“It is certainly easier to teach a programmer physics than teach a physicist programming.” — Alain Richardt Alain Richardt, founder of Atomic Tessellator, combines deep expertise in chemistry, quantum mechanics, and AI to accelerate groundbreaking discoveries in material science. From childhood experiments with toxic chemistry sets to building high-powered simulation tools capable of modeling complex atomic interactions, Alain’s journey highlights the transformative potential of marrying traditional chemistry with cutting-edge AI. In this conversation, he explains how AI-driven simulations now achieve results thousands of times faster than traditional methods, shares ambitious projects like developing fusion reactor materials, and reveals the exciting future of “declarative materials” where scientists specify desired properties, and AI generates the perfect material match.
Chapters
2:21 Alain’s Early Chemistry Adventures & Favourite Element (Molybdenum) ▶ 5:12 Mercury Experiments: Crazy Ideas & Influential Teachers ▶ 8:19 Why Alain Founded Atomic Tessellator: Fixing Broken Chemistry Tools ▶ 11:38 Force Field vs. Quantum (DFT) Simulations Explained ▶ 15:25 How Cloud Computing & AI are Revolutionising Chemistry Simulations ▶ 17:53 Promising Applications: Fusion Reactors & Plasma-Facing Materials ▶ 23:08 When AI Simulations Go Rogue: Unexpected (and Expensive) Results ▶ 26:04 Decoding AI Headlines: Schrödinger’s Cat & Quantum Computing Explained ▶ 35:09 Measuring Success as a Founder: Experimental Parity & New Material Discoveries ▶ 41:39 Hot Takes: Periodic Table Merch, Atomic Cocktails, and the Joy of Being a Rebel Founder ▶ Transcript Search or select a timestamp to jump ▾ 0:02 Georgie Healy: Founders scale faster on Deel. Set up payroll for any country in minutes. Hire anyone anywhere. Get visas handled fast and get back to building. Visit deel.com/dayone. That's D-E-E-L.com/dayone. 0:20 Alain Richardt: Over the, like, the last sort of 20 years, people have been doing chemical modeling using these force field models. And then only in the last kind of couple of years have we sped up this simulation method. You know, what used to take months amount of time can now be done in just a few hours. So— 0:38 Georgie Healy: Wow. 0:40 Alain Richardt: Humanity as a whole suffers, right? Because that idea is gone into the ether. So the question, the point is if we like pull everything together into a cohesive platform that's super easy to use, if you make anything easier, people will do it more. 0:53 Georgie Healy: Hot new thermodynamic chips. 'Could Trump Classical Computers.' So apparently there's this new chip to accelerate AI. What are your thoughts? Is this fake news? 1:04 Alain Richardt: We do like femtosecond simulations, and a femtosecond is like, one of the cool facts is like a femtosecond to a second is the same timescale as 1 second to 31 million years. Stop it. 1:21 Georgie Healy: Hello and welcome to In the Blink of AI, where I talk to the brightest AI startups and innovators each week. I'm Georgie Healey, and this week I'm speaking to Alain Richard. He's a New Zealand native and founder of Atomic Testlader. They're a molecular chemistry and materials science platform. Think diffusion models, LLM-assisted research. And for his day job, he's just, you know, casually finding new materials and molecules and helping scientists push the edge of discovery. In one of his many incredible previous roles, he modified NASA's Open Mission Control to drive trading bots instead of the Mars rover. Blue Origin could never. Before we recorded today's episode, Elaine said he was happy to simplify what they do at Atomic Testlabor, but when you have someone with the incredible AI and computational chemistry chops he has, you really just don't want to dumb it down. So for the deep tech fans out there, and even the chemistry curious, this show will be a real treat. Now, before we dive in, we do have our live show coming up on the 29th of May. Tickets will be in the description below. I hope to see you there. 2:39 Speaker: You're listening to a Day One FM show. 2:43 Georgie Healy: Hey, Elaine, welcome to In the Blink of AI. How are you this morning? 2:48 Alain Richardt: Hello. Hello. Thanks for having me. I'm great. 2:51 Georgie Healy: Look, I have so many amazing technical questions because having you on the show, I could not not take up the opportunity to pick your brain, especially because we've never had someone with a chemical expertise on the show married with that AI. I am so pumped. I'm going to dive right into it. But before I do, When did your passion for chemistry begin? 