Martin Shkreli on Drug Pricing, Being Hated—Rich, and His GPT-4 Doctor

Sure. It was a company called Vital Therapies. I applaud the company for making an effort to try to save people with acute liver failure — that's a huge, huge problem. And by looking at this company, the company did the world a bit of a favor by proving that their approach didn't work, which they should be applauded for. It's not — the people in that liver failure trial died. So it doesn't go like, yeah, we made money and the stock went down and we were cheekily celebrating it. It doesn't go beyond my notice that this is a terrible illness that needs a solution. So I'm not the hero in this story. Even though I think they did some things wrong — they kind of knew that their device wasn't going to work, but they marshaled ahead anyway, which is the part they shouldn't have done.

So what we saw is, to your exact description: they reroute the blood from, I guess, the portal vein to this machine and reroute it back. So it's just like a dialysis machine in a lot of ways.

The idea is: okay, your liver's not working, so let's put the liver in a box. Not a bad idea at all. It works for dialysis. It works for cardiopulmonary bypass. It works for plasmapheresis. There's a bunch of tools that work this way. So let's do it. And the problem is the liver is very complicated as a surface-area problem that you can't really replicate. It's vascularized, which you can't really replicate. The cellular machinery really has to work. You can't build an organoid in a box unless it really looks and feels like a liver — it can't just be a couple of liver cells that get perfused correctly. That's why the body looks and works the way it does, and why our organs are functional.

So the biggest hint they had wasn't in the published literature. It was actually in their own papers and presentations, where they did their first study with the box and showed that the product didn't succeed. But if they looked at a subgroup of patients who had — I forget what it was, like a MELD score that was too high — those patients did really well. But of course this is classical post-hoc, retrospective analysis that doesn't work. It doesn't tend to replicate. That's just knowing medicine and knowing statistics, knowing the history of this stuff.

They're not doing Bonferroni error correction and stuff like that. Once your primary hypothesis fails, there's really no statistical inference you can make on any secondary hypothesis. And they went ahead anyway and said, okay, this might work.

We did look at the history of all these liver-assist devices to try to figure out, is it even possible? And I actually thought one of the best parts of the story could be that we started a product that actually could work, because this would be a huge product. There are tens of thousands of people who have liver failure every year. So ultimately, a really important need — somebody should take this challenge on. This is big, big unmet medical need. But we made a bet. As sophisticates, we thought the bet was — we wrote a 40-page paper about it, which nobody on Wall Street does. And this was on Wall Street, by the way — this was just me and a couple of my coworkers using our own money. We thought the product would not succeed, meaning the people using it would not live longer than the people not using it. And survival here is a couple of months — if you're in late-stage liver failure, you're looking at mortality very quickly.

It did not work. The product did not work. The stock went down a lot. And this was kind of a kiss of death for me in a lot of ways, because there were a couple of friends we had who were also shorting the stock. Again, we thought the company was a very bad actor. We thought they were lying a lot, manipulating people, really glossing over the fact that their product proved not to work several times. We thought they were not good corporate actors at all. And so when the trial failed — or when I put out my paper — the stock went down a lot.

And I did this silly photo. This takes me back to — do you remember this doctors-in-bikinis controversy that happened recently?

I took this silly photo where I put these sunglasses on, a gold necklace, a watch. And I had a music video in the background of Flo Rida — a rapper who has a song called "It's Going Down" or something like that. And I did this very tongue-in-cheek photo, which looked like a frat bro would take. I made that my profile.

And then when the Daraprim controversy happened, everyone saw my photo and said, this guy is a jerk, look at this fool. The whole point was I was trying to dress up as a fool, because it was an inside joke we had over the stock. That was the dawning of the term "pharma bro." That disparaging term started with technology — there were these tech bros who were almost fraternity-brother-style wannabe computer scientists, wannabe programmers, but they don't really know much. They wear the same vest, the same clothes, and they're just these tech bros in Silicon Valley. So somebody saw that and said, oh, this is pharma bro — this is exactly that, in pharmaceuticals.

Sure. And it sounds like this will probably be the first time there's a sober, rational discussion about it — and maybe that can only take place seven years later.

So I'll zoom out even a little more and say there are probably at least two more pricing theories out there. One is unbridled market price — supply and demand, and there's an unseen hand that meets that supply and demand at equilibrium. That's happened in pharma many times, where a company doesn't have the foresight to price their product correctly, and they either price it way too low or way too high. That happens all the time. And what a lot of people don't understand is there is a standard of care before a new drug comes out. That standard of care can sometimes be as good, or is rarely better — but the distance between the new drug and the old standard is often not that big.

So if a company comes out with a drug that's, say, 20% better than the old standard of care, but they price it at some price the insurance companies don't feel is warranted, it'll turn out nobody will use the drug. There isn't this cry for rationality or pleading for the price to drop — you'll often see the drug company just mess up. This happened a little bit with Aduhelm when it came out for Alzheimer's. Insurance said, we're not going to pay for that, it's not worth it — why would we pay $50,000 a year for a drug that has almost imperceptible benefit? It just doesn't make sense. And yeah, it stinks for the patient with Alzheimer's who wants that imperceptible benefit — I'll take any benefit I can get. But there's also this very cold-hearted rationale of: well, if all we're doing is going back to the way it was last year — there were a million patients with Alzheimer's last year who didn't get out of the home, and if there's a million patients this year who don't get out of the home, it's really no different from last year.

