the things that really slow (03:15:56) them down. So we're specifically working (03:15:57) with the fastest growing companies on (03:15:59) the market. Like we've been working with (03:16:01) Cursor and Clay and Unifi, we've now (03:16:04) done this is over,200 contracts since we (03:16:06) launched and so and they look similar, (03:16:08) right? So they're kind of like they're (03:16:10) you see patterns, you see their (03:16:12) preferences. And so the way it works (03:16:13) today, and this is like it's like all (03:16:15) the pain that like I hated was like when (03:16:17) a lawyer sends you something back, it's (03:16:18) going to be an hour or two days, you (03:16:19) don't know, and they're going to ask you (03:16:21) questions, you have to clarify and kind (03:16:22) of solve things, and then you send to (03:16:23) the counterparty. Today, the way it (03:16:24) works is a salesperson at Cursor or at (03:16:28) Clay will just slack us a document, tag (03:16:30) our bot, we ingest it, we do review (03:16:32) obviously with some AI, some human (03:16:34) oversight, send it back within an hour. (03:16:35) Our median time is 58 minutes and the (03:16:37) salesperson sends it right back and we (03:16:39) have a knowledge base for everything (03:16:40) those clients care about. So, we know (03:16:42) the things that Chris is like, I don't (03:16:43) care. We don't need to worry about that. (03:16:44) We can let it go. And the things that (03:16:46) really matter and our our our like our (03:16:48) driving insight here was we can unlock (03:16:50) sales teams to be so much more, you (03:16:52) know, fast at their jobs. The things (03:16:53) that really matter because of this. Uh (03:16:56) you mentioned that you're building a uh (03:16:59) a word a Microsoft Word plugin. (03:17:02) Microsoft's obviously built co-pilots (03:17:04) into a few of their products. Are you (03:17:06) worried about getting paper clipped if (03:17:08) Clippy comes back and Clippy gets a law (03:17:10) degree? (03:17:12) Clippy with a Harvard law degree. That (03:17:13) is dangerous. I I I'm I'm joking, but (03:17:16) I'm somewhat serious. Like uh it feels (03:17:18) like the le the last thing on (03:17:20) Microsoft's roadmap, but they do have (03:17:21) GitHub Copilot. They do have a product (03:17:23) for software engineers. Is there any (03:17:25) risk of of Microsoft going into this (03:17:28) market just generally, even if it's not (03:17:30) directly? This is the question everyone (03:17:31) wants to ask. What if Clippy does this? (03:17:33) What if Clippy does this? Are you going (03:17:35) to get paper clipped? Yeah. Wow. That's (03:17:39) uh I don't think that much. All right. (03:17:40) So, I think that so you know the first (03:17:43) thing that like every founder in legal (03:17:45) tech is like I'm going to build a better (03:17:46) text editor. Like I'm going to build a (03:17:47) better Microsoft Word. And and it's like (03:17:50) this pipe dream and like it never work. (03:17:52) Like it's just you like you can't it's (03:17:54) impossible. If you move the image the (03:17:56) whole document always goes. That's just (03:17:58) the nature of word docs. I can't even (03:18:00) tell you how much time. So like I don't (03:18:02) know. We're I'm bullish on Clippy. Like (03:18:04) I think we're we're like we're working (03:18:05) heavily on top of word but we're not (03:18:06) compet like we the better that we do and (03:18:09) the more documents we review the more (03:18:10) the more word licenses that we need to (03:18:12) buy for our law. Yeah. And also I mean (03:18:14) you have you have a flywheel and it's a (03:18:15) completely different market but we're (03:18:17) not sell we will never sell against (03:18:19) Microsoft Word. Yeah. Yeah. Would you (03:18:21) ever would you ever release a a B2B to B (03:18:25) type you know product in terms of like a (03:18:27) co-pilot? It feels like a very distinct (03:18:29) decision to say we are our own law firm, (03:18:32) tech enabled law firm. We are not you (03:18:34) know just just allowing other law firms (03:18:37) to build on top of this. (03:18:39) So like (03:18:42) I think lawyers I mean I am one right (03:18:45) like I