The Delphi Podcast
The Delphi Podcast
José Macedo and Pondering Durian: The Birth of Delphi Intelligence
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Join Tommy Shaughnessy as he hosts Pondering Durian (Lead at Delphi Intelligence) and José Macedo (Co-Founder at Delphi Labs & Founding Partner at Delphi Ventures) to introduce Delphi Intelligence — Delphi’s new open research initiative focused on artificial intelligence. Learn why Delphi is going deep into frontier models, robotics, reinforcement learning, and the intersection of crypto and AI, and how this initiative aims to uncover transformative opportunities across emerging tech.
Delphi Intelligence: https://www.delphiintelligence.io/
🎯 Key Highlights
▸ Why Delphi pivoted to AI after years in crypto research and investing.
▸ Inside Delphi Intelligence: free, high-quality research on AI, RL, robotics, and more.
▸ How the team blends deep research with deal sourcing — like early crypto days.
▸ The three phases of AI scale: pretraining, inference, and the RL renaissance.
▸ José and PD on decentralizing RL and building “The World’s RL Gym.”
▸ The case for open AI: modular design, public reasoning, and collaboration.
▸ How LLMs are learning to “think longer” — and what it means for training.
▸ Open-source vs. proprietary AI — and China’s growing lead in open models.
▸ Why José believes AGI may already be here with GPT-4o.
▸ Where early AI startups can still win — from robotics to edge hardware.
▸ Why this AI moment feels like crypto in 2017 — and why it’s still early.
▸ The team’s views on AI safety vs. acceleration, including visits to Anthropic.
▸ A geopolitical view: US vs. China in AI, talent flows, and capital differences.
▸ What the next “Mag 7” of AI could be — and why Big Tech might not lead.
▸ How Delphi plans to find early winners in AI, like they did in crypto.
💡 Subscribe for more crypto & AI insights! 🔔
🧠 Follow the Alpha
▸ Delphi Intelligence's Twitter: @delphi_intel
▸ PD's Twitter: @PonderingDurian
▸ José's Twitter: @ZeMariaMacedo
🔗 Connect with Delphi
🌐 Portal: https://delphidigital.io/
🐦 Twitter: https://x.com/delphi_digital
💼 LinkedIn: https://www.linkedin.com/company/delphi-digital/
🎧 Listen on
Spotify: https://open.spotify.com/show/62PR1RigLG2YN5Pelq6UY9?si=18ac7ccf36ab4753&nd=1&dlsi=50105fd66e6c4124
Apple Podcasts: https://podcasts.apple.com/us/podcast/the-delphi-podcast/id1438148082
Youtube: https://www.youtube.com/channel/UC9Yy99ZlQIX9-PdG_xHj43Q
Timestamps
00:00 - Introduction to Delphi Intelligence
01:30 - Why AI Feels Like Crypto in 2017
03:30 - The AI Research Gap Delphi Aims to Fill
06:30 - Robotics Deep Dive: Humanoids vs. Specialization
10:00 - AI’s Impact on Traditional Tech Moats
14:30 - Can the Mag 7 Be Disrupted by AI Labs?
17:00 - Google, Meta, OpenAI, and the Race for Interface
21:00 - Crypto as a Hedge Against AI-Driven Inequality
24:00 - Safety vs. Acceleration: Anthropic vs. OpenAI
28:00 - China’s Open-Source AI Strategy
31:30 - Zero-to-One vs. One-to-X Innovation Models
35:00 - Open-Source Models: Strategic or Suicidal?
38:00 - Early-Stage Edge: Where Startups Can Win
42:00 - Verticalized AI Apps, Context Engineering, and GPTs
46:00 - US vs. China in Robotics: Software vs. Hardware
50:00 - The Role of Simulation, Sensors, and Real-World Feedback
53:00 - Delphi’s Dual Approach: Deep Research + Early Deal Flow
55:00 - The Vision for Delphi Intelligence 10 Years Out
Disclaimer
This podcast is strictly informational and educational and is not investment advice or a solicitation to buy or sell any tokens or securities or to make any f
You're now plugged into the Delphi Podcast.
SPEAKER_00Hey everyone, it's Tommy from Delphi Ventures, and today I'm joined by my partner Jose and pondering Durian, who's leading up our new initiative called Delphi Intelligence. It's one of the most exciting things we've ever done, and I'm really excited to talk to both of them about this. As a quick recap for Delphi's broad efforts in AI, we've invested over $25 million at the intersection of crypto AI over the past three years. We've hosted dozens of AI leaders on our podcast. Uh Pondering Durian has created six foundational reports on crypto AI at our research company. And Jose, Luke, Kevin, and the team have led two AI cohorts in Delphi labs. Despite this, I still think Delphi Intelligence is one of the coolest and most exciting initiatives we've ever done in AI. And I'm really excited for it. So Delphi Intelligence will be our free AI research hub where we're focused on sharing our own AI research and the research of other really smart folks in the field across all the domains of AI. We're going to include AI, robotics, frontier models, all the subsets, including text, voice, video, and a wide range of technical and non-technical topics. So today I'm joined by Jose and Pondering Durian, who I have the pleasure of asking questions to. These two guys are ridiculously smart at the intersection of AI, crypto, macro, and much more. Shameless shills here, but if you want to read the research, visit Delphi Intelligence.io to sign up. And if you're an early stage founder seeking funding, feel free to get in touch with me or anyone else on the team. With that, let's dive into the episode. Jose, PD, how's it going?
SPEAKER_01All good. Thanks for having us on.
SPEAKER_00Yeah. Pete, why don't you give a brief overview on Delphi Intelligence, then we'll dive into hot takes on AI.
SPEAKER_01Yeah, yeah, sure. So I guess the the Genesis or the playbook for Delphi Intelligence is kind of roughly similar to what Delphi did on the crypto side, you know, maybe starting back in in 2017, 2018, but um just seems like there is a gap in the market between very technical papers on the one hand and then kind of quick TLDR update newsletter types on the other. And it does seem like there's a gap that, you know, uh an institution like Delphi can fill by basically marrying the two and going deep in a bunch of different areas. So for instance, we just dropped a paper on or an essay on video models. We have another one on reinforcement learning, we have another one on AI browsers, um, and that'll just continue, you know, basically on a weekly basis, dropping large reports that are very similar to what Delphi has done on the crypto side, but just focused on topics in AI, in robotics, and other emerging fields. Um, and I think it's it's also going to be very complementary to Delphi kind of expanding into a lot of these domains from an investment perspective. So it is kind of classic Delphi DNA starting with research and going deep in the areas that we are excited about, and then also hopefully surfacing some interesting investment opportunities while we do that. Um, and so yeah, I'm super excited. You know, obviously I've been working with the partners quite a bit, um, have been building out our internal um AI focused team and launching this week was uh super exciting. So uh excited to hit the ground running.
