Loading summary
Transcript40 lines
- [00:00]
A
Anthony, you have increasingly been talking about open source versus closed source AI models. Why are you watching this topic right now?
- [00:07]
B
Well, I think that there's something called the FAT protocol thesis in crypto that a lot of people have forgotten about. And the whole idea of the fat protocol thesis was that there was going to be a protocol that was going to continue to get fatter and fatter by eating more and more of the stack. And eventually, whether it was Bitcoin or Ethereum or Solana or whatever that protocol is where all the value was going to accrue. What we saw in crypto is that's not true. Instead what happens is the protocol is a very important part of the infrastructure and the foundation of these ecosystems, but actually you still got consumer software, many other components of kind of the infrastructure. And so the fat protocol thesis, as I thought early on, was dumb and ended up not being true. We're seeing this play all over again in AI. Like the AI people should just be quiet and go listen to the crypto people. The fat protocol thesis is wrong and, and the fat AI model thesis is wrong as well. And so what you end up finding is that once you realize that the FAT model thesis is incorrect, then you start to realize that there's not only going to be fragmentation, so you're going to get consumer software, you're going to get specialized workflows, you're going to get kind of general purpose models and all the way through the stack, but also you're going to get rabid competition between closed source and open source, between American companies, Chinese companies, and many other different cuts of this. And so the reason why I keep talking about it is because right now everyone is like, oh my God, the American closed source models of ChatGPT opus Claude, you know, all these different models that everyone has been using are now getting their butts kicked on the playing field by the Chinese open weight open source models. And frankly, I'm pretty disappointed in the American companies because they're basically crying like little babies, going to the White House saying, oh, you should ban the Chinese open source open weight models. Like, no, the government shouldn't ban them, you should just build better models, you should compete on the field. And ultimately what I think we are watching happen here is there's a last gasp of desperation. And the reason I call it the last gasp of desperation is because these large language model providers are realizing, wait a second, I'm spending all this money to go and train these foundational models. We as consumers should be thanking these companies for doing it. Sam Altman, Dario Modi, all these folks, they have done an incredible service to the world by going and training these large language models. But what they now are realizing is that the Chinese companies are simply drafting off that work via inference and kind of all this stuff. And so a second that you start to see that there is a different way to do this, it's almost like, do you remember recently, Josh Kerr broke the 1 mile world record? Well, what a lot of people didn't see, didn't pay attention to as he ran those four laps and broke the world record, was when the race started, he ran out, but he was not number one, he was third. And there was two guys in front of him, and those two guys ran in front of him. The first guy ran the first two laps. At the end of the second lap, he stepped aside and he started walking. The second guy in line ran the third lap and then he stepped aside and started walking. And Josh Kerr was left with one lap left in number one place, ready to run. Now why is that? The first two guys, they were there to break the wind and let him draft off of them. So they basically knew, I can run as fast as I possibly can for two or three laps. I'm not going to break the world record, but I'm doing it in service of Josh Kerr, who's behind me because he called a shot. He's going to go break the world record today. That's basically what the Chinese models are doing to the American companies. They, there's a, that's so cute that you're spending so much time and money to go train these foundational models. Wouldn't it be bad if we simply drafted off of you and we were able to then go create these open weight open source models that are just as competitive and lower cost? So that's why the American companies are getting so upset. I, I hear them, I understand why they're upset, but they gotta figure out how to win in the market, not go and try to just yell and scream about, hey, let's ban the Chinese models. And so where I ultimately believe a lot of value is going to accrue is, is an American open source model. We can get into why I think the American open source models will be the Chinese ones. But that's why I'm talking about it so much is I think that open source, because of what I learned in crypto, with the FAT protocol thesis and the open source systems, et cetera, that is probably going to have a much, much bigger impact in the market than people give it credit for right now,
- [04:33]
A
what you're saying to me, it makes a lot of sense. One, but two, the outcome that I would expect on the scenario you're describing is that Chinese models will win out. Is that what you anticipate?
- [04:44]
B
No.
- [04:45]
A
Okay, tell me more.
