Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, AI in Life Sciences

Apoorv Agrawal 0:09 So today we're going to have two incredible guests, uh, from both sides of the aisle. Uh, the aisles being, uh, Anthropic, former OpenAI, former Meta, and now Chai. Uh, flow of class today, I'll, I'll make a quick introduction, but I'll hand it over to the speakers who will present about their work. They're all doing, both doing very, very exciting work. Um, and then we'll have some time for Q&A towards the end. Um, so maybe without further ado, I'll quickly introduce our, our guests and have you guys over. Um, our first guest today is, uh, Eric Abrams. Um, Eric has the fun job of convincing all biology to run on Claude. Um, he runs biology and life sciences at Anthropic, uh, launched Claude for Life Sciences last October, um, and they've got some really good partnerships with BenchLink, 10X Genomics, Novo. Uh, and I was just told that the last time you were here in this course was your, uh, final EE PhD scarred memory.

Eric Kauderer-Abrams 1:07 [laughs] It was a, it was a device physics final.

Apoorv Agrawal 1:10 Yes.

Eric Kauderer-Abrams 1:10 I'm still traumatized from it.

Apoorv Agrawal 1:11 Device physics final. Okay, good. Good, good, good. Well, thanks for joining us. And then Josh, um, you've spent your whole career teaching co- computers to do biology. Um, we missed you here at Stanford, but it's good to have you back. You were early at OpenAI, like early, early, early, like almost a decade ago at OpenAI. And then at Meta, uh, built the first, uh, life sciences product, which is almost half of all citations today are, are citing your work with, um, ESM-1. You started Chai Discovery to make dis- drug discovery an engineering problem, uh, as you call it, the CAD suite for molecules, and you've raised from the who's who, from OpenAI, from, from Anthropic, from, from Thrive at, at, at a, at a big price, so we're very excited to get into it. Thank you so much for, for joining us. [audience applauding]

Speaker 1:56 Thank you.

Apoorv Agrawal 2:01 Um, we'll start with you, Josh. Um, and I got your slides if you wanted to use them.

Speaker 2:07 Thank you.

Apoorv Agrawal 2:07 Tell us, tell us about your journey. You know, you, you were not a, a pre-med major, you were not pursuing a medicine degree. How did, how did you land up, how did you take the scenic road to, to Chai?

Speaker 2:18 So it's actually not totally correct. I, I think I did all the pre-med classes.

Apoorv Agrawal 2:21 Mm.

Speaker 2:21 I thought I would be a doctor. I come from a family of doctors.

Apoorv Agrawal 2:24 Right.

Speaker 2:24 Uh, I have a bunch of older siblings who were already in medical school when I was a kid, so I wanted to do something different from them.

Apoorv Agrawal 2:29 [laughs]

Speaker 2:30 So I learned how to code.

Apoorv Agrawal 2:31 Nice.

Speaker 2:32 Um, and the, the thing that I loved about programming, uh, that, you know, I didn't really see you could do in, in medicine at the time, was you could really distribute what you were doing, right?

Apoorv Agrawal 2:40 Yeah.

Speaker 2:40 You could write a piece of code, and then it was, like, infinitely scalable, right? You can just, like, send it to many people. Um, and it was late in high school that I actually discovered biotech, and I realized-

Apoorv Agrawal 2:48 Mm

Speaker 2:49 ... that drugs actually had a similar property. It was another way to scale medicine-

Apoorv Agrawal 2:52 Mm-hmm

Speaker 2:52 ... and medical discoveries.

Apoorv Agrawal 2:53 Yeah.

Speaker 2:54 Um, so, uh, you know, I've always liked coding more than the lab, but they're-

Apoorv Agrawal 2:56 Yeah

Speaker 2:56 ... they're pretty linked, I think.

Apoorv Agrawal 2:58 Nice, nice. And tell us, tell us a little bit about, about your work at Chai. What do you guys do? And, you know, feel free to take a couple minutes to walk us through it.

Speaker 3:05 Yeah. So, so you put it beautifully, like, we're building a computer-aided design suite, uh, for molecules at Chai. If you look at what's gonna happen, um... See if these slides work. Yeah. What, so one of the things we're gearing up to, I'd say in the, let's call it the medium term, because the long term is, uh, so much of the lab work that we do today is probably going to move onto the computer, and it's, it's gonna move as a, you know, kind of a goal that we can illustrate here, is we wanna be able to design, uh, antibody molecules. These are about half of the approved drugs these days are, are called antibodies, and, uh, there's a big trial and error process when you make a drug. You find some initial hit, you try to make a bunch, bunch of changes to it to get all these properties that you need for it to be a drug, and the big hypothesis is that, uh, someday we'll be able to just zero shot these molecules that are ready for patients right out of the computer.

Apoorv Agrawal 3:51 Yeah.

Speaker 3:51 So we started the company with this big goal-

Apoorv Agrawal 3:53 Yeah

Speaker 3:53 ... uh, thinking that this was going to become possible, and that when it became possible, it would be a no-brainer for most people to design drugs, uh, this way. So when we started, uh, the company about, about two, two and a half years ago, uh, everyone told us we were crazy to do this. I mean, both because, like, hey, is that technology gonna work?

Apoorv Agrawal 4:09 [laughs]

Speaker 4:09 And then B, uh, don't you need to make your own drugs? Uh, the ambition of, of the most ambitious biotech companies is usually to become a pharma company.

Apoorv Agrawal 4:17 Mm.

Speaker 4:18 And then the ambition that we were taking was like, no, like, we want most drugs to be designed this way.

Apoorv Agrawal 4:22 Mm-hmm.

Speaker 4:22 And we're not gonna make most drugs ourselves.

Apoorv Agrawal 4:24 Mm-hmm.

Speaker 4:24 Uh, we really wanna partner deeply with the ecosystem.

Apoorv Agrawal 4:26 Mm-hmm.

Speaker 4:27 Um, so we had kind of two contrarian bets in one, so-

Apoorv Agrawal 4:29 Mm

Speaker 4:29 ... I'm grateful that we managed to get those good names on the cap table-

Apoorv Agrawal 4:32 Yeah, yeah

Speaker 4:32 ... like you mentioned. Uh, but, uh, I think that there's, there's been a, a ton of progress, uh, since then.

Apoorv Agrawal 4:37 Got it. So you would provide your, um, the CAD for molecules to a pharma business like AstraZeneca, like Pfizer, and so on.

Speaker 4:45 That's right, yeah.

Apoorv Agrawal 4:46 Got it.

Speaker 4:46 So some of the biggest pharma companies in, in the world are already using, uh, Chai's models in order to design their drugs.

Apoorv Agrawal 4:52 Got it. Fascinating. Thank you. Um, Eric, over to you, sir. Please tell us about your journey. You've started a couple companies, ran a bunch of companies, and now at Anthropic, what are you up to?

Eric Kauderer-Abrams 5:01 Yeah, sure. So, um, I, I originally, uh, was trained in math and physics, and then, uh, got into AI research. The thread there was I thought that understanding intelligence was the great theoretical problem of our time. Um, and then it was actually here in, in grad school, I was in Kwabena Boahen's lab, the brains and silicon lab, uh, where, uh, starting as this arrogant physicist, I got my first taste of real biology as we were studying the brain to try to find principles to bring back to AI. Um, and, uh, it was love at first sight. I think ever since then, uh, I've just gone deeper and deeper into the world of biology, um, through the companies that I, that I started, getting into, you know, molecular biology, biochemistry, organic chemistry. And, uh, a- after grad school, I, I had started several different med tech and biotech companies and have been kind of in this world ever since. And I think, um, when I was running the, the last company, it was a molecular diagnostics company called Detect, um, and I was checking back in on what was happening in the AI world and seeing, you know, the, the very beginning of ChatGPT, and, and I'll never forget when Sonnet 3.5 came out- There was this step change in capabilities where I was asking it to do things that were coming up in my daily life of running a biotech company, of debugging some, you know, experiments that were failing in the lab and responding to FDA feedback and all sorts of things. And I was just shocked that it was useful at all back then. Um, and it, it became very clear to me in that moment that, uh, this was a technology that we could use to accelerate the whole process end to end of doing R&D in the life sciences. Um, and I think Josh and I have talked about this, what, what we're doing respectively, and I'll explain more in a moment, is, is complementary in that, um, at Anthropic, what we wanna do is figure out end to end how to accelerate the whole process of doing R&D in the life sciences. So this includes basic research, drug development, and within drug development, you know, I think what, what Josh is doing is, is a big piece of the puzzle of actually designing the drug. Uh, but after that, there's a lot more. There's all of the clinical development, regulatory processes, the transfer to manufacturing, right? And our vision is to train Claude to be able to do it all, right? So at the, at the model training level, we have a, a full stack model training program to make Claude state-of-the-art and meet or exceed, you know, human expert performance in everything that you can imagine. The hardcore scientific fields that underpin this, right, in bioinformatics, in chemistry, in structural biology and so on, and, uh, the, you know, clinical and regulatory aspects of designing clinical trials and responding to regulatory feedback, even the strategic aspects of running a drug development program, so picking targets and modalities, organizing an R&D program, deciding when to advance or kill programs. So we're, we're training the model to do all of those things. And, um, beyond that, we believe it's not enough to have great frontier models. We also need to make sure that the model capabilities are accessible to people and integrated into workflows and connect to all the tools that you need.