Again, that's too cold-hearted for me. I'd love people to get even imperceptible benefits. But there has to be some value. For $50,000, it just doesn't make sense for the healthcare system.

So there is that kind of — you pick a price, and ultimately the market does have to decide. There isn't this misunderstanding that pharma can hold insurance hostage. Insurance can say no. Insurance can say your copay is 50% — good luck. In that case, you destroy demand.

And then price happens the other way too. Pharma generally will slightly underprice their product and do price increases every year to combat inflation — but the price increases are usually much, much higher than inflation. I've seen some drug companies bump their price 10% or 20% a year. We went to Pfizer to buy one drug, and we were trying to negotiate it. They probably knew why we were going in there to buy it, so they told us they weren't going to sell us the product — and they went ahead and raised the price 50% themselves. A lot of drug company CEOs realized: when Martin comes calling, it probably means your drug is really underpriced, and you can make a bunch of money just holding onto it and bumping the price yourself. That happened about 15 or 20 times. So I do think there is a market. A lot of people feel that market's distorted by insurance just accepting anything, which is sometimes true too. But there are cases where insurance will say no. So there is a somewhat functioning market, sometimes.

Another theory of pricing, one that's really important and that pharma uses quite a lot, is similarity pricing. They'll look at other drugs and say, okay, that drug's worth this — so why isn't my drug worth that? That can be a very slippery slope, where we've seen some of the targeted cancer therapies go up in price to astronomical levels. Again, I support them in a lot of ways, but I do raise my eyebrow where I say: well, if your whole theory is "that drug's $20,000, so our drug should be at least $20,000" — and we did that with Daraprim, by the way. We looked at the hep C drugs — Sovaldi, Harvoni, all these hep C cures. We said, hepatitis C isn't nearly as bad as toxoplasmosis. Toxoplasmosis can end your life immediately. Hepatitis C often has no sequelae — you can live with that disease for 20, 25 years and get a little liver cirrhosis. Of course there are people who need their liver replaced, but the median patient doesn't have cirrhosis. So it's not a disease you want, but it's not toxoplasmosis. And we looked at that price and said, they're $80,000. We save your life — this is at least $80,000.

So the price of Daraprim literally is 750 — like, 100 pills you need to survive toxoplasmosis. That's literally the price of 401. That's how we came up with the price. We didn't do a whole lot more analysis. We said, there's no way we can be accused of having a bad price when Sovaldi's the same price. And Sovaldi had just come out, so a lot of people had sticker shock over Harvoni — but Harvoni sold, it was one of the best-selling medicines ever. So clearly that sticker shock didn't stop people from buying and prescribing it.

A lot of people don't understand pharma pricing in general, and I'll get to that in a minute — but I want to answer your question. Value-based pricing works really well because there's this "but-for" mechanism: but for your product, the cost would be this for treating this patient. So value-based pricing works. It's got flaws, no doubt. It's blind to how much money went into the drug. It's blind to any intellectual property. It's blind to when the drug got made. It's blind to so many things. But in a lot of ways I think it's the fairest.

I think the cost-plus theory is not a bad one either, and the sum of the two probably works best. The problem with cost-plus, as you know, is: it may have cost my company $100 million to develop this medicine, and maybe after I make $200 million I should drop the price to a very low price. But the problem is multifold. While I may have made that drug for $100 million, there might have been 50 other companies that failed. So in sum, there has to be a net return from this industry, or the industry won't be funded. Say there were 10 companies all chasing that drug, all putting $100 million in — that's $1 billion. And if that drug made $200 million back, well, it wouldn't cover the billion dollars that was spent. Sadly, when a drug is priced, it has to pay for not only that drug but all the other drugs — because the investors who invest in this industry invest in all the companies. There is this question of: well, if biopharma doesn't make a profit as a whole, then why invest in biopharma? Why not just let this industry die?

And then we'd have a big problem, because we wouldn't have the new cystic fibrosis cure from Vertex Pharmaceuticals, which is a wonder drug, a miracle drug that nobody ever thought could be made. But for their big gamble, their big bet on cystic fibrosis and this very weird small-molecule technology they had, they were able to break the barrier. I had a friend die of cystic fibrosis. It's a miracle medicine. They deserve a huge reward, in my opinion. In fact, I think they deserve a parade in Boston every year, for doing something more important than winning the NBA Finals and more important than winning the Super Bowl. Every person in America with cystic fibrosis — 90% of the genotypes are now cured. That's a miracle. And I think that drug company should make as much money as possible, and I'm very unapologetic about that. When you do something like that, you are doing — and I'm an atheist, but — you're doing God's work.

I look at media companies and say, okay, they make billions of dollars. I look at food companies, tobacco, gambling — there's all these industries where there's plenty of wrongdoing to go around. And you cannot tell me that making a cure for a dying person is anything but the greatest thing ever. I can't think of a company in the entire S&P 500 that is more virtuous — no matter how much they'll say they're virtuous — than Vertex Pharmaceuticals. And all they are is a drug company.

