0:00 spk_0
All right. Welcome to a new episode of Opening Bid Unfiltered. I’m Yahoo Finance executive editor Brian Sazi. Like I always say in this podcast, it will make you a smarter investor. You’re gonna get a lot smarter on this episode, and I, I think this is gonna be another one of those episodes, frankly, where we just blow your mind. Uh, Joshua Meyer is the, uh, co-founder of Chai Discovery.Uh, good to see you. For those not familiar, I’ve been following your story. I’m starting to see more from your company. What does your company do and why did you start it?
0:26 spk_1
Well, first of all, thanks for having me on the show. Uh, really excited to talk about this. I mean, this stuff continues to blow my mind. I’ve been working on this space, uh, since I was a kid, really, uh, and it is crazy to see it like finally working in such a big way.So our company Cha Discovery, we’re building a computer-aided design suite for molecules. Uh, what that means is, uh, imagine a piece of software that allows you to design drugs down to the level of an atom on the computer and to design those molecules so that when we go and test them in the lab, um, they actually work with, with pretty high success rates.
0:56 spk_0
Why,why’d you get intothis?
0:59 spk_1
Um, it’s a great question. I’ve been a, I was a programmer when I was a kid, and I actually went to a high school with a stem cell lab. I went there actually to join the math team, but there was a class again freshman year of high school on stem cells, and they called the process of making stem cells reprogramming. We’re able to program a cell into a stem cell, and I was like, I like programming. I got into that. So I’ve just been hooked on both programming and molecular biology since I was a kid.And then it was really exciting to find a way to bring them together.
1:24 spk_0
I’m justthinking back to all 240 episodes we’ve taped so far. I think you’re the youngest co-founder or founder we’ve ever talked to here onthe podcast.
1:33 spk_1
Well, there, there’s many younger ones than I in AI actually, so I can actually feel old among my peers right
1:38 spk_0
now. Well, you started, you started your career, what at Open, was it OpenAI and then you went to,to Facebook? That’s right,
1:44 spk_1
yeah, actually in the same building.
1:45 spk_0
Wow, 770Broadway shout outin New York City.
1:48 spk_1
Yeah, exactly. Um, so OpenAI was a was actually where these things came together because we were working on GPT1 and GPT 2 back then. So these were some of the first language models, right, where you take the whole internet, right? A lot of English language, German, French, whatever, and the models learned to speak that. And I realized like if this works for the natural language, why wouldn’t it work for the real natural language like.DNA and proteins and that’s actually what I came to Facebook to work on. We had a research lab in this building, uh, where we were just doing, you know, this was, uh, 2019 when I joined Facebook. So it was these pie in the sky ideas about what we could do on, on AI and back then applying it for science and we trained those first language models for, for biology and protein sequences.
2:25 spk_0
We probably rodethe elevator together. I started late 2018. We were probably in the elevator. You’re sitting a changing the world and I’m doing like, I don’t know, interviews and stuff. Um, what was it like? I want to get, of course we’ll get into your company, but we are in the midst of justI mean, AI mania and and open AI is just, I mean, they’ve dominated. Like what were those early days like? I’m gonna, I guess I’m gonna date you now. I just called you the youngest ever on this podcast, I’m gonna date you a little bit now.
2:49 spk_1
Yeah, well, uh, back then everyone thought this was crazy, you know, the mission of the company was still to build AGI back then, artificial general intelligence, but, and then there was also, you know, it was a nonprofit, we were kind of going through this transition into a capped profit. So the company’s gonna make money at some point and it was like, how are we gonna make money on this? And it’s like, well, we’ll build AGI and we’ll kind of ask it how to make money. And it was, it was almost this point about, look, the technology is gonna be so big, there’s clearly.Going to be massive value here if we can make it happen. So it was really just a group of people who were just like suspended disbelief about what was going to be, what was going to happen in AI and actually quite a lot of OpenAI success to that. Just that willingness to think extremely big, uh, even when there were a ton of naysayers. All the experts in AI were like, this is not going to happen. We need different methods. You can’t just scale things up. Um, and OpenA was like, we’re just going to try.And ended up working. And that was, uh, I think starting my career in a place like that has given me a ton of inspiration for what we’re doing in Chiy right now as well. To be able to do things like that in a molecular level now, you know, has also required us to suspend some belief of belief for the first, uh, you know, 2 years or so for the company and to actually see that now playing, actually giving fruits with what we’re able to do with the technology has just been incredible. Back then,
3:56 spk_0
didyou ever see the moment we are in now coming?