Apoorv Agrawal 7:48 Mm-hmm.

Eric Kauderer-Abrams 7:49 And so we also focus a lot on the product layer and, uh, making sure that, that we have products that are optimized for life science professionals. And we have along those lines something like a Claude Code for Bio that will be coming out fairly soon, uh-

Apoorv Agrawal 8:02 Ooh

Eric Kauderer-Abrams 8:02 ... that is, is our take on, on, you know, taking the, the power of, uh, Claude Code on the back end, but putting an interface on top of it that's designed for working scientists, right? So that you can visualize proteins and small molecules and play around with these things and have Claude know what you're talking about and spin up lots of compute to run lots of models like Chai and, and so on. Um, so we focus a lot on the model layer and, uh, the, the product layer, and we consider those two things together to be the platform that we wanna take and scale throughout the whole life science world to, you know, achieve our goal. And, and our goals here are, are to get this, you know, end-to-end order of magnitude or more acceleration on everything that's happening throughout the life science world. Uh, and I think that, uh, you know, the life science world is, is huge. There, there's so many different parts here. I think the, the first two goals that we're focused on are accelerating basic research and accelerating therapeutics development. Um, and, you know, in, in each of these things now, we have more specific initiatives that we're starting. [lips smack] So, for example, uh, we, we recently started a wet lab, uh, where, you know, we're taking our, our model and our product, and we're putting it to the test to try to see how can we maximally accelerate this particular field of research that we're choosing to pursue in, in metagenomics discovery. Uh, and, uh, I, I think it's important for us to be dogfooding what we're building and, and trying it out and, and finding the limits and figuring out, you know, helping us figure out where to point all of our model training efforts and product development, et cetera. And then on the drug development side, um, I, I, you know, could speak a bit more about this later, but we're trying to figure out, you know, in addition to having this platform of the model and, and the product, um, what else do we need to do as a society to deflect the trajectory that we're on-

Apoorv Agrawal 9:43 Mm-hmm

Eric Kauderer-Abrams 9:43 ... to alleviate the burden of disease and aging in, in a reasonable timeframe? I don't wanna get stuck claiming that we're gonna cure all disease in X number of years, right? We can get stuck there for a long time. Um, but I believe and Anthropic, we believe that, um, amazing things are possible. And, uh, we're certainly not gonna cure all disease in the next five years or anything ridiculous like that, but we're really focused on how do we deflect this trajectory and get us on track for, for outcomes like that. And so we have all sorts of more specific things that, that we're getting going there.

Apoorv Agrawal 10:10 Eric, I thought you-- I was hoping you were gonna say AGI implies cure of all disease, but maybe it'll take a little bit longer than AGI. Um, one follow-up on something you said, Eric, is, um, [lips smack] could you help us understand, the class understand, what is the process start to finish for, uh, a drug to go to market? Uh, it sounds like there's a bunch of different phases, and maybe if you could on, almost on the x-axis being time or, or median time, um, for a drug, uh, or pick an example if you'd like, and, and, and where in that process is AI, uh, i-if at all?

Eric Kauderer-Abrams 10:45 Yeah, great question. So, so just to set the scene, um, on average, developing a drug end to end from, you know, the moment you have the idea to the moment that you have FDA approval and it's on the market, takes about ten to fifteen years, something in that range, right?

Apoorv Agrawal 11:00 Wow.

Eric Kauderer-Abrams 11:00 There are examples where it can happen faster, and I think the world record was closer to the five, six-year mark, something like that, right? Um, but that, that's the median outcome. Um, so it takes a long time. Um, and there are a lot of, a lot of distinct steps in the process. I think one of my axes to grind is that, you know, when you look at the debate about what are the bottlenecks in drug development-

Apoorv Agrawal 11:21 Mm-hmm

Eric Kauderer-Abrams 11:21 ... and where are the opportunities to pull time out, you know, a lot of people are quick to say, "It's all in the clinical trials," right? Or, "It's all in the drug design." It's not all in any one of those things. It, it's distributed across, I'd say, five to ten bottlenecks. So it's not an infinite number of bottlenecks. It's not hopeful, but it's also not just one thing. That, that's, that's way overly simplistic. So if we walk through the process, the first step is, well, you have to decide what disease you're trying to, to cure in the first place. And, um, once you have identified the disease and the patient population that, that you're trying to work in, you need to select a target. So the target is the, you know, place, molecule in the body that, that you want to develop a drug for. And you're, you're saying with that, that you have a hypothesis that if I effectively drug this target, that it will be safe for the patient, and it will ... cure the disease or, or be helpful, right?

Apoorv Agrawal 12:09 Mm-hmm.

Eric Kauderer-Abrams 12:09 So the first step is identifying good targets.

Apoorv Agrawal 12:12 Mm-hmm.

Eric Kauderer-Abrams 12:12 Um, and there's a lot to say there.

Apoorv Agrawal 12:14 Mm-hmm.

Eric Kauderer-Abrams 12:14 Um, I think the, the first thing I'll highlight is that we have a big problem in the therapeutics industry, uh, that- that is referred to as target crowding, where if you look at in, in any given year how many net new targets are pursued-

Apoorv Agrawal 12:27 Mm-hmm

Eric Kauderer-Abrams 12:27 ... in the, in the clinical phase by the whole therapeutics market in the whole world-

Apoorv Agrawal 12:32 Mm-hmm

Eric Kauderer-Abrams 12:32 ... it's, it's on the order of about 30.

Apoorv Agrawal 12:34 Three zero.

Eric Kauderer-Abrams 12:35 Three zero.

Apoorv Agrawal 12:35 That's it.

Eric Kauderer-Abrams 12:36 So for that reason alone, we're not on track to cure a whole disease in any reasonable timeframe if we're only going after 30 net new targets every year, right? So-

Apoorv Agrawal 12:44 And, and how many targets are there in total? What is the universe of targets? What's the TAM?

Eric Kauderer-Abrams 12:48 Uh, many, many thousands.

Apoorv Agrawal 12:51 Okay.

Eric Kauderer-Abrams 12:51 Probably on the order of 10,000. So there's-

Apoorv Agrawal 12:53 Got it

Eric Kauderer-Abrams 12:54 ... this is overly simplistic, but there's 19,000 genes in, in the human genome, and, uh, so presumably any one of those could be a drug target, right? Not all of them will be, right? Uh-

Apoorv Agrawal 13:04 Right

Eric Kauderer-Abrams 13:05 ... and, and, and, and you also, you, you have one to many relationships, but certainly it's, it's on the order of, you know-

Apoorv Agrawal 13:10 Got it

Eric Kauderer-Abrams 13:10 ... thousands of them.

Apoorv Agrawal 13:11 So that's step one, and how much time budget of the 10, 15 years has already been used up in this, in this first part?

Eric Kauderer-Abrams 13:16 Well, I, the, actually, one of the, um, you know, somewhat pessimistic things is the clock doesn't really start until you've selected a target.

Apoorv Agrawal 13:22 Okay. All right. Well, that's different.

Eric Kauderer-Abrams 13:23 But, but that's an important part [laughs] of, of the whole process.

Apoorv Agrawal 13:26 Okay.

Eric Kauderer-Abrams 13:26 Um, and so once you've selected a target, then, um, it, it's Josh's time to shine of actually developing the drug, right?

Apoorv Agrawal 13:34 Mm-hmm.

Eric Kauderer-Abrams 13:34 And so you selected a target, and the next step is you have to figure out what, what type of drug am I going to develop. Will it be an antibody-based drug? Will it be a small molecule? Will it be one of these emerging modalities like molecular glues and other things or, or a genetic medicine? Um, so then you select the modality and, and it's not obvious. Sometimes there's many choices that you could make, but then you actually have to design the drug, right? So this is the, the preclinical phase where, um, you're, you're doing the fundamental science of trying to develop a drug that will bind your target, uh, and be safe for the patient, right? Not have off-target effects. Um, and, and not only that, but be processed by the body in, in a reasonable way so it doesn't get, you know, lost or converted into something else, et cetera. So this is, I think, what, what people would normally associate with the, you know, the scientific phase of, of, uh, of drug development. Um, and, um, this part, um, on average takes about four years if you look historically, uh, from selecting a target to, you know, starting clinical studies.

Apoorv Agrawal 14:30 Mm-hmm.

Eric Kauderer-Abrams 14:30 And so this is the first thing I think a lot of people get wrong, um, i- in saying that there's no opportunity here. There's huge opportunity here.

Apoorv Agrawal 14:37 Mm-hmm.