- [04:46]
B
The American open source models are going to be the Chinese open source models. And the reason is because I think that people drastically underestimate America's ability to innovate. Literally. We created the foundation models, right? They're drafting off of us. We have the godfather of AI. Jensen Huang is going all in on open source. He understands where this is all going. And I think that American open source models are going to be the winning strategy. Now that doesn't mean that open source OpenAI or anthropic or name your favorite, you know, Grok and Xai, I don't think any of them are like gonna go to zero or go. It just means that there's going to be more fragmentation in the market. And what you're going to find is that people are going to start reverting to these model routers. We've built one ramp, just came out with one that is for other people to use. But I think that really people don't care what model they use. All they care about is I want the most efficient completion of the task that I need done. So what does that mean? I don't care what the token cost is because token cost is only one input. What I care about is actually what was the total cost to complete this task. And so if you say to me, hey, there's a really expensive token, but it uses way less tokens overall, it still may be cheaper than maybe a model that has cheaper tokens but uses way more tokens. So ultimately, as a business owner, as a product developer, what you care about is what, what is the total cost to accomplish this task that has both cost but has token efficiency as part of it. And that's where these model routers and many things are becoming popular. And so what people are starting to realize is you're going to use multiple models in your product. You're going to route queries or tasks to the model that is best suited to do that job. Sometimes you need a general purpose model, sometimes you need a closed source zero data retention policy model. Sometimes you need an open source model. Sometimes, like for example, with the product that we're building with Sylvia, you don't want to use the Chinese open source models because potentially they have more socialist ideas baked into the weights. And if you're talking about finance in the Western world, then the cultural weight, which is something that you're going to start hearing a lot more about, the cultural weight should be more towards capitalism than socialism. And so again, there's a lot of nuance here and we're all figuring out together. If you had talked to me two months ago, I would have had way different understanding of this stuff. But I think that this is the rapid, you know, kind of group learning that's going on in this industry. And so I just think that open source is going to be way, way more valuable. And American open source is probably the, like, dark horse in the entire race.
- [07:17]
A
So I have a few friends that only use Deep Seek and they tell me they use Deep Seq because it's cheaper, it's as fast and as effective for the tasks they need, which is effectively, you know, search on steroids is what they're looking for. Do you think that that is going to be more and more popular just based on the cost effectiveness of something like a Deep SEQ or whatever, whatever else comes out? Because, you know, I pay, however, couple hundred bucks a month for various AI products, but because I pay, I don't use Deep Seek. There's probably a bunch of people that just don't want to pay.
- [07:55]
B
Of course, cost is always an important part. But guess what recently happened? Meta came out and they said that they're coming out with a cheaper model, right? Grok came out and they said they're coming out with a cheaper model. Nvidia and many other companies are starting to invest in American open source. It's like the price war is on. What a lot of folks I think don't really realize. And some of this is unfounded critiques, but some of it has probably truth to it is one of the big critiques against the big labs is are they marking up the tokens way more than they should be? Well, Jeff Bezos made the moniker your margin is my opportunity. Famous for a reason. And so there's a bunch of people who say, that's a cute business you have there. Wouldn't it be bad if I took it? So that's one thing. The second thing is when you are actually pinging an API, there's a lot of people who are wondering, are you actually routing my query to the highest value model that I'm supposed to be getting, or are you routing it to a lesser model and therefore you're actually expanding your margin at my expense? Again, I'm not saying that they're doing that or not. I just See people online talking about this. And so when you use a single model, like a deep SEQ or something, right, you know full well what is the cost of hosting. You know where your data is going, you understand what is the actual model that's being leveraged. Like there's all these components and you get to fine tune it. And so people who are technical can then take something off the shelf. They can start to fine tune it, they can start to mess with the weights, they can really start to make this valuable for their specialized use case. I think that's why you see adoption of these things.
- [09:25]
A
Okay, so Anthony, I have to ask you, Bitcoin as an AI play, how does that tie together? Because I know you've talked a lot about this, Jordy Visser talks a lot about this. A lot of people still see Bitcoin as its own ecosystem, its own asset in a vacuum, so to speak. But you have talked a lot about how it is actually a way to invest in AI. Can you explain that?