That's where the disconnect happens for me. And I'm deeply contrite, maybe even ashamed, that there's this black smear on pharma that maybe I contributed to. I wanted to remove that and show the glory of pharma — that it can change lives and save lives. You can argue about economic theories of value, about societal questions of how we pay for medicine. But pharma is a small part of healthcare budgets, which is something most people don't understand. When I went through the Daraprim fiasco, I talked to thousands of Americans using internet tools, and I asked them: what percentage of our healthcare cost is pharma? I'd get answers like 90%, 50%. And I said, what if I told you it was 12%? They were very surprised.

Then I'd ask: how long do you think a company like Pfizer gets to sell Viagra for? And they'd say, 20 years, forever. The real answer is like 11 years. They have 11 years to make money and then it's over — everyone gets Viagra for free after that. Obviously Viagra isn't the world's most important medicine, but pick any Pfizer medicine — you don't get forever to sell these things. The generics come in quickly, and once the generics are in, the prices drop dramatically. 90% of the drugs we use are generic drugs. Eventually the Vertex products will go generic too.

I've learned over the years that healthcare occupies a bit of a different place in people's minds than other products. It's a product category where you have to be very wise about how you establish your right to make money. Apple makes more money than any company in the world, and the question of whether they're quote-unquote earning their earnings is, to me, very different. I'm sitting here thinking: the company that's making the products that save our lives, they've deserved it, they earned it. If there's anyone we should permit to make lots of money, it's that, because we care so much about health. And of course people have the opposite perspective: our health is so important, we shouldn't let people profit off it. That's very counter to the way the world works in terms of needing to pay people.

You asked, is healthcare a public good? Now, there's a big difference between healthcare and pharmaceuticals. Pharmaceuticals are the product of a person or a team's invention and risk-taking. When somebody is assigned a task in a pharmaceutical company to, say, work on muscular dystrophy, you're asking a team to take on an impossible task. For that team not to be rewarded, it doesn't make sense — they won't do the job.

First, they need to be paid a salary — that's obvious. They need to be paid for laboratory equipment, materials, supplies, other resources like software and computers to run their experiments. And if they're lucky enough to have a lead, they have to manufacture that lead, which costs millions of dollars. So if you ask for it to be a public good — who's going to pay for it? The company is paying for it because they're hoping to make a profit. If the company didn't pay for it, it wouldn't happen. Universities don't have the capital to make a drug FDA-approved.

This is another thing that, after talking to hundreds or thousands of Americans, there's this big misunderstanding of: why do there need to be drug companies? And I said, there's no university that's ever gotten a drug FDA-approved. Not one. There's no non-drug-company entity that has gotten a drug FDA-approved. It's very laborious, very expensive, requires a set of expertise a university doesn't have. Universities have discovered drugs, but they haven't gotten them to FDA approval — ever, or to any other country's regulatory approval. So taking somebody's invention and saying "that's mine, that's for the public now" — it's not something that's ever happened.

A lot of people look at Jonas Salk, and this is maybe one of the worst mistold stories in history. Jonas Salk didn't quote-unquote "not patent" the polio invention because of altruism. First, he didn't believe you could patent an antigen. He said, it's as obvious as the sun — how do you patent the sun? For a vaccine, once you know the methodology Pasteur came out with — inoculate the person with the antigen — have you invented anything? Would it surprise you that giving part of the COVID protein to a patient results in COVID immunity? We've known that for hundreds of years. Now, technically there are details you can get a patent for, the specific idea. But generally, to get a patent you have to have some unanticipated invention, and Salk was talking about the fact that this wasn't unanticipated.

The other problem with Salk is what people don't realize: drug companies made polio vaccines and made a fortune. The only people who made polio vaccines were not Jonas Salk — Eli Lilly made Jonas Salk's polio vaccine, as did other companies, and they charged a fortune and made a fortune. So the only actor willing to actually make the product is doing it for a profit incentive.

Now, could that be different? Sure, why not? Why doesn't Bill Gates have a drug company that makes drugs for free for everyone? That would be cool. Or at cost. It would be a huge humanitarian endeavor, fantastic. If he doesn't do it, maybe our government should do it. I have no problem with that. Give me the rules and I'll play the game.

I think some people look at healthcare as a game. But from my perspective, this is a wonderful thing — when a company wants to make a discovery, wants to make a cure, and deserves a reward for it. They deserve a hell of a lot more reward, to me, than CBS, who's going to put on Keeping Up with the Kardashians. Why do they deserve billions of dollars of profit, but the company that makes a new treatment doesn't? And I'd extend it to physicians. If we want to have this fight, I'd say: why would I go to the doctor? Should the doctor charge me or my insurance? Why can't I have that for free? And the simple answer is: the doctor is a smart person, they may decide to have a different kind of career if that's the way it works.

It's a great question. I know this space better than anyone. When I was on Wall Street, people considered me an authority on pharma — the New York Times said I was an encyclopedia of pharmaceutical knowledge. So I'm trying to absorb your question in all the contexts necessary.