SPEAKER_00Zay, what do you how do you feel about intelligence? Um I'm curious what you would want to see posted on the site for you to read.
SPEAKER_02Yeah. I mean, I'm super, I'm super pumped about intelligence. I kind of said this in the thread that we did this week. Um, but yeah, I had my sort of the the moment that that pilled me on crypto was the this guy. Um, I actually don't think I can dux him, but he he he's on Twitter, Coley B. He kind of pilled me on on Ethereum, and pretty much since that week, I was just down the rabbit hole. And it was immediately I thought this is the most important and interesting tech I've ever seen. I need to devote my whole career and all my capital to it. Um and the only other time I've felt like that since was reading situational awareness last year. Um, like I was already pretty bullish on AI. Obviously, you were you were in it very early, Tommy, and helping us all get up to speed, but that that paper really like solidified it for me. Um, and since then kind of been thinking about like reading a lot, playing with it, investing in it, um, and just realized the best way that first of all, it's in a similar spot to I think where crypto was in 2017, despite um that might be weird to some people, but I think there's a lot of sort of experts in individual areas, but there's very few places you can go to get a good overview of what's happening and like kind of um actionable, like alpha alpha-packed research, I think at this point. Um and then yeah, so so I just think the best way to the the way we almost do things at Delphi, when once we get nerd sniped, you go down the rabbit hole, you write you you do research, that helps you get conviction, and then we can kind of deploy capital against that. Um, and I think that's the that's kind of the goal here to understand like AI really deeply, find some of the best researchers to to both uh write about it within Delphi and then also curate, because unlike Delphi research, this is gonna be free. So also curate some of the best research that we found find around the web uh on this stuff. But yeah, um yeah, the the thing I'd most like to see is just um it's a good question. It it sort of varies depending on the time. I'd love to see a report on robotics. Uh like I actually discussed this with PD maybe like a year ago, even or or definitely a few months ago.
SPEAKER_00Still waiting on it, PD.
SPEAKER_02That year ago is literally like last week, but yeah. No, no, no. I the when before the humanoid robotics became a thing, because there's a there's a whole thing, I think, an interesting one in robotics, just about humanoids versus like specialized robots and like the the trade-offs between them. So obviously with the humanoid, they can do way more stuff. The whole world is built for humans, you can get economies of scale on the manufacturing, get them really cheap. But the specialized robots can be much more efficient in specific things, right? And specialized robots is how most factories work now, um, like mechanical arms, you know, transport robots, all this kind of stuff. And so I think it's interesting to try and think like, okay, within manufacturing, how many what's will like what stuff is a humanoid useful for? And then sort of quantifying like what are the economies of scale you can get on producing a humanoid versus the efficiency gains you can get on specific tasks from specialized robots. Because yeah, I think it's an it's it's like there's smart people on both sides, I'd say. Um, so I think it's an interesting report, but there's loads of of ones like this in different verticals, I think. Um, yeah.
SPEAKER_01So so yeah, I guess I I would just say that over, you know, before Delphi Intelligence, I spent a lot of time doing tech investing. And it was basically, you know, a mix of infrastructure and enterprise software or infrastructure software or vertical application or consumer internet companies. And it was basically just like, hey, what do your unit economics look like? What does the TAM look like? Can we scale this business? And frankly, it was like super lame and cookie cutter, and until about like three to five years ago, that started changing. And now you wake up every day and there's you're like a kid in a candy store. There's like too many things to pick from in terms of really cool stuff that we could be writing long-form reports about. And so it's not like there's a shortage of topics from like robotics to bio to brain computer interfaces to AI and materials. Like literally, it's just nonstop across all of these different verticals, you're seeing so much change. And so that's one of the reasons I'm super excited about intelligence is like the the pace of change in every single industry and vertical is being transformed. And so, you know, it's impossible to be a master in all of them, but I do think you can, you know, with focus, um, get up to speed relatively quickly and help others do the same. And that's kind of what we want to do.
SPEAKER_02Yeah, you're gonna be you're gonna be busy, dude. I know we're gonna be busy. You you buy not just stop writing your your uh erotic fan fiction stuff, maybe just take a break. I mean, that's my that's my yeah, that's my that's my late night.
SPEAKER_00Hey, it's good fan fiction though. I need more chapters. Come on.
SPEAKER_01Yeah, not those kind of robots, please, PD. Excited for my first pod. Yeah.
SPEAKER_02This can count as a lot run for for your podcast. There's one there's one report actually that Anil's been talking about that I think is is a really cool topic of like what's the next Mac 7 basically. Obviously, there's been like more and more turnover in the Mac 7 over the last few decades. Uh actually, not sure if that's true. Um, he said that. Uh that doesn't seem right to me, at least in the last two, but but maybe it is. He was convincing. Um he was convincing, yeah. He was uh I guess Netflix and stuff, and then NVIDIA, there has been, yeah, I guess I guess there has been quite a lot of turnover. But um it would be, and like what's the next Mag 7 look like? Like assuming there's gonna be even more turnover. I think it was A16Z that like Mark Andreessen said back in the day, software is eating the world, and I think they've like adapted it to software is eating software now, like software ate the world, and now AI is gonna eat sort of trad software. And I really do buy that, like that you're gonna have entirely new apps, flows, like that it's gonna look entirely different in five years. And with intelligence, we really want to try and like map that from yeah, and and with all this stuff, like whenever you're going into something new, you just have to kind of start somewhere, right? And then build up nodes of knowledge and the the tree of knowledge, or like and you'll always talk about the elephant getting as many hands on the elephant as possible. His he likes his metaphors.
