Transcript 0:00 [on-hold music] Disclaimers. One, the content shared in this podcast is strictly for informational purposes and should not be construed as financial or legal advice. 0:09 Two, the opinions expressed by the hosts and guests are solely their own and do not represent the views of their employers or affiliated organizations. 0:14 Three, none of the projects or companies discussed should be considered as endorsements or recommendations. 0:18 Four, while efforts are made to ensure accuracy, the information provided may not always be up to date or entirely reliable. 0:23 It is advisable to conduct independent research and seek professional guidance before making any financial or legal decisions. 0:27 Five, the hosts and guests of this podcast shall not be held liable for any consequences or actions resulting from the use of the information provided. 0:33 Six, all content, including text, graphics, and audio, is protected by copyright laws. Unauthorized reproduction or distribution is strictly prohibited. 0:39 Seven, any discussions on sensitive topics aim to foster inclusivity and respectful dialogue, but differing opinions should be expected and not attributed to the podcast creators or sponsors. 0:46 Eight, by listening to this podcast, you acknowledge and agree to these disclaimers, and you are responsible for your own interpretation and application of the information presented. Hey, everyone. 0:54 Uh, welcome to the Intersection Four Podcast. Uh, this is our first episode. Uh, we're very excited to kick this off. 1:02 Um, the, the purpose we're putting together this podcast is, you know, we would have these conversations, uh, between the four of us as friends, uh, all the time, and we thought there were some very interesting insights there that we would love to share, uh, with, uh, everyone else as well. 1:19 So what you have on the podcast is a great set of, uh, different perspectives and backgrounds coming together to talk about whatever is on top of our minds and, and most important to us, uh, around us and to us, uh, in the world. 1:34 Uh, you have an investor, which is me, you have an engineer, which is Sam, uh, you have a founder, uh, which is Brendan, and you have a lawyer, which is Rich. 1:43 Um, so we'd love to kick this off, uh, with a few interesting topics for the week. Uh, but in terms of background, um, I'm the founder, uh, of an investment firm called Pudas21 Capital. 1:56 Um, Sam is an engineer, um, at, uh, Google focused on Web3 technologies, and he's also part of ATX DAO. 2:06 Uh, we have Brendan, who's the founder of Ethena Intelligence, and we have Rich, who's the head of strategy, uh, at Google for Web3 initiatives. Okay. 2:17 So the format that we're shooting for, for the pod is we have four people, four different backgrounds, as Hamiz was saying. 2:23 Uh, I bring more of an engineer in, in, in crypto, kind of like degenerate background and, and very in grassroots. Hamiz brings more of the financial investor background. 2:34 Brendan is running a really cool startup, and Rich brings more of the strategic and, and, and lawyer background. 2:40 So the format that we're gonna be shooting for moving forward is each of us brings one topic of discussion for, for every single episode that we're gonna have moving forward. 2:48 Each of us does the introduction for the topic that they wanna discuss, and then we have like a five to ten-minute discussion around that specific topic. 2:56 The four topics that we have for today, um, and I will not say who, who, who brought them. 3:01 It's up to you to, to guess if you want, and then, uh, as the pod goes on, you can actually figure out who actually brought each topic. But we have four topics. 3:08 We're gonna talk about whether AI, uh, and AI startups are overvalued and overhyped. We're gonna talk about, um, a recent claim by the IRS on the FTX bankruptcy procedures. 3:20 We're going to talk about quantum computing, and we're going to talk about growing financial concerns with the American debt ceiling and global financial crisis. 3:29 Those are our four topics, so I'll pass it on to Hamiz now to introduce the first topic that we have. So let's kick th-the, this off with the first topic of today, which actually, uh, I'm going to lead. 3:41 And the first topic is how we all felt that it was quite interesting to see, um, what we believe is a bubble in the AI space. 3:53 Uh, we feel that just like most technologies, you know, when the first version of it comes out and starts to make its way to the hands of regular customers, uh, a lot of times what you can find is, uh, that there's a big gap between, uh, what is expected and what people hope that technology could be, uh, and what probably the reality is going to be. 4:15 And so if you're spending any time on the internet, you will know over the last six months, uh, or really over the last twelve months, uh, people cannot stop talking about OpenAI, ChatGPT, large language models, and everything in between. 4:31 Um, and so initial thought as an investor is always, um, you know, whether, uh, the excitement and the interest in the space is, uh, is something that can be rationalized or something that might have gone into the realm of irrational. 4:47 Um, my belief today is that it has gone into the realm of ir-irrational already and very quickly. Uh, you're seeing some of the, the deals that are being done in this space. 4:58 You're seeing some of the price action for a company, uh, for any company that has AI in its name, and you can start to really piece those things together. 5:07 So I wanted to, you know, present this question to the group and see what, uh, their thoughts were. And so, you know, r-really anybody can, can pick this up and go. 5:17 Uh, but the first, uh, basic question is, are we in a bubble, um, or am I underestimating, uh, what this technology can do? Okay. 5:30 So my, my first thought, uh, obviously not biased running an, an AI startup- [chuckles]...um, is actually that those both can be true at the same time. 5:39 And- Okay...thinking about the, the space right now, I think an AI startup can be overvalued, and there could be a number of them that are currently overvalued, while the space writ large is still undervalued. Mm-hmm. 5:53 Um, few reasons for that is given the dry powder or investment decisions that are coming across the table in, you know, a number of different, uh-... spaces today, I, I wonder if there's a better opportunity. 6:08 And what I mean by that is, um- Mm-hmm... you know, we're going through somewhat of a recession. There's not a ton of, you know, fantastic companies that are coming across investors' decks. 6:17 And this is, uh, this is a opportunity even- Mm-hmm... even though it, it might not be the opportunity or correctly valued. Um, the one that comes to mind for me is, uh, vector databases. 6:28 I've seen an explosion of companies just discussing vector databases. One which I think, uh, is gonna be commoditized and eaten up by some of the, uh, the bigger infrastructure companies. 6:39 Uh, but I still think writ large, the, the space can be or can be and is undervalued. Mm-hmm. Do you feel, Brendan, that the commoditization will happen at the infrastructure layer? 6:52 Um, or do you think it will happen at the startup or application layer? Like, uh, where do you feel that the most commoditization happens? Definitely the infrastructure layer. I think we've seen-- Like if startups 7:07 win, it is typically because they 7:11 solve a customer pain point extremely quickly, and that pain point is felt most acutely, I think, at the application layer because that's where customers are typically interacting and where- Mm-hmm... 7:21 where, you know, decision-makers are typically touching that end product. The infrastructure is typically harder i-in many cases. 7:28 And if we think about the way that, um, businesses have currently bought software and infrastructure style investments, they have early in the year negotiated, uh, an Azure rate or a GCP rate or an AWS rate. 7:44 And for a startup to compete against, you know, a broad rate that all AWS products are given to a company, um, that would be-- You know, it has to be subsidized either by VC funding or, or some other mechanism to compete at that layer. 7:58 Mm-hmm. Mm-hmm. Yeah. And, and, and that makes sense. It makes sense that the commoditization happens at the model layer. 8:06 Um, one of, one of my biggest concerns with all of this is, um, as I speak to companies, uh, that, you know, you're right, are very interesting in a, in a, in an environment where very few things are interesting because the unit economics for most of the companies we see just don't make sense. 8:23 And one of the ways that you can make the unit economics make sense is you increase productivity so much using frontier technologies that now the unit economics do start to make sense, right? 8:35 But, but one of the challenges I have there is, you know, the way to monetize many of these models or to monetize at the application layer, uh, to me are, are really weak. 8:47 Uh, because we'll speak to companies with two hundred thousand, three hundred thousand active users, but less than a thousand dollars of monthly recurring revenue, right? 8:57 So to me, it's a lot of interest in the space for sure, you know, a lot of free users for sure, but they're not able to actually convert that into anything meaningful. 9:08 And so I think because we're up against a lot of open source, uh, competitors, uh, it-it's going to be very difficult to monetize this, I think, especially at the infrastructure level. 9:21 But maybe if you create a very specialized infrastructure layer for a very special use case, maybe there's some monetization there. 9:29 Maybe if you create a very special application and it's super specialized, maybe there's monetization there. But that's one of my biggest challenges is that this war is not between OpenAI and Elon Musk's next AI company. 9:42 Like this war is between OpenAI and all the open source models out there. 