think that we appreciate I mean I (03:18:47) say this sincerely like there's an art (03:18:48) to what we do but I think it makes us (03:18:50) really ambivalent about AI like we like (03:18:54) even if the work it does not all that we (03:18:56) we you know we talked to so many lawyers (03:18:58) and some of them are like the most AI (03:19:00) pills but most are like I'd rather write (03:19:02) it my way like I think the way that I (03:19:03) drafted that sentence like even if our (03:19:06) tools like I always thought that there (03:19:08) would a cursor for legal by now that (03:19:10) lawyers would just like use the way our (03:19:11) whole team is obsessed with you know (03:19:13) with with um you know better ID but I (03:19:17) don't think it's manifested which is (03:19:18) which is all to say I think part of what (03:19:20) we're innovating on here is not just the (03:19:22) software and the AI sort of workflow (03:19:24) tools we're building but also the way (03:19:25) that we train lawyers that we upskill (03:19:28) them it's a skill issue the way that we (03:19:30) get them better at what they do to use (03:19:32) AI the way that we want them to now if (03:19:33) we can fully automate this service one (03:19:35) day and it's basically like selling (03:19:37) software which I think we will do. I (03:19:39) still want to package it like a service. (03:19:41) I don't want people like opening up a (03:19:43) software platform to deal with this. (03:19:44) Like lawyers, that's just not how people (03:19:45) think about their lawyers. It's just a (03:19:47) problem you want to throw to someone (03:19:48) else. And so I'd love us to feel more of (03:19:49) that like service experience always. (03:19:51) Yeah. Like as an agent in Slack and you (03:19:54) can get that experience that you have (03:19:56) with a great lawyer of of um you know (03:19:58) just talking through things. Last (03:20:00) question. Uh for me at least um I don't (03:20:02) know how much you've studied the history (03:20:04) of like legal tech but uh Atrium's like (03:20:06) the popular story that everyone knows (03:20:08) but there was actually a company before. (03:20:10) Do you know Clear Spire? Have you heard (03:20:12) of this? Okay. So lessons from Clear (03:20:14) Spire and Atrium. I'd love to know how (03:20:17) you tell the story to maybe investors (03:20:18) who are asking about those companies and (03:20:20) what lessons you've learned and how (03:20:22) you're differentiating. Well, what's (03:20:24) interesting is like those companies were (03:20:25) all like sequentially sort of staggered (03:20:27) by like I'm amazing about Clear Spy. (03:20:28) That's like niche. There's this great (03:20:29) book about it. Do you know what I'm (03:20:30) talking about? I I I haven't read the (03:20:32) book. I I But there was like a Harvard (03:20:33) business case study on it that I read. (03:20:35) Yeah. PDF. I read the PDF. Yeah. Okay. (03:20:37) Fascinating. Yeah. Yeah. Yeah. So, it's (03:20:39) like So, it's like So, like I think they (03:20:41) they came up with a really clever (03:20:43) innovation which was you have a it's (03:20:45) just an it's it's a regulatory thing, (03:20:46) but you have a Ccorporation, you know, (03:20:48) typical startup and your PLLC and then a (03:20:50) binding contract. Right. Exactly. And (03:20:52) so, we we've done that and then Asia did (03:20:53) that. They took it a little farther. (03:20:55) They but and and they and they took it a (03:20:58) little farther and focused on types of (03:20:59) work that were maybe more automatable. (03:21:01) They focused on startup law. They were (03:21:02) like WC's law firm. Yep. And I actually (03:21:05) haven't spoken with maybe I spoke to one (03:21:07) person from Clear Spire, but I've spoken (03:21:08) to all the co-founders now of Adrian and (03:21:10) almost all of them except for one said (03:21:13) right idea, wrong time. Like we just (03:21:15) spent all of our money on NLP and now (03:21:16) that's a solved problem. Totally right. (03:21:19) And so and so that that's the and so (03:21:20) like that's the story is like it's a (03:21:22) really really good idea. Yeah. you go (03:21:24) after specific