马丁·施克莱里访谈：做空量子计算、大药厂的盲区与狱中领悟

Speaker 1: Um, he's been called the most hated man in America because he jacked up the cost of a lifesaving drug.

Speaker 0: Raised the price of a life saving drug 5,000 percent.

Speaker 1: He became the poster child for capitalist greed, spent seven years in prison.

Speaker 1: We definitely plan to raise the price. That's for sure. You can be prosecuted for not maximizing profit. And you have to do everything in your power to make as much money possible. And if I'm the enemy for you, good, unhappy man, I'm fulfilling some need in your life. Purchased the only copy of Wu-Tang Clan album with no immediate plans to play it. It goes the farthest...

Speaker 0: Please welcome Martin Shkreli.

Speaker 1: Welcome Martin, nice to come out, nice to meet you.

Speaker 3: You wore Brad's outfit today, exactly. Alright.

Speaker 0: Welcome, Martin. We've got so many topics to go into with you, but I want to start with one. You've basically taken down quantum computing. You said you think that this is a fraud, that people might be going to jail. So two-part question: Number one, um, maybe you could give us your thesis on why this is a fraud or fugazi.

Speaker 2: Whatever is, disgraced and a fugazi.

Speaker 0: And then two, your best advice for those CEOs when they do go to jail. You've got a little experience. Best advice in terms of—you don't seem to be the kind of guy who would thrive in prison, but from what I understand, Martin...

Speaker 2: You were a shot caller.

Speaker 1: I'll got shot caller for you.

Speaker 0: And you were running the yard, I understand. So break us through the quantum trade, and you know, best practices when you go to the big house.

Speaker 1: Yes. So I want to shout out my wife, Chelsea. So Chelsea, um, very few people know her, but is I think a big deal in technology. She was one of the first people at OpenAI, and her father, believe it or not, was one of the first people at Nvidia. In my family, I was one of first people in prison, so I thought I switched the gene pool, you know, shifted over a little bit.

But Chelsea studied at MIT, taught quantum computing algorithms and all that. You know, she got to teach me over many hours, penetrated my thick skull to teach me what exactly quantum computing even is. And I think a lot of people sort of synonymize it with faster computing, which sounds great. That's really not it at all.

Um, and I think the basic point of quantum computing is that there is one algorithm. In fact, the whole field was invented because of Peter Shor's thought experiment. Peter considered this idea that there are kinds of algorithms that you can run faster in different media. That was a revolutionary idea in computing, but it's not really a revolutionary idea for business.

Because there's one algorithm that runs faster in a quantum computer. And if you asked somebody which computer can sort a deck of cards—anybody who knows the famous algorithms books like CLRS knows the first pages are about sorting algorithms like bubble sort and merge sort and things like that. I asked quantum fans: well, which one can sort it faster? And they said, of course, the quantum computer. Instead, of course not! That's not the case.

The modern quantum computers run at clock speeds of hundreds of hertz, or thousands of hertz, not at the gigahertz that we're used to. They're single core, and they're single threaded when it comes to doing a job like a merge sort or something like that.

So the only thing it's really good at, which indeed is a magic trick, is going from exponential time algorithm—which is integer factorization or the discrete log problem, shout out to the ECDLP problem—the quantum computers can do those in polynomial time, which is remarkable. And to a nerd that's the greatest, coolest magic trick ever.

But to a business person, unless you're in the business of factoring numbers, there's not a whole lot to be excited about here. The other algorithms quantum could do are not much faster. So there'll never be an Nvidia machine. There'll never be a million Nvidia machines, and quantum doesn't have storage. There's no such thing as quantum RAM. So you have to design a problem that has a really, really small input. You cannot put the world's compute into it. Every time somebody pitches OpenAI from what I heard about quantum computing, they say, no, we're not interested.