3:17 Alain Richardt: I was lucky in that I grew up, I sort of started entering my teenage years at the end of the '90s, and they were still selling chemistry kits that were full of toxic materials and Bunsen burners and sort of semi-dangerous things. So my earlier memories of early teens and sort of those years were sort of out the back of my house building experiments, you know, with Bunsen burners and things exploding and things changing color and— 3:43 Georgie Healy: Oh, the good old days. 3:45 Alain Richardt: Yeah, they're good old days. 3:46 Georgie Healy: Oh my gosh. Do they not use Bunsen burners anymore? Because we had them in my high school. I don't know if they still do. 3:53 Alain Richardt: I think in high school they're still there, but I'm not sure about, you know, the old sort of chemistry kits that children used to buy. And to be honest, I was too young to be burning methylated spirits and, you know, things like that. But it was all good fun. 4:08 Georgie Healy: Yeah. You can confirm you've got all your limbs still. So clearly you had like a natural affinity early on. Look, a nerdy question. What's your favorite element on the periodic table? I have to know. 4:20 Alain Richardt: Uh, yeah, I was— it's quite a cool question. Um, I probably think molybdenum is because it's not very well known, um, and it's got some really wild chemistry about it. So, um— 4:31 Georgie Healy: Oh, tell me more, tell me more. Do you know what number it is, and is it in that weird middle section? 4:37 Alain Richardt: Yeah, yeah, well, some of the cool things about, um, molybdenum is like it's a cofactor for nitrogenase. So all of the nitrogen-fixing bacteria that feed all of our plants wouldn't be able to function without it, or nearly as well without it. So it's— and the reason I like it is because it's kind of like a, it's a, it's a not very well-known silent achiever in the periodic table. 5:00 Georgie Healy: Yes, you're like, I see you, I see you. I knew you'd overdeliver on that answer. My answer is a little bit more basic. But Mercury has always been my favorite. It's just such an unusual element, right? Just the way it acts and reacts. And also probably because it's so poisonous and deadly, there's some mystique around it. And I know you've got a bit of a story about Mercury from back in the day. I think the listeners would love to hear it, Elaine. Can you please tell them? 5:34 Alain Richardt: Yeah, so when I was a teenager, I had a really influential teacher at high school. And Mrs. Jane Clark was her name, and she really, really accelerated my love for chemistry. And I came one, one time I came to school with my laptop and I had a video on it of this pulsating mercury beating heart reaction. And I said I wanted to build this into an electricity generator. So we would make a— set up a standing wave, make all of these things that floated on the surface and the pulsing a mercury-beating heart would generate electricity through the mechanical actions of the waves. 6:11 Georgie Healy: Oh my gosh. 6:12 Alain Richardt: It was a crazy idea at the time. Who knows if it's possible, but my teacher really encouraged me and she like gave me keys to the chemistry lab and I'd be in there every weekend experimenting with mercury, building this apparatus and all of those sorts of things. And like you say, it's kind of wild to think about this because mercury is so toxic. And I didn't really know at the time, but I, you know, I knew enough to use a fume cupboard and Thank God. And I wonder, like, if that would even happen these days with all of the extra safety regulations and things like that. I feel a little bit sad for the young students of these times. You know, they might miss out on such crazy things happening. 6:54 Georgie Healy: Yeah, there was definitely something exciting about it. Admittedly, this was before my time, and I'm sure I've told you already, we were nerding about our chemistry backgrounds. I got a chemical engineering undergrad, but when I worked in a laboratory at a Transport and Main Roads government. It was my first job, actually. Yeah, the guys there used to tell me they used mercury thermometers to stir liquid bitumen, and as the bitumen would cool, they'd often crack and the mercury would be everywhere, and they'd be scooping it out with their hands. And just wild stuff. The good old days. Look, I'm really excited to hear more about what you're working on now. You've clearly always been passionate about chemistry and always been able to take take it to the next level. Tell us about Atomic Tesla later and tell us what you're working on. 7:47 Alain Richardt: Uh, yes. So, um, a couple of years ago I was working at a trading firm and my boss ended up being an ex-Stanford, uh, staff scientist who has a PhD in quantum chemistry, and we became instant best friends. So we were chatting chemistry all the time and, and, and swapping research papers and these sorts of things. So I started building Atomic Tessellator off the rekindling of my love for chemistry. It was a side project where I was studying catalysis and mechanical properties of alloys and things like that. So it was weird because I would, I would start work at— because the trading firm was at 3 in America, I would start at 3 AM. I would work until 10 AM and then I would sleep for an hour and then I would work all afternoon on Atomic Tessellator. 