types of law that can be (03:21:26) automated that like you don't care the (03:21:27) lawyer's name. There's types of work (03:21:28) where you're like, I need to know my (03:21:29) lawyer's name. I want to know their (03:21:31) resume. But for other work, you just (03:21:33) need it fixed. And so I I I really think (03:21:35) like they they figured out the hard (03:21:38) things for us and they set the stage. (03:21:39) And yeah, I mean the series A branding (03:21:41) was like we'll do your series A super (03:21:42) cheaply. It was like it was it was great (03:21:44) wedge, but it just didn't expand that (03:21:46) much cuz like you only do one series A (03:21:48) and then for the B it's like way more (03:21:49) complicated and you just bring in a (03:21:50) traditional. Totally. And that's a loss (03:21:51) leader for like that. (03:21:54) It doesn't feel like a loss leader when (03:21:55) I get a six figure bill, but uh but I (03:21:58) guess it is. Anyway, this has been (03:22:00) fantastic. I'm super excited. I mean, (03:22:02) there's so many different types of (03:22:03) contracts that don't make sense to that (03:22:05) that are important that need to be done (03:22:07) well, that don't make sense to, you (03:22:09) know, you you don't need to, you know, (03:22:11) spend tens of thousands of dollars on. (03:22:13) So, anyways, very excited for you. (03:22:14) Excited to follow the journey. Uh you (03:22:16) can be one of our new uh legal AI (03:22:18) experts and residents. So, uh, come back (03:22:21) on when there's news and, uh, congrats (03:22:23) to you and the team. Congratulations. (03:22:26) Cheers. Talk to you soon. On. Bye. (03:22:29) Great. Well, that's been a fantastic (03:22:30) show. Let's check in with Tyler on his (03:22:33) review of Windsurf. Did you get it (03:22:35) installed? How's it going? Yeah. So, I (03:22:38) mean, it's it's pretty good, I think. (03:22:39) Um, I So, I haven't used a lot of these (03:22:42) like IDEs yet. I I've used Cursor like a (03:22:44) little bit. So, you have like your (03:22:46) code's like handmade like farm to table. (03:22:48) Yeah. Yeah. I like to actually like to I (03:22:50) usually I write it up by hand and then I (03:22:51) just scan it in. Sure. I think it's more (03:22:53) of a Yeah, I feel I feel closer to the (03:22:55) code base then. Fantastic. Um but I I (03:22:58) think it's it's really useful. I mean (03:23:00) it's great for like front end stuff. Um (03:23:02) I think and obviously I've only been (03:23:04) using it for like you know an hour and a (03:23:05) half. Yeah, I'm sure there's a learning (03:23:06) curve. Um but I I think when you when (03:23:10) you start to have like okay now I I have (03:23:12) this function that runs on server. It's (03:23:14) like how does that connect? It might not (03:23:15) be in the exact same codebase. It's hard (03:23:17) to integrate these things. Um, but I (03:23:20) think for small like it's just running (03:23:22) locally, but it's like super great. (03:23:24) Okay. Uh, did you have to was there a (03:23:26) model selector because there was that (03:23:27) drama about anthropic uh like pulling (03:23:31) the plug on the windsurf connection so (03:23:33) that uh when you set it up? You used to (03:23:35) just give windsurf your credit card and (03:23:37) then you'd have access to claude and (03:23:38) they would handle the billing (03:23:39) internally. Now it seems like you have (03:23:41) to bring your own claude API if you want (03:23:43) to use uh sonnet which is the popular (03:23:46) one I believe. Yes. So, I'm looking (03:23:48) right now. What's your experience like (03:23:49) selecting a model? Um, so so I'm on the (03:23:51) like uh free tier, but I think I have (03:23:54) it's like a free trial. Um, but I I can (03:23:56) choose uh 3.5 sonnet, but yeah, it's (03:23:59) bring your own keys. Got it. So, so I (03:24:01) would have to go to the cloud website (03:24:03) and then do like export keys or just (03:24:05) paste it in. But even then, I mean, I'm (03:24:07) sure it's like pretty easy still. Did (03:24:08) you notice any The original question we (03:24:10) had was like if Microsoft got their (03:24:12) hands on this intellectual