9:47 And so the question I'll pose to, you know, our crypto bro here is whether, uh, we think that the way that Bitcoin, Ethereum, and some of the, uh, some-- the, the, the, the crypto networks have been able to monetize open source software, right? 10:06 Bitcoin is open source software, but they're able to monetize it and create an economic structure around it. 10:12 Do you think, Sam, that, uh, the companies in the AI space can borrow some of those learnings and, uh, can they also monetize open source software in a similar way? 10:23 There's, there's like several thoughts that I have here, so I'm gonna go on a bit of a, a spiel here. So first, I wanna contextualize the, the discussion about AI companies being overvalued. 10:33 I think there is public equity and then there's private equity here. And so far, we've been talking about private equity because kind of like the exposure that both of you have. 10:39 So Hamis being a private investor and then, well, running a fund, investing in private equity, and then Brendan being a, a sort of founder of a private company. 10:47 Um, but I think I wanna highlight that like in the public markets, um, since, since the opportunities to invest in AI startups are so limited, uh, anything that is minimally related to AI has been getting a huge boost lately. 11:00 I wanna highlight there is an AI ticker. Uh, there's a, there's a ticker in the New York Stock Exchange that is literally AI, which, uh, stands for C three dot AI. It's a company- Mm-hmm... 11:10 that does, um, just-- is a company specializing on artificial intelligence. I'm actually not super familiar with what they do. [chuckles] But I wanna highlight that- Nobody is. Yeah. 11:19 [chuckles] But year to date, they're up eighty percent. And then, um, and then late, late March or early April, which was the peak, they were up like over a hundred percent in their stock price. 11:31 Um, and that's just basically because everybody's trying to find a way to get exposure to, to AI. 11:35 But specifically on private markets, um, and, and just generally startups, I think we're seeing two really interesting things, uh, take place. Um, and, and, and one I wanna... Well, they're, they're the same thing. 11:48 They're, they're transpiring through two different entities. 11:51 Um, and I wanna draw this distinction here between what you were saying, Brendan, about commoditization at the infrastructure layer and then application-specific solutions. 11:59 So I think there's a lot of, um, a lot of interesting companies coming upBuilding solutions on top of the infrastructure that currently exists, and that infrastructure being, um, Microsoft's OpenAI, uh, and the APIs or, or Google's AI plans or, or any other big company that has infrastructure and has the models and is starting to offer APIs. 12:19 And then companies that are specialized in a specific subject, they grab that infrastructure and turn it to solve a solution, and then they go to market with that. 12:25 They move fast, and they solve a problem and, and the traditional sort of story. But I think this is really cool to see the trend that is the power, if you may, lies in the platform, in my opinion. Like the, 12:38 the, the really, really, really big opportunity lies in actually being able to have the platform, in having the models that people are interacting with or having the data, um, for that matter. 12:47 And most people, since they don't have the resources necessary to be able to train the models and operate the models at scale, it seems like most people are being relegated to just come up with AI-- with APIs, with wrappers, with plugins, with apps on top of that infrastructure to be able to somehow monetize some of that and, and, and distribute the, the, the value that is being generated through AI. 13:11 But bigger than this, I think there's a, a bigger concern here as we continue to move forward in this AI race that, um, 13:19 the power that is being concentrated by some of the platform players is just like slightly concerning. 13:24 Um, because it's really hard for-- even though the, the, the, the, the models themselves or a lot of the code for the models is open source, the energy requirements, and I went in a really deep rabbit hole on this before the pod. 13:36 The energy requirements, the compute requirements to be able to operate these, um, these models is huge. 13:42 Like, I saw an article before the pod that said that the daily operational cost of operating ChatGPT or GPT-3 slash four was around seven hundred thousand dollars a day. 13:51 Like, who can, who can, who can cover those costs, right? So I think we're seeing a very strong distinction here between platform players and then, um, apps or plugins or API integrators. Mm-hmm. Mm-hmm. Yeah. 14:04 That's right. To pitch slightly on that, I think, I think the costs are coming down at an exponential rate to train large language models. 14:12 We're noticing that you don't need as big parameter-- parameterized, overparameterized models as we originally thought. Emergent properties might not be there. There might be linear properties. 14:20 There's like research that's coming out week in, week out that I would guess in six months' time, you know, once Stability AI puts out their version of, you know, top ten language models, um, the smaller models are gonna start coming, and the smaller models are gonna start winning as well. 14:35 So I'm not as worried about like model training- Mm-hmm... not being able to be done by, um, startups and smaller companies. 14:42 In my view, like Microsoft also, uh, is incentivized to make it look more difficult than it is because- Mm. Sure... OpenAI is paid-- paying Microsoft the infrastructure cost in their, in their, their actual deal. 14:56 So there is some- Yeah... uh, some posturing, I think. Yeah. And, and I think that's the exact-- I think that's the exact point. Like that's what I'm, that, that's what I'm worried about. 15:05 I think that these really super specialized models, uh, that don't cost that much, you know, continue to improve over time, they're gonna chip away at this, you know, dominance that some of these larger players have. 15:18 And, and the larger players are going to front the cost of building this entire industry, but they will not be able to monetize it. Like, that's what my, that's what my fear is as an investor, right? 15:28 That, uh, uh, in OpenAI, they, they'll, they'll dump the billions and billions of dollars, right? 15:34 Like, according to the CEO of OpenAI, they think it's going to be the most capital-intensive business Silicon Valley has ever seen. 15:41 Like, they expect to spend a hundred billion dollars by some accounts to get this company off the ground, right? 15:48 Like, and so that's fine, and that's great for society, but that's not great for the OpenAI investors, right? 15:55 Because they're going to front that cost, and then they're going to get outcompeted by somebody who's training that, uh, super specialized model, uh, that everybody actually, uh, uses as an open source, uh, solution. 16:07 But how easy, though, how easy is to actually train some of these models on your own? And maybe this is more of a question for Brendan because- It is. Yeah... my understanding in, in training is like not only... 16:17 It's not only about like the, the compute resources, also access to GPUs. 16:21 And I think if you wanna do this at scale and, and not just build, not just build an application, a specific model, but rather build a generalized model that is able to serve applications on top of it, and that's, that's my argument, is who will be able to compete in the platform business moving forward? 16:39 Um, and one thing that I wanna highlight from crypto is there are some companies out there that are doing interesting concepts about bringing access to GPUs in a decentralized way. 16:49 So in a nutshell, they're just like building marketplaces of, of GPUs or, or- Mm-hmm... compute resources. For anybody who has an open source model, who doesn't have access to, to 16:59 may-a million GPUs or however, I, I don't know how many GPUs, uh, it, it took to train some of the GPT models. 17:04 But if you wanna have access to GPUs, you can go to some of these marketplaces and, and, and do it in a decentralized and open way. I think those are good things. 17:12 Um, but I wonder like what are the thresholds of access to compute power to be able to build a model that is not just unique to solve your use case but can actually be turned into a platform and compete in the platform business? 17:25 I'm interested to know a startup that has the amount-- the scale of data necessary to train a model from scratch and not fine-tune- So you think the addition is data?... from... 17:34 Well, I think the-- there's just not many companies that require... That have a data set that requires either fine-tuning or base model development, right? Like- Mm-hmm... there's just not a huge reason. 17:47 You can take the open source, uh, Facebook model, which startups can do this, but enterprises can't because of the licensing right now. So there's just a lot of, uh... 17:56 Actually, this is a good question for Rich to, to chime in maybe. What I'm noticing- Mm-hmm... is there was an, a letter that was leaked from Google, um- Mm-hmm... 18:03 about OpenAI, which saidBasically, hey, um, we don't have a competitive advantage, especially when models get leaked to the public, like Facebook's did. 18:13 Startups use those models to train net new pieces of work, and then they're not gonna be litigated typically because they're just too small and, you know, who's gonna go after them? 18:23 But a Google or an OpenAI can't do this, take this approach. So these startups are kind of getting like small headway, and then they're then figuring out their legal requirements, solving that at a later point. Mm-hmm. 18:34 I think this is like, uh, uh, a fact that, again, smaller models, less training required. We're focusing on fine-tuning. At least that's the direction the industry is going. Yeah. 18:42 I mean, like I-- Just a couple thoughts on that. Yeah. So Rich, what do you think about-- So, uh- Go ahead, Hamiz. Yeah. Yeah. I'd, I'd love your thoughts on two things. 