SPEAKER_00Zay, how do you feel though about like obviously open AI had like pure straight dominance? Now Gemini and Google has released like non-stop AI releases, like Zuckerberg's taking all the talent, Grok is like topping humanity's last exam. It it seems like the the new Mag 7 might just be the old Mag 7. Like, what what would change your views on OpenAI anthropic X replacing replacing them?
SPEAKER_02I mean X isn't an old Mag 7 though, right? That would be a new one. X X AI with with Grok. And I think um I think OpenAI has to be like a good contender to be in the new Mag 7. They're just building they're building so like they're yeah, they're models. I mean, I still think O3 is my favorite model. Grok 4 is pretty dope, and especially like a lot of the like the the 300 a month or whatever it is, tier is really cool with like the multi-agent stuff. But open AI is just crushing it, and also on product, like they have the the all the app integrations now. They're building this hardware thing with Johnny Ive, which I do think is the end game, like having a thing that's this listening to constantly. Um, yeah, I mean I I have I think the right way to play AI, or at least the way that I've settled on now, is to have some exposure just to the big liquid AI companies that are playing AI well AI well, which is not Mag 7, because I think Apple wouldn't be in there. Um it's like you know, the Googles, the Microsoft's, uh the Metas, and then having some exposure to sort of the private market like beta companies, so the labs um are the obvious ones. Then if you want to go further further down the risk curve, maybe some of the earlier stage stuff, some of the neo cloud stuff, whatever. And then I think having some true early stage exposure either via direct investment, which is obviously really hard, or just through like fund of funds or or something along those lines. That's kind of how I'm how I'm thinking about it. Um and there's some argument that you could just like do a lot less work and capture most of it with the first two, you know, like just ape the labs and the and the and big tech and you're and you're kind of kind of good. But uh, I was doing some research the other day and I realized like the the private the market cap of private uh AI companies total market cap is something around, I think it's something around 700 billion. Um with the yeah, which is like a fifth of crypto.
SPEAKER_00That's like right or a third of a meta or or a Google. Yeah.
SPEAKER_02And like it's it's it's it's a fifth of like total crypto market cap, which just I mean, obviously there's like publicly traded companies that you'd have to add in that capture some of this, but still, I think that number is gonna like 10 to 100 X over the next sort of decade or two. Yeah.
SPEAKER_01But I guess playing off that, like I I don't think it has to be a Mag 7. Like I think more companies can enter that tier. And like I guess that's the the COTU thesis. That's like instead of what is the Mag 7, like what is the Mag 25, and what are the companies that make the leap like the Mag 7 guys did, that get on the right side of you know, uh increasing model capabilities and have defensible motes that go from $500 billion companies to $5 trillion companies over the next like five to seven years. And yeah, I think at the later stages, that's a really smart way to play the market. Um, so yeah.
SPEAKER_02And it's interesting because AI doesn't get the new AI companies don't really have network effects in the same way as like a Google or a meta, like the social media companies, right? Where it's where it's just like impossible to unseat. They they sort of have more like traditional industrial like economies of scale, right? Where they have like hardware and and like data advantages and stuff like this, which is obviously also very hard to displace. Um, but I do think there can be more. Like you're right, that there could there can sort of be more than more than seven. And like, yeah, the the trillion or multi-trillion dollar question is like who is gonna be in there?
SPEAKER_00That that's the hard part for me though, with the web the the traditional fang. Like you're sorry, the traditional mag 7 is we always say that new companies will come and disrupt the old ones, but the new companies have so much or the old companies have so much cash, like Facebook, Google to just spend on hardware and then to pipe the end AR AI to like their billions of users. I don't know, Zay, do you have like a comparison historically to it's a hard question, but yeah, like I know you read the the tech bubbles papers a lot. Like, do you have a comparison historically where a an innovator was disrupted in a similar sense? Because it seems like all of their motes are playing to their advantage here with the capital have, with the revenue they have, with the user bases they have for AI.
SPEAKER_02Yeah. I think it's it's hard to generalize across all the Mag 7 because I think they they have different positioning um and and they they benefit from AI differently. In general, I really like the Carlotta Perez framework of like uh technological revolutions and uh and bubbles of sort of that you have if something's a truly disruptive innovation as opposed to like a sustaining innovation, then it's very hard for the incumbents to be the ones that capture it just because of sort of the innovators' dilemma, which is you know, you you have to sort of sepulku and like cannibalize so much of your traditional business by going after the new thing that it's just very hard. There's so many frictions, like organizational, internal, to to do so. Um and I mean Google, Google and open AI right now, right? Google, yeah, exactly. Google's the best example, I think, right now, of of of like a tough one there. Where if it's a sustaining innovation, something oh sorry, go. No, I'm saying if it's a sustaining innovation, something like mobile, um, then like the the incumbents can capture it and and and just like make even more money with it. And I think AI has some parts of it that are sustaining innovations, like you know, it's gonna make ads way more efficient, it's gonna make content way more addictive, that all benefits, you know, people like Meta and Google. Um, but there's some parts of it that are disruptive, I think. Like where you have to build entirely new applications where like the advertising model in particular is very challenged by AI because like necessarily you want answers, you don't want adverts, right? Like you don't want the 10 the 10 blue links. Um so I think um I think some of them are or like Google, I think is the is the most interesting one because they in order to really build to become an AI first company, they have to like sepulku their 200 billion a year revenue search business, right? Like you you just don't want um it's just like I don't use Google for almost anything anymore. Like I'm I'm mainly just asking open AI. Um and and I think like how do they adapt to that? And obviously they're they're leading in AI, like Gemini is one of the best models. Um, but it's it's gonna be like a much more competitive. I don't think they're gonna have 90% market share in AI. At least like that's not the base case right now, right? And it's also lower margin uh than like inference is just tradition. Obviously, we're assuming inference is gonna be the business model. They're there it's too early to say, but it's just lower margin than than search. So I don't know, it's interesting.
SPEAKER_01But but I guess I would push back and just say, like, I I'm I'm a Google bull largely because I think they will lose share, but I just think the markets that all of these companies are going after are 10x bigger than the ones that they've been serving. And obviously that's crazy to say when you think about the search and advertising market. But if you think about you know the cost of labor effectively going to zero, then the markets that these companies are going after start to look really big really fast. Um and so I think that's the case if they if they if they can get their shit together and if you think about all the data assets that they have, yeah.
