Crypto's Quantum Challenges & Optical as the True Quantum-Class Winner – Martin Shkreli

0:00Martin Shkreli:Today on the James Altucher Show.

0:30Martin Shkreli:tried to make the$100 bill curl. And it's like with optical, it's like God made this idea for this exact thing because light can do matrix multiplication in essence for free. And then the energy requirement's really low. We've really not tried other kinds of computing. And I think now that computing is literally threatening civilization, it's a good time to start.

0:57James Altucher:This isn't your average business podcast and he's not your average host. This is The James Altucher Show.

1:14James Altucher:So last episode, we spoke about whether this man, Martin Shkreli, is the most hated man in America. My conclusion was no, and I hope you come to the same conclusion by listening to the last episode. And in this episode, we talk about everything from Bitcoin to quantum computing. What's the future for AI and energy and the US? And here we go.

1:45I'm looking forward to this new thing we're doing together and hopefully will be a really big thing.

1:50Martin Shkreli:Let me ask two fundamentals questions.

1:52James Altucher:One is, you've been a big supporter of crypto. Crypto is obviously in this downturn right now. Well, yes and no. I mean, I've been a, I wouldn't say just a supporter.

2:00Martin Shkreli:I've also been a short and things like that. But Bitcoin itself, let's say Bitcoin and Ethereum. It's complicated. You know, I think I've always said that like these are, they're a conundrum, right? Markets are relatively efficient too. You know, you can see the good things about Bitcoin, but you can also see some very obvious bad ones. And one of the ones we discussed from time to time is encryption. And Bitcoin is encrypted by elliptic curve math. And there's no guarantee that this encryption will hold up. And it's especially guaranteed to not hold up in quantum computing regime. Now, we don't know if there will be a quantum computing regime necessarily.

2:38Martin Shkreli:But like most people, I think eventually there will be. And then Bitcoin sort of toast. And of course, you can patch it. But the problem is, you know, you have people like Saylor saying, well, this is the hardest form of money. And it's like, well, Mr. Saylor, you're not a cryptographer. No, I'm not saying I am, but I understand a little bit more than you do. And if you have to patch something every five or 10 years, it's not the hardest form of anything. You know, and to describe software of any kind, even the Linux kernel itself is not the hardest form of software. Bitcoin is programmable money, but it's very faulty.

3:11Martin Shkreli:And Ethereum is probably worse in some ways. So these are cool innovations, but are they locks like a stock? Oh, I don't think so. I mean, you know, they're speculative. I think they're almost like gaskets for excess money. If you have$10 billion you made from your tech company, you might put a dollar in Bitcoin because you need to do something with it. And it's kind of cool. But just because it's a gasket, does it make it a, you know, something that's like, you know, valuable?

3:39James Altucher:Okay. The one thing is though, like you look at it from, let's say Trump's point of view, his family got debanked when he was no longer president. And it's almost like your battle against pharma, like big pharma. They control more than people realize. Just like big banking controls more than people realize. And crypto became this sort of solution. Like if you can't get a bank account with Chase, this might be an alternative. And look at what's happening now with the Clarity Act. Like the banks insist on, it's a good thing that they keep giving 1 % savings rate on your savings account. When I could just stake Ethereum or whatever and get 3%, 4%, 5%.

4:15And the banks hate it

4:18James Altucher:and they're paying off the senators. and consumers are buying into it. It's like this unfair thing that crypto is trying to correct.

4:26Martin Shkreli:Yeah, I mean, there's a clear, you know, use for, say, stablecoins. And there was just a Fed member who just said, you know, what's the point of stablecoins when we have PayPal? And it's like, dude, you've never been canceled by PayPal, obviously. And you've never been, you know, you send, you know, a hundred bucks to somebody in Africa via PayPal. They think it's fraud and your account's canceled. Sorry. You know, that just doesn't exist. in most crypto schemes. And I think that, you know, crypto hat, you know, stable coins are amazing. You know, it's this unbelievable way to send money. But again, you know, to the state-sponsored nannies, you know, the Elizabeth Warrens of the world, that's apocalypse.

5:03Martin Shkreli:You know, how could somebody be sending money without their okay, their knowledge, their blessings? And, you know, they've even tried to make the$100 bill criminal. You know, and in the treasury manual somewhere, it says that the$100 bill is the tool of the criminal. and it's like, look, it's literally US currency. But if you go to a bank, withdraw your account and say, I want a million dollars and a hundred dollar bills, they look at you like you're crazy. And they make a bunch of phone calls. They take a bunch of notes and videos. They'll eventually give it to you. But it's just like, you know, so unusual.

5:38Martin Shkreli:Obviously, having a million dollars and hundreds is a little cumbersome, but there should be nothing suspicious or illegal about it. It's just, you know, if that's what you want, that's what you want. But we've evolved a society that's like very, you know, very Orwellian.