Speaker 2: I'm closer to you in my perspectives on quantum, but let me take the other side. Let's just take the other side. You know, the other side would say: look, we're in this a very early primordial phase of actually constructing transistors. And from the transistor, what we're going to do is like put a bunch of these things together, we're going to figure out gates, we're going to figure out logic, we'll then be able to build some reasonable operating system on top of it, we'll have some reasonable mechanisms to write software to these things. And none of that really exists. So do you buy that we're just at an early phase? And this is why we're frustrated that there's a computational limitation, except to a very narrow use case. I do believe that that could be possible.

Speaker 1: Yeah, no, of course. I think the issue though is: would you go long punch cards and vacuum tubes? Mercury delay lines, core memory? These are all things that disappeared and went bankrupt. Those were stocks at one point—CTR, Computer-Tabulating-Recording Company.

Speaker 2: But okay, is your critique business model and timing, or is it technological?

Speaker 1: I think it's both. You know, I think when you talk to Google or IBM or Microsoft and ask: why are you spending hundreds of millions of dollars a year on this? They say, well, someday there might be something that comes out of it. That's not a great business to me. What is very shocking...

Speaker 2: I mean, you can probably tell the story, but how many machines actually—how many machines has IBM and Google actually have? A very few?

Speaker 1: Google has a few prototypes. IBM has a little cloud. Chelsea and I have tried every single quantum computer in the world. We cannot multiply 300 by 200 on any of these.

Speaker 2: And you're shorting them?

Speaker 1: Yeah, I've been shorting them. Unfortunately, we lost millions of dollars shorting quantum computing companies, but I told my wife that we gained billions in insight, right?

Speaker 0: It's not easy to time the markets, yes, definitely.

Speaker 2: So where is the natural computational edge? I mean, people say it's going to be towards encryption and defeating encryption, because now we can actually use what it's actually good at to essentially brute force solutions to things that we can't solve today.

Speaker 1: I think it's crazy to even think about it, because we have a computational problem that we're trying to solve. And every time somebody uses the word AI, I actually think of the word matmul. These are matmul machines. And matrix multiplication is what is working, right? Multiplying two lists of numbers is what a GPU spends 99.9% of its time doing, but it does it in $n^3$ time.

Complexity algorithms are going to be paramount in the next 5-10 years as we move towards this society. $n^3$ means if you want to multiply 1,000 element square matrix versus another 1,000 element square matrix on an Nvidia machine, it takes a billion operations. That's a lot of operations.

As we learned about this computing stuff through my lovely wife, I was flipping through Nature—which is what we do as a husband and wife. I saw a paper from Microsoft that was called AOC: Analogue Optical Computing. In New York that means something different, but AOC in this case... What caught the corner of my eye was that matmuls in light are free. Yeah, they happen in instant time, which is a really remarkable thing.

So we're investigating, maybe we'll start a company or something like that. There's a couple of startups—Neurophos, OpticCore, Celestial—that want to make these photonic matmul cores, which would run circles around what Nvidia does. Now, of course, that's not the whole story. There's a lot of details and nonlinearity and storage.

Speaker 2: Beyond the hatred for quantum and your raging short that may or may not be winning: You were a hedge fund manager, correct? And the famous episode, just to catch everybody up, was you bought a drug, and then you inflated the price of the drug. Is that what happened? Can you just give us the story? What's the history, and what takes you now to where you are?

Speaker 1: Yeah, I started at 16 with Jim Cramer. Cramer is actually a tremendously effective and skilled trader. But Cramer's idea of investing is you buy and sell all day, every day, everything, and somehow squeeze profits out of it. It was the way you did it in the 80s and 90s, and he made 23% a year net of fees.

Speaker 2: And so you worked for him at 16?

Speaker 1: I worked at 16.

Speaker 2: How did you get that job?

Speaker 1: Just being in Manhattan, having a big mouth. You know, it's a great city, and loving stocks, loving investing. Bill Gates was my hero. And I found a way of getting into that. And Cramer one day pointed to me and said: genomics, you go look at genomics. And I said, okay.