I make fun of a lot of pharma for not being good at R&D. Their R&D spend and their R&D productivity are two different things. I remember talking to the CEO of Genentech years ago, who was on the board of Apple, and I asked him how much a drug company should spend on R&D. He had this very tongue-in-cheek answer: if it's Genentech, it's as much as possible. Because when Genentech spends a dollar on R&D, it gets back like $1.25. So for us, if we can find projects that give us $1.25 back, it doesn't make sense not to do any project. The problem with pharma is that when they do an R&D project, they usually spend a dollar to get back 95 cents. And that makes it difficult for them to want to do more R&D.

Pharma has increasingly externalized its R&D, and to me that's really sad and frustrating. It's hard to run a big company. When you have 50,000 people — I've never had the pleasure — it's not easy to have an R&D division of 20,000 people really be effective. They'll make some drugs sometimes, but I can't really point to others besides, say, Johnson & Johnson, maybe one or two other companies, that have truly effective R&D platforms that continuously pump out great new medicine.

Usually these drug companies go through a dearth. Pfizer went through a period where I don't think it made one new drug for five or six years. It struck out, struck out, struck out — not for lack of trying. Really smart people work there, brilliant folks. They spent billions of dollars and couldn't come up with anything. What they end up doing is they have this vision of internal and external R&D. Internal is the stuff they do in-house. External is basically the companies they buy out. And they do two other things with their capital: buybacks and dividends. I personally don't think they should do either until every disease has been fixed. I don't think that's a great use of capital.

I think the dirty secret in pharma is that we're actually getting very close to running out of new medicines to make. I know that sounds strange. I wrote an essay — I think I called it "the end of pharma" or something like that. I've been talking about this for a little while. The '90s boom of pharma, where we had the Lipitors of the world — a lot of those drugs are now generic. A lot of them have established a baseline, almost like a WHO essential-medicines list of now-permanently-free generic inventions for mankind. The drugs we're inventing now are for large rare diseases like cystic fibrosis, a couple of cancers. And what do we do next? The mortality has dropped for so many of these illnesses that it's not really clear where pharma can go next. Alzheimer's is probably the last big, big, big target for pharma. And after that, I'm not sure what's next.

There's a fairly high life expectancy now in the US and other industrialized countries where, after cancer — which is sort of halfway there with drugs like Keytruda, and drugs like Tagrisso for EGFR mutations — we're getting those pretty good. Those drugs will go generic soon. If a company figures out Alzheimer's, you have a couple of spots left — Parkinson's, certainly type 1 diabetes. There's a handful of places where there still needs to be innovation. But there's this problem people are calling target saturation. ALK-positive non-small-cell lung cancer — not many people have that disease, and there are about eight drugs for it. Is that really what we need? The drug companies are starting to eat themselves. There's still room — GLP-1s, where pharma's finding a way to make money — and I'm not sure they'll lose all their ways to make money, but there is that risk.

So R&D in pharma is notoriously ineffective. They take the money they make and buy out a drug company. Pfizer just bought Seagen — Seattle Genetics — a really innovative, amazing company, specialists in antibody-drug conjugates. Very tough technology, to make an antibody and link it to a drug and make it all work together. Pfizer bought them, but they spent $40 billion. That's $40 billion they could have spent on R&D. If they were good at R&D, they'd do that. Instead they said, well, these guys are good at R&D, let's buy them. The problem is, when you do that acquisition you may get some talent, but a lot of those people will leave — their stock options cashed in. They love the hunt of biotech and the excitement of it. They don't want to work at Pfizer. Half of those people are going to go to another biotech or start their own. So that doesn't work either. The pharma companies have this big block: how do you get the innovative streak?

When I talked to Art Levinson, he mentioned — well, there's Apple, Microsoft, Hewlett Packard, Dell at the time, and IBM. Those are five companies. He said, one of those companies spends very little on R&D but has the most productive inventions in the industry — and that's Apple. Microsoft spends a fortune on R&D. All these other companies, they just can't get it right. So you have to do R&D the right way, and getting that magic is really difficult.

One of the things I introduced to try to fix it, which hasn't taken hold in the drug industry: I wanted the chemists who were inventing the drugs to have a cut — basically a 1% royalty check, almost like a music artist, on the album they made, the molecules they discovered. I thought that was only fair. Why shouldn't the person who made Kalydeco or Orkambi or one of the Vertex drugs have 1%? If it ever became a blockbuster, that person would be really rich. And despite what we all want to believe about human interests, there is still a bit of financial incentive for people, especially scientists — they could go work and do something else. Scientists tend to find lots of things fascinating, but if a scientist gets a job offer for double the price, they often think about their family and say, maybe this is a good idea. So a percentage royalty is something we did for a lot of our chemists at my first drug company.

No other drug company did that, because there's this pecking order — the CEO should be the most highly paid person in the company. And I'm sitting there saying: hold on, the CEO of Merck doesn't do anything. The CEO of Merck shakes hands and kisses babies. The person sitting there trying to figure out how to make the next cure for Alzheimer's — he's doing something important, she's doing something important. For that team to get an incentive boost of "if you do your job just right and you're lucky, you guys might be billionaires" — that's really cool to me. That's the team that should make the money. It shouldn't be the CEO, it shouldn't be the board of directors — those guys I basically think don't do anything important.