Eric Kauderer-Abrams 14:37 And I think, I, you know, I imagine that you feel similarly. I think that we could take that four years down to, you know, near zero, right? In, in principle, in the theoretical limit, right? Um, and, and I, and I, I actually think that, my own opinion is it's not that important whether we get all the way to zero-shot drug design. One shot would be pretty good, too. If, if you just, you know, ask the model for a drug-

Apoorv Agrawal 14:57 [laughs]

Eric Kauderer-Abrams 14:57 ... and it gives you a few candidates, you do one round of testing and, you know, a couple weeks later you have your drug, that would be incredible, right? Compared to the four years that, that we're talking about.

Apoorv Agrawal 15:05 Mm-hmm.

Eric Kauderer-Abrams 15:05 Um, but, and I, I can let Josh elaborate on this, but even within the drug design phase, there's so many different parts, right? So you get your initial hit, which is like we're just getting in the right ballpark. We can hit the target, right? And then after that you do optimization to get to leads, right? And you're trying to get all these additional properties that you want related to how manufacturable is it, does it have off-target effects, is it stable in the body, is it processed in the way that you expect, right? So there's all this optimization that you do that I think is really ripe for, for AI. Um, but trying to accelerate through this a little bit.

Apoorv Agrawal 15:34 Mm-hmm.

Eric Kauderer-Abrams 15:35 So you design the drug, you have the drug candidate.

Apoorv Agrawal 15:37 Mm-hmm.

Eric Kauderer-Abrams 15:37 Um, a- and in the field it's often referred to as a development candidate. So you have your development candidate. You lock your manufacturing processes, and you have to meet all sorts of, of pretty, um, you know, high bar for, uh, for the rigor with which you're manufacturing this and documenting it, and then you're ready to start your, your, um, clinical studies in humans. Um, and so you have to get something called an IND from FDA in order-

Apoorv Agrawal 15:59 Mm-hmm

Eric Kauderer-Abrams 15:59 ... to start these trials. And so then one- once you get the IND, then it's the clinical phase and there, there are three parts of that. So there's a phase one. This is historically. There's all sorts of vari- variations that are possible, but there's a phase one focused on safety. Traditionally then a phase two, that's your first look at efficacy.

Apoorv Agrawal 16:16 Mm-hmm.

Eric Kauderer-Abrams 16:16 Uh, and then a phase three, which is the final validation study that, that produces the evidence that your drug is safe and effective in humans, and that's the basis of the ultimate, um, FDA clearance. In parallel as all that is happening, um, there, there's all sorts of work on the transfer to manufacturing side as well.

Apoorv Agrawal 16:32 Mm-hmm.

Eric Kauderer-Abrams 16:32 Um, so, so that's the whole process. You know, if you'd say like the developing the drug is four years, then the clinical phase, right, all, everything up until the FDA, uh, period is something on the order of, like, six to nine years or something like that, right? So that's how time is allocated. Um, cost is very much backloaded on the clinical phase.

Apoorv Agrawal 16:51 Mm-hmm.

Eric Kauderer-Abrams 16:51 Um, but I think there are really compelling and straightforward opportunities to use AI to pull time out of that process in both parts-

Apoorv Agrawal 17:00 Mm-hmm

Eric Kauderer-Abrams 17:00 ... on the preclinical phase and on the clinical side. Uh, uh, but that, that's an overview. Yeah.

Apoorv Agrawal 17:05 Super helpful. Thank you. Josh, you have clearly decided to focus, um, your efforts on the part that, that you found to be the most high leverage. In, in this framework, where are you, where is Chai's work, uh, and, and how are you gonna compress the timeline?

Speaker 17:18 So I actually give two answers to this. I think, uh, Eric did a great job, like, breaking down this into all these different phases, and creating a drug is so complicated and one of the reasons why the pharma industry has so many of these different stages is that how do you even know if you're on track?

Apoorv Agrawal 17:30 Mm-hmm.

Speaker 17:31 You do have to break it down into a bunch of different mini-games-

Apoorv Agrawal 17:33 Mm-hmm

Speaker 17:33 ... and win each of them. But at the end of the day, what we're really looking for is some molecular matter-

Apoorv Agrawal 17:39 Mm-hmm

Speaker 17:39 ... uh, that actually modulates some kind of disease. Like, that's what the process of making a drug is. And, you know, the high level, we have the preclinical phase, right, where we're trying to find that molecule-

Apoorv Agrawal 17:48 Mm-hmm

Speaker 17:48 ... and the clinical, which is where we're trying to prove that it's going to work.

Apoorv Agrawal 17:51 Mm-hmm.

Speaker 17:51 But if we just think about it at a high level, it's finding a molecule-

Apoorv Agrawal 17:54 Mm-hmm

Speaker 17:54 ... that does something in a patient, and one of the things that's so powerful about AI is that these things really can be linked again, right?

Apoorv Agrawal 17:59 Mm-hmm. Mm-hmm.

Speaker 18:00 If I come up with a better and more potent molecule-

Apoorv Agrawal 18:03 Mm-hmm

Speaker 18:03 ... hopefully that makes my clinical trial easier.

Apoorv Agrawal 18:05 Mm-hmm.

Speaker 18:05 One of the really sad things about the space right now is there are so many drugs that are, like, expanding, extending lifespan by, like, one month or two months or something like that.

Apoorv Agrawal 18:12 Mm-hmm.

Speaker 18:12 And again, for those patients that need it, that's, that's amazing, right? And we, we should keep doing that work. Um, but when you're looking at changes that small, clinical trials are really difficult Right? And being able to design a trial such that you can actually see that with statistical significance and get your drug approved is very difficult. If you have some amazing data, your trial becomes a lot easier all of a sudden.

Apoorv Agrawal 18:31 Mm-hmm.

Speaker 18:31 And I think we can actually look at code generation as, like, a lens of what's possible here.

Apoorv Agrawal 18:35 Mm.

Speaker 18:35 If you would have thought, if you look how people used, like, LLMs for code gen even, like, a year ago, you would get some initial solution, you would try to debug it-

Apoorv Agrawal 18:43 Mm-hmm

Speaker 18:43 ... you'd put in some other LLM to, like, ask some questions about it. This whole iterative process, and then eventually you have some code that works. Now you can, like, zero shot an app.

Apoorv Agrawal 18:50 Yeah.

Speaker 18:51 Right? And I think that's what, like, zero shot drug discovery promises to potentially unlock.

Apoorv Agrawal 18:55 Mm-hmm.

Speaker 18:56 It's not just taking, you know, four years and getting it down to zero years, I think it's also just about getting better medicines overall. We talked about targets-

Apoorv Agrawal 19:03 Mm-hmm

Speaker 19:03 ... and the TAM, right?

Apoorv Agrawal 19:04 Yeah.

Speaker 19:04 And there's, like, 30, you know, a small number of targets that people crowd around, but, like, 80% of named diseases, like, don't have an approved medicine.

Apoorv Agrawal 19:11 Mm.

Speaker 19:11 Right? And that's a huge problem here as well. So I think just creating better medicines, creating more medicines, uh, that's what it all comes down to. So with all of that said, the place that Chai focuses on right now-

Apoorv Agrawal 19:21 Yeah

Speaker 19:21 ... because, again, as you're, you know, you're building a startup, you need to focus somewhere. So we've started on that molecular generation process, 'cause we just see that as, like, the apex for all of these other things that can happen around that.

Apoorv Agrawal 19:31 Mm-hmm.

Speaker 19:31 And one of the reasons why we're so excited about your work, for instance, is that if we can speed up the whole outer loop of, like, right now those iteration cycles, then we can just get more shots on goal coming out of the model. Uh, but I think as both of these things get better, they're just gonna reinforce one another and we're, you know, I think the whole thing's gonna be exponential in the next couple of years.

Apoorv Agrawal 19:48 Fascinating. Um, so the obvious question is why now? This is a problem as old as time. Uh, we've wanted to live forever, forever. Um, what has changed? What, what, what new, um, uh, uh, technology's unlocked and why is that gonna compress n- now?

Speaker 20:09 Well, so I, I've been trying to do this stuff my whole career, honestly.

Apoorv Agrawal 20:11 [laughs]

Speaker 20:12 So you mentioned the story before, you know, OpenAI to, uh, to Facebook. Um, and I think it's, it's, you know, shouldn't come as a surprise to anyone that AI has gotten a lot better. Uh, the model architectures have gotten to a point that they're a lot more scalable, a lot more compute has come online.

Apoorv Agrawal 20:26 Mm-hmm.

Speaker 20:26 Everything is just moving faster. Um, and there's a lot of new data sets that, that we can build as well.

Apoorv Agrawal 20:31 Mm-hmm.

Speaker 20:31 And I think the other part of this too is that, um, I think there's also a lot of willingness to experiment with these things. If you think about how, how Chai orients ourselves, right? We, we bring our models into pharmaceutical companies. There's also the other side of this, which is, like, you can build the thing-

Apoorv Agrawal 20:44 Mm-hmm

Speaker 20:44 ... but people also need to be willing to experiment.

Apoorv Agrawal 20:46 Mm-hmm.