And it was a lot of fun. I turned it into a Tiger Cub gig, started my own fund. I realized I was a much better entrepreneur and private equity guy than a trader. We found that the marketplace for molecules had huge amounts of alpha. Whereas the marketplace of public companies—we could buy a drug from SmithKline for $2 million, and that same company is public today at $6 billion.

So that was my first startup. And it was a really inefficient market. Pfizer would sell you a drug for a million bucks that was worth a billion dollars. They don't care.

Speaker 3: You bought the rights to the drugs, so you could optimize the price?

Speaker 1: Yeah, absolutely insane.

Speaker 2: Why didn't everyone knock on Pfizer's door? They're like: take what you'd like for a million dollars.

Speaker 1: We saw this. And we built two very similar companies. Our biggest constraint was finding people to vet these assets. Just reading papers, reading patents, going to universities and small drug companies and big drug companies, making relationships. It's called business development in our field, and our biggest bandwidth problem was reading these papers. So I started a small NLP company on the side in 2015. We tried to read every paper that was published in medicine—that's four million papers. And having humans do that is a really tough thing.

Speaker 2: How do you take a molecule and flip it from $25 million to $6 billion? What are you doing in that 6 or 9 months that you own it?

Speaker 1: What I did with Sparsentan—a drug that I named, which is now approved and a blockbuster drug—Bristol discovered and developed it in hypertension, which is sort of a dead, sleepy market that nobody really needed for high blood pressure.

And so we moved it into a disease called FSGS, which is a disease that had never had a drug approved for it. So you sort of angle it in the way that makes sense. And in that case, the drug got approved. Thirty thousand people who have focal segmental glomerulosclerosis and never had a drug before now have a drug. I can be a good guy too. And I think that is the art of that field. It's a great way to make a billion dollars.

Speaker 2: Is this also what Royalty Pharma does? They come in...

Speaker 1: There's a bunch of different players. Ingenious guys, obviously, made famous that they took what was getting this big royalty from Remicade and bought it for like one times earnings.

Speaker 3: And then that royalty 20x. Buy the cash flows at a bigger discount.

Speaker 4: Martin, take this framework now. The big levers in the market today are what's expected in terms of compute, what's expected in terms of this CapEx build-out. Are you spending much time looking at that? And do you see any boogeymen? What are you worried about? What would you be short in? What would you be long in that complex?

Speaker 1: I mentioned photonic computing before, because I do think in 20 years, these A100s will be in the scrap heap, H100s in the scrap heap, B200s in the scrap heap. We're going to find ways to do this better.

I think it's insane that we're thinking about space, thinking about nuclear. Sam Altman finally said publicly on a podcast that he thinks photonic computing could be the future. So I'm just wondering why we think this is a sane idea to cover the earth with computers just so we can make a faster one.

I'm not long, nor short compute right now. Jevons always wins in the end. Alexandr Wang told me Jevons is going to take care of all this, it'll be a million-x compute improvement.

There are guys in the Valley talking about better transformers that are a thousand times better. If there's a thousand times more efficient transformer, that's bad for Nvidia. There's no doubt about that. You're not going to have a million-x demand. There aren't a million times more people.

My guess, like anybody else, is probably what people guessed in 2000 with Yahoo. So it's a tricky place.

Speaker 4: Do you run a hedge fund still?

Speaker 1: I don't.

Speaker 4: Okay, so you're managing your own money. And when you think about portfolio allocation, are you thinking about it in the way you did when you ran your fund? Looking for specific high-alpha events or startups or disruption and going long behind those, not really thinking about competing with the market?

Speaker 1: Yeah, certainly. This Dario trade was like a great reason to go long or short...

Speaker 2: What did you think about his essay?

Speaker 1: It was ridiculous. I want to pay deference to him creating a multi-billion dollar company, but I kind of hate the guy. I think he's a fear-monger.