5:53James Altucher:And now, okay, let's talk about optical computing. So I think this is where we very much agree is that Moore's Law has hit its limits, which is kind of the wall that just basic CPUs have run into against. They're too small. They can't get really smaller. They're not going to get faster. But then there's Huang's Law. There's Jensen Huang as Huang's Law, which is that don't care about the speed of the chip. compare, care about the computation that's actually the product, which, so how fast is your data set or how fast are your interconnects, your GPUs, all the things that come into one piece of computation.

6:28James Altucher:Wang's Law is that computation itself, forget about the CPU, computation itself will get twice as fast every two years. And that's where optical computing comes in because suddenly now you can replace, or if it works, and this is a big if, if you can replace electricity with light or lasers, you're going to have chips that are 100 times faster, 1 ,000 times faster with one-tenth or one-100th of the energy. And life's going to get, computation will get faster and more energy efficient. And we're sort of, the technology is sort of there. We're like right on the edge of it being there.

7:00Martin Shkreli:Yeah, absolutely. I know Elon is supposedly very, very excited about it. And there's a lot of startups in the space, including basically ours. But I think that, you know, I'd add a few things to what you said. I'd say that like the energy efficiency is more like maybe even on the order of a million times. I'm not an energy guy. So like I have two minds about this. I think that's great on one hand, and we could debate, let's put a pin in that for a second because I also think it doesn't matter, but let's put a pin in that for a minute. I'm a speed guy. You know, this is what I'm a junkie for. And I think that, you know, NVIDIA is a formidable competitor.

7:33I think that in essence, they've been able to kind of beat Moore's law

7:38Martin Shkreli:through some cheats, you know, one of which is massive parallelization. So you're right, the clock speed's not getting any faster. They are squeezing more transistors on there, but they make the chip bigger, and they make it more, especially, they're putting more energy through it. So as the chips have gone in generation, they take more and more wattage. And the next chip is going to be like 50 % more wattage. Well, basic physics tells you 50 % more wattage gets you 50 % more compute. And so you don't have to add, you know, you add another, you know, 50 % geometric multiplication here, but you add whatever the exact multiplication is worth of transistors and you've doubled your speed.

8:16Martin Shkreli:And so just putting more power into it will do that. Now, you need better cooling too. So it's not that simple. It's a lot of great...

8:24James Altucher:That's why Hawaiian's law, if you want to call it that, it's the whole package. Because people don't care about the chip. They just want their answer faster. Exactly. And this is your point. I don't care about energy either. I just want my answer twice as fast.

8:39Martin Shkreli:Now, the problem with that is, so I've never believed this because I've always just said instinctively, okay, we can get more energy. Mankind has always been able to, since the dawn of time, we've never run energy. Now, so like who's crying about this? Now, as I think about it more, there's actually a reasonable claim here. So first, the grid itself is not globally interconnected, right? There's kind of like these siloed pieces of the grid. So like building a power plant in Nevada, some of that energy may end up in Florida, but generally it's going to end up in its surrounding area. And it costs money to transport it.

9:16Martin Shkreli:And there's kind of this localization of energy and electricity generation that's utilities people understand this way better. But then if you zoom out and say, so it's not so simple to just add to the grid, right? Like you have to get local establishments to like use resources that are nearby. you know but part of me feels like maybe this is a good thing that you know we can put 10 million middle class workers to work you know making coal plants and maybe making the first refinery since 1970 and things like that let's let's upgrade our energy generation capability but if you look at the amount of energy it used to be like bitcoin for example was using it was comparable to like a small country and now if you look at how much ai is using it's comparable to like relatively big countries like France.

10:01Martin Shkreli:And so if you believe, like I do, and I think almost anybody with a brain believes that AI is going to keep getting usage, especially with test time compute, this notion that the longer you let the chip cook on the question, the better your answer will be. And for some questions, people will wait and let the chip cook. And in fact, if you're the CEO of Boeing or McDonald's, and you say, I'm the CEO of McDonald's, what's the next hit sandwich we can make. And you want the world's best thinkers to think on this. Well, in the past, you got to go find, you know, Johnny Ive or something like that. Today, all right, what do 10 million GPUs think of this question?

10:42Martin Shkreli:And I'm just going to go with what GPT-9 thinks. Well, this could be a multi-billion dollar W if the GPU gives you a fresh idea. So you might say, I'm willing to put 10 million in that. I'm willing to put 100 million on that. Because what if it comes up with an idea that my guys couldn't think of. And more importantly, what if it comes with the reasoning that makes sense more than my guys have come up with the idea, but I didn't believe their reasoning. So you're going to see corporations want their own GPU farms to answer questions that they have. You know, Jeff Bezos would always say, my job as CEO is to make one or two good decisions a year.

11:14Martin Shkreli:And that's totally true. You know, everything else, anybody can pick a healthcare plan or, you know, hire relatively well or whatever. But, you know, that one big decision, should we do cloud computing? It's not so easy. And, you know, GPUs may be able to answer that better than us. So I think you're going to see an explosion in demand for AI as well as just the average person. And not, this is not even to mention things like robotics and whatnot. So I feel like you're going to see this insane explosion. So if we're at France right now, is AI going to be the size of China in electricity usage?