SPEAKER_00I was talking to Paul last night, um, and who's who's on the intelligence side, and one of the main concerns I have with Google is it doesn't feel like they're doing a really good job productizing their really good tech. Like I like I don't use Gemini daily, like I don't feel like it's in my workflow. And I feel like the user just identifies Chat GPT with AI. So despite me being a Google bull, and I and I love all their dozens of AI releases, I'm really struggling on the product side. And Zay, to your point, like are they really gonna throw Gemini and replace search? Like that'd be kind of bearish, right?
SPEAKER_01Yeah, I I guess, I guess my view you is that you have to believe that they execute. But if you think about all of the touch points that they've had on you for decades, between you know, the cookies from the ad side, the browser, all of the browser assets that's starting to look pretty interesting. Gmail, YouTube, um, you know, Google Maps, you know, I don't know, uh a gazillion of these things, like the the workflows that we use at Delphi. Like if you can wrap all of that into an interface that's basically like a personal assistant and Android and potentially coming out with a hardware device, um, then I think they they have a like a very, very strong bid to basically be the verticalized integrator that services that basically liaises on behalf of you with the rest of the internet. And that's what OpenAI wants to do. And I think on the chat side, they clearly carved out a really good position. But I still think Google has a ton of assets that are very, very difficult to replicate if they can manage to um coalesce them into you know kind of a single bundled offering.
SPEAKER_02Yeah. I mean, so does Apple though, and they're still fumbling hard, right?
SPEAKER_01Like Apple seems like dropped the bag so so aggressive. Yeah, don't come in here, don't come in here with Apple, dude.
SPEAKER_02I mean, Apple's Apple's super uh Apple's by the same thesis, right? They just like own the the endpoint for the consumer. Like they should be able to just be the main connector between you and and like they should be the best position to build this, but they've just sort of fumbled.
SPEAKER_00I mean, Zay to your point on picking the Mag 7, like I think you outperform just by excluding Apple.
SPEAKER_02Yeah. Like at least right now.
SPEAKER_00What what's the bull case for Apple and AI? You'd say. I mean, remember, they've dropped the ball on intelligence, they they ship summaries, they haven't acquired anybody, they didn't take it seriously. Like, what would be the bull case for them?
SPEAKER_02They like acquire open AI or something, or acquire like anthropic, or like I mean it's it's such a small part of their it's like the they could acquire anthropic for like two and a half percent of their market cap or something, you know? It's like does does acquiring something like that make Apple two and a half percent more valuable? Like it's a no-brainer, right? I think the stock would pump like 10% on the day. I think if they acquired open AI, the this this the stock would pump on the day. That's my uh that's what I think anyway. But um, yeah, it's also interesting to think where we were talking about that before the show went live, like where crypto fits in here. Cause uh for me, like high-level AI is sort of a bet on just an abundant future, like this, this uh or or not, you know, or or we blow ourselves up, but but but in the in the bull case at least it's like an abundant future, right? Where we just have free labor, robots doing everything, like you know, five to ten percent GDP growth, um and and just yeah, like these DI companies just both capture and create more and more of the of the world's value. And crypto to me is it's a bet on debasement, but it's also just a hedge against like currency debasement and and uh and like all the sort of historically pretty predictable consequences of that, of like capital controls, wars, just social upheaval with with AI sort of accelerating wealth inequality. And so I almost think like there's just two parts of of like the I don't know, like the Chris Call Dragon portfolio, you know, you have like some some some crypto and gold, and then you have just the AI assets, and you're like the you know, the the bear case and the bull case for the world.
SPEAKER_00I Zay, I ha I have a question for both of you on the Anthropic point you brought up. Um we're seeing this a lot with the the funds we look at, just the safety versus accelerationism debate. And it seems like anthropic is six months behind on releases because they have such crazy safety. And the accelerationism guys like Grok and now OpenAI just release rapidly. And the thesis, I guess, is if you focus on safety, you could be way further ahead if the acceleration guys blow up and the government shuts them down. Yada yada. I'm curious where you both land on this debate. Are you pro safety in the Anthropic camp or are you pro accelerationism in the other camp?
SPEAKER_02Yeah. Um I mean, I think before answering that, I think Anthropic has like the best coding model, and it's not that close. Like uh, and and people sort of are even replacing cursor with it, which aggregates all models just because Claude Code is so dope. And I think that's a giant, giant market. Like I think the like I don't it's it's hard to quantify just how how big that is, but if you can have like software engineering agents, like most of the world's uh value right now is created by by software, right? And I think that's like more and more uh is is the case. Maybe robotics will will sort of change that again. But um, and so I think that's a huge market where they're pretty pretty um like a head in. So uh yeah, I think that's super valuable. Um on the safety angle, yeah. I don't know. I I went to we we went to the anthropic office, right, when we were in SF. Um and I think it's I I think it's nice to have uh people like that that are super smart, uh, you know, pretty autistic, and and very and and really care about safety. Like I I think it's nice that someone is doing that. And it gave me the feeling that if you're a really smart engineer, uh with with like a strong kind of ethical code around this stuff too, that's kind of the place you'd want to be. Like they all are very serious, um, like they seriously care about this and are doing things the right way. I think that's a cultural advantage that you just don't get somewhere like meta, right? Where you're where you're paying up for it for just paying a billion dollars, whatever, for engineers. So I think that's a moat. And then also I do think at some point safety will become like just required. Like there'll be some some basic safety standards that you have to follow, like some kind of testing framework, some kind of like interpretability. You have to know what's going on with the model in some way. And maybe Anthropic is sort of you know ready for that beforehand or whatever. I don't know. But uh definitely I definitely hated whatever their release was where they're like ratting on users and stuff. Like that's the wrong kind of thing.
SPEAKER_00Oh, like calling the FBI and when you do it.
SPEAKER_02I I get where it's coming from, but that that's like yeah, I don't want my model ratting me out to the FBI, and neither does pondering for sure with some of the stuff he gets up to of his models.
SPEAKER_01Yeah, thanks. Thanks, Mama, for listening. Really, really love, really love working at Delphi.
SPEAKER_00Don't check my computer.