Imagine you and I started a business, and I said, I'm going to show you Pandora's box here, look at this. And you say, wow, that's amazing. And he says, well, Martin down the street is going to open Pandora's box too, so we have to do it safely... I would probably sue you for trade secrets or something. This is exactly what happened.

Speaker 2: So you're saying because he was at OpenAI, and then he and five or six co-founders leave...

Speaker 1: Would you be friends with that?

Speaker 2: I think in Silicon Valley that kind of happens a lot. It's sort of hard to pinpoint what is a pattern versus retained knowledge. I'm not excusing it...

Speaker 1: But the Waymo guy went to jail for it.

Speaker 2: Well, that was because he was found to have actually stolen physical documents. So you have an issue with the origin story?

Speaker 1: Probably the fear-mongering, the ridiculous hyperbole. He says AI is going to help drug development. But what about Effective Altruism? That is even worse.

Speaker 0: Why? What do you find so offensive about this group of people? Let's unpack it.

Speaker 1: I helped popularize that, and I even think I helped start the e/acc movement, which was this counter-insurgency.

Speaker 2: So there's EA, and there's e/acc—Effective Accelerationism. Which means what?

Speaker 1: It means sort of "move fast and break things." Zuck came out the other day and said, yeah, we should move. And I think the fear-mongering is ridiculous. There'll be time to panic, but it's not now physically.

For now, you can turn off the computer. It's not that big of a deal. I think nerds get carried away. The jocks know what to do when the spotlight is on them, the nerds don't. And some of them, instead of saying everything's fine, want to draw attention to themselves. They want to say: well, the thing I'm working on is really amazing.

The reality is that this is just software. It's not the end of the world. It is the most important kind of software, but it's not going to kill us all.

Dario also said that we'd be fully unemployed within 1 to 5 years, about two years ago. Well, he's got three years left. He also said it would cure all diseases. I don't think there'll be a single AI-relevant drug approved in the next ten years. AI does not help drug discovery very much. If you've ever built drugs—raise your hand if you've made a drug and got an FDA approval for a drug? Okay, I'm pretty sure that's nobody else here.

It's pretty hard. We've had computers for a long time. I built a Cray cluster to analyze drugs. Now I can use GPUs, great. But there isn't much AI can help us with in drug discovery.

Speaker 2: What is the revenue cycle that's driving valuation? Unpack it from a hedge funder perspective. You see these run rates. Jensen mentioned it yesterday: billions of dollars that have gone into venture have been redirected into AI. Do you interpret that as VCs funneling LP money to buy tokens effectively? Is there a valuation trap in there?

Speaker 1: I don't think so. There is infinite demand for intelligence and infinite demand for work. There's no limit towards intelligence.

My buddy left OpenAI and started Cognition, which we're really excited about. He invented the reasoning model at OpenAI, and he's got a startup. I think you're going to have AI starting businesses. Part of the plan is to have within OpenAI: OpenAI Pharma, OpenAI Motors, OpenAI Aviation. You're going to have a whole S&P 500 within OpenAI or Anthropic. The opportunities are boundless.

One last thing: McDonald's can afford to spend $10 billion and say, OpenAI, give me 10 million GPUs of minds, I want to know what the best next sandwich launch is. And that's an NPV-positive trade, if the intelligence is worth it and not asymptotic.

Speaker 2: What is a good analogue for the AI marketplace? Handful of providers of a critical resource, incredible valuations attracting capital, breeding competition. Jensen made the analogy yesterday of like water—you have $8 water and free water. What's your mental model? How do you equate this to previous markets?

Speaker 1: It has to be like the Internet. The only true winners will be consumers and businesses. They get this tax-free highway. Cisco and some other guys will get a little bit out of it, but the main highway that we're all going to benefit from is that we have free intelligence in our pocket.

Speaker 2: So we have to be guessing right now around who is going to overearn, and the duration of that overearning?

Speaker 1: Sure, and I think it's a matter of timing. We're still Day 1 of AI. The chatbots—we're going to look back and laugh and say these were the early prototypes.

