Transcript: How AI Is Unlocking Materials We’ve Never Been Able to Build | Radical AI

Host: What are the typical customers who are looking for new materials? I mean, every single company on Earth.

It doesn't matter if you are an aerospace automotive manufacturing, defense, climate, semiconductors, energy, or even something like athletic apparel and consumer wear, but all of them feel this problem of very long timelines, typically 10, 15, 20 years, and incredible amount of cost north of a hundred million for a single material system.

Host: We don't just test a few materials a year. We test tens of thousands of materials per year.

Host: Yesterday is Joseph Kraus. Uh, he's the co-founder and CEO of Radical AI, a company that is trying to change what's possible with material science and how we discover new materials.

Host: This vision to accelerate material discovery attracted a $42 million fundraising round from RTX Ventures and Nvidia's VC arm.

Host: You mentioned general and DX speed in the flywheel. Where does the speed advantage coming from it?

Guest: In materials, this is challenging. That is why making them are incredibly important.

Host: Then startups come up with, you know, big ideas and go pursue very big ideas, it's easier to fundraise than in proposing a new mobile app.

Host: You got some funds from different companies or from Ale Corp, you have a good founding thing.

Host: But what the product that Radical AI creates, what was the next step?

Guest: I think there's a third option, which is very unique, and only a few materials have this opportunity.

Guest: This is something like a room temperature.

Guest: So the industries that will impact or benefit from that are endless. (Sound effect: Explosions)

Host: Hello everyone, welcome to Startup Project.

Host: In startups, we have this famous saying called software is eating the world.

Host: I think today AI is eating the world, and it's eating every sector, industry, and potentially continue to be that.

Host: Uh, today I'm a podcast, we talk about a deep tech company called Radical AI.

Host: A company that is trying to change what's possible with material science and how we discover new materials.

Host: Uh, yesterday is Joseph Kraus, uh, he's the co-founder and CEO of Radical AI.

Host: Uh, the goal of Radical AI is to accelerate material R&D by integrating AI engineering and robotics.

Host: Um, their vision is to create a self-driving lab, uh, where AI doesn't just assist scientists, but it also autonomously designs tests and discovers new materials.

Host: Um, this version that to accelerate material discovery attracted a $42 million fundraising round from RTX Ventures and Nvidia's VC arm.

Host: Uh so in this episode, we'll dig into their uh material's five-bleed concept, how are they increasing the speed of material discovery, their open-source engine and more.

Host: With that, Joseph, welcome to the show.

Guest: Absolutely, thank you, sir.

Host: Joseph, welcome to the show.

Guest: Absolutely, thank you so much for having me.

Guest: I appreciate it and excited to be here.

Host: So you have a kind of a diverse uh background from material science, working in Army National Guard, um little venture capital at Alicorp, who I'm assuming is also an investor in Radical AI.

Host: How did these different sort of experiences converge and led to building or starting Radical AI?

Guest: Yeah, great question.

Guest: I always love to talk about the quote from Steve Jobs of you can never connect the dots looking forward.

Guest: Only looking backwards do all the dots kind of connect in a single line.

Guest: And for myself, I I very much feel this present today.

Guest: So, going back to when I was in graduate school, I was pursuing a material science PhD at Rice University.

Guest: I was working on a lot of what we call functional materials or materials that can be used in real applications.

Guest: And I was frustrated with my lack of ability to push things into industry.

Guest: And that I actually isn't a uh in you know, a poor performance from academia.

Guest: It's just not what academia is truly focused on, right?

Guest: Academia's job is to drive our fundamental understanding of science, and for me, I wanted to take that fundamental understanding, work on applied applications in materials and really put them into products in the world, and I wasn't seeing the connection between those.

Guest: And at a similar time as I was going through my PhD, I was separately serving in the US Army National Guard.

Guest: Army National Guard is a kind of a part-time military service where you're kind of can get called up into active duty if needed.

Guest: And I've always been a big fan of the military.

Guest: I always wanted to serve in the military, so it was a very important thing for me.