Yeah. In technology, Moore's law has worked not just for processing power, but for sending a megabyte of data from me to you. It's worked for storage, where storing a gigabyte has gone from thousands of dollars to almost nothing. So why isn't healthcare getting this? Why are healthcare costs going up nonstop? Why is the number of drugs flat? The number of new FDA-approved drugs has been around 30 to 40 per year since the start of the FDA. So the amount of medicine we have access to every year has not grown one bit — even though our population has grown quite a lot, the global population has gone up enormously, and the cost of making a medicine has gone up enormously.

So what the heck is going on? Why isn't the great advantage of technology making healthcare subject to Moore's law phenomena? Health is one of these goods for which we have infinite marginal utility and marginal demand. There's nobody watching or listening to this who doesn't want a little bit more health. Nobody would say, I don't need that much more health, thanks, I'm good. But if I said, would you want an even nicer T-shirt, a lot of people would say, my T-shirt's good enough. At some point the clothing you wear, the food you buy, the home you live in — do you want one extra bedroom? Maybe if you're Bill Gates. But eventually — or Mark Cuban, let me pick on him, or Donald Trump — eventually you don't need that extra bedroom. But healthcare is something we can't get enough of.

I think there are two reasons we're not getting Moore's law effects in healthcare. First, we're not applying technology effectively enough, or that technology wasn't available. Second is our insatiable demand. Our insatiable demand results in a lot of bad things around the margin. One really bad thing is healthcare labor. Because of our huge demand for healthcare, at least in America — it's an innate human interest, it is everywhere, it's just more suppressed in other places — we have a very finite labor pool for healthcare, and it's a very distorted labor pool. That's not to sound conspiratorial, but I think there's a vested interest, an element within healthcare that limits the amount of graduates in the physician labor pool.

I've graphed over time how many physicians per million people there are in America, and it doesn't vary very much. How many physicians per million people above the age of 50? That has actually dropped — which is interesting, because those are the people who consume healthcare the most. So you have this artificially limited supply pool of labor. We've tried to fix that with NPs and PAs to bolster the physician workforce. But on the doctor side — let's get back to drug discovery in a minute — I think we have a terrible labor-supply problem, because we won't take doctors from other countries, we won't do telehealth from other countries. We have an AMA that's very strict about its quality, ostensibly. But I think there's financial interest in that too. When you limit supply of something — I know enough about economics to tell you the price goes up. There's a benefit to that for physicians in America.

I know a lot of physicians will roll their eyes and say, I'm not making as much money as you think. But there is that vested interest: if we flooded the country with physicians — whether by easing medical-school requirements or allowing foreign physicians to practice more easily — the fear from the physician lobby is that quality would drop and we'd all be sorry for it. But the counter is that the price stays very high. And again, only 12% of healthcare expenses are pharma. Where is the other 88% coming from? Very few people want to ask that question when you have a boogeyman you can hit with a stick named Martin Shkreli or anybody else. The other 88% is as much to blame, if not eight times more to blame. In fact, I don't think pharma is to blame at all.

Now, in terms of drug discovery itself — it should be easier to discover a drug, right? I think that's easily dispelled: low-hanging fruit is easy to pick. As the fruit goes higher, we need — sometimes logarithmically higher. It's not a linear tree, it's more of a log-scale idea, where the next piece of fruit is 10 times harder to get. You look at Vertex — okay, you're telling me this 3,000-amino-acid protein is messed up in cystic fibrosis, how the heck am I going to fix that? It's like landing a person on the moon in a soda can. I don't know how you do that. It seems impossible. I could do it with a space shuttle, but with a soda can, that seems crazy. Vertex found a way, and it took deep breakthroughs in understanding to get there. Those deep understandings don't come easy — they come with a lot of pain.

But — and you're a physician, and many of your listeners are, and I want to be careful I don't sound anti-physician — the biggest cost of clinical trials is the clinical-trial administration period. The biggest cost of pharma R&D is the clinical-trial process. The actual discovery people, the people who do the lab work and make the molecule, are a large cost, not a small cost — but they're about 20% of the process. The other 80% is everything that comes after. My hope is we can get that cost down over time.

I can't blame the physician for saying: if you want me to not do my day job — to see a patient — and instead fill out all these forms and do this clinical trial, that's really hard to do. These are not easy. You've got to see the patient way more than normal, take care of all these measurements and blood samples. If you want me to do this clinical trial, I need to get paid a lot more than I normally would. I'd rather just go see a patient. Unfortunately, I know physicians who, the first thing they ask when presented with a clinical trial is two words: "how much?" Because why would I mess up my practice? I'm making this much money a month — you want me to stop and help your science experiment? I'd rather keep doing what I'm doing. If I'm going to study this new HIV drug or hep C drug, I need to know how much I'm getting paid to sign up off the bat, and how much I get paid for every patient that comes in.

That's reasonable. I really don't think that's a problem. But when I sign the checks to pay for clinical trials, that's where most of the money is going, and there has to be a better way. Some centers do clinical trials based on merit — is this drug really exciting, etc. But typically it's about money. The more money you pay to enroll a patient, the faster your trial gets enrolled, and the faster everything else in this big cog and wheel is happy. So if we could come up with a better clinical-trial infrastructure — some people are talking about decentralized clinical trials.