SPEAKER_01Um great guys, really great team. We love you. Um so so so so back to safety versus acceleration. Yeah, I kind of feel like at Delphi Intelligence I'm at like a very interesting juncture where on the one hand, you know, Delphi and kind of particularly our DAI um investments have like basically leaned very heavily into um, you know, uh basically keeping things open and acceleration um and basically trying to pull things out of the big labs. So I kind of feel like there's a little bit of a tension where safety is like a little bit more authoritarian, and you basically just have to end up trusting that a very few parties make the right decisions and that can be good, but it also comes with risks, right? And you know, power corrupts, absolute power corrupts absolutely. So there's a little bit of a tension between, hey, do we want like a very few smart people that we think are good to have an insane amount of power over the rest of humanity? Or is it better to take the, you know, kind of roll the dice and take an open source, more pluralistic approach that lets like a lot more of humanity have a say in how these systems actually evolve? Um, but that also comes with risks and there will be irresponsible actors and things like that. And so, you know, you're kind of stuck in a little bit of a catch-22, um, where clearly the like any technology, it's it's going to be dual use and it's going to have, you know, incredible benefits, but also there will be actors that choose to use it negatively. Um, and I think we just have to be cognizant of that and try and build systems where that becomes difficult, but not acknowledging that you will never be able to nip all of the risks in the bud and you still need to go forward anyway.
SPEAKER_00Yeah, no, that that is a good you you Petey, you have like very unique, a very unique view here given how much time you spent in China and in the US. And historically, I think in America, we always viewed China as you know a very controlling uh country, but they are releasing literally all the top best open source models. We're getting freedom from China and centralization from the US. Like, how do you view the China US debacle or just in AI, right? Like your video model report you just dropped on intelligence, like China is really killing it across the board here, too, not just in text-based LMs.
SPEAKER_01Yeah, so it's it it's a very interesting question, and there's a lot of ways I could take it. I actually just did a quick post on our feed that kind of touches on US versus China and the different approaches to AI and kind of the pros and cons of each. Um, but yeah, frankly, one, I would say China has been killing it. They have a ton of talent. If you think about AI researchers and leading AI researchers, probably 50% of them are in China or ethnically Chinese. If you look at the Zuckerberg list of people that he's been poaching from the top labs, there's a lot of Chinese names on there. Um, so yeah, that it's definitely not a shortage of talent that's holding back China. I would say their biggest, I would say politically, you know, they are taking an approach that is much more coordinated. Um, and you know, you you do have organizations within China that are trying to pull uh pool compute, they're trying to pool data, they're trying to keep things open source so their efforts can compound, you know, across the major labs. And also I think there's a very big focus on diffusing the capabilities through the rest of society, right? So into public services, into government, into their industrial base and industrial capacity. So I think there's like less of a focus on building, you know, quote unquote God or you know, the supermodel, and much more on, hey, how can we harness these capabilities and diffuse it throughout our entire society and economy? Whereas in the US, I think like capitalist incentives have pointed towards just like a, you know, a fragmentation of efforts that tend to be closed because that is what it takes in order to garner this investment. But on the pro side, you get a ton of investment, right? And so I do think in the paradigm where the amount of compute matters a lot, and the US has by far more data center capacity than any other country on earth, you know, those incentives do matter quite a bit. So even though they're fragmented, um, I would still give the US the upper hand just because they have so much compute and China is pretty constrained by their chips, their access to chips, and frankly, the capital markets that just haven't been able to attract the same amount of FDI and investment globally because of some of the policy decisions that they've made recently. And so that's kind of the interesting conundrum where China has less access to compute because they're taking a more coordinated approach, but the compute that they have, they are using quite efficiently and they do have a lot of talent and they're letting that compound via opening up a lot of their learnings.
SPEAKER_02What what do you think of the? Because there was this report, um, it was actually linked in BG Squared, it was a really good analysis of like Chinese industrial policy and how that's worked really well for fields where you're going from one to X. So the innovation's been made, and you just need to make it more and more efficient. And you have this, you throw subsidies at it, a bunch of companies compete, and you know, and solar panels obviously worked really well, electric vehicles, batteries, but it didn't work as well in fields where it they needed to go zero to one, right? Like uh chips, there was there wasn't really a big success from the subsidies program. I think they had the and the biotech was was the same thing, and they kind of talk about this in the report. Do you think um yeah, is is this unfair? Do you think China has done some some some zero to ones or or do you think it's a fair charge that like generally better at the one to X?
SPEAKER_01So so I would say historically for a long time that that was largely the case. I think increasingly that is no longer the case. Um but I would say I yeah, like I think.
SPEAKER_02And what's an example you point to where they've they've like truly like innovated, I guess, built something new.
SPEAKER_01Um, I mean, I think there's like if you think about most industrial fields now, I would say like robotics or drones on open source foundational models. Yeah, drones is a good example, I guess. Yeah, I mean, uh, if you go to any emerging market now, they will like they kind of did leapfrog on cars and they do have the best EVs per price point in the world.
SPEAKER_02Um, but that's like but I I feel like that's a one I don't know, they made they made better EVs, right? But the EV was kind of invented by by Tesla, I'd say, and then they made them much better, and like obviously POID is crushing it. Would it yeah? I it's hard for me to think of an example of like true maybe Deep Seek, but that also seems like they found a way more efficient architecture. Again, I don't I'm not in the weeds enough to know whether you could consider that a zero to one innovation, but it seems like a way more efficient architecture, right? Which is what the Chinese are insanely good at. Um, and maybe that's all you need now that AI sort of you have the transformers and and it's just like scaling that up. I don't know. Well, do you have to be first UX is something new, huh?
SPEAKER_00Zay, are you strong that you have to be the first mover to win here? Like if China is not the first mover, do you view them as not winning?
SPEAKER_02Or no, I don't think so. I I just wonder how many more, yeah. Because you might need more zero to one innovations in in different fields to to win, right? Uh or or not. Like Chiny, China has way more like uh power capacity. Uh they're they're coming up with really sick innovations on the on the model side.
SPEAKER_01Um maybe that's I mean, pretty this is this is actually this is I mean, this is pretty lame and it's not really like deep tech, but I would say like anyone who's spent time in China's consumer internet ecosystem, like there's a lot that happens there that was that would never came from the West, right? And frankly, like I mean, TikTok is also a pretty good example.