- [09:48]
B
Well, I think that most great investment themes have a barbell to them, and sometimes those barbells are opposing themes. So there's a bunch of hedge funds in the public market, for example, they go long, the disruptor, short, the disrupted. And so it's like a pair trade, right? Well, that's what we're watching play out with Bitcoin and AI. Well, what's the pair trade? You're getting abundance of intelligence and you're getting scarcity of value. So what you're getting is AI is delivering this abundance of intelligence to people at a very low cost. People have been searching all around the world for the highest intellig to mankind for decades and centuries, and now all of a sudden it's available in an API, right? And so of course they're going to want that, but that brings enormous abundance. When you have tons of abundance, what do you get? You get abundance of content, abundance of code, abundance of creativity, abundance of data and information, etc. Well, in a world of abundance, what becomes valuable is scarcity. And so when you get abundance of intelligence, you then want scarcity of value. And I think that's where bitcoin becomes really interesting. And so it kind of comes back to this idea of like, we have abundance of dollars. Those don't seem to be getting more wealthy, right? They're not getting stronger. And so I think that is ultimately the pair trade is abundance of intelligence and scarcity of value. AI and Bitcoin are actually all part of the same trade.
- [11:05]
A
You know, it's Funny to think several years ago you've been in bitcoin for a long time, buying it now as a quote. AI play is not how it started originally, but now I think that's going to increasingly become one of the theses behind why to buy Bitcoin. As more people catch on to what you're talking about, what Jordy Vizard is talking about, I think that's pretty compelling.
- [11:28]
B
I think that it's more related though than people think. A very early argument for bitcoin was there's going to be abundance of dollars. And my view is that AI is deflationary and robotics is deflationary. And so if you get deflationary forces, then you are going to print more money, which we already see. Like people for forget. Since 2019, inflation has been averaging 4% a year. That's a big number. That's double the Fed's stated target. Well, right now, if inflation is 3.5%, why since December has the Fed been expanding their balance sheet to the tune of $200 billion? They know what's going on. They're not going to admit that they are changing their inflation target. But the only reason why you would run inflation hot at over 4% per year since 2019 is because of either external shock Covid, or you are okay with it being run hotter because you understand these deflationary forces of AI, robotics, deportations, tariffs, et cetera. And so if you have inflation at 3.5%, why are you expanding your balance sheet? Is because they realize that they're going to have to devalue the dollar. They're going to deal with the national debt, the deflationary forces, et cetera. So in that world, which is now being kind of accelerated by AI, that leads to more value to Bitcoin. Because if there's abundance of dollars, there's scarcity of bitcoin. And so it's all the same argument. It's just that AI, it's not that AI is like a new thing. AI just became an accelerant to the story that was already there for bitcoin. The problem is most people, they're not thinking, you know, kind of second and third order effects. They're just thinking about AI, you know, bitcoin, right? Like they, they just think of them as two separate industries.
- [13:07]
A
That makes total sense. And I'm on board with you. I will say I don't think the Fed, I don't know if we should give them that much credit that they are positioned for an AI world. I don't think they're running inflation hot for AI. I think they just happen to be doing it, and it might line up in the right way. You know, I think the Fed's made a lot of mistakes over the last few years. A lot, A lot.
- [13:30]
B
Almost all mistakes, yes, can be traced
- [13:33]
A
back to the Fed, which is why I don't think they're planning for an AI ecosystem.
- [13:37]
B
I don't think that they are. I think they know how to spell AI. That's probably about the extent of their knowledge. But I don't think they're sitting there being like, oh, AI is coming. I think what they see is that there's more deflationary forces in the economy than most people realize. And so tariffs. I think they've kind of waved the white flag, like, hey, those weren't as inflationary as we thought they were going to be. Deportations. They're not dumb. They realize that if we have, you know, no net growth to the population, that's deflationary. Right. And then I think that they understand technology in general is deflationary, and they see the investment dollars flowing into technology. Now, do they go break out AI and robotics versus everything else? Probably not. But I think that they. They're not dumb. They understand that there are these deflationary forces, and so that's probably why they're a little bit more comfortable with inflation running hot. You just can't tell the people that, because the second you say, hey, we're cool with 3% or 4%, all hell breaks loose.
- [14:28]
A
Yeah. The economists would go crazy, too. Okay.
- [14:31]
B
And they're already crazy.
- [14:33]
A
Fair enough. They're usually wrong. You are a public company CEO, which I also work for the public company as chief market strategist. And this company's main product is. Silvia, tell me about this product. And also I think maybe more interesting right now, the stock is down about 80% from its high. How are you thinking about this? Because it is your job.