Speaker 20:47 And A, like, people are just really open to AI right now.

Apoorv Agrawal 20:49 Mm.

Speaker 20:50 But then, uh, but then B, uh, there's also a lot of pressure. If you look at what's happening geopolitically in life sciences-

Apoorv Agrawal 20:56 Mm-hmm

Speaker 20:56 ... and, uh, you know, actually any of these pharma forums, and whenever I go to a conference about pharma, there's two topics these days. There's AI and there's China.

Apoorv Agrawal 21:03 Mm.

Speaker 21:03 And one of the things that, uh, you know, the, the US is struggling with right now is that China is just outrunning us when it comes to drug discovery.

Apoorv Agrawal 21:10 How so?

Speaker 21:10 They are, they're just much more efficient.

Apoorv Agrawal 21:12 Mm-hmm.

Speaker 21:12 They work harder.

Apoorv Agrawal 21:13 Mm.

Speaker 21:13 It's cheaper to discover drugs there.

Apoorv Agrawal 21:15 Mm.

Speaker 21:15 So this is even on the preclinical side, not to mention the clinical side.

Apoorv Agrawal 21:18 Mm.

Speaker 21:18 Uh, the regulatory state there, it's a lot easier to just get a drug-

Apoorv Agrawal 21:21 Mm-hmm

Speaker 21:21 ... into first in human in China.

Apoorv Agrawal 21:23 Mm-hmm.

Speaker 21:23 Um, but one of the reasons why, if we just think about, you know, the US' role in biotech, we, like, uh, desperately need AI to work here.

Apoorv Agrawal 21:30 Mm-hmm.

Speaker 21:30 Uh, because China can run faster than the US, but, but China cannot run faster than AI. Uh, and I think this is gonna be a huge, like, democratizing function in terms of-

Apoorv Agrawal 21:39 Mm

Speaker 21:39 ... um, you know, like, our ability in the US to actually come up with better drugs. So I think there's a lot of, like, why now.

Apoorv Agrawal 21:44 Mm.

Speaker 21:44 There's both, like, why now, and then also, like, we really need it to work now.

Apoorv Agrawal 21:47 Yeah.

Speaker 21:47 So we're kind of lucky-

Apoorv Agrawal 21:48 Yeah

Speaker 21:48 ... that it's happening-

Apoorv Agrawal 21:49 Yeah

Speaker 21:49 ... in terms of what's happening in the world.

Apoorv Agrawal 21:50 Yeah. And is there a technology aspect to the why now, uh, for either of you that you might say like, "Hey, obviously the demand has, exists now because of AI and, and China." Uh, but has, has something changed? Ha- have the, uh, large language models given us a leap forward? Is that a superpower? And what is the potency of that superpower, uh, in your estimation? Is this like a we're gonna bring this down from 15 years to, to, to, to 15 minutes or is this 15 years to, to five years? Um, help us size that.

Eric Kauderer-Abrams 22:19 Yeah, so I, I think, um, by far the, my, my first answer to the why now is the existence and fast progress of, of large language models.

Apoorv Agrawal 22:29 Mm-hmm.

Eric Kauderer-Abrams 22:29 Um, I think, but I actually think that there's a few trends that are all converging right at the same time, and we're very fortunate because it didn't have to be the case. Um, but it's the, the large language models is one. There's all sorts of other large models like Chai models, right? These foundation models, and they're complementary because the, as Josh said, the large language model I think of as the outer loop. It does what you do, right? As a human, you use Chai's model to get some designs, and you think about it, and you test them in the lab, right? And you keep iterating, right? And hopefully you don't need too many iteration cycles, right? But that whole process can now j- just be done by Claude and, uh, and it's getting better and better, right? So you have these two trends of the underlying foundation models are getting better, so you need fewer iterations in the first place, and you have the large language models that are available to do the outer loop and make that whole thing faster. So that, I think that's a strong trend, and I was giving an example of, um, using these in the, the preclinical phase, but for the LLMs, the same thing I think can be said to, to some extent in the clinical phase as well. Um, so tha- that's one trend. The other trend that, that we're very fortunate that is happening right at the same time is the massive scale of data generation that's possible in biology, right? So we have all of these different measurement techniques that are, are becoming, you know, increasingly mature. Sequencing at this po- point is, is old news, but on top of that, right, we have single cell sequencing and proteomics and, uh, very high throughput techniques for measuring antibodies a- and all sorts of other, um, you know, assays that, that we're using to just keep fueling these. So I think these trends all, all happening at the same time is, uh-

Apoorv Agrawal 24:01 Mm-hmm

Eric Kauderer-Abrams 24:02 ... is a big piece of it. As far as where the floor is, right, if we're in the 10 to 15 year zone-

Apoorv Agrawal 24:06 Mm

Eric Kauderer-Abrams 24:06 ... um, I think that we have clear line of sight to bring in clinical trial, uh, sorry, the whole drug development timeline down to certainly the five-year range, and I would put that-

Apoorv Agrawal 24:15 Wow

Eric Kauderer-Abrams 24:15 ... as an upper bound. Um, I think, you know, you can imagine the whole thing happening in, in a few years.

Apoorv Agrawal 24:20 Mm-hmm.

Eric Kauderer-Abrams 24:21 Ultimately, there are lower bounds that are going to be established by the duration of clinical trials.

Apoorv Agrawal 24:26 Mm-hmm.

Eric Kauderer-Abrams 24:26 But here too, I think there's a lot more opportunity than, than people think for a few reasons, right? Like someone will point out, for example, um- Let's say you're, you're running a, a clinical trial, uh, to get at, um, uh, osteoporosis and, you know, other, you know, kind of classically you have to wait and, um, see how many people's bones break over a certain period of time, and so to get enough statistical power, you need to wait a year or something like that, right? And so that, that establishes naively this floor-

Apoorv Agrawal 24:53 Mm. Right

Eric Kauderer-Abrams 24:53 ... of a year for, for that clinical trial. But there are other methods that, that can, in principle, work. For example, if we can develop proxy measurements, right? If, if we take the right-

Apoorv Agrawal 25:03 Mm-hmm

Eric Kauderer-Abrams 25:03 ... assays and we learn more about the biology and we can tell from all the things we're measuring about the body that, that the intended effect is happening, right?

Apoorv Agrawal 25:09 Mm-hmm.

Eric Kauderer-Abrams 25:10 Uh, you can imagine having different endpoints for the clinical trials that, that bring the time down.

Apoorv Agrawal 25:14 Mm-hmm.

Eric Kauderer-Abrams 25:15 Um, and also as Josh said, um, I think this is a really important point, the larger effect size is in the first place. So the number of subjects that you have to enroll-

Apoorv Agrawal 25:22 Mm-hmm

Eric Kauderer-Abrams 25:22 ... depends on the effect size, and the more effective the drugs are, you know, the fewer subjects and the faster the whole thing can go. So that, that's where I back into that, that time estimate.

Apoorv Agrawal 25:31 Yeah. Historically, there has been framing in life sciences of either you're developing the full drug and, and, and selling the drug and enjoying the revenue that comes from selling the drug, uh, or being a, a tool, or you're selling a tool, right? Um, and at least, you know, most traditional biotech investors will, will value life sciences businesses as the, call it, revenue stream or free cash flow coming from their drugs, pr- uh, forecasted back to today and have some discounted cash flow effect, and that's your market cap or, or, or, or, or, or, or the valuation that VCs are willing to pay. You have, you have decided to build a platform of tools that you'll be selling to these people, and it sounded like, Eric, you said wet lab. You're gonna build a wet lab, so does that imply you're gonna develop the full drug and monetize the drug at Anthropic?

Eric Kauderer-Abrams 26:22 Yeah, good question. I should, I should, uh, correct that, that misconception. So, uh, no, that is not our plan-

Apoorv Agrawal 26:27 Okay

Eric Kauderer-Abrams 26:27 ... to, to develop drugs. We also are very much building tools that we want-

Apoorv Agrawal 26:31 Of course

Eric Kauderer-Abrams 26:31 ... the rest of the industry to use. Um, the purpose of our wet lab, so I've been talking a lot about drug development-

Apoorv Agrawal 26:36 Mm-hmm

Eric Kauderer-Abrams 26:36 ... but our other primary objective is to accelerate basic research, right?

Apoorv Agrawal 26:39 Mm-hmm.

Eric Kauderer-Abrams 26:39 As an end in itself, and so the wet lab is our sandbox for pursuing that goal. So we actually have a basic research team, um, that, uh, you know, is, is just doing science, putting y- you know, our tools to the test and maximizing use of Claude through everything that you can imagine. Um, and so the wet lab does serve a different purpose.

Apoorv Agrawal 26:56 Gotcha. So one of the biggest things that we talk about in this class is where will value accrue in AI. Certainly so far, a lot of that has been in the semiconductors layer and, and not as much as got into the app layer. In your world in biology, uh, historically that has been at the, at the, at the layer that is selling the drugs and not at the tools. Um, correct me if I'm wrong, uh, but you're both building tools and probably have a reason to believe it's gonna be different this time. Why might it be different this time?