SPEAKER_02Yeah, TikTok is a great example too.
SPEAKER_01Yeah. Yeah. So so I I I feel like, yes, there are, but yeah, I still think China is like really good at scaling up stuff that works. Um, but I do think, yeah, to your point, in a world where, you know, uh software development is starting to get commoditized, then those variables do start to matter quite a bit. But I also think that China um does have challenges in terms of efficient capital allocation. Um, and so they can do things really, really good, but that also comes at a cost of other things that might not get made or innovations that might not have happened because they kind of crowded out the private sector. And they said, hey, we're gonna build semiconductors, we're gonna build EVs, and we're gonna be the best in the world at you know, robotics or yada, yada, yada. But you know, they're uh that comes at a cost of other things that don't get funded. Like, you know, they cracked down really heavily on the gaming ecosystem. And obviously, NVIDIA started out as like a gaming chips company and then became the most important. So you yeah, you can never like you never know the the path that a lot of these companies will take. Um, and so I am still skeptical of like you know, the centrally planned mindset, even though it does lead to some very good outcomes in China and in specific industries.
SPEAKER_02Yeah, that's kind of the argument of the of the paper that you can't direct innovate zero to one innovation at least, like incremental innovation you can direct with subsidies, but you you can't um yeah, I think I think that's pretty interesting.
SPEAKER_00What what's your take though? Like China seems like they're killing it on the open source front, whereas we're not. Does your views change given how open AI is in China that you could you don't have to self-direct the innovation when it's open enough to let anybody sort of build and iterate? Does that change the view at all or no?
SPEAKER_02I I I think uh I don't know. I'm I'm skeptical. I've always been skeptical on open source. I think the only incentive to release it is if you're behind. Like, I think once you're ahead, like I don't think the Chinese companies will continue releasing their models if they're if they're ahead. Like the the game theory just changes, right? If you're behind, it makes sense, you're like uh directly attacking your your competitors, you're building an ecosystem around your thing, deflating the cost of of these models, bringing attention to yourself, and they've crushed it at that, and clearly have the best open source models. Um but the the model capabilities are uh like behind the closed source labs, right? By like quite significantly at this point. Um, and I think if China gets ahead, which uh it's definitely possible that they will, they're they're they're right up there, then I think they would stop open sourcing. Um I don't know. What do you think, PD? I well it doesn't seem stable, a stable equilibrium, a world where the the best open source models are Chinese. That that's like crazy, crazy state of affairs.
SPEAKER_01Yeah, I mean, I I don't know. I guess I do I do see elements to why senior folks in China would both like be scared of open source, but also simultaneously understand why this is a super powerful public good or has the potential to be a super powerful public good. Um, and so, you know, outside of like to your point, outside of Meta, who was like, you know, open like spending tons of money to open source their models largely because they were behind and they didn't want to be dependent on somebody else. Um, I actually think there is a case to be made that having very, very good open source models is like, you know, incredibly good for your economy and not having extraction by like, you know, two or three juggernaut companies is something but you don't have to have like especially in China, you you don't you don't have to, you could just like mandate that they can't extract, right?
SPEAKER_02Like they they don't have to open source them to to do that. You could yeah, I don't know. There's probably you you can do price controls or whatever. That's the advantage, right, of being in in in China. Like open sourcing, I feel like the the if you're ahead, it would just be leaking all those like advancements to the rest of the world in this like very competitive like arms race. I don't know. Um yeah, I I honestly I have I have no idea. Yeah, yeah, it's hard to hard to say. But clearly there's like insane. I mean, we've been speaking to a bunch of um like fund managers and and people in China, and obviously PD is very well connected there, and it's super exciting. I think it's it's definitely one of the most underrated, misunderstood areas in the in the in the West. And like, yeah, I'm I'm I'm super excited about about making some investments there.
SPEAKER_00Um Yeah, it it's a good dovetail to a question I had for for both of you. Like, we're all looking at early stage AI fund managers, we're looking at early stage AI startups and crypto and in and web two as well. And it it sort of begs the question you know, should we just buy the mag 7 open AI anthropic and call it a day, or can we see real innovation, right? It it kind of feels like you know, Bitcoin, Ethereum Sol, and then altcoins in a sense. Um so curious where you both see like early stage either managers or projects able to like effectively compete with trillion dollar companies, unlimited compute spend, mass data. It just seems really hard. Uh, I have my own views here, but but I'd rather ask you guys first.
SPEAKER_02Yeah. Uh you you want to start, Pete, or should I?
SPEAKER_01Yeah, yeah, I mean, happy, happy to go first. I would say like I do kind of see it as just different risk reward profiles. Like, I'm I'm obviously quite bullish on large tech companies that have good AI capabilities in both the US and China. Like, I think there are a lot of returns on those assets. And basically having distribution and compute and plugging in new capabilities to your massive ecosystems that basically expand your CAM. Like, I think that will continue to happen. Um, but at the same time, like those companies are relatively small portions of their overall economy. And I think, you know, there for every Mag 7, there's 493 companies in the S P 500 that are not those companies, right? And those companies tend to be a little bit slower moving for the most part and probably less likely to adopt these new tools. And so I just think there's a ton of white space for very smart, motivated people that are building AI native companies that have maybe a lot less employees, that you know, have a bunch of geniuses in the data center that they figure out how to harness, um, that will also build really sizable companies that are very difficult to predict. Um, and so yeah, like I think both of those things can be true, where hey, big companies that have a lot of AI assets that figure out how to use the technology and also sell it to everybody else will do well. But I still think there will be a ton of pockets where smaller players that leverage these, you know, basically these mountains of GPUs and these increasingly capable foundational models to build, you know, novel services can also build very sizable companies with a very small footprint. Yeah.