property that (03:24:14) would be incredibly valuable like does (03:24:17) that feel like a reasonable narrative to (03:24:19) you or or are we leaning more like this (03:24:21) is more about data aggregation in the (03:24:23) front door to AI? (03:24:26) I mean if I had to guess I assume it's (03:24:28) the latter like I I I think there were (03:24:31) some rumors at some point about OpenAI (03:24:33) trying to build their own kind of IDE (03:24:35) and that it failed. Interesting. Um, and (03:24:37) that was like maybe the reason for the (03:24:38) acquisition. Yeah. Um, but I I mean I (03:24:41) can't imagine that it's really the the (03:24:43) IP. Yeah. Um, because I mean at some (03:24:47) point it's like just it's kind of a at (03:24:48) some point it's like a VS code fork, (03:24:51) right? Yeah. Yeah. I mean Microsoft owns (03:24:53) VS Code Code. So yeah, you'd think that (03:24:55) they would be able to iterate towards (03:24:57) like the best and practice UI patterns (03:24:59) essentially if that's what really makes (03:25:02) Windsurf stand out. I'm wondering like (03:25:04) there was a ton of heat on Windsurf on (03:25:06) Axe for a while. It was really popular (03:25:08) and yet I saw that uh at uh the AI dev (03:25:12) world fair uh that Swix put on. Uh they (03:25:14) had everyone raise their hands and it (03:25:16) was like 80 or se 70 80 90% cursor (03:25:19) users. So cursor is clearly broken (03:25:21) through in a really meaningful way into (03:25:23) just like broad adoption. Um, and that (03:25:26) seems really really valuable because (03:25:27) you're getting all that training data (03:25:28) that then can be used in reinforcement (03:25:30) learning contexts and uh and derive (03:25:33) better AI models. So, um, yeah, it seems (03:25:36) like it it it's kind of unclear to me (03:25:39) exactly how much of a real issue this (03:25:41) is. We'll have to ask some more people. (03:25:43) We have some other uh AI folks coming on (03:25:45) the show later this week. we'll we'll (03:25:47) try and dig into you know is Microsoft (03:25:50) after just the cash flows or the equity (03:25:52) or the data or the IP or the models (03:25:56) because they have like different claims (03:25:58) on different pieces of open AAI and uh (03:26:00) all of those are probably negotiating (03:26:02) points as they kind of separate these (03:26:03) two entities out and so um it's a (03:26:05) certainly a unique situation because (03:26:08) most investors if you get them on your (03:26:09) cap table maybe you give them a board (03:26:10) seat but they're not getting access to (03:26:12) your GitHub they're not getting access (03:26:13) to your code or your IP can you imagine (03:26:15) if that was like, oh yeah, like uh you (03:26:17) know, I took money from this tier one VC (03:26:19) and then they just gave all my code to a (03:26:21) competitor that they invested in. Like (03:26:22) that would be a wild deal. It really it (03:26:23) really is a wild deal in hindsight. I (03:26:26) mean, it it was a wild deal at the time, (03:26:28) but the sort of 49% profit share. We (03:26:32) don't get equity. We don't have a board (03:26:33) seat, but you know, we have sort of a (03:26:35) right to um IP here in the individual (03:26:39) models. So, it's going to take a long (03:26:41) time to shake out. And Sam is fighting (03:26:46) wars on on at least a couple fronts. (03:26:48) Yeah. Uh I mean we we we didn't get a (03:26:51) chance to dig into the full Ben Thompson (03:26:53) article on on this, but uh he says the (03:26:55) Wall Street Journal doesn't indicate who (03:26:56) its sources are, but there is little (03:26:58) doubt in my mind. They too are on the (03:27:00) OpenAI side and I get why. And he says (03:27:02) his biggest takeaway, this is Ben (03:27:04) Thompson writing a stratey which you sub (03:27:06) should subscribe to. Uh he says, "My (03:27:08) biggest takeaway from this story is that (03:27:10) I have underestimated the amount of (03:27:12) leverage that Microsoft has over open (03:27:15) AI. I mostly wrote about this slow (03:27:17) motion train wreck last month. I still (03:27:19) think that his his take is broadly (03:27:21) correct and note that