18:51 One, that licensing-- I'd love your thoughts on the licensing. I'd also love your thoughts on the reg-regulation of the space, right? 19:01 Where are we-- One of the biggest risks I see as an investor is, you know, a bunch of, uh, uh, you know, a bunch of, uh, rule makers get together and say, "Oh, actually, this is gonna hurt jobs, so we're just gonna shut this down." 19:13 Right? So, uh, talk to us about what you th- Yeah. So on the first thing, like it's kind of funny, like what's old is new again. 19:20 If, if anyone was like following a lot of the facial recognition, uh, startups that existed a couple years ago, they were like really hot, especially pre-COVID. Um, and 19:32 the, the, the fundamental problem was exactly the same. 19:34 It was in order to train what we call like a machine vision algorithm to be able to recognize a face, basically do something like pairwise comparison, where I say, "This is a picture of Sam. 19:44 Now find me Sam again in a million images." Um, what you have to do is you have to train that, that artificial-- uh, you know, really it's a, it's an AI ML model. You have to train that on a bajillion different faces. 19:56 Um, and so what ended up happening was people started asking questions like, "Well, where did these faces come from?" And it led to a bunch of multi-billion dollar lawsuits. 20:06 Um, and the reason was like nobody had really thought about the fact that, um, I don't know, using people's faces in training data that gets put into a black box of a machine vision, you know, neural network that says these are the, these are new faces, like that's actually gonna cause a problem. 20:22 And so like I think that's-- On the licensing side, you know, the, the letters about like, you know, the, uh, being an incumbent and why it's challenging like are absolutely right, and you saw that with facial recognition technology or what we call face matching, right? 20:35 Facebook settled, I think it was like a billion-dollar lawsuit in Washington State over this. Um, and it had nothing to do with like the ethics of the model. 20:43 It had nothing to do with like them using the model to like, I don't know, uh, help law enforcement or anything like that. 20:48 It had to do with someone's face ended up in the model, and then the model was used without their permission. Um, and so I think like you-- that L... 20:58 Like that question in the LLM context is times a gajillion because it's not just y- it's not just an image, right? For, for generative purposes, but it could be like, you know, copyrighted materials. 21:09 Um, any sort of, you know, information or IP that you have around your business. Could be patents, could be anything. So I, I think that's like we're gonna see that to a magnitude of a hundred. 21:20 And, um, and the funny thing is like we have this concept called synthetic data. Do you think though- Yeah. I was gonna say, do, do you think they know this and they're just okay because 21:36 traditionally the litigation costs and the fines are worth the squeeze for now? Or is that like-- Are they actually- I mean, they're, they're j- yeah, I think that they're the cost of doing business. 21:46 Like I think this ties into Hamiz's second question, which is like the regulatory stuff, which is like a civil litigant suing you. 21:52 If you're, if you're operating in a heavily regulated space or a space where there's a lot of plaintiffs, like you build that in, right? 21:58 If you, if you're the CEO of a large broker-dealer and you sell equities, you just know someone's gonna sue you for a securities fraud. It's just gonna happen. 22:05 Somebody's gonna say, "Oh, well, you sold me this IPO, and the IPO didn't do as well. And it didn't do as well because you didn't have the right disclosures." 22:13 And you kinda just say like, "This is the cost of doing business. 22:15 If we wanna do this at the orders of magnitude of billions of dollars a year, every year we're spend twenty million dollars dealing with someone suing us for securities fraud." Um- Oh, we lost Rich. We lost him. 22:27 We lost him, man. Yeah. In the middle of it, man. Yeah. W- And I was just in the middle of the spiel around the cost of doing business. Yeah. Yeah. Okay. Go ahead. So, okay. 22:39 Um, yeah, so like when you're doing any of these-- when you're building a business like this, y- you're, you're building the litigation costs and the cost of doing business. 22:47 The, the trick, though, is that what you're not building into your business is that there isn't some sort of like ex machina moment where Congress just says like, "W-- You know, we're putting a full stop to this." 22:57 Or, you know, like what happens in sports gambling, right? 23:00 Which is it was like a cottage industry run by states, and then all of a sudden they said, "Hey, look, sports gambling's illegal," and it went state by state by state by state. 23:08 And eventually, like there is no cost of doing business that you can build into that. 23:12 It's just that you're running a criminal enterprise- [chuckles] -in the eyes of the regulators, and so you have to stop it, um, until you can lobby for a change. 23:19 So like I think that that's, that's like the thing that I'm interested to see here is, one, what does the, you know, we'll call it like the litigation around, um, model training look like for LLMs, and like how aggressive is it? 23:34 How high stakes is it? And who are the, who are the targets? I think Getty's already sued some of these people for use of their images around the generative AI stuff, I think in stable diffusion and mid-journey. 23:43 Uh, and then the next is like what do the regulators choose to do, um, in terms of, of regulating this? 23:50 And like I, I think there's like a really interesting Section 230 argument that you could start to go down here, which is, you know, arguably, you know, these things like ChatGPT and Bard, they're platforms to Sam's point. 24:01 And if they're a platform, you really have a hard time, you know, managing the content that gets published there without violating 230, and so- YeahMaybe there's, maybe there's an angle there for these guys to, to prevent that kind of like sui generis regulation that would shut it down. 24:18 In a more kind of like philosophical thought too is, um, I remember seeing a post, I think I heard about this in the All-In Pod, um, of a-- an animator that was just super disheartened with Midjourney and said, "I used to love my job, and now after Midjourney, like, I feel completely useless. 24:33 All I do right now is, like, creating prompts inside of Midjourney, and I, I've pretty much lost all the passion for my job." 24:39 But I think he closes with a statement saying, "I think it is very tragic to think that all the art that I've done, uh, throughout all the years went in to feed this model that is, that is now displacing me from what I used to love to do." 24:53 Um, so kind of like in a deeper metaphysical, metaphysical sense or philosophical sense should I say, it's a little, [chuckles] it's a little tragic honestly but- Yeah. 25:01 Like we were, we were, like, playing around with AutoGPT yesterday. 25:06 Um, and I've never been, like, so scared that something was gonna take over my computer and just kinda like run my life because I had like no idea what it could do. I... 25:16 Like, it was asking me before doing things, but I wasn't sure if it needs to ask me, right? Like, it, it has access to my, my computer and the internet. 25:26 Um, and I, I really started thinking like, I think the only, like, majors that will be left in college are all going to be like liberal arts majors. Right. Yeah. 25:34 And it's going to be all, like, people who can explain things to a kid, and whoever gets-- like becomes the best at explaining things to a kid will be able to explain it the best to these models. 25:47 And there will be no, like, hard skills anymore. Like, it'll be... A-and I think all of our jobs, like not just like some of us, I think all of our jobs... 25:55 I was thinking yesterday, I was like, "Hey, w-the stuff that I do, like, i-is, does-- is it necessary? Like, uh, should I just be [chuckles] also a prompt engineer?" Like, I don't know. 26:06 Like, uh, the AutoGPT, especially yesterday, because ChatGPT, I used to be able to tell myself, "Oh, yeah, it doesn't know what's going on in the world. Like it's not gonna be able to replace me." 26:17 But after playing with AutoGPT yesterday, I was like, "Man, this stuff is, this stuff is serious." 26:24 I think there's a deeper, there's a deeper philosophical discussion here, um, that I think we should actually explore maybe on a, on a future, future episode and probably bring a philosopher as a guest as well. 26:34 Um, but what I wanna say is there's two things. So one is, one is a comment and the other one is a question for the group. 26:41 I think the comment here is that regardless of what happens with the future of the technology in terms of whether we reach AGI and whether this leads to, like, the destructions of society in, in a very kind of like, um, pessimistic and, and, and, and exaggerated way, regardless of it, we're definitely gonna see a huge shift in, in, 27:02 in demand for skills, a huge shift in terms of displacement of jobs. Um, new jobs are gonna be created, but I think this is going to force us to reckon with the meaning of purpose for most of our lives. 27:15 A lot of people who found purpose or who find purpose in, in, in their work, especially in Western societies, 27:21 feeling that the purpose is suddenly ripped away by something that can do the job that you used to do infinitely cheaper, infinitely faster, infinitely better, that is really, really hard. 27:32 And I feel like all of us need to, like, start to come to terms with it, um, and think how that's gonna impact kind of, uh, the feeling of, of purpose that we, that we get from work. 27:40 And then the second point that I think I wanna pose it as a question, a closing question on this topic for the group is I, I, I think we can all agree that we don't know exactly what we are... 27:52 Or, or, well, I, I, I should say, I don't-- I think we don't know exactly what we're getting ourselves into when it comes to, um, gen AI and, and where this may lead us. 28:00 And one concern that I've heard several people make, um, I was listening to a pod with Lex and Max, Max Tegmark, Max being the founder of, uh, Future of Life Institute. 