SPEAKER_02I agree. I agree. Like I think there was some point initially where I was like, maybe all this just gets captured by the big tech companies. They're like so perfectly well positioned. But I I've changed my mind there. And obviously it's hard not to seeing the pace of just you've never seen private companies grow this fast, right? Whether it's cursor getting to 500 million RR or like lovable. Um, like yeah, some of these private companies are growing at a astounding rates. And I just think that uh ultimately the question you want to ask is does this thing uh are they like uh uh sort of excited and happy when the models get better? Like it is it a is it a tailwind for this startup or not? Uh if it if it's not, then I think you're you're you're fucked. You're sort of fighting the scaling, scaling lows, and you just don't want to be on the other side of that trade. But there's a bunch of project of products that just get way better as the models improve. In fact, there are a bunch of people building things that literally don't work with the current models, but they're they're just betting on the scaling laws that the models will improve enough for them to get there. Um and so, and I think there's loads of stuff there, whether it's hardware, uh, there's a bunch of people building different uh chip architectures, like obviously etched is is probably the most famous one, atomic semiconductors, there's there, there's a bunch of others. Um there's and then one I'm really excited about is sort of the the thing that's been in a derogatory way referred to as the the GPT wrappers. Um I really think that the models already today can do so much more than what most people are using them for, uh, including us. Like even the people on the on the on the frontier. I think we we we struggle to use them in the best way just because of the the missing, like both context engineering and like just making products that make it easy to get that out of the models. And I think in every vertical, you're gonna have uh like multi-billion dollar businesses that are just optimizing around a particular vertical, building a UX around it, uh, making context engineering around it really easy, like adding your files, ingesting all that, and and building like massive, massive businesses there. Um yeah, so I'm I'm pretty bullish on the private company opportunity. Um, I do think it's hard to to invest in right now. Like there's definitely some negative selection if you're not deep into AI, where the deals you have access to. And I think one thing you get being in crypto for so long is just like a paranoia about your edge, right? Like you always want to know why you have edge in this trade, like why you're not just gonna get get dumped on, like why like you always are obsessed with that. And I think in AI it's pretty hard to have edge right now. I think maybe China, like if you if you wanted to move to China and invest there, maybe there's some edge, although there's great fund managers there too. Um, so I think that is a challenge, but there's definitely gonna be, I think, some massive opportunities in in private companies, and also the labs are gonna do really well too. Like I think that everything is just gonna go up. Like, I I don't think any yeah, I just don't think any of it's priced in. No, nor nor could it be really.
SPEAKER_00Yeah, I mean, one of the things sorry, you gonna be no, I was gonna well, I was just gonna turn the tables.
SPEAKER_01Tommy, you're you're always the the interviewer, and I know you have a lot of good takes as well. So I wanted to hear um, you know, if you if you had any thoughts around kind of value accrual and the areas where you were particularly excited.
SPEAKER_00Uh yeah, no, I appreciate it. Um I think uh comment Zay made a couple of days ago in in New York really stuck with me that if you it's hard to compete casually in all of these fields, like and cover everything, like there'll always be somebody who's hyper focused on out competing and winning in a specific area, and I think. When you couple that with just the absurd, never-ending overflow of talent we're seeing on the early stage AI side with the fund managers and with the projects themselves, I find it really difficult to imagine Meta, Google, Facebook, OpenAI, Anthropic, SSI thinking machines capturing every single vertical in a in a big way. We're seeing OpenAI win like the general side, meta on companions and probably social media dystopia, but like the long tail of people hyper focused on ideas is just crazy. It feels like DeFi Summer, but on steroids, um, so yeah, I'm I'm pretty bullish on the early stage side. I definitely go back and forth quite a bit though. Um like when you see something like Google's I.O. release with just 40 AI products across the board, you know, topping charts for text-based LLMs, video, medical, and everything else across the board, it it becomes exciting. Um, but then when you think of all of the startups that can use all of those in a specific field, uh, I think I get re-excited. Uh, but long term, I think if we invest on the early stage side, we end up with Google and OpenAI stock anyway, when they when they buy out all of the early stage projects. So we'll see. Um yeah, so we'll we'll see what what happens on on that side. Um I think we should probably talk a little bit about robotics though. What do you guys think?
SPEAKER_01Yeah, um, yeah, yeah, let's uh let's do it. Lot to talk about.
SPEAKER_00So we have uh we have the US and then we have China's hardware dominance in Shenzhen. You could walk across the street at any part you want in the world. So it seems like China is just universally winning on robotics, but but PD, I want to hear your take. You're you're on the ground.
SPEAKER_01Yeah, I I guess I would say like I think China clearly has a lot of advantages in terms of the supply chains and the components that you know are built up around Shenzhen. And um, you know, I I think those agglomeration effects are real, and there's just a lot of iteration and implicit knowledge that comes from being the manufacturing base of the world. And if you look at what happened in EVs, like, you know, I think the there's very large swathes of that supply chain that are directly applicable to robotics. Um, but I also think geopolitically, there's a lot of markets that will not be excited about having Chinese robots running around just uh for strategic and national security reasons. And that's going to be like a pretty big challenge for China, who subsidized you know, a lot of manufacturing areas to try and outgrow their property bust or at least kind of reinvigorate growth. But they have such a trade surplus already, um, and they're starting to run into trade barriers from their biggest customers, like the US and the EU. And so I think I I largely think that you know the US will have a domestic industry that services the US demand, and they'll have more global supply chains where they source like different component parts out, and then China will end up winning large swathes of the global south with very, very cost competitive high-quality robots. And so that's, I think, you know, both markets will have large winners. Um, I think in the event that it was just like if the market stayed open, I actually think purely on price per quality, China seems like it would have a lot of very real advantages. But I think realistically, that will run into political considerations and trade considerations that will stall the number of sales that they make globally.
SPEAKER_02Well, what what can it's interesting, like what Kang says is, and like clearly China is is like unparalleled on the hardware supply chain side. When I spoke to Andrew Kang about this, he says that the Chinese companies just don't have the the intelligence level, as in not the people, sorry, the the the robot intelligence level. Like uh they don't have the software side anywhere near as strong as the as the the sort of big US companies, I guess figure Aptronic and and Tesla with with Optimus. I don't know if that's true or not, or how you would or how you would even how we would even find out at this point, but that is that's his thesis, and that like robots have been around for ages and they're really nothing without the without the software side. Um so yeah, I don't know. Do you have any take on that?