the right to make (03:27:23) OpenAI models available to other cloud (03:27:25) providers is one of the points of (03:27:26) contention in the story. But the fact (03:27:28) that OpenAI seems to be reduced to (03:27:31) issuing fantastical threats through (03:27:33) press suggests that the decision about (03:27:35) the end point of these negotiations is (03:27:37) almost entirely up to Microsoft. To that (03:27:39) end, I can understand if the company is (03:27:41) pushing for more than I suggested. (03:27:43) Indeed, if anything, these threats (03:27:45) through the press might make Microsoft (03:27:47) want to hold on to more since they (03:27:48) probably have uh since they'll probably (03:27:51) have to fight OpenAI again in a few (03:27:53) years. I still think however this that (03:27:55) this is all the more reason to give more (03:27:58) ground now in exchange for an ironclad (03:28:00) agreement securing Microsoft's exclusive (03:28:03) Azure access to OpenAI models in perpetu (03:28:05) in in perpetuity. Uh that is a (03:28:07) fascinating (03:28:09) uh thing because if OpenAI (03:28:12) somehow compounds and runs away with it (03:28:14) like they they have this massive (03:28:16) consumer engine that could be so it it (03:28:20) feels like right now Anthropic and and (03:28:23) OpenAI on the developer side are neck (03:28:26) andneck and Anthropic even seems a (03:28:27) little bit ahead on codegen and just (03:28:29) developer adoption. But if you if it (03:28:32) winds up being a capital fight, if it (03:28:33) winds up being biggest data center wins (03:28:35) and we need to build the trillion dollar (03:28:37) data center and Stargate becomes (03:28:39) extremely real and extremely material, (03:28:41) well then having a consumer app that's (03:28:44) printing $10 billion a year is going to (03:28:46) be way easier to underwrite as an (03:28:48) investor to go and build the next model (03:28:50) and then you might get higher scale (03:28:53) which could make the developer API more (03:28:55) valuable. And if Microsoft has exclusive (03:28:58) right to resell that that and they are (03:29:00) the B2B player they're the B2B front end (03:29:02) to open AI models that is extremely (03:29:06) valuable extremely valuable. So yeah I I (03:29:09) was not I was not aware of that. He also (03:29:11) talks about his his interview with uh (03:29:13) cursor co CEO Michael Truel. It is it is (03:29:16) it is the the the perfect irony with uh (03:29:20) with OpenAI is that it's it's one a YC (03:29:22) company run by Sam Alman who uh was (03:29:26) president of OpenAI and if YC was (03:29:30) advising a startup that came to them and (03:29:32) they said we're going to enter into this (03:29:34) like really complicated commercial (03:29:36) agreement and we're going to structure (03:29:37) as a nonprofit. We're going to take all (03:29:39) these donations and then in the future (03:29:41) we'd like to go public. They would be (03:29:42) like, "You should just try to not add (03:29:44) too many people to your board. You know, (03:29:46) take as little dilution as possible. You (03:29:48) should you should basically run like a (03:29:50) dictatorship." Yeah. And talked about (03:29:52) this. I would not take my I would do not (03:29:55) do as I say, not as I do. Yeah. So, so (03:29:57) to h to to create one of the most (03:29:58) important new companies of of of uh a (03:30:02) generation and then to have it just be (03:30:05) ins snared in in all of this uh drama (03:30:07) from the nonprofit stuff and the battle (03:30:08) with Elon all the way now to a battle (03:30:10) with Satia is just absolutely brutal. (03:30:13) Well, the last story we should we should (03:30:14) close it out with is uh the Trump phone. (03:30:17) The Wall Street Journal says Trump's (03:30:19) smartphone can't be made in America for (03:30:22) $499 by August. That's because we got (03:30:25) Tyler over here assembling iPhones in (03:30:27) America. If Trump gets a hold of him, I (03:30:30) think all bets are off. But Trump's (03:30:32) mobile phone shows some specs that would (03:30:33) be that would beat Apple's biggest, (03:30:35) priciest iPhone models. Doesn't want to (03:30:37) just