28:09 He was one of the leaders of the, the open letter that came out a couple months ago to pause, um, large language model training for six months to allow society to catch up. 28:20 Um, but one of the things that, that he was mentioning in that podcast was that, um, maybe this is a technology that we don't want open source. 28:27 Maybe this is a technology that we don't wanna make really easy for people to have access to, because the more people have access to really... 28:33 Not, not necessarily have access to, but have access to actually creating some of these models. 28:37 Because the more people have access to creating these very powerful models, granted, yeah, the value is democratized, the access is democratized, but at the same time, the risks are also democratized. 28:47 So that's kind of like a question that I, I wanna pose to the group here is like should we be pushing to open source and, and give more access to this technology and to the tools to create some of these models or not? 28:57 Mm-hmm. Like the UN treaty on- Go on. I think all of us have something to say on this, right? Yeah. Uh, I, I think like you, you can't pick and choose. 29:09 Like, they have said that about every technology that's ever existed. Like, they've said, "Don't open source this, it will change the world." No, but dude, no, I think- And it's the same thing... 29:17 this is objectively different. Like it's, it's pretty much like should you open source, uh, access to nukes? Because I-- like I, I see- Well, that's... Yeah. Yeah. 29:25 That's, that's, that's almost- It's nuclear proliferation, right? That's the same thing. Yeah. It's the same... 29:28 Like that was-- that's-- If you're gonna take that heuristic, you have to actually take the, the mental model of this ought to be treated like, you know, sort of the, the UN's treaties on nuclear proliferation and pro-prohibition on nuclear proliferation. 29:44 Um- Yeah. Like the unfortunate thing about human society, Sam, is like we probably won't see that until you see AGI, um, powering like swaths of drones attacking civilian cities. Like that's actually- Yeah. What you... 29:59 That, that will lead people to... It's the same thing. 30:01 Look, you blow up, you, you know, you have Trinity in the desert, in, in the Alamo with Oppenheimer and all those guys, and that's, that should have been enough to say, "Hey, we shouldn't do this." 30:10 And they weaponized it immediately, dropped it on two cities, and then everyone was like, "Oh, maybe we shouldn't do this anymore." 30:17 So-I mean, it's unfortunate, but I think that that might be, Sam, like the way that we have to think about it if it goes that way. 30:23 And you can, you can draw a pretty close line to the printing press and the Catholic Church, right? Mm. And you basically said, what does it look like to open source, uh, access to, you know, reading and writing? 30:37 Um, they would make the same arguments. Mm-hmm. But- But who... Yeah. I think the question is, who gets to decide and what makes them decide that, right? 30:48 So do we-- can we decide for the first time in the history of technology, uh, to do something like this together before things get so bad that we're forced to d- make the decision? Right. 31:01 I made the argument before, and maybe I think I, I've spoken to some of you personally, that I think this is very much... 31:07 Like, this is very similar or I see it very similarly to the Cold War, but I see it critically different in some very specific and, and dangerous points. 31:16 So the two main points that I see being critically different and making the situation a lot worse is that during the Cold War, the impacts of a negative outcome were very clear and were devastating. 31:28 So everybody knew that if any nuke went off, the world was blowing up. So everybody was very clear on that. 31:33 Here, we don't know what's the danger of one of these, one of these really powerful models leaking or, or some sort of agency being developed. That's kinda-- that's very unclear. 31:42 And then the second thing that I think is different from this and the, and the, and the, the Cold War was here, the incent... 31:50 Well, back in the Cold War, the incentives of the key decision-makers were not directly tied to the outcome. 31:56 Granted, everybody didn't want the world to blow up, but you didn't have any sort of personal incentive, uh, in a specific outcome, uh, in, in the Cold War. 32:04 Here, the key decision-makers, uh, the people who are saying, "Yes to this test, yes, deploy this, yes, launch this product ahead of time," have a direct economic incentive tied to the outcome of that specific race. 32:15 So I think those two little differences there, the outcomes being a little bit more murky, and then the incentives being directly tied to personal, personal gain, make this a lot more dangerous. 32:25 And, uh, I am, I am personally, like, kind of concerned, um, but it's also kind of a fascinating, fascinating thing to be witnessing from the technological side. But- Yeah. Interesting. Should we move on or...? 32:36 Yeah, let's move on. Let's move on- Okay. -to the next one. Uh, the next one is, uh, going to be the state of quantum computing. 32:45 And, uh, our resident founder and, and machine learning, quantum computing, and AI expert is going to lead this. Go ahead, Brendan. Quantum expert. 32:57 But, uh- [laughs] -well, I thought it would be fun to do a recap of where quantum's at today. 33:03 I think, uh, there's a ton of hot fun topics in the media, but things that are making progress, you know, in the underbelly, uh, we, we wanna touch on one-- some of those and, and kinda like do a quick review of where it's at. 33:17 So as a, as a quick, um, refresher for, for some of us who maybe aren't as familiar with, uh, quantum computing, quantum and kind of like what it-- what this looks like, I would, uh, make it like an analogous position to energy production. 33:33 There are, uh, energy production through solar, through nuclear, through gas, oil, a hamster wheel. Um, all of this is energy production. Uh, similarly, there is a, uh, multiple approaches to doing computable, um, work. 33:50 Um, quantum is a platform approach to doing such work. Uh, there's multiple approaches to quantum: superconducting, annealing, topological, trapped-ion. 34:01 Um, various startups and, uh, c- companies are taking different approaches here. All this to say is, um, it's not a one-to-one mapping to classical computing. 34:10 However, there is strong promise that this will be, um, revolutionary for a certain set of problems, primarily optimization problems, um, which are, um, very apparent in machine learning. 34:22 And because of this, we wanted to, A, just touch on the, the topic, review where everyone's at from, uh, the key players and kind of what's coming down the pipeline. 34:33 So from a key players perspective, uh, we have Google at fifty-three qubits, I believe. IBM just went from a hundred and twenty-seven to four hundred and thirty-three. 34:43 There's D-Wave, Microsoft, Rigetti, IonQ, are some of the major ones. The US is still the-- known as the leader. China is fast approaching, I believe, on this topic. 34:54 Um, qubits, uh, are composed of not only their computational power, but also, um, how high of quality they are. So we can't take a one-to-one mapping. 35:05 Uh, and then in recent news, uh, twenty nineteen, uh, Google announced quantum supremacy, so just kind of a, a landmark, um, in which a quantum compu- computer was able to do a task that would have taken a-- our best supercomputers roughly a hundred thousand years to co- to complete. 35:23 So this doesn't mean that they are, uh, fully available today and, uh, useful, but, um, one such task, they were able to produce, uh, results which were quite promising. When I was, um- Yeah. 35:36 Just a quick question on, on the, on the state, on the state specifically with IBM. 35:39 When I was doing some digging on the state of quantum computing, um, I was actually surprised with the, with the qubit, with the qubit processing power of... Like, on the, on the lower scale. 35:48 I thought we're a little further ahead. I think, like, I saw IBM having like a one twenty-seven, you said, uh, qubits processor available. 35:53 Not available, like publicly available, but some of their, um, research and so on. 35:58 But I remember seeing when I was doing the prep, um, that they had promised or they had claimed that by twenty twenty-three, so this year, they will have a one thousand qubit quantum computer. 36:10 Any insights on that or anything that you've heard, like any major breakthroughs on huge, on a huge number of qubits, um, com- becoming available anytime soon?Yeah. So I think the, the industry 36:24 in the last few years has switched from raw numbers into, um, qubits of higher quality. So you can- Mm-hmm... 36:32 keep jamming more qubits into these superconducting, um, arrays and lattices, um, but you lose, um, computability on top of them. 36:40 So I think there was a race to quantum supremacy, which IBM and Google, Google, we would argue won, um, and IBM is close following. Um, the next one is actual, like, applicability in industry. 36:53 So I think we care less nowadays about the number of qubits and more about, um, real-world use cases, which there are still few and in between, but are starting to crop up. 37:04 I think, um, one, one, like, area that I wanted to pose is 37:09 this feels a lot like crypto in that we have a technology and we're trying to find a use case, and there's gonna be a lot of swinging and a lot of misses- Mm-hmm... but it just takes one to really hit. 37:21 Do you guys have the sort of same, same feel, or, or how do you think about this market compared to other frontier markets? Maybe Hamiss to start. Hey, that's not fair. We, we already have a use case. [laughs] Right? 