Guest: So I had done that where I was doing that as I was in in my PhD.

Guest: I had this opportunity come up from the Army Research Lab, which is the US Army's corporate development research lab.

Guest: One of them is located in Adelfi in Maryland where, hey, we are doing materials research, but we are thinking about problems for the army and how these things can at one point impact the future war fighter and the future technology of defense.

Guest: And so I actually got really interested in, wow, I can do science in kind of technology that I care about the impact of, right?

Guest: I'm separately serving in the National Guard and I care about the war fighter having the best technology.

Guest: In this sense, the war fighter was myself and and others and including friends and colleagues in the in the military.

Guest: Now, let me go do that research there.

Guest: And so I took a leave of over or kind of went to do a journeyman fellowship at Army Research Lab, working on neuromorph computing and how do you apply this work.

Guest: And even here, it's still fundamental, because again, it's still at the core cutting edge of research.

Guest: And so I still felt this understanding or this lack of ability to execute on translation and pushing things out of research into the market.

Guest: So I took a leave of absence for my PhD, I joined Alicorp, which as you said is a VC based here in New York City, and I started investing in material science, semiconductors, 3D printing, informatics, a wide range of areas inside materials, really to try to get at how do I actually just try to commercialize and help commercialize materials.

Guest: And finally, while at Alicorp, me and one of my other co-founders, Horacous, who was also at Alicorp at the time, kind of came up with this inception for Radical AI.

Guest: We found found our third co-founder, Huder, and we rolled out and started the company from there.

Guest: So again, looking back, very linear and kind of makes sense trajectory.

Guest: At the time, I had no idea the next step I just walked through was coming when I was in the current phase of what I was going through.

Guest: So that was the the process to get to the founding of Radica.

Host: So Paul, looking back, what about all conditionally so going things to be at a army uh you know, institute that were preventing this commercialization of new materials, right?

Guest: Yeah, I think two things.

Guest: One is a focus, as I mentioned, the primary purpose is not to commercialize application.

Guest: It can be an outcome of academic research and this is why there are tech transfer offices and a lot of commercialized science coming out of academic institutions.

Guest: It's it's an opportunity or an option coming out of research, but it's not the core driver.

Guest: The driver is understanding and discovering new fundamental phenomena.

Guest: I mean, that is truly what academic research is focused on.

Guest: So that's number one.

Guest: And then number two is the commercialization of science typically, maybe always, ends up in a product.

Guest: And so if you think about silicon or semiconductor transistor technology and the materials research behind that, it goes into a semiconducting chip, which is then put in an advanced processor or packaging facility to put a bunch of backend connections to that, so we can use it in a smartphone or a computer.

Guest: And so if you work on transistor technology, which I did in my PhD, I worked on neuromorph based material systems, you're still a couple of steps removed from the product that will actually utilize that technology or those materials specifically.

Guest: And so that disconnect is hard to bridge and it's quite removed from fundamental to application.

Guest: And so because you don't have this focus and because you're not really building the end product, you're more focused on the science of what's going to drive that end product, you just end up at a place where you can't really solve through commercialized problems.

Guest: And the ones that do, think about nuclear fusion, uh, as an easy example.

Guest: Of course, they take that scientific invention and they start using it in a real product, right?

Guest: Like a reactor, a nuclear fusion reactor.

Guest: And so we've made this discovery of how to make fusion power, and now we're putting that inside of a nuclear fusion reactor and that technology is being utilized in that sense.

Guest: So bridging that gap, in my opinion, really requires private enterprise.

Guest: I think that's why you've seen people like fusion all the way through to space, with space X and blue origin and other companies there, building private companies on really advanced scientific and technological capabilities.

Host: I think I mean the proposition is very attractive, right?

Host: Uh when it's category like, you know, discovering new materials, sounds a little bit impossible, but also very exciting.

Host: It's almost I would say in startups come up with, you know, big ideas and go pursue very big ideas, it's easier to fundraise than in proposing a new mobile app.

Host: That least got my thesis.