8:32 Georgie Healy: Oh my gosh. 8:34 Alain Richardt: I did that for a year. 8:36 Georgie Healy: Wow. Some of us watched movies, but yeah. Okay. Amazing. 8:41 Alain Richardt: Yes. Yeah. Yeah. So yeah, it was, it was cool. It started off and, uh, we started studying these chemical systems. I found like that all of the tooling to be really, really poor. So I thought, well, we can do better than this kind of as an industry. And I've got history as a programmer. So I started making the tooling better and, and more compatible and easier to use. That sort of thing. 9:04 Georgie Healy: Wow. So it really just started as like a passion project and you're like, oh, let me try and fix it kind of thing. 9:10 Alain Richardt: Yes. 9:11 Georgie Healy: When did you kind of start to get a sense for, I would really love to start leaning into this more? 9:18 Alain Richardt: Yeah. So one of the cool things is that over the, like, the last sort of 20 years, people have been doing chemical modeling using these force field models. And then only in the last kind of couple of years have we sped up this simulation method. That means that, you know, what used to take months' amount of time can now be done in just a few hours. So that kind of, that kind of speed up really means that we can start to study much more broadly lots of different systems that are out there and just experiment with crazy ideas much faster and much, much more aggressively. 9:54 Georgie Healy: Wow, that's so incredible and so exciting. You know, we may have lost some of the danger and drama from chemistry, but to be able to discover things that much quicker must be exciting. And just, just for the listener, you know, we're talking force field models. We're already getting really deep on the periodic table. High level, why is this important? What are we doing and why is it important, Elaine? 10:18 Alain Richardt: Because this tooling is so generally so hard to use, You can imagine, like, there's thousands and thousands of material scientists and chemists that graduate every year. All of these people have great ideas for new experiments and sort of new explorations in their head. But the problem is, like, what actually concrete happens when somebody has a new idea. So if you're a material scientist and you have one of those crazy shower ideas, right? Oh, what happens if we try molybdenum, for example? How do they actually evaluate that? So, they get out of the shower, they go to their computer, right, with their crazy new idea, they start trying to use the tools, the tools are very bad, they can't figure it out, they can't get the environment up, they don't know how to describe the system that they're trying to model, and then even if they get that far, they usually give up, and then humanity as a whole suffers, right? Because that idea is gone into the ether. So, the point is, if we like pull everything together into a cohesive platform that's super easy to use, If you make anything easier, people will do it more. So we want to capture all of those crazy shower ideas and let the scientists experiment better. 11:25 Georgie Healy: Wow, that must feel genuinely really rewarding. You know, like, there's a lot of rewarding parts, but that's, that's really exciting and really incredible and probably gives more longevity to those people that pursue that kind of a career. So you've told us about kind of the, the, the tooling and how it was quite archaic before. How do you model chemistry with the different methods and trade-offs? Like, I'm sure, you know, I'm sure people would all love to use new tooling, but just kind of map it out in a competitive map a little bit for us, please. 