4:00 spk_1
I mean, I think we all
4:01 spk_0
did. I should sayback then, I mean, it’s not too longago, right? Look, it’s,
4:04 spk_1
it’s the kind of thing that you write it down on a piece of paper and it’s like, yes, like I guess this makes sense. We have all these scaling laws like this is going to happen, but there’s one thing to like, you know, uh, intellectually think it’s going to happen, and there’s another to experience it, and there’s just, there’s nothing that prepares you emotionally to actually see something you’ve been working on for so long actually finally start to work in such a big way.
4:24 spk_0
How did you knowit was time to, to pivot from doing what you were doing?Opening eye, then you went to Facebook and now you’re a chai. Yeah, I
4:31 spk_1
mean it was, it was honestly a bit of like a bunch of even false starts, I’d say. Like the whole AI drug discovery field has been around for a long time. You can probably find like 100 companies that are of AI drug discovery, exactly. And people have tried many times to get this off the ground.And I think what’s different this time is, is the technology has really started to work in a big way. So if you just follow my path through this, so we started with the open AI, right? Let’s do language models. You know, GPT1 and GPT 2 were not that interesting to talk to, right? Uh, you didn’t have these like AI friends, right, who could actually be super entertaining nowadays. There were a couple of, of, of leaps that were missing, but we did the first attempt, uh, back here at Facebook actually with the language models for protein sequences, and that kind of established something would probably work here. The same ideas of like we showed scaling.Laws. We put more data into the models, more compute, bigger models. Uh, it’s starting to learn biology in a more interesting way. I actually had a stint between Facebook and my current company, Chai. I joined a company called Absai, um, which was a, actually it was a cell line development in the company. We were making these little bacteria to actually manufacture drugs, and, and I joined them to help do this pivot into drug discovery using AI. And it was an incredible place. It was one of the first places where we wereUsing AI to make those molecules and test them in the lab. And we started to see things like uh we were designing parts of the molecule back then. We could design 3%, 3% of the of the molecules would work, then it was 5%, then it was 10%. So it’s clearly again another scaling law. Things are getting, things are getting bigger and better. Um, and, uh, and that’s around when we, we decided to start the company afterwards because we realized that if this technology was going to work in such a big way, it would actually enable.new kind of business model. Uh, if you look at the co-founders that came together, we have 4 co-founders in the company with a bunch of different experiences. So I’m kind of working at this molecular biology AI, uh, uh, intersection. Uh, Jack was working at Stripe for many years. Uh, so actually, you know, financial payments, security, and infrastructure is so important. Sorry, molecules. Uh, well, well, it’s, I mean, the point is like there’s a lot of shared problems across technology like the infrastructure and security to make this work for pharma is huge.So we’re taking, you know, bank grade security and trying to bring in here, uh, Jacques and Matt were, were leaders at this intersection of AI research in, in, uh, in biology. Matt had actually built some of the first deep learning models for protein folding, uh, and Jacques was doing a lot of work on AI for chemistry, uh, and actually building molecules together with, with pharma companies around that. Um, so we kind of came together, uh, realizing that the technology, we made a bet that, that the technology would start to work in a big way, right? And, uh, going back to this point.that you asked before, you know, intellectually, we’re like, yes, like this will probably work. It didn’t, we didn’t expect it to work as well as it did today. And also when it actually happens, it is just like ridiculous to think about all the second order effects. So it’s this kind of thing where your ambition just keeps growing and growing and growing over time because once one thing works, you just set a new benchmark for yourself, and that’s been our story for the past few years. I think
7:17 spk_0
to understand what you’re, what you’re doing, it starts with understanding your typical day. Like what do you wake up doing and how does your day?