37:34 We, we have multiple use cases. We have currencies, we have meme coins, right? We have- [laughs]... NFTs. Like, we have use cases, man. Like, don't, don't be, don't be like that. But yeah, I agree with you. 37:47 I think that, um, uh, at scale, uh, crypto has, uh, at least on the store of value, shown that it can attract between five hundred thousand and a billion users on a daily basis on the Bitcoin network. 38:02 Um, it can have more than a hundred million accounts on that network live. Um, and that system has been running successfully for about thirteen years. 38:11 Um, the, the reason that I got exposed to quantum was through the crypto channel because it was one of the arguments that we would hear all the time about how these systems could be broken, uh, using this, uh, type of, uh, computing capacity. 38:27 Um, so, um, that, that's really all my, um, interaction with, with quantum computing. Uh, but that's what I would say on the crypto side. 38:36 I would say that, um, there are actual live use cases that might not make a lot of sense to us, uh, today, given, you know, our socioeconomic background, our environment, but do make a lot of sense 38:51 in a lot of the countries that we spend time in and, and, uh, speak to people and where, um, this type of, you know, crazy inflation or this mistrust in, in local government, uh, has really made this a real use case. 39:07 And through thick or thin, through up cycles and down cycles, um, at least that store of value use case has actually had pretty consistent, uh, daily active users. Yeah. To your question, Brandon, I think, um... 39:21 What's a use case of this technology? 39:23 I think in the early days of quantum, and by early days I mean, I don't know, like I, I remember hearing this even before I went to college, like twenty fifteen or so, people claiming or, or dreaming about one day, um, quantum processors being commonplace in every single computer and every single laptop out there. 39:37 So the same way that your computer has X number of CPUs and then, um, X over eight number of GPUs, it will have like X over, I don't know, thirty-two number of, of, of, uh, quantum processors. 39:49 Um, I don't think that's gonna happen mainly... Or I wouldn't say necessarily I don't think it's gonna happen, I think like that's kind of like a far... 39:57 I see what you meant-- what, what you mean by, by we're looking for a use case here, because I don't, I don't necessarily see a use case in which 40:06 normal people going about their lives using computers for any sort of problems that they have or even like general developers building applications have the need to have such an immense, um, computational power that, that comes from, from quantum processing. 40:22 Um, so I don't, I don't, I don't see it becoming commonplace, but I definitely see a really strong use case for, for just generally, like research, research problems. Mm-hmm. 40:32 I don't know, I don't know if there's a huge business to be built out of this, but I'm excited for the research implications of this. Mm-hmm. 40:37 I think, uh, closing the gap between NP complete and NP hard, uh, is kind of like a fascinating prospect of, of, of quantum computation. 40:45 And to your point too, Hamiss, of, of, um, cryptographic, cryptographic vulnerabilities from, from quantum computing, this is also a really big concern that I had. 40:53 Um, but the more I digged, the more I saw that not only are we kind of like far away from, from 41:01 quantum computer becoming a, a realistic threat to elliptic curvature, um, cryptography, but also even if it ever gets there, you can always make the key size longer, like significantly longer, and just make it a lot harder to, to break in terms of brute-forcing, um, brute-forcing a private key through the existence of a public key. 41:22 You can always make that public key like twice, three times lo- longer. Um, so I think we're, we're kind of safe in the infrastructure, in the mathematical cryptographic infrastructure that is powering blockchains. 41:32 But, uh, that is not to say that I don't think, that I don't think quantum computing is really exciting, um, really exciting from the research perspective. And just- Can you use-... just to close one of... 41:43 Just, just super last comment. Um, one of my favorite kind of like, um, reference in the quantum computing or quantum computation field is David Deutsch. 41:54 Um, and I remember when I was reading his book, The, The Fabric of Reality, in which he talks a lot about like quantum computation, he-- I, I, I got really curious about, "Oh, what does this guy think about crypto?" 42:05 And I went to look some stuff up, some comments I had made in crypto, and I remember seeing a tweet that he made, um, that basically said... Somebody asked him about, "Oh, wouldn't quantum computer break all of crypto?" 42:15 And he basically said, "Okay, what am I-- what do I wanna trust more? 42:19 That governments can maintain, um, economic stability or that somebody's gonna be able to have a huge quantum leap, uh, and be able to build incredibly powerful quantum computers and widespread proliferation of quantum computers?" 42:30 And- [laughs]... it feels like the, the, the, the latter is a little bit more challenging.But sorry, Rich, what were you gonna say? 42:36 I was just gonna say like, what-- so we were just, we just got off of the AI computational issue. Mm-hmm. Like how come nobody's really talking about- Yeah. 42:45 I mean, could you imagine, you wanna talk about scary nuclear power type stuff, Sam, you know, some sort of auto GPT, uh, sort of general artificially intelligent thing that just hoovers up as much data as possible, powered by a quantum computer connected to a nuclear reactor sitting in space, right? 43:02 Like you got Skynet right there. [laughs] Um, but like is that, Brendan, is that something that people are even talking about? 43:07 'Cause the only thing I've ever heard quantum computing, like, like Hamiz and Sam said, like I've only heard it in the context of two things. One is kind of like this broader sort of cyber defense deterrent, right? 43:19 So like US government, Chinese government, like really, you know, trying to signal their prowess there just as like, almost like a, again, like a defense mechanism. 43:28 And then the other is cryptography, like just bre-breaking encryption and being able to essentially, you know, read all your signal messages or double spend Bitcoin. Mm-hmm. 43:37 Um, but like, I don't, I don't know if it- Let me... Yeah. It's different. Let me add a third technology to that. 43:41 So what if there was a, there was a AGI model that was being powered by a quantum computer that had a crypto wallet that could pay humans, right? To do what they wanted, right? 43:55 Like if you put all these frontier technologies together, that's how you make Skynet, right? You know what you get? You get Clippy. You get, uh... Have you heard of Clippy? 44:03 [laughs] Have you ever heard of, uh- Microsoft Clippy? You haven't heard of Clippy? Like [laughs] No, like, like the, the AGI idea of Clippy. No. [laughs] So, so- No. It's, it's a mental experiment to, to-- 44:14 yeah, paperclips. It's a mental experiment to illustrate the dangers of AGI. Basically say you have a model that is, uh, super intelligent, and then you give it the goal to maximize the production of paperclips. 44:23 It starts by 44:25 buying all the paperclips that it could possibly get its hands on, and then it starts manufacturing its own paperclips, and then eventually it keeps like optimizing and optimizing until it realizes, "Oh my gosh, actually, humans, uh, have a lot of the core, um, materials and, and, and, and, and atoms that I need to actually make paperclips." 44:40 So it starts just gobbling the entire world just to make paperclips, and the entire universe turns into a paperclip. Um, that's what you get when you connect all those three technologies. What does Elon Musk say? 44:50 He always says the most entertaining outcome is the most likely. Yeah. [laughs] Like, could you imagine? [laughs] Yeah. So the problem is like- What happened to all these people? 44:58 Is that like, don't, don't you think that's, that could be the biggest, uh, risk, the same way to crypto. The-- that's one of the biggest risks. Uh, it could be the biggest risk to OpenAI, right? 45:08 Like, if, if they can-- if they spend ten billion dollars training their models, and then a quantum computer can train that same model for two dollars, right? They're obviously going to lose. 45:20 So d-does anybody talk about the combination of these technologies? There's like a lot of interesting discussion. I would say they are so disjoint today that we wouldn't train models anywhere nearly in the same methods. 45:36 And what I, what I'll say by that is not that we can't, it's that the only reason in some ways that OpenAI existed is because the transformer architecture and the GPUs from NVIDIA just at the right time were both like available and someone had the idea. 45:53 Mm-hmm. So I think it-- that's gonna get cheaper. There's Ce-Cerebras, there's tons of companies that are already making this like significantly cheaper. 46:00 By the time quantum's available, we wouldn't be thinking about the same problem sets to solve for. We would be solving for a brand new set of problems, and that'll look like a different platform. 46:12 So I-I'm less worried about quantum in the term of AI being a real AGI method because we just don't even have an architectural thoughts- Mm. -about how we would do it. 46:24 And even if we did- What do you think are the- What comes first? Yeah. Like, do you think we'll solve AGI first? Do you think we'll solve quantum computing first? 46:34 Or do you think we'll solve like a global reserve cryptocurrency first? [laughs] Or, and you can put it in order. That's a great question. Like, everybody can put this in order, like, and, and go for it, right? 