Host: So I'm doing Uber for salon, like Uber for something else, which is technically like I'm going to do Uber for material grooms.

Host: And you kind of pitch a bunch of inggies, and that's going to be a tougher fund race than, um, you know, I'm going to redefine an entire category whether it's material science or something private space.

Host: It would become initial but I feel like you can quickly ton the momentum.

Host: On your said because ambition isn't really good when you're going for this, you know, moon shot scene and in summary, you are in sort of like a moon shot way.

Host: How does, you know, do you have, I'm assuming let's say, let's skip the fundraising part because you got some funds from different companies or from Alco, yeah you have a good founding thing.

Host: Now, how do you go about what's the next step?

Host: What did you create a new word?

Host: And if it's a Uber for something, I will go and create a new app.

Host: What what would the product that Radical AI creates?

Host: You got your three people, what was the next step?

Guest: Yeah, so great question.

Guest: This is a super important thing to identify.

Guest: For us, the most important thing to think about is technology, and then from that technology, what is the product you develop.

Guest: So what we create is new materials, and we want to commercialize those materials, so actually sell those materials at scale.

Guest: And the area that we are working in is very much an aerospace, defense, um and and energy where we use these structural metals to actually build these components.

Guest: And so our product, the output of our technology is novel materials that enable new applications.

Guest: And so that's where we're focusing.

Guest: The technology piece of that is how we do that.

Guest: And I think that's really important.

Guest: If you think about SpaceX, I'll just use them, I love them as an example because they're a great company, they focus on bringing things to orbit and at one point getting to Mars.

Guest: And the way that they do that is really advanced to rocket capabilities that are reusable, that are more cost-effective to build, and that have different payload capabilities from Falcon 9 to Falcon Heavy to then Starship.

Guest: And so that technology is the mechanism or the means of how they actually get things in the space and get and get and build a Mars colonization.

Guest: And so if our end goal is to develop incredible materials that enable new applications, the way that we do that is our materials fly with.

Guest: And this is where this AI modeling and engine and our fully robotic self-driving lab come in, they make up that materials flywheel and it allows us to do materials discovery faster, 370 times the speed of a human scientist, more interconnected.

Guest: So bringing in different data sets and different sources of information like patents or scientific literature to make new hypotheses.

Guest: And then lastly, actually build a high throughput capability where we don't just test a few materials a year, we test tens of thousands of materials per year.

Guest: And so when you bring all those together, you have this materials flywheel that allows you to very quickly discover novel materials, and then from that, really go into applications that will directly use those novel materials for new improvements or advancements in performance or cost or something else.

Guest: And so that's what radical is working on and again, our end product is these novel materials that we will sell.

Host: I'll say two follow-up questions.

Host: One is what are the typical customers who are looking for uh new materials.

Host: And one obvious answer is, yeah, US Army going research, but commercially speaking outside army, who are the uh customers who are generally thinking about new materials.

Host: And the second question is, you mentioned, you know, 300x uh 370x speed in the in the flywheel.

Host: Where does the speed advantage coming from?

Host: Is it because uh yeah, I was getting.

Host: I think Zuckerberg was talking about this, if you can reproduce a virtual cell or virtual molecule, then you can do a lot more experiments that would otherwise be manual, step-by-step in real world, but you can simulate that in a biological environment and I think they're revolting their whole nonprofit around this idea of using AI and simulation to do real life experiments that would otherwise be in real life parallel.

Host: So tell me about those two things.

Guest: Yeah, great question.

Guest: So the first one who buys materials.

Guest: My favorite question to answer because it's every single company on Earth.

Guest: It doesn't matter if you are an aerospace or automotive, manufacturing defense, climate energy, semiconductors energy, or even something like athletic apparel and consumer wear, every single industry in the world is a direct result from novel material advancement.

Guest: But all of them feel this problem of very long timelines typically 10, 15, 20 years, an incredible amount of cost north of a hundred million per single material system.

Guest: And so in short, you arguably have one of the largest markets in the world with novel materials.

Guest: We just cannot execute on it fast enough.