Speaker 27:26 I think I can give, like, the simplest answer to this, is that the tools are becoming more valuable than they've ever been before.

Apoorv Agrawal 27:30 Mm-hmm. Mm-hmm.

Speaker 27:31 If you look at where value has accrued historically, like, the drugs that we have today on the market are, are histor- are, are extremely primitive.

Apoorv Agrawal 27:37 Mm.

Speaker 27:37 It's, it's honestly a miracle at all that we can discover drugs with the tools that are available to us today.

Apoorv Agrawal 27:42 Mm.

Speaker 27:42 Uh, people often ask Chai, like, who our competitors are, and it's, like, literally the yeast and the mice and, like, you know-

Apoorv Agrawal 27:47 [laughs]

Speaker 27:47 ... the traditional tools in the lab, uh, that are being used.

Apoorv Agrawal 27:50 Mm-hmm.

Speaker 27:50 And the drugs these days are as simple as, like, trying to jam up, like, a target, for instance, uh, on a cell.

Apoorv Agrawal 27:55 Mm.

Speaker 27:56 Um, and they have all these side effects, and most drugs are actually, they have many off-target effects that they, they probably shouldn't have. Um, so I think as the tools become more powerful, it makes sense that more value should accrue there.

Apoorv Agrawal 28:07 Mm-hmm.

Speaker 28:07 Why should you pay that much for a tool that doesn't impact your probability of success? If everything that we're talking about here, if what Eric is saying, uh, you know, comes to pass and timelines go down, probability of success goes up, then by adopting these AI tools, if you're a pharmaceutical company-

Apoorv Agrawal 28:21 Mm-hmm

Speaker 28:21 ... for instance, you should probably start trading at a higher multiple.

Apoorv Agrawal 28:24 Right.

Speaker 28:24 'Cause if your probability of success goes up-

Apoorv Agrawal 28:26 Right

Speaker 28:26 ... uh, and your cost goes down for, for getting things to market, uh, and the timelines go down, uh, already just-

Apoorv Agrawal 28:32 Mm

Speaker 28:32 ... by adopting these tools you become more valuable-

Apoorv Agrawal 28:34 Mm

Speaker 28:34 ... and therefore a lot of value should accrue to the toolmakers.

Apoorv Agrawal 28:37 Yeah.

Speaker 28:37 And the tools are just way more scalable, uh-

Apoorv Agrawal 28:39 Yeah

Speaker 28:39 ... than, than actually the end products as well.

Apoorv Agrawal 28:41 Yeah. One of the other things that, you know, we observed is the largest biotech companies m- m- in the West were founded decades if not centuries ago. Like, they're in the 20th century, there's a couple in the 19th century, and, um, there's a couple of them that were f- formed in the last 25 years. Do you think one of the impacts might be more democratization, more, more, um, startup, uh, forming 'cause the tools are so, uh, accessible?

Speaker 29:11 I think it can go in, uh, and, and there's a, there's gonna be a couple of second-order effects here. So-

Apoorv Agrawal 29:15 Mm-hmm

Speaker 29:15 ... the first thing is today you could start a biotech company just by being a little bit more efficient than a pharmaceutical company.

Apoorv Agrawal 29:21 Mm-hmm.

Speaker 29:21 Pharma companies are built big organizations, and there might be some target insight that you get out of Claude, right? And then you rush a, a, a molecule to, to clinical trials. You get some proof of concept in a patient, and then a pharma company buys you up.

Apoorv Agrawal 29:33 Mm-hmm.

Speaker 29:34 What happens when two years from now you can just zero shot that molecule?

Apoorv Agrawal 29:37 Mm.

Speaker 29:37 And you can, like, zero shot a biotech.

Apoorv Agrawal 29:39 Mm.

Speaker 29:39 Right?

Apoorv Agrawal 29:39 Mm.

Speaker 29:40 I think, does that mean biotech goes away? I don't actually think that happens. I think biotech just evolves around that.

Apoorv Agrawal 29:45 Mm-hmm.

Speaker 29:45 The same way today, you know, if you can zero shot an app, yeah, you can't just, uh, sell a tool that's as simple as, you know, that zero shot.

Apoorv Agrawal 29:52 Mm-hmm.

Speaker 29:52 You need to figure out either some other mode or some more complex, uh, software that has bigger value.

Apoorv Agrawal 29:56 Mm-hmm.

Speaker 29:56 Um, so I think it's very hard to predict what's gonna happen here-

Apoorv Agrawal 29:59 Mm-hmm

Speaker 29:59 ... because, again, this is happening exponentially.

Apoorv Agrawal 30:02 Yeah.

Speaker 30:02 There's a lot going on at once. There's a lot of second-order effects. Um-

Apoorv Agrawal 30:04 Mm-hmm

Speaker 30:05 ... so I, I think on the balance, uh, if everything stays, you know, stagnant, you might see accumulation to, um, uh, to the incumbents, but in reality, people are always gonna find new opportunities. There's gonna continue to be startups, and-

Apoorv Agrawal 30:17 Mm

Speaker 30:17 ... the outcomes will just be bigger than they've ever been before-

Apoorv Agrawal 30:20 Mm

Speaker 30:20 ... because we have better tools to make these things happen.

Apoorv Agrawal 30:23 Fascinating. You guys have a very special vantage point, and you probably know all that's on the f- frontier of, of development. What are the most, uh, the biggest unsolved problems, um, in your fields right now that you're, that you're looking forward to being solved with either the models or otherwise?

Eric Kauderer-Abrams 30:41 Yeah, um, I think there's a few that I would highlight. So the first is, um, you know, what companies like Chai have done for antibody development. I think doing that in other modalities is very much an unsolved problem, right? So antibodies are probably the most effective class of drug right now, but there are, are plenty of targets and diseases for, for which antibodies aren't immediately applicable.

Apoorv Agrawal 31:02 Mm-hmm. Mm-hmm.

Eric Kauderer-Abrams 31:03 So we have small molecules and these other modalities that are emerging, and, um, I'm really excited about taking these, these, um, modalities that AI... And I- the way I like to think about this is there's a certain frontier of targets that are drugable today-

Apoorv Agrawal 31:17 Mm-hmm

Eric Kauderer-Abrams 31:17 ... with our current methods, right? And, and we need to advance that and turn those into engineering disciplines, right? In the way that antibody design has gone from what used to be, you know, somewhat more like an alchemical craft into more of an engineering discipline. And, uh, so I think that is, is well within reach with AI.

Apoorv Agrawal 31:34 Mm-hmm.

Eric Kauderer-Abrams 31:35 Um, but, but is a hard, important problem to work on.

Apoorv Agrawal 31:37 Mm-hmm.

Eric Kauderer-Abrams 31:37 Um, and the other thing that I would highlight is, um, scaling up the discovery of high-quality new targets. So again, we're still crowding around-

Apoorv Agrawal 31:46 Mm-hmm

Eric Kauderer-Abrams 31:46 ... a relatively small number of targets.

Apoorv Agrawal 31:47 Mm-hmm. Mm-hmm.

Eric Kauderer-Abrams 31:48 And I think as a field, we're all trying to figure out, what's that n- next big scalable way that we can unlock a huge number of, of high-quality new targets to pursue. Um, and you know, there are a lot of efforts going on, um, here, for example, in, uh, virtual cell and cell perturbation models, which is this, you know, new technique that is emerging, and, um, more traditional genetics-based approaches for looking at human genetics data at the population scale and correlating it with, with, um, health records. Um, but I think that's a really important frame of, um, as a society, we need a scalable way to find many, many more good targets.

Apoorv Agrawal 32:23 Yeah. Anything you'd add to that, Josh?

Speaker 32:25 I mean, fully agree with those. Uh, maybe the one last one I would add is, uh, more sophisticated medicines.

Apoorv Agrawal 32:30 Mm-hmm.

Speaker 32:30 So I kind of alluded to this before, but a lot of the drugs today are, are, are very simple.

Apoorv Agrawal 32:34 Mm-hmm.

Speaker 32:35 And I think as we get more control over them, you can think about, um, you know, what we are... One of the things we do with Chai is, is we, we can fold up a lot of the proteins in the drugs that we're designing.

Apoorv Agrawal 32:44 Mm.

Speaker 32:44 So think of this almost as like an atomic level microscope of like what's going on in your system.

Apoorv Agrawal 32:48 Mm-hmm.

Speaker 32:49 And if you have that, you can now start to dream a lot bigger with the kind of molecules that are possible. So I think this is something that's also gonna play out.

Apoorv Agrawal 32:56 Makes sense. The, um, productivity of a big unlock in your field is incredibly high. There are companies that have single, uh, drugs that have led to hundreds of billions of revenue, uh, certainly many even in the tens of billions of dollars of range. Um, is there a, a program or two that you guys are watching that you're like, "Hey, this is gonna be the next GLP, uh, level success, or has the potential?"