7:25 spk_1
And there’s, there’s no typical day in this world. Yeah, I mean, the technology is moving so quickly. Like literally what we are capable of today, we were not capable of two months ago, even in Chai. Uh, the technology is moving that quickly, and there’s always another bottleneck for us to figure out. So one day that can mean, uh, working with one of our partners like Eli Lilly, uh, to go and deal for you guys. Um, yeah, it’s really exciting to, to announce that. Yeah. So Lily’s using our technology, uh, working on a, on a, on a large number of, of projects using.Our tech. Um, so it could be sitting down with them actually flying to San Diego tonight and spending some time with the team there. Uh, so it might look like something like that. We’re spending time with our partners. Uh, it might look like, uh, uh, we have to go and build up a large new data set because we’ve identified a new bottleneck with our model, and we think that by scaling data in a certain direction, that could work. So that’s really, uh, we actually had a meeting, uh, uh, we were so busy, we, we sat down, I think it was like 10:00 p.m. till midnight to do a kickoff meeting for this like new data set build that we’re doing.Uh, so it looks like something like that. It could be, uh, sitting down with the research team and when we’re debugging, uh, uh, certain things to make the models train in a more stable way so that we can get, you know, bigger models that take more compute power to get better results. So it’s really a ton of different things happening. Uh, andI think this is what’s so exciting about AI right now. It feels like there’s something new happening every day because it’s so interdisciplinary. There’s either something happening with the data or the models or the partnerships or the labs, um, and, you know, just every day there’s, there’s something new happening and a new challenge to face.
8:48 spk_0
If there’sA new breakthrough in healthcare. Does it come from a company like yours that is using AI in ways that have never been used before?
8:58 spk_1
I mean, I think there’s hopefully a lot of breakthroughs going to happen in this space. There’s also all these second order effects. You know, if we can make molecules now with really high success rates in the lab, there’s still a lot that has to happen before we end up with, you know, a ton of molecules going into people, for example. It’s like we’ve figured out this potential solution to a problem, and now we have to have all these folks go and actually, you know, verify that, start to bring it into their pipelines, for example,The work we’re doing with Lilly, um, but, uh, you know, if we think about what does that unlock next, there are a bunch of problems that people haven’t been able to go after before. Things like undruggable targets. So there’s certain parts of biology we just don’t know how to make molecules against. And if we can use AI for that, then we can start to learn new biology as a result. We can take, oh, I have these 50 ideas I’ve never been able to try before, and now I can try them. Um, so there are things like that that it’s, you know, AI actually opens up opportunity to do things that are even outside of AI.
9:47 spk_0
To the super average human being out there, why the focus on molecules?
9:54 spk_1
Well, I, I think there’s no better thing that we could be working on, right? Like, put yourself in my position, right? Like, been working in the AI space for a while. There’s a ton of action happening across the landscape. What is the most important thing you can be doing with your life right now on this if we can use this to like advanced human health and the human condition? Like there’s nothing more visceral than that. No, but I mean, like, you know, really helping people at the end of the day. Um, there’s a lot of questions in, in AI. Like people are starting, I mean, I have a bunch of friends building companies right now. They’re all doing like really exciting things.Um, but a lot of people would think about like what is the right problem to work on, and for me it’s always been clear since I was a kid. It’s like if there’s some way in order, I thought I’d be a doctor when I was growing up, right? Like to me that’s just a value that’s so important and to be able to work on a technology that’s both very cool, uh, also very good business, but then also to be able to marry that with something, uh, that could really, you know, change the lives of many, many patients. I think there’s no like better cause to be working on.
10:46 spk_0
All right, hang with us,uh, Josh, we’re gonna be right back here on Opening Bin Unfiltered.All right. Welcome back to Opening Bid Unfiltered. Having a fun chat here with, uh, Joshua Meyer, uh, Chai Discovery co-founder. Uh, 12 minutes in, you’ve definitely blown my mind on what’s possible. Yeah, I remember, I was thinking back, I recently talked to a Pfizer CEO, uh, Doctor Albert Borla, and he said, because of the advances of technology, the next big breakthrough will come in cancer. Like if you had to pick vertical by vertical, like where, where is the technology likely to, to unlock like.Just getting rid of these diseases, like where, where’s that coming from?