Guest: And so for us, anyone can truly be a customer, that's from the RTXs and Boeings of the world, through to the Lulu Lemons and Vioris of the world who also use materials.

Guest: And so that's kind of the opportunity.

Guest: On the second piece, really good question on what I would call digital research, right?

Guest: Or digital twinning research or for lack of a better phrase.

Guest: In materials, this is challenging and we cannot only do digital-based research.

Guest: And the reason why is so much of the know-how in making a material at scale is not just how to make the material, but it's how to make the material in large quantities and with the properties that you need.

Guest: What we call the processing of a material.

Guest: And when you go and look at some of the biggest material companies in the world, most of their IP is trade secrets around the processing of those materials.

Guest: Yes, they file patents on compositions and the ways that they make those materials, absolutely, but it's really the know-how on we make this material better than anyone else in the world and at a better cost than anyone else in the world.

Guest: And so that's kind of where the value in materials lie.

Guest: So if you never make it in a lab, see the properties it has for an application, and then start to understand the conditions to actually go scale that, temperature, pressure, oxidation concentration, all these different mechanisms or or I should say like variables that you need to consider when you go to manufacture, then you truly can't capture the real value in material science.

Guest: So we do a lot of stuff on the digital side.

Guest: We use a lot of machine learned models, like machine learned potential which allow us to simulate crystal structures.

Guest: We use generative technology to really do inverse design.

Guest: So we'll use a diffusion-based model to take a bunch of properties and then inversely design structures from those properties that we think will have them.

Guest: But we still at the end of the day need to go make that material in a lab.

Guest: And that's kind of even a huge challenge of material science.

Guest: To give a very specific example, if you think about an end customer, and thinking of an automotive company.

Guest: They want to put a new lightweight aluminum so that their cars are not as heavy, so battery technology goes farther.

Guest: I can show them the coolest aluminum replacements in the world on a computer screen, but until they really have that, they can put it through their processing and manufacturing line, and then they can run the test required from a crash perspective, a performance perspective and make sure they can really sell automobiles that use this material, they haven't discovered that material.

Guest: And so this this gap in materials is the hardest thing to understand and where all of the value sits in the material's industry today.

Guest: That is why making them are incredibly important for us.

Host: So well, one things um I was curious about is, how do you pick?

Host: Because when you're trying to, you know, come up with the new material, are you let's say you, you have this specific assignment from US Army to do yeah new alloy of some sort.

Host: They might have certain needs what this new material should do, right?

Host: and a specific combination and specific performance yeah if it's on heat survival or core survival uh in a different dimensions.

Host: So you can pick different scenarios like if if I'm a Lulu Lemon and you know, trying to get a material which is not sweat resistant which on kind of that.

Host: So which direction are you pursuing?

Host: There's obviously a direction when someone gives you a contract to pursue in a certain direction, you'll go and so for that direction, but otherwise are you only going into directions where you have pre-commitments from customers who are looking for certain types of materials or then you have your own ideas of okay, I think we need to find, I don't know, completely new material that's going to change, you know, how Nvidia does its chips.

Host: Like which directions do you pick?

Guest: Great question.

Guest: A very important question because as an early stage startup, focus is very important.

Guest: You cannot boil the ocean and so many companies fail or struggle to find success in their application because they do not focus on actually getting the market and showing the value what the company can build.

Guest: So, for us, it's actually both of those things.

Guest: And let me explain why.

Guest: We make sure the materials that we work on or the classes of materials that we work in are directly driven from customer.

Guest: We do not want to randomly work on materials with no customer and then problem from that customer identified.

Guest: In that sense, academia is a better vehicle for that, right?

Guest: In terms of fundamental discovery.

Guest: So that's the first thing that you mentioned there is these problems that we're getting, they are directly received from customers.

Guest: We do not um arbitrarily pick those problems.

Guest: Second reason, the reason why the answer is both, the second thing you mentioned is it's up to us to come up with what the new material is to solve those problems.

Guest: And so a perfect example of this, we work in the hypersonic space.