- [14:58]
B
Well, you're being nice. You should just say, I'm the public company CEO. That stock is down 80, 90%, and the entire world's betting against us.
- [15:06]
A
I love.
- [15:06]
B
I love every second of it. Right. If you go and you look, the company went public in December of last year. We raised quite a bit of money, and we put a lot of that money into Bitcoin, which we had told investors we were going to do. They were excited about us doing that. At one point, we actually have a press release that went out that we were up $100 million in unrealized gains on the bitcoin purchase. So for a while, people were like, you guys are geniuses. This is incredible. Look at how smart you guys are. And then bitcoin went down 50%. And all of a sudden everyone said, you guys are idiots. And I like to remind them, I say, well, if I was smart and dumb, but I never did anything different, then maybe it's the market now that doesn't absolve the situation. Right. We still have a stock that is down. And so you have two options at that point. Well, really three options. You can quit, you can do nothing, or you can try to turn around the business.
- [16:01]
A
Right.
- [16:01]
B
And so what I've chosen to do is third option. We're gonna go turn it around. And I keep calling it the impossible comeback when your stock drops 80, 90% coming right out. I mean, we went public in December. It was down 80, 90% within three months. You just get punched in the face and everyone's sitting there and they're laughing at you like you're an idiot. You're, you suck at this. Blah, blah, whatever. Okay, well, who. How many people in the world think we can make it work? How many people think we could turn around the business? Not many. There's people inside this building, and then there's a couple of handful of shareholders. I love those odds because guess what that means. That means there's tons of asymmetry. It means that you have something that is non consensus. And if we're right about it, we can create a lot of value for all of the owners of the business. Now we got to go make sure that we're right. We got to go make sure we actually do the thing that we say we can do. And, and so I thrive in these situations where basically my back's up against the wall. Everyone's betting against you. There's very low probability of success according to the outside voices. But you have a small team of people who. I call them the misfits. Right? You got a small team of misfits who all believe, like, what else do you want when you're building a company?
- [17:11]
A
I mean, I'm very excited. I think you're one of the misfits. I know things are moving in the right direction. Let's talk about Sylvia. For people who don't know this is a personal financial product, tell us about why you think this is valuable.
- [17:26]
B
Well, I just talk about my personal experience when AI really started to become popular. I was like, this is awesome. I want to take this new superhuman intelligence. I want to use it for things that I'M interested in. What am I interested in investing? Well, if I have superhuman intelligence, like I have the smartest person in the world, why don't I use this to help me make more money? That's like, you know, tale as old as time, new technology, how do I make money?
- [17:46]
A
Right?
- [17:47]
B
And so I went to ChatGPT and these different services and I said, you know, what do you think I should do? And it kept telling me, I don't know anything about your situation. So I started telling it some stuff. Then I started screenshotting my, you know, bank accounts or brokerage accounts or whatever and just trying to feed it information. There's a pain in the ass. I had to constantly keep screenshotting stuff and uploading it. I had to keep telling it stuff even though I'd already told it that information, whatever. It's just like it wasn't a good experience, it wasn't very valuable. And so I started talking to Shane Knorr, who was by far the best engineer we had any of our companies.
- [18:17]
A
Cracked Engineer.
- [18:18]
B
Yeah. And I just was like, look man, why don't we just build something to help solve this problem? And you know, he really understood, I think a lot of the power of the AI and what we could do there, et cetera. And I said, look, I'm willing to bet if we build a product that's valuable for me, there are millions of other people who want this product as well.
- [18:32]
A
Right.