Eric Kauderer-Abrams 33:23 Yeah, good question. Um, I would highlight there's some emerging targets coming out around, um, increasing lean muscle mass that, um-

Apoorv Agrawal 33:31 [laughs]

Eric Kauderer-Abrams 33:31 ... I just think from a commercial perspective. I'm making no comment on the, the value to society, but, uh- [laughs]

Apoorv Agrawal 33:37 [laughs] The Incredible Hulk.

Eric Kauderer-Abrams 33:39 [laughs] But I, I will say I think there are some really, uh, important medical use cases for those. But that aside-

Apoorv Agrawal 33:45 Mm-hmm

Eric Kauderer-Abrams 33:45 ... uh, you know, if there's a drug that you can take that, that makes you ripped, I think that's gonna do pretty well.

Apoorv Agrawal 33:49 [laughs]

Speaker 33:50 Yeah, my, my team is gonna kill me for saying this, but this is actually our internal joke at Cho- at Chai as well. You look at like Renaissance RenTech, for instance.

Apoorv Agrawal 33:58 Yeah.

Speaker 33:58 Uh, they have their internal medallion fund.

Apoorv Agrawal 34:00 Mm-hmm.

Speaker 34:00 So one of the ways you get compensated if you go to Renaissance is you get to put money in, in like the best performing fund. So we're all like, "We should... If we, if we ever make, uh, drugs at Chai, we should start with that."

Eric Kauderer-Abrams 34:08 [laughs]

Speaker 34:08 And it, it's just for Chai employees. So-

Eric Kauderer-Abrams 34:10 [laughs]

Speaker 34:10 ... you come to Chai, you get jacked and, uh- [laughs]

Apoorv Agrawal 34:15 Nice. Um, any that you're looking at, Josh?

Speaker 34:18 Um, I think that, uh, things for sleep are gonna be really interesting as well.

Apoorv Agrawal 34:22 Mm.

Speaker 34:22 So I think what the GLP-1s and obesity drugs are showing is they're almost like consumer medicine essentially because of how big the patient populations are.

Apoorv Agrawal 34:28 Mm-hmm. Mm-hmm.

Speaker 34:28 And I think sleep is also something that leads to-

Apoorv Agrawal 34:31 Right

Speaker 34:31 ... so many of the disorders that are out there. So if we can find ways for people to like sleep better or get rid of sleep disorders-

Apoorv Agrawal 34:36 Mm-hmm

Speaker 34:37 ... uh, I think that's also interesting. And actually as a similar, um, kind of, uh, uh, economics, if you will, to the GLP-1s, it started with, uh, with diabetes and kind of moved to obesity.

Apoorv Agrawal 34:46 Okay. Mm-hmm.

Speaker 34:46 And same thing here, you can start with sleep disorders and then maybe that turns into like... I don't know, if I can sleep more effectively, like who wouldn't want a drug like that?

Apoorv Agrawal 34:53 Mm-hmm. Yeah.

Speaker 34:53 Um, so, uh, I, I think we, this might be the start of these like consumer medicines, uh, and, and really change the way that we live.

Apoorv Agrawal 34:59 Fascinating. Wow, those are two, uh, pretty remarkable ones. Everybody in the class is wondering what, what should we go invest in for getting ripped and, and sleeping better.

Speaker 35:08 [laughs]

Apoorv Agrawal 35:09 Um, maybe, maybe we'll start with you, Josh. Um, pick a, pick a business. Pick a, pick a, pick a business or idea or startup that you're bullish on, um, based on everything you know about the world. Uh, and the other way, what are you, what are you, what are you skeptical of? Which you think is more hype than reality?

Speaker 35:25 Yeah, the, the two categories I'd, I'd point to, first, I mean, it shouldn't be a surprise, like pharmaceuticals. Like, uh, you know, if we're gonna have better drugs, then, um, you can almost think of many pharma companies almost as like capital aggregators.

Apoorv Agrawal 35:36 Mm-hmm.

Speaker 35:36 Uh, right? They have these drugs that are creating billions of revenue.

Apoorv Agrawal 35:39 Mm-hmm. Mm-hmm.

Speaker 35:39 They have a mandate to reinvest that into more drugs, and they might just become way more effective.

Apoorv Agrawal 35:43 Mm-hmm. Mm-hmm.

Speaker 35:44 Another thing also, uh, and I just want to give us just a little bit more contrarian, uh, but is actually lab experiments themselves.

Apoorv Agrawal 35:45 Mm-hmm.

Speaker 35:45 So a lot of people, uh, when they talk to us, they assume that like, oh, Chai wants to like make experiments dead or, or something like that. But I actually think if you can design your experiments better, the ROI in experiments goes up.

Apoorv Agrawal 35:45 Mm-hmm.

Speaker 35:45 So I think it's gonna be a Jevons paradox actually for a lot of lab work-

Apoorv Agrawal 35:45 Mm-hmm

Speaker 35:45 ... where we go into this renaissance where, uh, and we can actually pull like Dario's remarks, right, from like the incredible essays that imagine like ten years of biomedical discoveries happening in the next year.

Apoorv Agrawal 35:45 Mm.

Speaker 35:45 It's gonna be a lot of lab work for that to happen, right?

Apoorv Agrawal 35:45 Right. Right.

Speaker 35:45 And AI is gonna be driving it, so hopefully a lot more of like, you know, Chai usage as well to make that happen-

Apoorv Agrawal 35:45 Yeah

Speaker 35:45 ... and cloud usage. Uh-

Apoorv Agrawal 35:45 Yeah

Speaker 35:45 ... but, uh, uh, yeah, so those are the things I'm, I'm most excited about.

Apoorv Agrawal 35:45 Yeah. Anything you're bearish on?

Speaker 35:45 I'd be bearish actually... So it, I, I mentioned these things are kind of in the physical world, right?

Apoorv Agrawal 35:45 Yeah.

Speaker 35:45 The things I'd be more bearish on are things that are purely in the software world-

Apoorv Agrawal 35:45 Mm

Speaker 35:45 ... but are not frontier.

Apoorv Agrawal 35:45 Mm.

Speaker 35:45 So for instance, a lot of the, uh, the methods, uh, that are used, uh, today in the bio world, uh, there's actually, people have been using like physics-based methods and computational-based methods for a long time in drug discovery. There's companies that have been around for, for 30 years, and, and they've had tremendous impact in the space. Uh, but, but AI is becoming quite powerful-

Apoorv Agrawal 35:45 Mm-hmm

Speaker 35:45 ... and is gonna eat a lot of that stuff as well.

Apoorv Agrawal 35:45 Mm-hmm.

Speaker 35:45 So I guess you could summarize this as like, you know, AI in the physical world-

Apoorv Agrawal 35:45 Yeah

Speaker 35:45 ... is gonna be really big.

Apoorv Agrawal 35:45 Yeah. Yeah.

Speaker 35:45 Um, so, uh, I, I think we, this might be the start of these like consumer medicines, uh, and, and really change the way that we live.

Apoorv Agrawal 35:45 Fascinating. Wow, those are two, uh, pretty remarkable ones. Everybody in the class is wondering what, what should we go invest in for getting ripped and, and sleeping better.

Speaker 35:45 [laughs]

Apoorv Agrawal 35:45 Um, maybe, maybe we'll start with you, Josh. Um, pick a, pick a business. Pick a, pick a, pick a business or idea or startup that you're bullish on, um, based on everything you know about the world. Uh, and the other way, what are you, what are you, what are you skeptical of? Which you think is more hype than reality?

Speaker 35:45 Yeah, the, the two categories I'd, I'd point to, first, I mean, it shouldn't be a surprise, like pharmaceuticals. Like, uh, you know, if we're gonna have better drugs, then, um, you can almost think of many pharma companies almost as like capital aggregators.

Apoorv Agrawal 35:45 Mm-hmm.

Speaker 35:45 Uh, right? They have these drugs that are creating billions of revenue.

Apoorv Agrawal 35:45 Mm-hmm. Mm-hmm. Mm-hmm.

Speaker 35:45 They have a mandate to reinvest that into more drugs, and they might just become way more effective.

Apoorv Agrawal 35:45 Mm-hmm. Mm-hmm.

Speaker 35:45 Another thing also, uh, and I just want to give us just a little bit more contrarian, uh, but is actually lab experiments themselves.

Apoorv Agrawal 35:49 Mm-hmm.

Speaker 35:50 So a lot of people, uh, when they talk to us, they assume that like, oh, Chai wants to like make experiments dead or, or something like that. But I actually think if you can design your experiments better, the ROI in experiments goes up.

Apoorv Agrawal 36:00 Mm-hmm.

Speaker 36:00 So I think it's gonna be a Jevons paradox actually for a lot of lab work-

Apoorv Agrawal 36:03 Mm-hmm. Mm-hmm

Speaker 36:03 ... where we go into this renaissance where, uh, and we can actually pull like Dario's remarks, right, from like the incredible essays that imagine like ten years of biomedical discoveries happening in the next year.

Apoorv Agrawal 36:13 Mm.