Speaker 4: Would you go out today and buy OpenAI and Anthropic?

Speaker 1: Yes, we have OpenAI obviously through a few different ways. And I think that Sam's hedge with hardware is really brilliant. Being able to deliver that model is probably going to be more valuable than just having a creative product.

Speaker 4: What did you think about Elon's suggestion of competitor peer review as kind of this third way, a pragmatic solution to balancing safety and speed?

Speaker 1: You know what they said of the Wright brothers: Warren Buffett said if he could go back in time, he'd go back and shoot the Wright brothers and stop the aviation industry from ruining capitalists. Nick Bostrom wrote this book Superintelligence that scared the hell out of Elon.

It would have been better if some of the EA kind of folks never had this narcissistic self-importance that is driving this fear-mongering. OpenAI doesn't truly believe it; they're in this just wanting to not be too different from Anthropic.

Thankfully, Meta has the spine to call it out and say this is nonsense. Jensen has the spine to do the same. It's just computer science, it's not that big of a deal.

Speaker 3: Some of these EA guys are pretty brilliant. Dario's no dummy, Dustin Moskovitz, Sam Bankman-Fried...

Speaker 1: Sometimes you could be really smart at one thing and really dumb at another.

Speaker 3: That is my point. Some of the smartest people are unconventional thinkers, and you are clearly a brilliant person. People that are brilliant are generally brilliant because they're not conventional, they're heterodox in some way. But in the same vein, if you have that brilliance, perhaps it's also hard for you to be conventional in the ways that are, call it, law-abiding.

I want to ask your opinion on Sam Bankman-Fried. You've met the guy, right?

Speaker 1: I gave him advice.

Speaker 3: You gave him advice on how to tough it out. But can you give us your sense on the guy? He was early in Anthropic at the beginning. You look at the portfolio of investments he made across the board in 2020, 2021—he looked like a genius.

What got that guy off track? Is it that he was so out of bounds as a thinker that got him all the alpha, but in the same way got him in trouble? Or was it speed?

Speaker 1: I think it's both. One of the problems, if you grow up as an intellectually gifted person, is you have this big albatross. That happened as well with Elizabeth Holmes. Raj Rajaratnam was another example of a hedge fund manager that was a decent investor.

You have this self-consciousness that you have to live up to something. And sometimes wanting to live up to that big thing can be very detrimental. And just failing at admitting that you are not that great at that thing.

I had a hard time admitting I was not the best hedge fund manager in the world, but I was a wonderful entrepreneur. That simple refusal to admit that put me in prison.

Speaker 3: Have you changed your views on your behavior? What has changed about you from prison? What's the change?

Speaker 1: Psychology is a really powerful force. All our actions are motivated by innate drives that we don't always understand. I didn't realize how petrified I was of not being seen as a great trader. What a stupid thing!

You can be a great long-term investor, value investor, entrepreneur, great scientist, or a great dad. But if you fixate your mind on "I have to be this," that can be really detrimental.

Speaker 0: Putting aside your case, valid or not, what did you learn during your time in the big house, and how did it change you as a person? Because you seem to have gotten a lot sharper, a little bit more methodical in your approach to life. Take us through what you learned in that time.

Speaker 1: One of the interesting things about prison is, no matter where somebody's from or what they did, everyone sort of has a lot of these same real common features that bind us as humans. Everyone is upset when their family is in trouble, or longing for that person they miss.

And it's really refreshing to remember that, no matter what heights we achieve, at the bottom, like you said earlier, we all want to be the best dad we can, or whatever the case is.

Speaker 3: I think you're just a very independent thinker. And you're willing to say the things and communicate in the way that other people aren't. That's why we wanted to have you here, and obviously this was a great showcase of that. We really appreciate you taking the time to be with us. Thank you.

Speaker 1: Thank you, yeah, that's great.

Speaker 2: Thanks, nice to meet you.

Speaker 0: Man, come back anytime.

Speaker 1: Thank you, thank you.