make it in America. Uh he wants to (03:30:39) release a gold Android phone that will (03:30:41) be proudly designed and built in the (03:30:43) United States. Um fascinating (03:30:45) development. This is the first time I (03:30:46) heard about it was when the Wall Street (03:30:47) Journal was saying it can't be done. Um (03:30:50) but uh yeah uh apparently Trump's son (03:30:52) told the podcaster Benny Johnson on (03:30:55) Monday morning, you can build these (03:30:56) phones in the United States. So we'll (03:30:58) see where how got to dig into that. Uh (03:31:03) the the display is going to rival (03:31:05) Apple's two uh $1,200 iPhone 16 Pro Max (03:31:09) in Nobody Nobody ever thought to turn (03:31:11) the White House into an incubator. (03:31:14) It's happening. Nobody ever I mean it (03:31:17) was surprised hundreds of years of (03:31:18) presidents. Nobody thought I'm going to (03:31:20) turn this into Jimmy Carter had a (03:31:22) venture studio venture studio. Jimmy (03:31:24) Carter had a peanut farm but he divested (03:31:26) an idiot before he took the presidency (03:31:29) because he didn't want he wanted his (03:31:30) hands clear. He didn't want any (03:31:32) conflicts of interest. So he sold his (03:31:33) peanut farm and we are now in a (03:31:35) different era building 50 megapixel (03:31:38) iPhones. The the hottest new venture (03:31:41) studio is the White House. The White (03:31:43) House. Anyway, I mean the velocity of (03:31:45) new C corps and LLC's that they just (03:31:49) firing out of out of an absolute cannon (03:31:52) canon is Bitcoin treasuries and social (03:31:55) networks and crypto. We didn't really (03:31:57) cover the story of of this uh of the (03:31:59) Tron IPO. Yeah. Not enough people are (03:32:02) are are which Eric Trump is now denied (03:32:05) having a role. Okay. Um (03:32:08) I mean this is the first CEO founder of (03:32:11) a unicorn company in the White House. (03:32:14) Yeah, it's huge. People talked about (03:32:17) Zuckerberg maybe running for president. (03:32:19) They never say unicorn tech founder (03:32:21) Donald Trump. Yes. Yes. They never they (03:32:24) never introduce them that way. Maybe (03:32:25) they should really put the focus on what (03:32:27) matters. Social network truth trading at (03:32:29) truth. Figure it out. (03:32:32) Oh, wild times. Wild times. Uh a cool (03:32:36) 5.1 billion. 5.1. That feels down off of (03:32:40) a off of a little bit of a high. Yeah, I (03:32:42) mean it it it's been generally down and (03:32:45) to the right. Um (03:32:48) in uh in 2022 it peaked at 92 90 92 Well (03:32:54) the the intraday high was $99. It's (03:32:57) sitting at $18 now. Okay. So it was like (03:32:59) a $30 billion stock at one point (03:33:00) something like that. Yeah. You know, if (03:33:02) if if you sell enough Trump mobile (03:33:04) devices who knows and then maybe it's (03:33:07) catalyst. Maybe Trump media DJT true (03:33:10) social is just pre-installed and maybe (03:33:12) they they have their own search. They (03:33:14) build out their own search. I think they (03:33:16) should get into foundation models. (03:33:18) Really really take a run at the big (03:33:19) labs. Hire some researchers. Start doing (03:33:22) some reinforcement learning. (03:33:24) Let's see it. I want I want I want some (03:33:26) some research papers. Maybe they should (03:33:27) do an open source model. (03:33:29) I want some I I want some thought (03:33:31) leadership on on artificial intelligence (03:33:34) AGI. We've suspected that he's AGI (03:33:37) pilled, but we haven't seen it in the (03:33:39) research papers. Let's make it happen. (03:33:41) Anyway, that's our show, folks. Thanks (03:33:44) for watching. We had some range today. (03:33:46) George was wild. You had some spicy (03:33:49) takes. That was fun. I wish we had (03:33:51) another 30. We went all over the place. (03:33:52) We went over to Chinese history. We went (03:33:56) into the NFL. technology investor Saquon (03:33:58) Barkley, some series A's, some series (03:34:00) B's. Really good range. Fantastic show. (03:34:03) I think that's exactly what people come (03:34:04) here for. Thank you for watching, folks. (03:34:07) Focus.