- [18:32]
B
And I just happen to have a direct distribution channel to a lot of them. And so the way that Sylvia works today is you go in, you attach all of your financial accounts, your bank account, brokerage, crypto account, credit cards, et cetera. You can upload your private investments, real estate, cars, collectibles, whatever. That data very important to me. Right. My data is on the platform, is it is, my PII is encrypted, anonymized, and then all of the Data itself is SOC 2 compliant, you know, in terms of the company, et cetera. And so it was important to me that if I'm going to put my data in, I don't want the engineers on our team going and seeing that, you know, I bought this stock or I got this much money in my bank account or whatever, so then you can start talking to Sylvia. And the beauty is that we've built a bunch of proprietary technology, we've built file systems, we've built memory, we've built multi agent orchestrations, we've built a model router. All these different things that help Sylvia give you better answers. But Every time you talk to Sylvia, it comes with the context of your portfolio. So the best example is if you go to Google or ChatGPT and you ask, how do I get my taxes down? How do I pay less in taxes? It'll give you super generic information that's frankly not that valuable. If you ask Sylvia, Sylvia will go asset by asset in your portfolio and tell you exactly what things you should consider are based on your portfolio. That personalized information is really valuable. If you want a kid to learn, what do you do? You give them personalized one on one tutoring. If you want to have very healthy person, what do you do? Personalized one on one private healthcare. Well, for finance, what should you do? Personalize one on one insights into your personal finances using superhuman intelligence. That's what Sylvia does. The big bet we're making is the exact opposite of what the big model labs are making. They believe in the fat model thesis. They believe that general purpose models are going to be able to do this stuff. I vehemently disagree. We're literally betting our entire company on the fact that specialized workflows, specialized applied AI is going to be much better at serving the client at the end than the general purpose models. And so far, based on the objective view as to how Sylvia is going and how some of those products are going to. I like the fact that it looks like the data is showing that we're making the right bet.
- [20:32]
A
So I will say I use Sylvia every single day. And I was smart, I tried. Well, I know I'm not smart, so I have to use Sylvia. And I was using Sylvia before I joined the company. So now I joined the company. Of course I'm going to know it a little more intimately. I understand the product better. But even when as an outsider, I used it every day, it was very valuable. Helped me with my taxes this year. You know, I have business, I got a job, I got a podcast, you know, a bunch of different income streams that frankly are too hard to figure out in my own head. So Sylvia Baller misfit. Yeah, well, well, Sylvia is the way to figure everything out with all your context and has all the meme stocks I buy everything.
- [21:14]
B
One of the most valuable things you can do. I tweeted it recently. Just go on Sylvia, attach your accounts and then tell Sylvia, roast my portfolio.
- [21:21]
A
See, I'm too scared to do that.
- [21:23]
B
And she will literally go through your portfolio and tell you why you're stupid. And she'll just be like, what are you doing? You got too much concentration here. This is Dumb, you don't have enough cash, whatever, right? It's super valuable because if you go to somebody and you say, hey, you know, you should, you should roast my portfolio, right? That's usually not that valuable because guess what that person does is they basically bullshit you. They don't tell you the truth, right? They don't hurt your feelings like, oh, you're doing a good job, but you know, maybe we don't need as much of this stock. Like, nah, just tell me the trut truth. What do I need to do, Sylvia Roast my portfolio is a great one. Another one, go look at what rich people do to get pay lower taxes. Then tell me what I'm not doing that I should be doing. Easy. Another good thing that you should do, run a Monte Carlo simulation and tell me where I'm likely to be in 10 years. That's pretty eye opening because she looks at what are you good at? What are you not good at? How have you been growing your portfolio? What is your cash flow look like? Etc, and then she runs a Monte Carlo simulation, tells you in 10 years, here's where you're likely to be. You start doing this stuff super, super valuable.
- [22:22]
A
Even just looking at how you could be making money will make you more money just by paying attention to it. Of course, it's, I mean, it is an unbelievable tool. And I'm not just saying that because
- [22:32]
B
the day the data objectively shows that people who are regular users or heavy users, so people who use the product a lot, they grow their net worth faster than people who do not. Now when you go and you look at it, that is true regardless of your net worth or regardless of your income level. But those people have grown their net worth somewhere between, depending on how you cut the data, 16 to 40% their total net worth, 16 to 40% in the last six months drastically outperformed the market, drastically outperformed most other things. And so you look at that and you say to yourself, well, why is that the product 100% is valuable, right? In terms of we see people who use it more often, et cetera. But also there's an element of if you're using the product a lot, you're measuring what's going on, you're paying attention, you're looking at, oh, this is working, this is not working. You're getting information, you're making decisions, you're making changes, you're improving, you're iterating all this stuff. So it's the combination of you're somebody who cares and you got a product now that's really valuable. You put those two things together again. I like our chances.
- [23:28]
A
I like them, too. Anthony, thank you so much for your time, and we'll do it again soon.
- [23:32]
B
Thanks for having me.