46:46 Like, what do you think will happen? I think we're closer to, um- So I'll make a quick comment on the, on AGI. Yeah. AGI to me is so far off than anyone expects. I think like significantly further off than people think. 46:58 And the biggest reason I think is, um, we don't have enough sensor data for anything near AI. Like we're already ingesting tons of images, tons of the internet. 47:10 Like there just doesn't exist enough sensory input for AGI in my current-- given the current architectures to be realistic. And that could change if there's new architectural- You're assuming that AGI needs sensory data. 47:24 Like, like you-- when you mean sensory, you mean like, like human senses? Is that what you mean by sensory data? Like r-raw, just like weather data, uh, image, uh, olfactory data? 47:34 Depends on what you mean by s- What do you exactly mean by sensory data? Depends on what you mean by AGI then. 47:37 Like that's my response is like, if you want AGI to talk to you, it needs enough sensory data to like do the talking task, which is language and then audio. 47:45 If you want AGI to run around with us, you need mechanics data, and Boston Dynamics is like, has a lot of interesting work on like how hard that problem is because there just doesn't exist an open format that's been collected over twenty years of the internet that's like you can just scrape and train on. 48:00 I think like the- Do you think a text- Like the real one. Do you think a text interface is enough for AGI? Like if, if the issue is the interface, I feel like text... 48:09 I, I feel like the, the interface just, just solves for how do you communicate with it. Doesn't necessarily solve for a model that can, that can reason. No, I don't think it... 48:22 I don't think you'll get AGI given the current- You think both are related? So like the, the, the interface and, and the reasoning are, are related to each other? Yeah. 48:31 Well, there's pr- I think there's probably another point that maybe, I don't know, Brendan, if this is kinda what you're thinking, but it's kicking around in my head, which is-The definition of AGI is actually a subjective definition. 48:42 In other words- Yeah... like it depends on what the, the viewer, the, the, the recipient of the thing is feeling or seeing or thinking. 48:50 And so there's probably a decent point that if it is just restricted to text, no human would ever interpret that, or humans would have generally a hard time interpreting that as AGI as opposed to proxies for intelligence. 49:03 Kind of like when people talk about like sentience when they're like, "I talked to ChatGPT, and I said, 'How does it feel?' And it said, 'Great.' And so it must be sentient, and it like it fee- it has a soul." 49:13 And people are like, "No, no, no, those are, those are kind of proxies for those things, but it's just telling you what you wanna hear. It's not really-- doesn't really actually feel anything." 49:22 I mean, who the hell knows if that's true. But [chuckles] you know, like it is kind of interesting. Uh, but I, I wonder if that's kind of the, the issue is that it depends on us. [chuckles] So let's- Okay... 49:31 rank it to your definition, to your definition. Uh, global reserve currency built using cryptography, AGI to your definition, quantum computing. What comes first? I'll go first. 49:43 I-I'm very certain that the global reserve currency is gonna come first. It's going to be a factor of both that we already have that technology and also a factor of how poorly governments are run, right? 49:55 So that's going to be like a fast one. Then I think it's going to be, um, with just how much money is going to be poured into it, uh, I think we-- then we'll get AGI. 50:08 Um, and then I think the last thing that will break is, is quantum. Because I think the problem with quantum is that because people can't really-- the, the VCs can't really see it and feel it and play with it. 50:19 It, it-- you have to be like a really like technical hard tech VC to, you know, get excited about it. But AI stuff, all the VCs have already played with it. It's like more familiar. It's like a friend. 50:30 It's like somebody that... And so you're more willing to put billions to work there than something that, okay, I don't even understand like what the use case is gonna be. So I think- Agree with the ranking... 50:40 that's my order. I think it's going to be world currency, then it's going to be AGI, and then it's going to be quantum computing. Agree with the ranking. 50:49 Because the only thing I'll push back on is didn't OpenAI get funded the exact way you said quantum computers can't be funded, is dumping a billion dollars into something that doesn't have a use case at the, at the start, on the onset, and has like no idea of what it's gonna do? 51:03 Like, that was their original investment thesis was like, "Put a billion dollars into us." That's right. "We'll, we'll, we'll give you guys something interesting in like four years, seven years." Yeah. 51:11 And but, but now- Maybe even call an investor... there's not gonna be... 51:14 But now there's not gonna be any, you know, there's gonna-- Because the, the use cases we're talking about on quantum, like, you know, figuring out the cure to cancer or whatever, right? 51:23 Like that type of stuff, like you have to be from a very specific background to appreciate that problem and the solution. But everybody appreciates that they've played with ChatGPT, and it's a cool solution, right? 51:35 So I think that more people are willing to invest in that a-at least today. You're right, maybe six month-- uh, five years ago, it was in a similar state. 51:45 But I think today there will be more money flowing there, um, and all the attention is going to be there because it's something that is familiar. So- Brandon, what's your ranking? Sam agrees with my ranking. 51:55 What about Brandon? Quantum, AGI, reserve currency. No way. You think reserve currency will be la-- is going to be last? I think it's the last thing that we as... we have control over. 52:14 Mm. Like, I think quantum's gonna happen. China and the US are both running at it. One of those two is gonna do it. AGI, I think startups, governments, I don't know, someone's gonna get to it thereafter. 52:26 I think re-reserve currency is the one thing that the current elites, we'll call it- The coordination problem... the, the... [chuckles] Yeah. Like, I think that's the one- Mm... where we actually have somewhat- Yeah... 52:38 chokehold on. Um, the other two are just- Mm-hmm... gonna happen by like sheer time. Mm-hmm. Which way you think? Interesting. 52:48 I mean, the, the reserve currency thing is, is got some recency, and it seems like it's a lot easier to pull off because it's- Mm... 52:57 it's just kind of like a memetic effect, and that just, uh, with the internet, I just think that's so easy to disrupt now. So I would say reserve currency. I think quantum computing comes first and then AGI. Um- Okay. 53:13 I do think like, you know, without trying to get over my skis here on what the hell I know what I'm talking about, but like honestly, the quantum computing side thing, like I feel like we just have a, a better idea of the very finite niche use cases it could be used for, and I think what people are just gonna do is just like hyper-specialize, hyper-focus. 53:30 The AGI thing is a little bit of like, yeah, you know, it's kinda like fusion. I feel like it's gonna feel a lot like that, where it's like, well, what does fusion look like? Well, what could it do? 53:39 And it's like, well, it could power the sun, or it could power your automobile. Mm-hmm. 53:42 But we're just gonna keep doing this thing at the Princeton Fusion Laboratory, and, you know, eventually we'll know it when we see it. 53:48 Um, so I think that's, that's kinda my order, but I do think the reserve currency is there. But you know, you know how we will not reach, uh, the status of reserve currency with crypto? 53:58 If we keep having explosions like FTX, which is a great segue into what's probably gonna be our last topic, uh, because we've been here probably for two hours already. Yeah. Which is, uh, the situation of FTX. 54:10 Um, I don't think we're gonna get to talk about the, the debt ceiling and the, and the banking crisis and all that, but that's probably gonna be for our next episode. But Rich, this was your topic. Yeah. 54:19 Um, ta-tell, te-talk to us about, uh, the FTX bankruptcy procedures and the IRS. Yeah. I mean, I'm gonna keep it really short 'cause I think everyone knows about FTX. 54:27 I think, uh, you know, for those who don't know, they're a giant crypto exchange. Um- Were... 54:32 they had a spectacular, uh, we'll call it like rise and then fall, um, you know, within a matter of weeks, completely unwinding and, um, essentially unraveling like what, what appeared to beA, a pretty nefarious scheme of insiders and self-dealing that led to, you know, billions of dollars in losses. 54:53 And so, um, like any big company, like En-like Enron, right? Um, these companies go into bankruptcy, and so we've been watching the FTX bankruptcy pretty closely. 55:02 I mean, I think that there's a couple things that are really interesting about this that I thought w- you know, is worth sharing with, with listeners. So FTX is actually one of the largest bankruptcies by claimants. 55:13 So the largest bankruptcy in all of U.S. history w-- usually is referred to as Enron, and it had a, uh, you know, about a million claimants. Um, and FTX is looking at, like, close to ten million, [chuckles] 55:26 uh, which is unheard of. Rich- So not, it's not just the value, but, like, the number of people who have an issue with this or, like, need to get paid money is, is quite high. 55:35 Be-before we go there, um, do you wanna, like, start by explaining how bankruptcy procedures work? Um, just, like, for anyone who may not be familiar. Like, what's a claimant? How... Like- Yeah. 55:43 I have money in an exchange, it goes, goes bust. What happens to me? Well, this is g-- I mean, we can keep this conversational, right? I think this is fine. Um, so, like, when you, when you have a bankruptcy, right? 