The FDA is very specific about how they want these things done, and they don't compromise. I love the FDA. I'm kind of a libertarian, but in terms of government structures I don't think should be burned down, the FDA deserves rewards and awards for the great job they do. They're really people who love medicine and love pharmaceuticals and love studying them. They've called almost every drug that's come in front of them right, from what I can tell. They're really benevolent people. They're not deep-state people — they're people who love medicine. When a new cancer drug comes out that's saving lives, they'll approve it in three months. When a new diabetes drug comes out that's the 43rd type 2 drug, they'll take 18 months, because that's what they should do.

Yeah. So drgupta.ai — that's our website, I have to plug it for a second, I hope you don't mind. It's the start of what I think is going to be very important. This is my full-time job. I have a software company focused on this. Dr. Gupta is our main product.

I think that in 50 years — and take that number with a large standard deviation — I don't think we'll have physicians anymore. I think our children will ask us what it was like to talk to a doctor, and they'll be curious, no different from the way we might ask our grandparents what it was like to be in a horse and carriage. The answer comes from computing technology — it's not just AI, it's this explosion of information. If you're a physician who is, let's say, a neurologist — can you really honestly keep up with every paper and every academic meeting in neurology? The best doctors I know truthfully say they can't. It's very hard to treat patients and also become an expert in everything in neurology. For my PKAN disease, most neurologists don't know a lot about it, and I don't blame them — it's extremely rare. There are hundreds of neurological diseases. Can you really master all of them? No. Should you? Probably not.

The physician's job has become impossible due to this explosion. There are now 38 million articles in PubMed, a couple million added every year. It used to be a couple hundred thousand a year. The scientific and medical knowledge has exploded so much that being a doc is really awfully difficult. Can you remember everything from med school? Are there going to be days where you have bad days? Times when you're burnt out and want to quit? Times where you wish you could see ten patients today but you have to see a hundred or two hundred? Times where you wish you could spend an hour more with your patients, but you also have to spend an hour with your family? It's really hard. You're a human. It's necessarily limiting.

If there were a machine that could do it — which up until now I would say there's no chance — a doctor's job is very, very difficult, it takes training for a reason. And if there's even a part of that job that could be done by a machine, I'd be impressed. Well, I'm impressed. A group from OpenAI and Microsoft put out a paper in the New England Journal — which I never thought those words would come out of my mouth, that Microsoft would be publishing in the New England Journal. But they did, and everyone's been talking about it. I know physicians who use OpenAI's ChatGPT every day instead of Google. If you haven't spent time on ChatGPT — forget my website, just go to ChatGPT and ask medical questions. You're in for a treat. These products are scoring equivalent to good doctors on USMLE exams. Well, the USMLE exam is not all you need to be a good doctor, but I do think some percentage of the physician's job can now be automated. And that's a little scary to some people. You might call them Luddites or anti-progressives. But we have to accept technology and use it to improve the world.

I got an email from an Iranian group yesterday who said Iran has rural areas that are very poor, and we'd like to take Dr. Gupta and apply it to huge parts of Iran that don't have medical expertise. These people have almost no options for healthcare. And we think your product would be incredible for a big chunk of Iran. Some of the people who've come to me — we've gotten overwhelmingly positive responses, including from physicians, 80, 90% of the physicians we've talked to think this is wonderful. The 10% or so probably don't realize there are lots of people in the world who don't have access to healthcare — including in America, Canada, and the UK. There are poor parts of those countries too, places where there aren't good doctors. If you want the very best specialist in Lennox-Gastaut syndrome, he's probably not on your block, not in your metropolitan area — she's probably somewhere else. Charlotte Dravet discovered Dravet syndrome. She doesn't live here, I can't talk to her. But if I can talk to a machine that's embodied most of the knowledge she's written, maybe that's good enough. Maybe that's a good start — at least informative for a physician or a healthcare worker or a patient to read about and move forward.

These trillions of dollars we spend on healthcare — something has to give. America's going bankrupt, which our politicians seem to ignore. We spend trillions of dollars more than we take in for tax revenue, and we spend it on healthcare. And I think we should spend it on healthcare — in fact, I think we're going to spend more, because we demand that. But who's going to pay? A lot of that money is being spent on wasteful physician interactions that are not particularly pleasant. When you go get a doctor's appointment here, you're sitting in a doctor's office for an hour, then you're shuttled around, you talk to NPs and PAs for 10 minutes, and you see the doctor for 90 seconds. The doctor is trying to make it 85 seconds — how can I shave off any time with this person?

So I think about the person in the Bronx whose first language is Spanish. Finding a patient doctor who will answer all your questions in your native language — I'm not surprised that minorities utilize healthcare less than English-first speakers and people making lots of money. I've seen it in my own family. So making a tool that equalizes access to healthcare information — of course not treatment. Dr. Gupta can't treat you, can't write a prescription, can't perform surgery. But he is a healthcare simulation of a doctor. When you enter Dr. Gupta, we actually don't let you use the service until you acknowledge you understand it's not a real physician. A lot of people have that mixed up — well, not a lot; we're on pace to have hundreds of thousands of users this month, all satisfied. Nobody said Dr. Gupta got it wrong. Nobody said Dr. Gupta told me to take an antibiotic with high blood pressure or something.