Speaker 36:13 It's gonna be a lot of lab work for that to happen, right?

Apoorv Agrawal 36:14 Right. Right.

Speaker 36:15 And AI is gonna be driving it, so hopefully a lot more of like, you know, Chai usage as well to make that happen-

Apoorv Agrawal 36:19 Yeah

Speaker 36:19 ... and cloud usage. Uh-

Apoorv Agrawal 36:20 Yeah

Speaker 36:21 ... but, uh, uh, yeah, so those are the things I'm, I'm most excited about.

Apoorv Agrawal 36:24 Yeah. Anything you're bearish on?

Speaker 36:26 I'd be bearish actually... So it, I, I mentioned these things are kind of in the physical world, right?

Apoorv Agrawal 36:29 Yeah.

Speaker 36:29 The things I'd be more bearish on are things that are purely in the software world-

Apoorv Agrawal 36:33 Mm

Speaker 36:33 ... but are not frontier.

Apoorv Agrawal 36:35 Mm.

Speaker 36:35 So for instance, a lot of the, uh, the methods, uh, that are used, uh, today in the bio world, uh, there's actually, people have been using like physics-based methods and computational-based methods for a long time in drug discovery. There's companies that have been around for, for 30 years, and, and they've had tremendous impact in the space. Uh, but, but AI is becoming quite powerful-

Apoorv Agrawal 36:53 Mm-hmm

Speaker 36:53 ... and is gonna eat a lot of that stuff as well.

Apoorv Agrawal 36:55 Mm-hmm.

Speaker 36:55 So I guess you could summarize this as like, you know, AI in the physical world-

Apoorv Agrawal 36:58 Yeah

Speaker 36:58 ... is gonna be really big.

Apoorv Agrawal 36:59 Yeah. Yeah.

Speaker 37:00 Um, so, uh, I, I think we, this might be the start of these like consumer medicines, uh, and, and really change the way that we live.

Apoorv Agrawal 37:04 Over to you, Eric

Eric Kauderer-Abrams 37:06 Yeah

Apoorv Agrawal 37:06 Long, short

Eric Kauderer-Abrams 37:07 So, so my, my category for, for long is similar to what Josh said at, at the end. Um, I'm, I'm very excited about this class of companies that make it possible to, um, for Claude to run experiments. So there's two ways for, for Claude to be able to run experiments. Um, one way would be to hook up Claude to lab instruments and work through all those integrations-

Apoorv Agrawal 37:27 Mm

Eric Kauderer-Abrams 37:27 ... and we're absolutely working on that.

Apoorv Agrawal 37:29 Mm.

Eric Kauderer-Abrams 37:29 But Claude can run experiments right now-

Apoorv Agrawal 37:31 Mm

Eric Kauderer-Abrams 37:31 ... by just, uh, placing orders with contract research organizations or CROs, right?

Apoorv Agrawal 37:36 Mm.

Eric Kauderer-Abrams 37:36 Claude, uh, what would I do if I wanted to do that? I would draft a protocol, and I would email somebody, and we'd talk back and forth.

Apoorv Agrawal 37:41 Mm.

Eric Kauderer-Abrams 37:41 Claude can do that right now-

Apoorv Agrawal 37:43 Mm

Eric Kauderer-Abrams 37:43 ... and is doing that right now. And, uh, and so these aren't quite CROs, but these are what I would call, uh, the class of business that I'm really excited about are these really scalable sort of AI native real lab, you know, wet lab companies, right? That, that are, um, manufacturing-

Apoorv Agrawal 37:59 Mm

Eric Kauderer-Abrams 37:59 ... uh, materials or running experiments with high scale, um, in a way that is easy for AI to interface with, right? So some examples of companies that, that I think are, are great in this space, uh, Plasmidsaurus-

Apoorv Agrawal 38:12 Mm-hmm

Eric Kauderer-Abrams 38:12 ... on, um, all sorts of sequencing that you wanna get done and Adaptive-

Apoorv Agrawal 38:16 Mm

Eric Kauderer-Abrams 38:16 ... and, uh, Twist, right? These are companies that, that make it really easy to do these things at scale and execute, you know-

Apoorv Agrawal 38:22 Mm

Eric Kauderer-Abrams 38:22 ... phenomenally well on, on the wet lab part. Uh, so that, that's a, a class of things that I'm excited about.

Apoorv Agrawal 38:27 Awesome. Things that you're, uh, skeptical on?

Eric Kauderer-Abrams 38:31 Yeah. So I, I have maybe a kind of generic comment here, and it might sound strange with two people that have a business selling tools-

Apoorv Agrawal 38:37 [laughs]

Eric Kauderer-Abrams 38:38 ... to, to companies that are doing therapeutics, but I'm, I'm generally bearish on, on, um, companies with business models that are reliant on selling tools to pharma. Now, obviously, you can make- [laughs]

Apoorv Agrawal 38:49 [laughs]

Speaker 38:50 [laughs]

Eric Kauderer-Abrams 38:50 ... big businesses successful there, but I think it's very hard. There's very few examples, um, uh, that, you know, of groups that have been successful doing that compared to just doing the thing itself, right? The barriers to actually going and starting to develop drugs, um, have, have never been lower and are coming dramatically down to the point where we like to call it pipeline in a person. You could have a single person using lots and lots of Claude and foundation models actually running a portfolio of early-stage drug programs.

Apoorv Agrawal 39:18 [laughs]

Eric Kauderer-Abrams 39:19 Right? And soon, I think that frontier will advance so that a relatively small team can be running several clinical-stage programs. Um, so I'm just highlighting, you know, of course, you can build great businesses on the tools model, but I think that path is really, really hard. And alternatively, uh, I think democratizing actually developing therapeutics in its, in itself, um, i- is, uh, just getting easier and easier and more accessible.

Apoorv Agrawal 39:42 Wow, what a savage response. You're making sure nobody starts a competitor. [laughs]

Eric Kauderer-Abrams 39:47 [laughs]

Speaker 39:47 [laughs]

Apoorv Agrawal 39:47 Um, do not compete with this guy.

Eric Kauderer-Abrams 39:49 [laughs]

Apoorv Agrawal 39:49 Um, okay. Um, the, uh, you know, you guys have a phenomenal setup, as you said, very, very special spot. Um, w- what would you be doing if you were not doing this? What's your next best idea? Asking for a friend who's not gonna start a competitor to Anthropic. [laughs]

Eric Kauderer-Abrams 40:07 [laughs] Um, well, I think the first thing I would be doing is what I said at the end, uh-

Apoorv Agrawal 40:12 Yeah

Eric Kauderer-Abrams 40:12 ... you know, trying to run a bunch of drug programs as efficiently as possible. Um, because I already said that, I would throw out one more, which, um, I think figuring out how to connect AI to the wet lab is, is a big frontier also, right?

Apoorv Agrawal 40:26 Mm.

Eric Kauderer-Abrams 40:26 So today, Claude can run experiments by-

Apoorv Agrawal 40:28 Mm

Eric Kauderer-Abrams 40:28 ... communicating with CROs. That's great for today-

Apoorv Agrawal 40:30 Mm

Eric Kauderer-Abrams 40:30 ... but we'll reach a whole new level of efficiency-

Apoorv Agrawal 40:33 Mm

Eric Kauderer-Abrams 40:33 ... when Claude can actually directly interface with lab instruments, right? And I think-

Apoorv Agrawal 40:37 Like instrumenting it. Yeah, giving it the arms and legs to run experiments. Yeah.

Eric Kauderer-Abrams 40:40 Yes, exactly, and I think there's a lot of parts to this problem. There's, you know, the hardware layer. There's the kind of software protocol communication layer on top of that. Um, but it feels to me like there's something productizable there to do.

Apoorv Agrawal 40:52 Fascinating. So that's like almost like a programmable chemistry.

Eric Kauderer-Abrams 40:57 Yes, exactly, like the kinda, you know, lab in the box. Like, you know-

Apoorv Agrawal 41:00 Mm-hmm

Eric Kauderer-Abrams 41:00 ... you, you plug this thing in, and all of a sudden, Claude can, you know, control your wet lab.

Apoorv Agrawal 41:04 Is, is there an instantiation of this? Is there evidence of, a business that looks like this?

Eric Kauderer-Abrams 41:09 I think there are, there are a number of companies that, um, are pursuing, you know, mostly startups that are pursuing things like this.

Apoorv Agrawal 41:15 Mm-hmm.

Eric Kauderer-Abrams 41:15 Um, but I think it's the early days, you know, we're maybe a year or two into what I expect to be a many-year journey.

Apoorv Agrawal 41:21 Got it. This sounds more promising than a Anthropic competitor.

Eric Kauderer-Abrams 41:23 [laughs]

Speaker 41:24 [laughs]

Apoorv Agrawal 41:24 Um, what about you, Josh?

Speaker 41:26 Um, I, I think actually very close to that. It's like, uh, how do you actually set up autonomous drug programs, right?

Apoorv Agrawal 41:31 Yeah.