Guest: And hypersonic missiles have a problem with alloys that they use today.

Guest: Essentially, the alloys that we're using cannot keep up with the CMC composites that we have inside these missiles.

Guest: So there's this huge problem of how do you drive higher heat and better mechanical performance at high heat or high temperature for these systems.

Guest: That's the problem.

Guest: And it's from the end customer.

Guest: But the solution that we think and that we have come up with and the industry has has thought about is what is called a high entropy alloy.

Guest: And so we are actually designing novel high entropy alloys for this application.

Guest: And other people have tried different types of materials.

Guest: People have looked into moving the more aggressive CMCs, ceramics on their own or ablatives, alloys generally and comes with super alloys, nickel-based alloys or something similar.

Guest: It's both of those things where we are taking the problem from an end customer, absolutely, and then we are thinking about in which way or what material system can we work on that we think will kind of solve that problem.

Guest: And we think there's huge opportunity and a huge market that if we do solve it, if we do discover a high alloy, there will be hundreds of billions of dollars of value created from that.

Guest: And so that's where the both those things come together.

Guest: There is a third option, which is very unique, and only a few materials have this opportunity.

Guest: This is something like a room temperature superconductor, right?

Guest: So the industries that will impact or benefit from that are endless.

Guest: This is things like losses nuclear energy, floating trains.

Guest: I mean, you name it, right?

Guest: Like you come from our type superconductors.

Guest: There's really not a market for those things today.

Guest: There are not no saying, I want to install this in the next two years, in the next 18 months, in the next six months.

Guest: But if you had an our tap superconductor, obviously all of these companies would be directly interested in purchasing and acquiring and using that in their applications.

Guest: We kind of put those as what we call enabling based technologies.

Guest: And we do look at and are thinking about where we will use or where we will work on enabling based technologies.

Guest: However, focus.

Guest: And that's why I started with this point to your question of focus.

Guest: While we know and are interested in our temperature super conduct uh sorry, room temperature superconductors is an application for Radical, that is not the first place we are starting.

Guest: In the next two years, Radical AI needs to show the world that we have the best scientific discovery flywheel in the world, and with that flywheel, we drive real commercial value.

Guest: Otherwise, we're not a good business.

Guest: Once we're able to do that, then we can focus on more enabling type things like an our tap superconductor.

Guest: And so we describe this as a stepwise based approach.

Guest: You start with the first and you can move to the second as you develop capability and as you become the industry leader in scientific discovery and implementation.

Guest: That's how we pursue that.

Host: I think a couple of years back, there was a paper from uh, I don't know if it was a Chinese or research team that started became viral and once people started taking that theory and testing, the people claim that, you know, we could solve reproduce a super conductor at room temperature.

Guest: Are you talking about OK 99 specifically?

Host: I think so.

Host: I don't remember the name, but there was a whole boom around for a couple of days.

Guest: that's going to be it.

Host: It's excited.

Host: Um so you're talking about the similar uh superconductor, right?

Guest: Yes, and that's a perfect example of where experimental validation of a material is very important.

Guest: I talked to you a little bit earlier around in materials, you had this hard problem where you really have to make it.

Guest: This is a great example of this paper, I believe it was out of South Korea, I think.

Guest: I'm not I'm not 100% sure.

Guest: Made this LK99 and they claimed it had room temperature superconducting ability.

Guest: And we had run or people had run, not radical, people had run in DFT and other simulations to try to confirm that this was truly an our tap superconductor.

Guest: And some of those confirmed or had thought they confirmed, yes, this really seems to be an our tap superconductor.

Guest: When it came to rebility and experimentation, of course, we did not confirm or we could not replicate that material and there's a whole report that's been put out of what people actually think it really was.

Guest: Um you can go read that, but there's a perfect example of where experimental work is equally as important as the simulation work that we can run.

Host: So uh what time?

Host: I mean, because you're trying to combine technology sort of you're creating a new process and how we're making discoveries.

Host: What type of people are you hiring?

Host: Are you hiring?

Host: AI people?

Host: Are people?