I dispute that very quickly.

It's not a medical device, that's for sure. And the FDA said as much. If it were a medical device — the FDA commissioner didn't tell us directly, but I'm reading the FDA commissioner's public statements on OpenAI, for instance. We're a small, tiny piece of this. We're a startup with five people. We're not OpenAI. You go to ChatGPT, it'll give you medical advice. They take all the same disclaimers we do. But they have 100 million people using their product; we have 50,000. I'm sure we'll grow to millions. But the point is we live in a new world where regulation has to get around and think about AI in general, and it has to do that with a careful hand.

If you read the FDA commissioner's comments, he was very sharp about how to do this. You don't want to take a sledgehammer to this promising new technology — that would be a big mistake. You'd allow economic growth in every other country other than the US. And trust me, it would take me 30 seconds to ban the United States from Dr. Gupta. I wouldn't like it, but I'd press that button if that was what was asked. And every other country in the world would have access to Dr. Gupta except America. Does that make a lot of sense? I don't think so. If Germany or the UK said, this is a medical device, you've got to register — okay, no problem, click. And I'd also say, by the way, make OpenAI do that too, because they're basically the same product. We just take their product and wrap it in a different bow. It's a better bow, and I could explain the technology of why we give better answers than OpenAI's product — which I love, they're great, they're our partner. But we tweak the technology quite a bit to make it better.

Yeah, I'm so happy you mentioned that. Hallucinations have dropped with GPT-4. GPT-4 is very expensive and very slow, so we still use it, but hopefully technology will improve. Hallucinations are down like 95% with GPT-4. We think we reduce hallucinations 99.9%. We use a tool called introspection — an approach I stumbled upon very early on in my own playing with this stuff. I've been playing with GPT since GPT-2. I created a library called BookGen, which tries to write a novel from scratch using GPT. It does a terrible, terrible job until I stumbled upon this idea of introspection: I want you to read what you just wrote and tell me how you would fix it. What's remarkable is that the book quality jumped up dramatically. It's not Dostoevsky just yet, but it was able to take its former writing and say, hold on a second, I would have adjusted this and this. And I said, do it again — and slowly you got this remarkable self-improvement. On all kinds of tests, like standardized tests, this reflection approach — introspection and reflection, some people call it actor-critic or critic-actor — seems to improve the results dramatically.

Dr. Gupta is a bit slower than ChatGPT for this reason. Not a lot slower — like another 5 or 10 seconds. But in the background we're firing off like a dozen different screening and analysis requests of what Gupta wants to say, making sure it makes sense, making sure it's not hallucinating, making sure it's solid through a couple of different lenses. It's not going to ever be perfect, and that's why our disclaimers exist. We're not advertising that this can diagnose you or treat you, because it can't. We're advertising that this is a healthcare information tool. It's like Google, but a lot better presentation. And Googling is not a medical device. Going to WebMD is not a medical device — if you read an article on WebMD, they don't have to register that article at the FDA. GPT is a search engine, that's really all it is.

In terms of sources, LLMs are completely inscrutable. They have these trillions of weights, and the calculations done to go from your query to an answer are inscrutable even to the companies that make them. So what we've tried to do — and we'll be rolling this out on Monday or Tuesday — is the first-ever product where if you talk to Dr. Gupta and say, I have a stomachache, it feels in the lower part of my abdomen, and you keep conversing, he'll usually ask you a bunch of questions — have you been eating something different lately, are you traveling — and ChatGPT doesn't do any of that. ChatGPT just spits out an answer. Whereas Gupta wants to engage with you and take a history. GPT just says, there's a lot of causes for stomachache, here's six of them. It's a very smart answer, a good answer — but it doesn't have any citations or sources.

So our system, at least in a couple of days — you might have seen a figure lurking around here; we cooked up this demo over the last 48 hours, we've been in this room instead of partying in New York City, Saturday and Sunday working our butts off — I woke up to this guy who hasn't been sleeping. We're programming the system where you're going to get citations, and all the citations will be in a box that Gupta provides. When you see a doctor, we all know the doctor went to medical school, we all know the doctor's responsible — but you also don't get a stack of references. When he says, I'm going to give you this drug called sumatriptan, he doesn't say, here's 20 papers on sumatriptan. Dr. Gupta will do that, starting in a couple of days, where there'll be that citation list.

How do we get to the point where we can confidently say Dr. Gupta has actually read, digested, and crystallized every paper? I'm working on that too. It's a different problem. What we found is that these LLMs seem to contain all this knowledge anyway. I was very distraught, because I wanted a physician to truly understand this. Broca's area in our brain is the area important for speech processing — when you have a stroke injury or a physical injury to Broca's area, you lose the ability to speak. LLMs to me are like Broca's area. But we have the rest of the brain other than Broca's area for a reason. There are other parts of the brain that do real thinking, real agency, real planning — the whole prefrontal cortex. So in AI, we need a prefrontal cortex before these tools can really be as dramatic as we'd hope. And I'm saying 50 years from now there'll be no physicians — we're not even close to that yet.