11:25 spk_1
Yeah, I, I mean, first of all, great, great call to Albert. Actually, Michael Dolston, who was the CSO of Pfizer, uh, joined our board last year. Uh, so we spent a lot of time even thinking about these questions like where is the value going to go? And to be honest, we haven’t figured out an answer because the models don’t care. Like AI can be used across immunology, across oncology, their neuroscience, there’s so many places where we could be applying this, uh, and the models just don’t really seem, seem to care.
11:48 spk_0
Eli Lilly, that partnership you, you announced with them, how does Eli use what you’redoing?
11:54 spk_1
Yeah, so most companies, uh, in our shoes, go to a pharma company and say, Hey, let’s work together on like 2 or 3 products. Give us the hardest things you have. We’ll spend a year or 2 years working on it. We’ll come back with a molecule.And this is something I think that that made, made sense a couple of years ago, right, when the technology was barely working, and there was a lot of handholding you had to do around it, and you had to do a lot of back and forth between the AI and the lab. You, you couldn’t just go to an Eli Lilly and say, hey, here’s my AI, like, go do something interesting with it. It just really wouldn’t have worked. But now that we’ve had such a, such a step change on the, on the capability side, it’s enabled us to do these new kinds of collaborations where Lily’s not just working on two or three projects, uh, but really.Working on dozens across their portfolio, uh, and it’s really exciting to work directly with their teams and empower them, uh, with the newest form of AI. I think overall, if you just look at the pharma landscape, you know, a lot of, uh, the way pharma has engaged with AI before has been like, uh, you know, we need to be at the forefront. Just backing up for a moment, if you’re in pharma, uh, there’s this thing called loss of exclusivity, you know, if you don’t keep innovating on your drugs, eventually you will go out of generic. Yeah, exactly. So you always need to be at the forefront of innovation.So farmers always try new things. They’re trying to see what’s working. Um, and they’ve been doing that for the past 10 years in AI drug discovery. They finally reached a point where it’s working, and they actually need to start internalizing some of these capabilities. They need to have these models, uh, being used by the teams themselves. It’s the same thing with like, look at, uh, people using chat GBT these days. You’re not just going to go to an AI consultant and have them use AI to do your work for you. Everyone’s using that as a daily tool, and that’s about to start happening in drugs. It’s happening now in drug discovery.So that’s why that partnership with Eli Lilly, it’s focused on actual broad deployment of the capabilities. We’re actually even working together to co-develop a frontier model together using our frontier AI models together with their historical data as well. So these are the kinds of opportunities that are now opened up by models that are just working so much better than they did even a year ago.
13:41 spk_0
If we’re having this conversation 12 months from now, how will your models look different than they do today, you think?
13:47 spk_1
Yeah, well, we’re, there’s a lot that we’re going to figure out this year. There’s things.Like coming up with, uh, models that have higher success rates, just to contextualize this, a year, literally a year ago, um, we had a 0.1% success rate for making these molecules in the lab with AI. So, 1 in 1000 or 1 in 10,000 molecules designed by AI would work. Uh, now we’re at 1 in 5. and I don’t really see a reason according to the physics why we can’t get to 1 in 1, right? Uh, and that’s where we’re actually doing a lot of that data build right now because even though again we’ve reached that point where this is really useful, this technology shows no.Signs of slowing down. So I think we’ll have better predictability. So not only just better molecules coming out, but can we have AI models that predict what the impact is going to be in the lab? Can we build, like, you know, simulations of the experiments we’re doing in the lab? Like we’re getting to this pretty crazy thing, ideas right now. If I said this on your podcast a year ago, none of the, I don’t know what he’s talking about. Well, well, no one would take me seriously, right? It would be like, uh, there, there’s no way that that’s possible. Actually, even when we were doing like customer discovery for the company and asking people, do you think this capability would be useful.Be like, no, I, I don’t think it’s useful. We’d ask why, and they’d be, oh, well, it’s impossible for the models to work that well. And we’re like, no, just like, again, suspend disbelief for a moment. What if it would work? Uh, and people are like, yeah, I guess so, but there’s no way, you know, it’s actually going to work and therefore we wouldn’t buy it. And it’s, it’s crazy to see that working now. So, uh, actually one of the things I told the team at our, our company holiday party, uh, was we have to think bigger this year. Like we told our investors in our Series A that we would get to a 1% success rate in any body design because the state of our year ago was 0.1%, right? So it’s saying we’re going to get to a 10x improvement.Within like 2 years, that was already pretty crazy and most investors thought that it wasn’t going to work. Um, and within a couple of months, uh, we managed to get to 10 to 20%. So even 10 times higher than we shot, and it happened 2 to 3 times faster. So I think there’s, if that, if that was the slope in this past year, then you can imagine like what’s going to happen in this year. Is
15:38 spk_0
a big part of yourjob finding out ways to raise more money, like to achieve what you want to achieve. I mean, you’ve raised money, you’ve had success doing your unicorn status.Don’t you need more of those funds to scale up this operation as fast as possible?