Host: Are you hiring more researchers or also to extend that are you also what type of AI are you using?

Host: Are you building your own models?

Host: Is it traditional machine learning models or generative AI models.

Host: A little bit about like the AI part of Radical AI.

Guest: The people that we hire spend five technology buckets, machine learning and AI, software engineering, automation or robotics, mechanical engineering, and then material science, both computation experimental.

Guest: We hire across all five of those technical buckets, and building an interdisciplinary team is imperative to Radical AI.

Guest: One of our huge differentiators is that we hire across all of those buckets and we work the intersection of all of those fields to drive advanced scientific discovery.

Guest: It's a very important piece and thesis of the company.

Guest: And so that is where and how we've hired the team today.

Guest: On the AI front, we use a multitude of different AI technology, um that really kind of forms into for lack of a better phrase, a multimodal based approach.

Guest: We have an agentic system that sits at the top layer.

Guest: And this is LLM technology.

Guest: We do not build LLMs from scratch.

Guest: We use the big AI lab's models and fine-tune them for our application, and that is what orchestrates our entire AI engine.

Guest: And those agents have access to different tools that we have built.

Guest: They have access to graph neural nets, which in our field, we use for machine learned autonomic potentials.

Guest: They have access to generative-based models.

Guest: We use mostly diffusion today to actually do this inverse design problem I mentioned.

Guest: They do use LLMs themselves for crystal structure prediction, although most of that that's more of a novel field and we haven't seen as good results from the LLM prediction as we've seen in something like a diffusion-based prediction, but we do look at that technology.

Guest: And then we'll use kind of I guess what you would call old school.

Guest: It is old school AI.

Guest: On the computer vision side in our lab, we are tracking how we do experiments and results of experiments with cameras.

Guest: And we are building ML models to very quickly analyze that data.

Guest: We use open source computer vision models that have been built a long time ago.

Guest: Like not cutting edge technology here, as well as something like basing optimization, a perfect example.

Guest: Now, we look into replacing Basing optimization with LLM-based technology today, but in our design of experiments loops, which is this hypothesis loop that we make, Basing optimization has been used for the last couple years in science to really find the right next best material to go make in the lab.

Guest: And so our AI engine really spans a wide range of technology from cutting edge LLMs or generative, you know, diffusion-based technology through to some of the older architectures like BO, which have been around for a while are well understood and we don't care about reinventing those.

Guest: We care about using them to drive our flywheel faster.

Guest: And so it's a mix of technology across the board.

Host: You know, uh cool uh deep mind and I think they're sort of had learning to be a big tech company called Isomorphic Labs.

Host: Uh and they have Alpha 4 3, which is sort of like the new AI model and that both companies are developed and it sort of predicts the structures of proteins, RNA, and other things and how they interact.

Host: Is there like an opportunity in even materials and space or is anyone pursuing that problem or uh do I have model?

Host: I don't know, I might be out of my depth here, like predicting crystal structures or predicting new material structures and that prediction maybe you can take that and test it in real life and yeah created is that like a real thing you see?

Guest: Absolutely.

Guest: That's a that's a very real thing.

Guest: And that's been deeply starting to be deeply supported in material science.

Guest: So we do that internally.

Guest: We have our own models that do that directly.

Guest: So we are predicting crystal structures.

Guest: Big tech also has that.

Guest: So Microsoft has a model called Matter Gen and that allows you to use generative technology to predict new structures.

Guest: They also have a Matter Sim model, which is more on that atomistic modeling side, although relevant when you're thinking about structure prediction.

Guest: So that's theirs.

Guest: Google has a model called Gnome, which is for structure prediction.

Guest: that is that is open as well or that they put out, they published paper on as well.

Guest: So that's another one.

Guest: Meta has a lot of MLIP-based models that they use with their data sets they put out.

Guest: They have this open catalyst project and they have the open materials data set.

Guest: They just put out a new one this year.

Guest: We use a lot of the that data in our training models, but but they have models as well that are built for for prediction.

Guest: So I'd say the field is is growing here rapidly.