12:01 Alain Richardt: Yeah, so the way that we, the way that we used to do it for the last sort of 20 or 30 years was called force field modeling, and the And then, and then what you also have at kind of the other side of that is called DFT. And so these two different models are kind of competing technologies. The force field modeling on the right is where you have a model that's been parameterized by experiments. So people do experiments and they do some simulations and they feed data into it. But the problem is whenever you're fitting, just generally, whenever you're fitting a model to a set of experimental data, you have to do an enormous number of experiments for the model to perform well. And then on the left-hand side, you have this DFT, and this thing is based on quantum mechanics itself. So instead of being sort of a model that's learned from a few different experiments, you have this thing that's based on fundamental scientific laws. And DFT for a very long time has been a very good workhorse of the industry. But you need to have a supercomputer to do it. And breakthroughs just in the last few years have meant you don't have to have a supercomputer to do it. Even a researcher with a small number of GPUs can now outdo what used to take a supercomputer months. 13:15 Georgie Healy: Oh, it's so exciting. And you went there, you started saying quantum. So for the listeners at home, quantum mechanics, people might start thinking of like the Hadron Collider or even not know what what, you know, quantum versus atomic versus anything else size-wise is. Take us back to high school for a second, Elaine, and tell us what that means. 13:38 Alain Richardt: When we're talking about quantum mechanics, we're talking about actually going from the Schrödinger equation through density functional theory, and then there's a whole bunch of sort of approximations that you can make on top of that. Very broadly speaking, as an overview into the industry, you have like the Schrödinger equation, which is like a mini body equation for all of the electron activity. The trouble with that is that it's really super hard to compute. And that basically means that it would give you the most accurate answers if you could do it, but there isn't even a supercomputer that humanity has that could do it right now. So we take this next step of doing DFT, which is this kind of intermediate step. It's more of an approximation, but it allows us to do stuff that's intangible, physical stuff. So There's kind of like, and then very broadly speaking in the industry, if you've got like super accurate over here and you've got medium term accurate and you've got low accurate here, it's all for the scientists to choose the kind of trade-offs that they're willing to make. But what we want to do is push away that lower accurate one because this kind of DFT modeling is becoming more practical for much larger system sizes. So, And because it's based on a fundamental law, you don't have to make so many shortcuts or approximations. So we're sitting in that spot. We're kind of doing scientific computing that's based on physical laws rather than trying to fit some model to abstract data. 15:09 Georgie Healy: There is genuinely a bit of a sweet spot there. That's really exciting. Look, I definitely am thinking in the back of my mind, I really need to invite you to trivia next time because like, I feel like you would smash all of the math and science questions. But we're on the In the Blink of AI podcast now, so I'm dying to hear how you incorporate perhaps LLMs into this system. So, you've got experience with very large-scale systems at Google. Am I correct in saying you're applying some of that experience to how you design Atomic Test Lab's cloud systems? Tell me about that a little bit. 15:48 Alain Richardt: Yeah. 15:49 Georgie Healy: Yeah. 15:49 Alain Richardt: And I think one of the really exciting parts about just the last year is that now that we can do these simulations, in some cases 100,000 times faster than previous, one of the things that harms material scientists is that they study very simple systems at like 0 degrees Kelvin, perfect crystal systems, and they work in all of these kind of idealized conditions and reality is not like that. So, so— 16:21 Georgie Healy: I haven't been in 0 degrees Kelvin myself, so— 16:26 Alain Richardt: Yeah, it makes the math a lot easier, but it's not great for simulations. So what we've done is like now with the rise of like, you know, cloud computing, Dockerized computing environments, large-scale on-demand computing, these sorts of things, we can make it so that when you run a simulation on the platform, you don't just describe some single idealized system and run a simulation and get a result. What we do is we inject a certain amount of disorder and we run a large number of simulations in parallel. And what that does is that mimics— That mimics reality. So if you have a slab of copper or a slab of any alloy material, it's not just a perfect little slab of atoms. There's all this disorder at grain boundaries and certain other macroscopic features. And if you, and if you can really easily run simulations, you can inject a huge amount of disorder, run them 100,000 times faster than you previously could, and then get a much more accurate modeling of reality and all of this complexity of reality that, that you couldn't really do before. 