15:52 spk_1
Yeah, so the, the GPUs are, are, are expensive. Uh, that’s the reason why I shout out a video valuable company in the world, yeah, uh, and the data sets are expensive too. There’s this cool thing happening now where we actually use the AI models to make the data sets. So we kind of ask the AI like what data do you need, uh, and then go and, and, and use the models to, to generate that data with higher efficiency. But all this costs a lot of money, um, so it’s one of the reasons why we’ve raised a lot of money here. I think the.Side of that though as well is that uh hopefully we’re able to start doing more and more of this in partnership as well, right? So this is one of the reasons why uh we, we love working with, with companies like Lily, right, where we can actually kind of pioneer uh these things together because there’s a lot you can do when you bring, you know, the, uh, the expertise and, and data of a big pharma company together with like the fast moving AI and GPUs and all that that you have in a company like Ji.
16:39 spk_0
You must havesome general view on where is this all, where is this all going, AI, where, where, what’s, is there an end?Place we’re all going to be at
16:48 spk_1
Well, I don’t, I don’t see why we should put an end on this, right? Like the, the tech is going to continue getting better and, and like I was saying before, every time we reach a, a certain blocker or a certain frontier, there’s always the next benchmark to move forward. So, take what we’re doing with, with molecule design. So, we wanted to hit that 1% success rate on, on antibody design because that would allow us to uh stop doing like high throughput screening. You wouldn’t have to screen 10,000 molecules. If you want to screen 100 molecules, you can kind of do that by hand. So that’s what we were thinking originally.But then once it worked, we realized, wait, there’s all these really hard problems that we can’t go after previously because there’s just no way to screen for 1000 molecules against this target. We can only screen 10 at a time. Um, so, you know, AI enables us to do that. We can start to think about what are new, uh, biology problems that were impossible to go after before. Uh, maybe there’s a certain disease area that we don’t even know how to tackle. It’s not even finding the molecule, we just don’t even have the idea yet.Well, can we come up with ideas by just brute forcing lots of things using AI? Can we actually put AI on the outer loop of that and have it even tell us like what are the kinds of experiments and ideas to do there? So I, I think that that’s just been the story of technology, uh, you know, in, in all of like human history. Like whenever we find something, we find ways to build on top of it and go solve the next problem. So I, I don’t think anything is, I don’t think that’s going to stop anytime soon.
18:01 spk_0
What isAnd we have a lot to be thankful for in this country, but what, what is broken in drug development?
18:08 spk_1
Well, drug development is a, is really hard. Like the fact that we can discover drugs at all with the technologies that we have today, uh, is honestly one of the most inspiring things ever. Like you go talk to, uh, folks like, like Michael Dolson on our board who’s discovered like dozens of medicines that have been approved, right? And, and you talk to Michael and like just the intuition he has about how to make these molecules just incredible. So I love spending time with him, and I think it goes visa versa because it’s like he has all these ideas and then the AI models can come up with other ideas that are interesting too.Um, so, so what’s like broken in drug development, I think it’s just a really hard problem. And when things are really hard, then you end up with a bunch of inefficiencies along the way. So, even something like being able to discover a molecule faster, if it takes 2 years to discover a molecule, versus take 2 weeks, um, that changes the kinds of ideas that you’re willing to go after. There might have been something that was so hard before that it just, uh, actually, let’s put it on a financial level, right? Uh, if you look at the NPV of many new drug programs, actually negative. Like most drug programs, if you put it on aACF, you just should not invest in it.