Guest: There's been some push back on some of those models in the scientific community that some of the predictions are not novel, so we've seen that material before or they're not valid.

Guest: We actually cannot make those materials in the lab.

Guest: And so in our technology, as well as other technologies that are coming out now, these two constraints are really important.

Guest: What is the novelty of the material and what is the validity of the material?

Guest: We make sure to vet against those two constraints when we're thinking about structures to predict.

Guest: But yes, to your question, absolutely.

Guest: This is a huge burning field of material science and it's probably going to be one of the more impactful things in material scientific discovery over the next five years.

Guest: There's a big problem with a lot of those models and a lack of experimental data.

Guest: And so those pushbacks I just gave you come from this lack of experimental data.

Guest: Those models are trained almost exclusively on computational data.

Guest: And that is good because it is relevant and we do run a lot of DFT and MD here at the company, but you really need the ground truth in materials as well if you want to get best performing models and that exists in the experimental lab.

Guest: Now problem in materials is that we don't really capture experimental data.

Guest: 90% of what work that a scientist does does not work.

Guest: It is a process of trial and error and actually using that 90% on experience to get to the 10% of things that do work.

Guest: We don't really capture that scientific data, right?

Guest: It lives in a in a lab in a scientist head or in their lab notebook.

Guest: We don't really publish on that information because it's all the mistakes, all the negative results that we don't want people to know about.

Guest: And then lastly, even if you and all are working on the exact same material problems.

Guest: So a high entropy alloy.

Guest: I have no idea what you've done in the lab already and you have no idea what Radical has done in the lab here over the last two.

Guest: So how can we build off that?

Guest: We can't.

Guest: We don't get to share that information.

Guest: We don't get to stack and compound that knowledge and kind of are building the knowledge concurrently.

Guest: And so this lack of experimental data has been a huge problem in material science, that's why we build self-driving labs.

Guest: We capture every single data point in our experimental process, temperature, pressure, oxidation concentration, even the humidity in New York and how that impacts the room, the temperature and the experiments that we're running.

Guest: We capture all of that data.

Guest: And we take that experimental data and we feed it back into our AI engine so that we do have the most proprietary data set in the world that includes both computation and experimentation.

Guest: And so that's how we're trying to get around just identified where they're not novel or these materials aren't valid.

Guest: We think experimental data will help reduce those errors.

Host: Why pick New York?

Host: Like is New York the best place to build this or or something else in the US?

Guest: Absolutely.

Guest: New York is one of the capitals of the world.

Guest: It's the financial capital of the world and starting to become one of the innovation capitals of the world as well.

Guest: We love New York for two specific reasons.

Guest: Number one, the talent density in New York is very high and can continue to grow.

Guest: Meaning, there are not many elite performers who do not want to or would not move to New York City.

Guest: That's number one.

Guest: Number two, and it builds on this concept.

Guest: If you want to work on advanced scientific discovery and deep technology, there are not many places to do that in New York.

Guest: If you go to the Bay Area, there are or 500, a thousand different companies that you can go work on with science and deep technology and this field of startups.

Guest: In New York, there is not.

Guest: There are very few to we might be the only place where you can do AI and science, for example, in New York.

Guest: Maybe in bio, there's a couple other, but there are not many options.

Guest: And so if you want to be in New York City and you want to work on really hard challenging problems in the industries that AI are going to impact, then Radical is one of the few places that you can really go for that.

Guest: And so of course, those two are linked together.

Guest: We think it continues to be an attractive hub for great talent and when that great talent wants to come here, we're the only place that they can come.

Guest: And that's why New York is super important to us.

Host: I think that's a good note uh to end the conversation and um have one final question.

Host: What is your um, you know, best AI use case for you?

Host: Um for you personally or, you know, imagining it something?

Guest: Yeah, research, knowledge game.

Guest: So I'll give you a perfect example.

Guest: I might talk with a customer and they have a very specific materials problem to something they use in their in their product.

Guest: So they give an automotive company.

Guest: We use this material in the dashboard or in our battery chemistry

---