So I'm trying to figure out — I joked, our codename for it is corpus callosum. We traverse this idea that part of these machines should have structured, explainable, traditional database data, and the other part is this LLM that's a black box. Do we want to make this connection back and forth? We've done a lot of this type of programming before with LLMs. The hard part is that the LLM seems to be as good or better without it. And coaxing it to explain its answer is something GPT is making a lot of progress on. So we're just trying to get that sense. If you ask Dr. Gupta about Mounjaro, for instance, it doesn't know that drug because it just came out. That's a big gap — we want people to get the most up-to-date information. So how do we implement that is another big question. There's a lot of architectural issues we have to improve upon.

Sure. On the first question — I grew up with, I had my first symptoms of panic disorder in my early teens. I had other issues like depression. I had this anxiety-panic-attack disorder that I started getting treatment for when I was about 17 or 18. I started using venlafaxine. And part of my love for medicine and pharma was that that product saved my life. And that doctor saved my life. I still owe everything to Dr. Sahlein here in New York and the people who made venlafaxine. Because I used to have two or three panic attacks a day, and I used to sit in the New York subway station. The room would be shaking, my heart beating out of its chest. I'd sit down in the subway station, and people would look at me like: is this guy homeless? Is he about to detonate a bomb? What is this guy doing? I really couldn't move. It was so scary. And I took this drug and it all went away. I know that's a lucky case, a really lucky win. But that miracle for me was like — wow, I love this doctor and I love this medicine. Some people have that experience with medicine; some people have the opposite, where they take a drug and feel it really hurt them.

As this strange life I've had has continued — it hasn't been strange just because of Daraprim. I had a very tumultuous high school experience where I was both dropping out and far behind, but I also finished school early and went to college early and then worked at the hedge fund early. I was working in this big-time Wall Street thing at a very young age. So there was always this weird life that wasn't exactly normal for me. Starting my own fund, starting my own drug company, taking that public at 28 or 29, being a millionaire very young — all the tumult that comes with that. And certainly getting indicted, going through the Daraprim experience, wasn't pleasant. All of that is a combination that's an interesting tapestry of life.

The mental health part — in general, it can be very depressing, very frustrating. There's all kinds of emotions. I'm very lucky that I have some resiliency. I'm not neurotic — there are people that, if they ran out of mayonnaise, it's the worst day of their life. I'm not that kind of person. I'm very resilient, almost stoic. I think that's benefited me in large part. I've had a huge fan base, as you may be aware. Some of the biggest people in Silicon Valley, in politics and culture, are friends of mine at this point. And untold people — I started YouTubing and podcasting, and people said, gee, I actually kind of like this guy, and he's not what the media made him out to be. Some people see it and say, I still don't like this guy. And that's okay. You can't make everyone happy. I know the choices I've made in my life are, I think, the right ones. I haven't made every choice perfectly, like anybody else. But when you have that kind of inner peace, I think it helps a lot.

In terms of wealth — I'm going through some interesting litigation that's subject to appeal and all kinds of stuff, guarantees and indemnities. So I'm not too worried about the recent litigation. I'm still a very wealthy person. But ultimately — if I could decide — math is one of the things I love. If I could prove a math theorem that would have my name on it for the rest of time, or have all the money I've ever made and all the money I'll ever make taken away, I'd take the math theorem without a second of thought. Because being rich is not what it's cracked up to be. Warren Buffett says it all the time — I sleep in the same bed as you do, I eat the same hamburger you do. There are very few things more comfortable for me than for you, just because — and he's literally the world's richest man. The time you spend with your family, the scientific challenges you work on — those are fun for me. When I'm programming the computer, I'm having fun.

There was a big party in New York the other night that I missed, that everyone was trying to drag me out to, and I had to work on Dr. Gupta — and that was 10 times more fun for me than going out to this party that was supposed to be great. That's not what I do. Put me anywhere on God's green earth with a computer and internet connection, and I'm going to have a great time. Sure, I'll make ends meet financially, and I'll find people like me who want to do stuff like that. The really wealthiest people are the people who have good friends, good family, good peace of mind. They're not chasing the next thing. When I was younger, 22 or 23, I wanted to be a millionaire, I wanted to be rich. And then when it actually happened, there isn't a whole lot to do other than join this class of people that are less exciting than you maybe thought they were. The people I really like to be around are scientists, programmers, physicians, people who straddle a couple of those things. Those are the most exciting people to me.

People won't remember who Steve Ballmer was 100 years from now. But people do remember some of the early physicians. People will remember Pasteur, or people like Jennifer Doudna — she'll be remembered probably 100 years from now. Whereas the average billionaire business person is probably not even remembered right now, 100 years from now. So to the extent that that's a metric of importance, I think it's really relevant. You may be able to have a big house or something. But — there's a plaque in the background there of a US patent I got. If you look around my place, it's pretty much minimalist. But that one little plaque, I'm really proud of making that invention of the phosphoramidate prodrug. Those are the kinds of things I live for, and sometimes they require money, sometimes they don't. Ultimately, whether I have 50 million, 100 million, 2 million, 0 million — to me it's — I also have a lot of faith. I think a lot of wealthy people don't have faith that they can make money, so they get very protective of it. For me, I have no doubt I can make a billion-dollar new business very quickly. Very quickly is not overnight, of course — it takes years. But I have a talent that can't be taken away from me, and a passion that can't be taken away from me.