SPEAKER_01Or so I I do. I from what I've looked I'm not a robotics expert. From the research that I've read, I do think that's true. And there are, you know, companies in the US like Google that are pretty unparalleled in terms of like the the brain, the simulations that they can run and things like that. Um so I think it'll be very interesting to see, you know, how much you can scale purely via simulation and software, and how how much the feedback loops between hardware and software deployments matter. And I think both will matter quite a bit. Um and so yeah, I I think you're you're seeing several different buckets on the robotic side. Some is basically just like integrators, like you know, your Teslas and probably your BYDs or your Xiaomi's. Others are like Google, and then you partner with someone like an Aptronic, and then others are like physical intelligence, where it's like purely software that hopefully can work with like any modular hardware provider globally. Um and I think uh, you know, it remains to be seen which approach proves dominant in the end. But if you look at kind of the, I guess, the the recent NVIDIA robotics team, it does seem like the the approach that they're using is a mix of as much video data as they can get their hands on, with as much simulation data as they can get their hands on, with as much like high quality real world robotics data that they can get their hands on and trying to like piece it together, noting that the last part is in super dearth of supply. I actually talked with one company. I talked with one company that was super interesting, but they were basically like uh a bunch of AI engineers that work with industrial robotics companies in Korea and Japan. And they basically, hey, where the talent will be your effectively outsourced lab, and they basically go in and help them to uh collect all of the data by just setting up like really high quality uh cameras and sensors and basically go through what each of the robots do in a particular um, you know, uh job or task. And and so like I think data sets like that are really unique where you you clearly need to have some AI talent, but you also need to be in proximity to the actual tasks and the actual like robot manufacturing themselves to to also do it really well. So I kind of feel like no one has the complete picture. Um and it'll be interesting to see which approach ends up ends up winning. I I I don't know. Yeah.
SPEAKER_02I think that's a very good contestant. Like I think that'll be a robotics company in the Mag 7 for sure. In the new Mac 7.
SPEAKER_00Like I totally agree. Yeah, yeah.
SPEAKER_02Maybe a few, maybe even like more than one, as as you said, PD, maybe in the Mag 25 there'll be like a few of them. I think that's pretty likely.
SPEAKER_00The Mag 7 seems like such a slim number for the number of companies we're proposing replace what's in there. Like uh OpenAI anthropic robotics company, that's three alone. Yeah, it seems like they're yeah, it seems like they kind of get completely replaced.
SPEAKER_01Well well, hopefully, hopefully Andrew Kong invites us to his private island once all these these humanoid companies crack the Mag 7.
SPEAKER_00You gotta frequently stay in touch with them. Make sure you get that.
SPEAKER_01Uh yeah, yeah, get get some allocation.
SPEAKER_00Yeah. Maybe let's cover what we want from Intelligence, Jose and PD for for people listening, maybe like across intelligence ventures, and uh, and we can close out there.
SPEAKER_02Yeah. Um go ahead. Yeah, you start, PD.
SPEAKER_01Okay, yeah. I mean, I I guess um to me it kind of seems like we we're taking both an interest, like a top-down approach on the research side and a bottoms-up approach on the investment side. And I think it is quite complementary to Delphi and kind of their networked approach to both research and investing. Um, but yeah, I mean, uh on the research side, like the goal is basically to just dive deep into particular subdomains where we're excited about and hopefully can surface interesting insights and investment opportunities, both for our subscribers and for, you know, uh the ventures fund that Delphi has in-house. So um, you know, I think in the coming weeks, you will be able to get a flavor of the types of deep dives that we will be consistently putting out. Like we mentioned, we had one on videos. I'm gonna have another one on data. We're having multiple on reinforcement learning, we're having one on AI browsers. Obviously, there's one in robotics that's uh very, very in in the oven, Jose. Um, and then on the and then on the on the bottoms upside, like I do think we've been working hard as a team about finding the right venture partners and scouts, you know, in Silicon Valley, in Shanghai, um, in London to try and like, you know, uh basically meet with entrepreneurs that are in the right circles and then also uh build out our network of early stage GPs, both for access and kind of deal flow. Um, so we uh can kind of marry those two things, where building up our expertise on the research side and going deep in particular subdomains, and then um also kind of building up privileged deal flow um on the venture side via relationships with early stage managers. So so yeah, that's uh kind of the the approach. Um but Jose, feel free to tell me what I missed.
SPEAKER_02Nah, you you you covered you covered it very well. I think I think for me, like I guess on a feels basis, it's sort of if you read Delphi research from the beginning, you would not only be smarter, you would have done pretty well, right? Like you would have you would have been in Bitcoin very early, calling calling the the the bottom back in back in like 2018, right? When Jan did that that great report. Um, you would have been in stuff like you would have found about about Thorchain very early, you would have then found out about Axie super early. Uh in 2022, you would have had Sol from Cetaris, right? At like 10 bucks, like a Bible on on why Sol was was was dope. Like and I think for me for intelligence, it would be great if it was if it was the same, right? Like you're not just getting smarter, but there's actionable stuff in there that kind of can change people's lives if they to have them look into something that they wouldn't have otherwise and and potentially allocate capital towards it. Although obviously none of none of none of this is investment advice. But I do think that's the uh that would be a a a really cool goal. If in if in 10 years uh we had managed to do even even a fraction of what I think Delphi did for for people who read it in crypto, and obviously it's gonna be free, so I think the impact can be so much larger, which is which is super cool. Um, that would be a really nice thing to have happened. Just to see us write about stuff early and then and then seeing it play out, um, is is really satisfying. That's what I would love for Delphi Intelligence in in 10 years.
SPEAKER_00Yeah, I I love that guys. I mean, for for me, it it's everything you guys said. I I do think that given how we started Delphi, even though we're not experts, you know, leading experts yet in AI, I think we'll get there. I think we can look at a lot of these early stage investments from a a pretty solid first principles lens and are willing to take the risk in new people and and new ideas that can really change the world outside the preordained views of like Web2 AI Labs and what is the the norm right now. So I'm I'm pretty excited and hopeful that we can we can do that. So if you are that person, obviously get in touch with Mizay or PD. Um and uh shameless chill here for these guys. They're starting a new podcast called 2035, which will live right on the intelligence site. So right where you read all the reports, you can get access to to these two chads, talking AI uh with with some really smart people. So uh pretty excited for that. Um, but guys, thank you for hopping on. I know time zones here are are horrible to plan, uh, so it means a lot.
SPEAKER_01I appreciate it. It was super fun. I really appreciate it.