11:45Martin Shkreli:Could it be the size of North America? And then obviously you can't have it be much more than, one more log order than that. We're running on a planet. So maybe there will be an energy crisis. And again, Elon's trying to say, well, I'll put them in space. And that sounds great, but maybe we should just make a more energy efficiency. And I think that's where optics comes in because my calculations are you could do the same inference potentially as much as a million times less energy. Now, again, I care about speed, but if this ends up really being a problem, which it very well may, and some people are so strongly opinionated on this that they believe it's certain, then yeah, optics is a great way to go.

12:26You don't even actually need the speed up.

12:30Martin Shkreli:You just win on the energy. And even if you're equivalent or arguably even slightly less fast, who cares? Because I can't even deploy a chip if the grid's congested.

12:42James Altucher:So I'd rather... I could throw more money at it. If I'm truly saving a million times on energy, okay, I'll spend a million times more. for energy and have a much faster chip.

12:52Martin Shkreli:Well, I mean, I think it's that. And then also, you're just cut off at some point. If the government says, sorry, Google, you get two gigawatts and you can't spend a watt more than that. Then they say, well, we have demand for 50 gigawatts using a GPU. But if we use an optical or photonic computer, we could actually satisfy all that demand. But we got to switch out our stuff and they'll do it in a heartbeat. So, you know, we may get to that point where we're rationing energy. And look, utilities are still regulated. Like if they said, we want to draw two gigawatts tomorrow, the local utilities would say, hey, no, no, we're not doing that.

13:26Martin Shkreli:We need people to light their homes. Forget your AI. So I think the more I think about this, the more I see it as being realistic. Now on the flip side, as a GPU programmer, I can tell you that these are getting more efficient. And the rumor is DeepSeq is about to drop a 10x or 40x more efficient model. Now today's models are still, they are like hundreds of times more efficient than yesterday's. But Jevin's paradox is showing that we're using them way more than 100x. Even another 10x increase in efficiency, which if you're in video, you're like, wait, you can do with 10 GPUs what used to take 100.

14:05Martin Shkreli:And then maybe in another year, you could do with one what used to take 100. Well, Jevin's paradox says, yeah, but you're going to need 100x more thinking, 100x more video generation, 100x more robotics. All this stuff will take that efficiency and eat it up. I think that that might balance and equilibrize, reach equilibrium in a much better way than maybe we are giving you credit for because there's always scientists sitting around saying, how do I make this GPU faster? What can I do with this instruction set to like skinny down the transformer so that, you know, maybe do I need this? Do I really need that?

14:40Martin Shkreli:Let me tinker with this. And lo and behold, you tinker enough and, you know, it's 10x better. And Claude just did that OpenAI's just done that. And so they'll pass those savings across to the customer or the capacity. And those things will really open up and we could have a GPU Armageddon where it's like, actually, we found a way to be a million times faster just on GPU. So the grid's not a problem. And NVIDIA's, you know, going to have its own problem with demand. So I don't know what's going to happen. I mean, I think that Jevin's paradox could win. Or... It always has. That's why it's a paradox.

15:15Martin Shkreli:That's a good point. You know, Intel demand for trad compute x86 compute did not scale with Moore's Law. And in fact, you know, the trad compute scaled faster than Moore's Law, so trad compute prices went down a lot. And, you know, that's probably going to happen here. But the demand, we've never seen something like AI before, where if you throw more compute at Windows 95, it doesn't get better. It has a certain amount of compute. It doesn't get exponential. It's the same Windows 95. AI doesn't work that way. You throw more compute at AI, it gives you better answers. It gives you better output.

15:54Martin Shkreli:So I think there is something a little different here, but we'll have to see. This is the exciting part about technology. Nobody knows.

16:02James Altucher:Yeah, and look, I would say if you compare kind of these competitors to Moore's Law, like let's say optical computing, DNA computing, quantum computing, optical computing seems to be the front runner. Like they know a laser can get from point A to point B. A laser probably can trigger logic gates. Like, you know, there's already success with interconnects and 6G communications. So this is probably the closest winner. I suppose like quantum computing, which is the most exciting in the space, optical is probably the closest.

16:34Martin Shkreli:For AI, it's perfect. You know, I would say for other types of computing, it'd actually be terrible. So, but the funny thing is the only kind of compute that matters right now is AI. So I was talking to, you know, this big quantum investor and I said, okay, say this company makes a strip. What's next? What do you do with it? Unless you're ready to crack some Bitcoin or crack some state secrets. I mean, you know, there's not much you could do with it. And he said, yeah, that's the catch with quantum is like we have to figure out, hopefully something will come up that will be really useful. And it's like with optical, it's like God made this like idea for this exact thing because light can do matrix multiplication in essence for free.