19:07 spk_0
Discountedcash flow for all you financial analysts out there.
19:10 spk_1
Yeah, yeah, exactly. So, uh, it actually does not make sense to work on most drug programs, uh, but, uh, if you imagine now using AI where the probability of success goes up, the timelines go down, there’s many things that didn’t make sense to fund before that now do make sense. It might even have positive NPV just from the idea alone. So there’s a lot of dynamics in this space that I think will really start to change because of AI.
19:31 spk_0
Two fun ones, uh, here at the end. One, what, now that you, if you can look back, is there one moment that changed the arc of your career?A conversation, a meeting, a day, something.
19:45 spk_1
That’s a great question. Oh, there’s so many, uh, to me, I’ve just been fortunate to work in so many of these incredible places. So, you know,
19:52 spk_0
it’s hard. What, what’s like it’s hard
19:53 spk_1
to pinpoint one
19:54 spk_0
idea? What’s the best advice on us, Sam Open Eyes back to your company. Has Sam given you, Sam Altman, given you like amazing advice that you still adhereto every single day? Yeah,
20:02 spk_1
actually, maybe I can take a story there. So, so back at OpenAI, you know, when I was leaving to join Facebook, we actually had some of these conversations with Sam about should we go and do something around like proteomics, like and, and actually use AI to do protein design, you know, back, back then.And kind of decided that the technology wasn’t ready yet. So I think there’s this, uh, maybe the lesson there, and it’s the same with AGI, it’s like you have to think really big, but you also need to hit the iron when it’s hot. So it wouldn’t have made sense again to go and put these AI models into people’s hands like 3 years ago because it just didn’t work yet. And I think one of the reasons why we think this will be the year of deployment in AI drug discovery of AI models actually being, being deployed is that the technology is finally working.One thing that uh Ilya had actually said to me, this is um I think my first week of the job at OpenAI, and he said that uh any experiment you start, you should try to make sure it finishes that day.And I, and that’s something I’ve taken with my throughout my career. It’s not even just like training AI models, but you need fast feedback loops. You need to be able to try things and get feedback from people. And this is why drug development is so, so hard, right? If it takes 2 years to discover your drug, how are you getting those like feedback loops? And if we can just, you know, kind of compress those timelines and go after harder problems, we can just iterate through a lot more and then hopefully come up with much better solutions.
21:13 spk_0
One more before yougo. I mean, you seem like a pretty intense, energetic dude, which I can appreciate. Like I’m there with you. What do you do to relax?
21:19 spk_1
Um,
21:19 spk_0
it’s not even possible.
21:21 spk_1
Yeah, I, I like to go running, to be honest, uh, and I, I feel bad saying this because my co-founder Jacques is literally like a marathon runner. Uh, I think his, uh, marathon time is like 2 hours 40 minutes or something like that, probably even faster now.
21:34 spk_0
But you gotta do a highrocks. I’ve gotten big into high rocks. I’m doing my first race in New York City this summer. I am jacked,
21:39 spk_1
man. Interesting,
21:39 spk_0
yeah, yeah,
21:40 spk_1
yeah. We’ve got, I think we’ve got a lot of runners on the team as a result.You know, it’s the one thing that it’s like, OK, you take that intensity and you channel into something else you just can’t think about, uh, you know, work anymore if you’re doing that, and
21:49 spk_0
it givesyou data like I did the boops job. You can collect your data, like see what you’redoing. Yeah,
21:53 spk_1
we’re AI researchers love data, but like it’s, it’s hard to relax with everything going on right now. This is not the time to relax. This is the time to like go really hard because this is, this is the year when things are happening.
22:03 spk_0
If I was.Asking you that on an interview, that is the right answer. Uh, I’m excited for you. I can’t wait to follow, uh, continue to follow your story. I think you’re working on some absolutely amazing things, transformational things, important things. Joshua, good to see you. Thank you. I
22:14 spk_1
appreciate you having me. All
22:15 spk_0
right, and that’s it for the latest episode of Opening Bit Unfiltered. Uh, please continue to hit us with all that love on YouTube and the podcast platforms. I love it all. It makes me better at doing these interviews. We’ll talk to you soon.