55:52 The idea is you, you, you file what they call, uh, some sort of a Chapter seven or eleven or thirteen. There's different chapters. In this case, it's a Chapter eleven, so you're gonna try and restructure the company. 56:02 It triggers what's called an automatic stay, so everything freezes globally. Like, no one pays debts, no money leaves the accounts, no one's authorized to do anything. Everyone's supposed to freeze. 56:12 Um, and this whole process of, of a bankruptcy is really designed to do a couple things. One is it's designed to kinda, like, stop payments so that it can right size the ship. 56:23 In other words, like, if this company's viable long term, 56:27 maybe we just need to restructure its debts, um, based on its revenues, and then, you know, maybe forgive some of those debts or maybe write some of those debts off or whatever it might be. 56:36 Spin them off, sell them, um, liquidate some assets to pay them. And then ultimately, the goal is, like, you know, you patch the ship up and you set it out to sea again, and it becomes the next... 56:46 I mean, this is, like, you hear about, like, RadioShack, like, going under, like, five times, right? Like, these... There's a lot of companies that have done this that people have heard about. Um, 56:54 in this case, though, like, be... You know, having it a c-crypto exchange is very different than, than running, like, a retail business that goes under and then, you know, gets p- new life breathed into it. 57:03 Specifically because, well, one, there's a fourteen billion dollar hole, which isn't [chuckles] traditionally an issue that you, you get with normal bankruptcies. 57:10 And a lot of that, I think, is the root case of the, the fraud and the other stuff that was going on. 57:14 But two, the sheer network effects that a crypto-- a global crypto exchange has in terms of its reach is much greater than I think what, you know, we're traditionally used to seeing in the bankruptcy context. 57:25 And so, um, if you were somebody who had money on the FTX exchange, in any of their products, and you were n-unable to get that money out, essentially, you get what we, what we would consider to be like a... 57:37 You have a bankruptcy claim against the estate. So you have, like, a little ticket. It's an IOU that says, like, "Hey, I had a hundred dollars here," or a hundred million dollars. 57:45 Um, and the bankruptcy court's job is to figure out, how do we make those people whole, um, while, you know, not totally, you know, sort of, uh, destroying the-- whatever's left of the company? 57:57 And so so-- a lot of this is just gonna be selling these assets. But the thing that was really interesting is so everyone's getting in line. That's the bankruptcy process. 58:04 You saw-- So, like, Hamiz is a, an exchange client. I might have sold FTX software, for example. We sit in different positions, right? Like, I have a contract. I had a debt. 58:14 Maybe I secured it with something that FTX owned. Um, like, I would be considered a secured creditor. I'm at the top of the waterfall. I get paid first. 58:22 Hamiz, who just put, like, a hundred bucks in there and lost his hundred dollars, he's an unsecured creditor. He's all the way at the bottom. 58:28 And be-below that are the equity holders and the employees who said, like, "Oh, well, I had a retirement package, and you were gonna pay me a hundred million dollars 'cause Sam promised me this was the, you know, job of a lifetime." 58:37 Like, that's even lower. Um, well, somebody who comes in over the top, which is kinda crazy, and just did recently, was the IRS. So you have all these people pairing these claims out. The IRS basically said, 58:49 "Hi, bankruptcy judge and trustee. Um, this estate, the FTX estate, owes me something in the ballpark of forty billion dollars in back taxes." 58:59 Um, and so, like, that's, that's the big so what today because if you're the person standing in line, you're basically saying, "Okay, well, they keep pouring cups of water of assets until that cup above me gets filled." 59:12 So when the creditors, you know, secure creditors, that cup gets filled, it overflows into the unsecured creditors. And then when that cup gets filled, it overflows. 59:19 But the IRS just took, like, a basically a giant twenty-gallon bucket and put it above everybody. And the question is, like, is there [chuckles] ever going to be anything that overflows for the rest of the claimants? 59:30 Rich, um- But how is that even possible, Rich? Like, how is it possible to have forty billion dollars of back taxes? Like, uh, there's no way- That's the question... 59:38 that they could have paid that much to themselves in compensation. So it must be some ridiculous penalties, you know, five, six, seven hundred percent penalties or even thousands of percent of penalties. 59:50 It just doesn't-- The math doesn't make any sense. Well, it, it, it, it can because here's-- there's a couple issues. Like, first off, so even unlawful activity you have to [chuckles] pay tax on. Like, no joke. 1:00:02 So, like, if I stole all of your money, and then, you know, the IRS caught me, they would be like, "Hey, Rich, you never paid income tax on your, on your stolen money." And, like, I would, I would... 1:00:12 Actually, I could be brought into- Yeah. I could-- They could care nothing about the theft. [chuckles] They could just care about my not paying the taxes on the stolen funds. 1:00:20 So that's one part of it, which is arguably, you know, the estate is, like, all these executives and p-perhaps the crimes that the, the corporate entity itself engaged in which, you know, we don't-- I don't know what the size of that is. 1:00:30 I think everyone's still sizing that up, and we'll find out more in the, in the, you know, the criminal trials. 1:00:35 The second part of it is, um, if you look at most of the, uh, obligations that FTX had, um, or its-- or the assets that it had as an entity, they're mostly related party assets. 1:00:47 In other words, FTX doesn't really have a whole lotAlameda had a lot of it, and Sam and some of the executives who got, you know, all these crazy related party transactions. 1:00:57 The IRS might just be looking at, like, how many times is Alameda passing money to FTX and FTX passing money to Alameda, and each time that's income. And like, they're not really doing a true up, and who knows how... 1:01:08 Like, by the way, this is crypto, so if I send ten Bitcoin to Alameda and it's at sixty thousand dollars, well, I'm paying tax on that right there and then at sixty thousand dollars. 1:01:19 Neve-never mind that Bitcoin's now only thirty thousand dollars. So th-that's why I think you get this. But just to kind of tee the-- like end this, right? The IRS is always gonna name a bigger number. 1:01:28 They're gonna come in with forty billion, and then what they're gonna do is slowly they'll probably negotiate a settlement that's much less than that. 1:01:35 But the idea is they wanna stick their flag way far in the other side's territory so that even walking it back halfway, they're getting a twenty billion dollar check. I have a, I have two questions here, Rich. 1:01:44 Um, first is how big is the estate estimated to be right now of FTX? Because I know they keep money-- they keep finding money every day that they forgot under the, under the mattress or whatever. 1:01:54 How big is the estate right now? Yeah. So the estate right now, like, again, this is based on some of the more r-r-- like this is the balances at the time of the petition. 1:02:05 So the estate had f-about fourteen billion, um, in liabilities. In other words, things they owed people, debts, money they owed customers, et cetera. 1:02:15 In terms of assets, they, uh, allegedly also had fourteen billion in assets, but thirteen billion of it is what we call related party, uh, [chuckles] related party assets or receivables. 1:02:27 In other words, it's money they're owed by a related party. Who the related party is, is it's Alameda. Alameda, yeah. 1:02:32 So FTX, technically, if you, if you get rid of that giant chunk of thirteen billion, FTX has about a billion dollars in stuff that they can liquidate. Uh- But even- And, and some change...e-even if you have... 1:02:44 Okay, let's just say that all of FTX has fourteen billion in assets to, to serve, uh, fourteen billion in li-in liabilities, which are all the creditors. 1:02:53 Um, my question then is, let's just assume that the IRS is going to get half of the check, right? Let's just, just twenty billion. Um, so two questions here. One is, 1:03:03 w- is-- do you think there's any sort of like second order incentives for the IRS or the government to be pretty much completely extinguishing the entire bankruptcy procedures or any other claimants? 1:03:14 Because this pretty much renders every... 1:03:16 I-if the IRS gets a twenty billion dollar claim, this pretty much renders everybody else, uh, th-who has a claim in FTX, which as a matter of fact, are very crypto-friendly people because are people who were either doing business with FTX or were, um, directly users of FTX, it renders them pretty much insolvent. 1:03:33 Yeah. Do you think there's any kind of like hidden incentives here by the IRS to like kill e- or, or hit everyone else who was involved in that? I mean, the-- it's just, it's bad PR for them, right? 1:03:42 Like, so it's the opposite. Like b-basically the IRS coming, coming in and doing this, like to your point, they're kinda robbing Peter to pay Paul in a sense. 1:03:49 In other words- Yeah...they're taking money from FTX that would've gone to victims of FTX, their schemes, right? And they're saying, "Actually, we're gonna get paid first, um, because this entity owes us taxes." 1:04:01 I don't... 1:04:01 I mean, look, the IRS does this like I worked on the FIFA World Cup, uh, bribery schemes for five years and, um, we got judgments against a lot of bad actors for hundreds of millions of dollars, and I will tell you, almost every time we'd be like, "Wow, we recovered fifty million dollars for our client," and [chuckles] you know, like that's a, that's like victim funds. 