17:30 Georgie Healy: Wow. So you're marrying like everything you've learned technically about cloud computing and being able to run all of those iterations a lot more effectively, right? And accurately. 17:41 Alain Richardt: Yeah. And also the other thing is the material scientists themselves want to be focused on the materials science. You don't want to have to be a distributed cloud engineer as well. 17:51 Georgie Healy: Oh yeah, such a good point. Such a good point. Actually, few of us do, right? It's a niche interest. So, so can we get to some juicy stuff? What are some of the most promising applications or materials you might be able to discover? Are you looking for something? Do you have like a white whale of materials that you would love to discover? 18:15 Alain Richardt: Now that we can do these simulations a lot faster and a lot bigger, it means that we can model systems that were traditionally very, very hard to model previously. So some of the really exciting stuff we're working on at the moment is high entropy alloys for the plasma-facing walls of fusion reactors. 18:34 Georgie Healy: So— Say that 10 times quickly. Yeah. Okay. 18:39 Alain Richardt: Inside a fusion reactor, you've got the metal that is facing like the actual fusion itself. These get bombarded with of like neutrons and different, different high-energy atoms, and they have really crazy thermal requirements and just radiate. 18:55 Georgie Healy: Wow. 18:56 Alain Richardt: And things like that. And you end up with— and some of these things happen at kind of like tens of nanometers of material, and that way you need to be able to simulate 30,000, 40,000, 100,000 atoms and a neutron bombardment sort of happening all the way through that. And traditionally So an example of this is like in 2001, there was a few good, really good papers on doing this type of simulation, and it took a whole cohort of, of global nuclear science engineers to run these types of simulations and get the results. Whereas even now, just a few years later, we can do all of these simulations on a very small cluster here and literally in my house, right? And we can— So the whole amount of work that would in a few hours that would take globally the scientists to do a year in 2021. So— 19:46 Georgie Healy: That is incredible. 19:48 Alain Richardt: Yeah. 19:49 Georgie Healy: Sometimes I hear stories like that and I'm like, I feel really blessed to be working in tech or like in this industry at all, just to hear about these stories. This is just such an exciting time. I do have to ask too, plasma is something that I do find very mysterious and exciting and sexy and cool. Are you interested in plasma in general as well, or is this something that's not as mysterious and sexy as it was, you know, 15 years ago when I first kind of started learning a little bit about it. 20:17 Alain Richardt: We have to learn enough about it to be able to model the interactions with the matter. Because also the other thing as well is like in fusion research as a whole, it's like the problem used to be with like starting the fusion reaction and, and maintaining it and these sorts of things, which are still, which is still very, very challenging. But actually, I would argue that one of the most difficult parts about fusion now is actually the materials challenge. They actually know how to start the fusion reaction and they know enough about magnetic confinement. But what, like the growing kind of next big problem in fusion is making sure that the actual reactor walls don't disintegrate themselves when the reactor is running. 20:57 Georgie Healy: Why would they do that? Due to the heat and the pressure? Or like, why would, why would they melt? 21:03 Alain Richardt: Yeah, it's definitely, it's, it's just that the the plasma is actually giving out so much radiation that— so what happens is a neutron atom hits the material and it causes what's called these, these Frenkel pairs, which is like just where a neutron hits an atom, it dislocates it, and then it causes a vacancy in the actual lattice structure. And also, and the atom that it's pushed out the way causes another strain inside the crystal structure. So What happens is with these AI models, we can actually predict the amount of strain that this is doing inside the material, and we can actually tell ahead of time whether this is more strain than the material can naturally handle, and therefore it becomes brittle and breaks and all of these sorts of things. So we do all of this actually just only in software, and that saves having to build an entire lab and then do all of these expensive simulations. 