Speaker 41:31 I think the dream will be how far-- It's almost you should start of it, think of it as an eval even of like, "Hey, can we actually come up with a, a set of benchmarks and see how far can the models push on each of these things?" Uh, and I think that converges into, okay, at some point, the models get past a certain mark-

Apoorv Agrawal 41:46 Mm

Speaker 41:46 ... where they've done something really useful. So I think you, you'd even take the frontier problems in drug discovery, many of which we've talked about here-

Apoorv Agrawal 41:52 Mm

Speaker 41:52 ... and then you would keep putting the frontier agents on them, and then eventually you'll get some breakthrough, and you try to push that to market as soon as possible.

Apoorv Agrawal 41:58 Mm-hmm.

Speaker 41:59 One of the interesting things about pharma is it's, it's one of the most competitive industries that's out there.

Apoorv Agrawal 42:03 Interesting.

Speaker 42:03 Right? Because if you think about, uh, what these companies are trying to do-

Apoorv Agrawal 42:06 Mm

Speaker 42:06 ... they're trying to treat disease, right?

Apoorv Agrawal 42:08 Mm-hmm.

Speaker 42:08 And many of the d- it's not obvious how to treat a disease or even, you know, doing the commercial work of, like, which is the best to go after just 'cause it's so complicated, but, like, treating disease is, like, the most obvious business model. That-

Apoorv Agrawal 42:18 That's right

Speaker 42:18 ... mo- most obvious business case. That's actually one of the reasons why I g- just get excited about this stuff. Like, a lot of friends-

Apoorv Agrawal 42:23 Yeah

Speaker 42:23 ... in college were like, "What's the next Snapchat to build, and what's-

Apoorv Agrawal 42:25 Yeah

Speaker 42:25 ... the best company?" I'm like, "Guys, there's a lot of disease out there, right? Why don't we just try-

Apoorv Agrawal 42:28 Yeah

Speaker 42:28 ... to try to treat that?"

Apoorv Agrawal 42:30 Yeah.

Speaker 42:30 Uh, that's gotta be valuable to folks. Um, so I think just writing these things down and then, uh, just seeing how far can the models get there-

Apoorv Agrawal 42:37 Mm-hmm

Speaker 42:37 ... uh, and then just being able to be, to be the first to do that-

Apoorv Agrawal 42:40 Mm-hmm

Speaker 42:40 ... is, uh, probably the other most interesting thing, uh, to do. You wanna be-

Apoorv Agrawal 42:43 Right

Speaker 42:43 ... first in class, and people often call, say b- uh, best in class-

Apoorv Agrawal 42:47 Mm

Speaker 42:47 ... but being the last in class, uh-

Apoorv Agrawal 42:48 Fascinating

Speaker 42:49 ... is, uh, something else someone mentioned to me, uh, recently-

Apoorv Agrawal 42:51 [laughs] Interesting

Speaker 42:51 ... which I think is a really exciting concept.

Apoorv Agrawal 42:53 Fascinating. Um, we'll open it up for questions if folks have any. Oh, here we go. Go ahead.

Speaker 42:57 Um, so I come from a pharma background, small molecules, all manner of delivery. You hear a lot about discovery. You hear nothing about development. Can you elaborate on why you think that is from your vantage points and, you know, what the opportunities might be?

Eric Kauderer-Abrams 43:18 Yeah, I think, um, I think there are a few reasons for that. So first, the o-opportunities in discovery are pretty straightforward. They're, they're hard to execute on, but I think everyone who looks at it can name what they are. The opportunities in development, you know, it's not any one or two super compelling, easy things. Um, I think it's also a lot of things that, that aren't obviously blocked by model intelligence, but are sort of operational, and you need to have all the right connections in place. So my opinion is that, first, I mean, it's less glamorous to begin with, um, and it's less obvious, and it's less obviously related to model intelligence. I would put all those together. I also think the AI community tending to be more, you know, scientific in nature, has less experience with, with that phase of things, and I think that, that plays into it. As far as what the opportunities are, I think there are many. So, um, uh, you know, going through a, a few of the ones that I'm excited about are optimizing patient recruitment, right? So you can imagine, uh, using AI to be intelligent about site selection and looking at prevalences and, and enrollment and things like that, and, uh, you know, ma-making sure that you're choosing the right si-sites and enrolling patients as quickly as possible. There's the trial administration side of things, where, uh, right now there's a lot of operational overhead that goes into site monitoring and, you know, entering and checking records in the, you know, electronic database and all that. So that should all be automated, like, right away and something that, that, you know, Claude should do. Um, and on top of that, I think there's the more scientific components. This is where discovery and development interact of... Well, you wanna make sure that the trial has a higher probability of success in the first place, and you wanna make sure the effect size is as large as possible. So those are, I think, two very significant opportunities that, that, um, relate more to the discovery side, actually. Yeah.

Apoorv Agrawal 45:03 Go ahead.

Speaker 45:04 I had a question. Um, so I guess there are two questions here, but there's talk about these training runs getting bigger. Like, now we're in the billion-dollar phase. We'll go to ten billion, possibly a hundred billion. What would that unlock, um, on this-- on, like, everything you're doing? Like, when we think about LLMs, it's like, well, you give it a book, and then you mask out, like, a paragraph, and it learns to fill it in. But what is that equivalent for the models that you're trying to build? And if you can do a hundred billion dollar training run, like, what would that unlock? I know it's a really general question, but.

Eric Kauderer-Abrams 45:43 Yeah. So I think, um, there's a steady stream of model improvements with, with every model generation. Um, and so the way that I think about it is there's just certain capabilities that we're targeting, and, um, as we do more training, you know, the capabilities get better, and sometimes there are scale effects where there are capabilities that are kind of flat. You know, you don't have them at all, and then at a certain scale, they start turning on. Um, and so, you know, just to give a few examples, like, we found that our models didn't understand much about proteins for, for a while, right? And then with Opus, uh, four point six for the first time, we started to see that, that the models were starting to learn more about, about proteins, right? And that has to do with things that we're doing in training and, and the size and all sorts of things like that. Um, but I think those are, you know, kind of the two trends that are happening here of the incremental improvement in, in many different capabilities and then also the scale effects that we've seen throughout many domains out-- even outside of the life sciences and AI, where sometimes you just don't see something until, until a certain model generation in the first place.

Apoorv Agrawal 46:42 Go ahead.

Speaker 46:44 Um, even in software, the, um, single shotting works very well for so-called shallow stacks like product services and so on. But, uh, our experience has been that, uh, it does not work well at all for deep stacks like databases and distributed systems. And biology seems like the very extreme of that deep stack. Uh, what kind of applications do you, uh, see both in, uh, single shotting actually working?

Speaker 47:14 I, I-- Yeah. That's a, that's a great question. I mean, I think part of this is a question of how, uh, challenging the task is that we can go after, right? And that the, the frontier will just keep getting pushed there. So what you said about, you know, not being able to zero-shot a data-- I mean, uh, that might-- I don't even know if that's true anymore. I wouldn't be surprised if you guys have already achieved that, right?

Apoorv Agrawal 47:31 [laughs]

Speaker 47:32 Uh, and I think the same thing applies in biology. A lot of people-- It's funny, when we started the company, um, uh, one of our, uh, investors did a, a lot of work for us, like, kind of talking to-- doing all the expert networks and saying, "Hey, like, how much would you pay for this sort of thing?" And people were like, "Oh, we wouldn't, we wouldn't pay much for, like, you know, zero-shotting a molecule." Um, and they're like, "Why?" And they're like, "Because it doesn't work. Like, why would I pay for it?" And they're like, "No, no, let's assume it works." And like, "It's impossible." Uh, and they're like, "Okay, so it looks like we're onto something," right? Uh, so I think a lot of these things, it's, uh, it's more a question of-- I think that's what's so exciting in AI right now. You really can dream big. It is possible that what the stuff I have up on the, on the slide over here just, like, is not possible, right? We don't know. But you see scaling laws for these things, and as you build more datasets, as you bring on more compute, they start to become in-- come in reach. Uh, so, uh, I think it's more a question of, like, when rather than if. But, you know, we-- To be honest, we don't know for sure. Uh, and I think that's, that's part of the, that's part of the adrenaline, uh, that comes along with building in this space.

Speaker 48:26 But it seems like the loop has to have, uh, sustained, uh, regulation. It's, it's like a, it's the mega, mega loop source loop. Um.

Speaker 48:39 Yeah. I, I think actually it relates to the last question too. I think one of the reasons why you hear a lot more of, like, AI in discovery is 'cause the feedback loops are shorter. So for instance, if it's in clinical development, it's like, okay, are we gonna go do a whole clinical trial with AI? That might take a long time, versus just to design a molecule and test it in the lab. You know, you can do that in a couple weeks.

Apoorv Agrawal 48:57 Awesome. Well, everybody's gonna be single shotting the drug to get ripped and sleep better. We're gonna wrap it here, guys. Feel free to come find them downstairs. Thank you so much. Really appreciate it. [clapping]