1:04:20 It's like tax-free. It's like this is what you're gonna get paid back for the harm that was caused to you. And then the IRS would like pop up and be like, "Oh yeah, that guy owes, uh, fifteen million in back taxes." 1:04:32 And they would take a giant chunk out of it, and then you'd say, "All right. 1:04:35 Well, I can't, I can't give the Football Association in the Caribbean, who by the way, whose annual budget was a million dollars a year and they can't even afford uniforms, I actually can't pay them out that million dollars anymore in restitution because the IRS [chuckles] basically took a giant chunk out of the fund. 1:04:50 So they get two hundred thousand." And is this the revenue that they're after? Like th-like will the IRS does this for other, for other bankruptcy procedures or other, uh, restitution claims? 1:04:59 Is it just like, "Yeah, we just want more revenue"? They're, they're... Yeah. I mean, they're just... They're a tax agency, right? They're, they're, their whole goal is to collect taxes. 1:05:07 Um, it's, it's kind of a double incentive too, right? Which is like, you know, there's always a saying, "Don't break the law while you're breaking the law." 1:05:14 [chuckles] It's very common among legal circles, which is kind of like, uh, makes you scratch your head. 1:05:18 But the point is, um, if you were a criminal and you were thinking about taking a bunch of money, you're gonna get whacked both times. 1:05:24 Which is not only will you get charged criminally, and you'll have to pay penalties and restitution in the Southern District of New York or from, you know, in a federal court, but then additionally, the, the IRS is also there. 1:05:36 And so you get pinched both ways. Um, and ultimately, like the, the government likes that. It's a double... It's like almost like getting, uh, you know- Double penalty...two bites of the same apple. Yeah. Yeah. 1:05:47 Is there... I know there's, uh, private markets for some of these claims, right? 1:05:51 Um, for people who have claims in, in FTX, kind of like exchanging for, uh, quicker liquidity at a fraction of whatever they have a claim for. How have the markets, the claim markets reacted to this news of the IRS? 1:06:03 Have the claim value gone to like zero cents on the dollar? Ye- I mean, it, it's, it... [chuckles] The question is, is like how much is the IRS's number? How much are they gonna be able to extract? 1:06:14 Will they settle for less? And they probably will. And then what does that look like? So like th-these claim markets, I mean, um, as you probably see this because 1:06:21 it's very interesting and like from an investor standpoint it's an area that a lot of people play. I think like the, the... for the short term, like the value of these claims is gonna get take like a pretty serious hit. 1:06:33 I don't think it'll go to zero 'cause there's just so much to happen. Like, I mean, honestly, who knows? Sam, Sam could have two billion dollars squirreled away in Panama for all we know. 1:06:42 Um, in which case like- That still wouldn't do anything, right, Rich? Like that's the, that's the problem. 1:06:46 The number that they've come out with is so large- So, so large, yeah...that like if we take all the best case scenario, we say, "Okay, they only settle for twenty-five percent," right?But nobody still gets paid, right? 1:06:58 And, and Sam finds it $2 billion, people still don't get paid, right? 1:07:02 Like it just-- Uh, I, I remember, uh, I, I don't follow this very closely, but I remember the FTX claims were selling for about twenty to twenty-five cents. Mm-hmm. 1:07:11 I would imagine they could really only sell for a cent now, and that'd be somebody taking a real like, you know, [chuckles] a real bet, like a, a real bet to see like what happens. 1:07:22 But I, I wouldn't pay anything more than [chuckles] a cent for it. Like, uh, not even close. Maybe. I mean, like, you know, there's gonna be people who wanna play. Like, I mean, it's rolling the dice, right? 1:07:32 It's a lottery ticket. I mean, it's kinda like the Mt. Gox thing, right? It's like all, all your Bitcoin's gone, guys, and there's only like a fraction of it left. Why would you even bother holding onto this thing? 1:07:41 Just sell it. And then Bitcoin goes parabolic, and all of a sudden the estate's liquid. It like, not just like you're made whole, like it's liq-- they have too much money. 1:07:51 [chuckles] And that's, you know, like crypto's a crazy place. So like I wouldn't be surprised if a lot of the firms are looking at this, went through Mt. 1:07:58 Gox one time, saw that, you know, that economic kind of like force of nature come and realize like, "Oh, wow, okay. If I just stuck this out, it might've been worth it." 1:08:07 And by the way, what's the opportunity cost if the claims are so cheap? You know, you park a couple million dollars there. I mean, geez, funds make bad bets every day that are millions of dollars. 1:08:17 [chuckles] Um, I mean, investing in FTX was a bad bet as, as we know. So like why not just double down on the claims and say like, "Well, I think the crypto ecosystem's here to stay, and if Bitcoin goes 10X-" Mm. 1:08:28 "Well, that's good." So. Mm-hmm. Yeah. Okay. Okay. Any, um, any other remarks here? Oh, no way. There's no way this is true. I'm looking at one of the markets that I kinda watch. It went up. Is it X Claim? The claims? 1:08:43 Yeah. No, it's like claims-market.com. What is it trading at? Like twenty, uh, bid is twenty cents, ask is twenty-five. Oh, wow. How does that happen? I mean, these-- But, well, 1:08:58 the, the other thing is you have to pay-- Like the markets for bankruptcy claims are not very liquid. And so- Mm-hmm... 1:09:05 it's kinda like when you see your favorite, you know, meme coin and it's like, "Wow, it's like going up like so much." And like the volume's like $10,000 a day. Yeah. 1:09:14 It's like one guy decided to buy [chuckles]- That probably is what it is. Yeah. Because in my- X Claims had it. Uh-huh. What is it trading at? Yeah. So I'm looking at it now. Um, 1:09:28 feature-- So X Claims has it at about twenty-four and a half cents on the dollar. Come on. There's just no way. Is it, is maybe- Guys, have we just found the largest arbitrage in the history of arbitrage? 1:09:44 [laughing] The big short, right? Like... You should start this. Yeah. You should start a, a three-letter company to s- to make some trades in this arbitrage opportunity. [chuckles] Yeah. Yeah. 1:09:54 Sell a bunch of these claims, buy a bunch of them at a cent, and sell them at twenty-five cents and [chuckles] It's like- I mean, this is what people... 1:10:01 You know, people said, like, bankruptcies, like all litigation is kinda like this, right? Like it's, it's a saga. It's like a- Mm... soap opera. 1:10:09 Like when, when FTX first went under, um, a business partner of I and, and, and mine and I bought like a bunch of claims. Like people were like, "This is dead. There's no money here. Sam took it all. He lit it on fire." 1:10:20 And it was like three cents on the dollar. Mm. Like everyone was just running for the hills. 1:10:25 Um, and then, you know, it was like six months later, this is crypto, everyone's like, "Hey, I think we could turn this into an exchange again. Like we could bring this thing back to life." 1:10:33 You know, people tweeting about it. And it's like those same people were literally like, "There's no such thing as crypto anymore. It's all dead." And then six months later, they whip lash. 1:10:40 But you know, actually, like I would say that, uh, the technology itself is, was legitimate. The company itself was legitimate. Like there is no doubt that the company itself was actually a profitable company. 1:10:53 Like even to this day, after all the investigation that the f-- uh, that the new FTX management did, like their billion-dollar, their claim of a billion dollars of revenue, that was real. 1:11:03 Their claim of a couple hundred million of profit- Yeah... that was real too. The reason they failed was not because the exchange itself wasn't a profitable, highly successful business, right? 1:11:14 It was a three-year-old company making a billion dollars of revenue. That's a successful business. It was because of these conflicts of interest and this like ins- insider, uh, dealing that, that caused them to fail. 1:11:27 Uh, their hedge fund was a terrible hedge fund, but their exchange was, was a phenomenal company. Yeah. I mean, it's kinda splitting hairs 'cause you, you need both to run a successful company. 1:11:37 It's kinda like Home Depot was a great place to buy hardware until they pilfered the entire company and stole all the cash. [chuckles] And it's like, if they just hadn't done that, if they just had good management. Yeah. 1:11:47 So it's, it's, you know, part of running a financial services company is having good management that can suss out the, the conflicts and things like that. 1:11:55 But, I, I mean, I wouldn't be surprised, and you're seeing a couple of, of people who are in or around FTX like starting new businesses, right? Where they're like, "Hey, like this worked at a global scale. 1:12:06 So we found product market fit. We know it's there. Now we just need to pick these pieces up and like solve for the d-don't be a criminal, and maybe we can build another billion-dollar exchange." Yeah. Yeah. Yeah. 1:12:18 I would, I would agree with that. Okay. Um, let's- Should we- Let's- Should we wrap and, and leave the banking crisis and that saving for our next episode? That will be the first topic next time, I promise. All right. 1:12:30 Sounds good. Perfect. And we'll all go after you. But thank you everyone for joining. This was a very interesting discussion. Sweet. We hope it was for you as well. 1:12:40 And, uh, we will try to do these every two weeks, so you'll be seeing a lot more content from us. Uh, please give us your feedback. 1:12:47 This is the first one, so we're, we'd love to hear from you to see how we can make this better, uh, what format makes more sense, whether there are certain guests we should have on. 1:12:57 But we're excited to get this, uh, off the ground. Thanks. [outro jingle]