21:56 Georgie Healy: Oh my goodness. And wasting the team's time and budget and energy and just breaking it constantly and then being like, oh, damn it. 22:06 Alain Richardt: And the, like, the space of possible materials is so huge. 22:11 Georgie Healy: Oh, the space. How much space physically? Like, when you say wall, what are we talking here? 22:17 Alain Richardt: Yeah, and we're talking like, also the number of things to explore, right? So if you have like, you know, you've got the whole periodic table, you might have 20 or 30 elements that have reasonable radiation profiles. You might want to mix between 3 and 6 of them together, or you might, and then you've got like, in what ratios, 20% this, 10% this, you know, these sorts of things. So, you can imagine like, if you do all of those combinations, suddenly it's just this huge number of things you've got to try. It's not really practical to do it outside of simulation at all. We just need to be able to do huge grid searches on clusters of computers to actually identify candidates. Yeah. 22:56 Georgie Healy: Oh, I really want to see it in action. Before we move on to the next section, I would love to hear, you know, I'm guessing atomic particles are their own beast, but have you ever run an experiment and gotten, even with the AI simulation, gotten a completely different result? Or the AI simulation gave you a completely different result? Like, you've got so many years' experience, I'm sure your hypotheses are quite on the like market, but have they ever been super wrong? 23:29 Alain Richardt: Yeah, yeah, it's interesting because there's all sorts of really cool stories in AI about how a model cheats, you know, so you hear about all these funny stories about how the model has learned some video game exploit or it's learned some other way of gaming the system and we're not immune to that either. So, One of the recent examples that we have had is that we have a model that can produce, it can like target various types of properties. So what you can do is you can say, oh, I'm really looking for something that has a bulk modulus of 700. And so you, which means that the material is very, very hard, like it's an ultra strong material. If you target, and then we say to the AI, okay, give us some candidate materials for that. It churns away and it spits out like 50, 100 different crystal structures. And then we look at them and then what we've— what we see that it's done is it's just cheated and it's added osmium to all of them. Right. So like all of these structures have osmium on them. And that's— although that's a good result in that, yes, the material is super strong. Also, osmium is currently €3,000 a gram. Right. 24:43 Georgie Healy: Just add gold to the— yeah. Thanks. Thanks, AI. Yeah, no worries. I'll just do that. Cool. Oh my gosh. 24:51 Alain Richardt: It's going to be one expensive building or something strong, right? Yeah. 24:56 Georgie Healy: Yeah. Yeah. Try get that through the VCs. Like, I just need a couple more mil capital because the AI told me that I needed this. Oh my gosh. That's so funny. We had a previous guest last week who used to be a professional Go player. And he was like you said about the AI playing games. They'll do a move that no one can understand. Like that's the most strange slash idiotic next step to move, and then you realize, oh, actually they're 100 steps ahead. And then, and then maybe there's like this AGI, ASI next level thing that we just aren't even aware of. So maybe the AI model knows something you don't, Elaine. 25:36 Alain Richardt: Like, I don't know. Exactly. It could be way ahead of us. 25:38 Georgie Healy: Yeah. Maybe like through tariffs, it's gonna get real cheap or something. I don't know. 25:43 Alain Richardt: Yeah. 25:45 Georgie Healy: Look, you've been such a good sport, so I'm gonna punish you more and make you unpack headlines. For us because most of us see AI headlines around, you know, chemistry or deep tech, and we just kind of take it at face value. So I want to get your amazing analytical brain on some of these headlines and tell us what you think. So the first one, all I'm going to say is, you know, you brought up Schrödinger before, UNSW's Schrödinger's cat. Can you tell the listeners what is going on here? I tried to read it. I didn't really understand what was going on. Please help us unpack this headline. 26:26 Alain Richardt: This kind of backs onto quantum computing as a whole. So, which is a kind of a really exciting frontier that we need to stay on the edge of. So when I just previous, just a few minutes ago, I touched on having like this Schrödinger's equation and then there's DFT. And then there's the force field models. 26:47 Georgie Healy: Yes. 26:48 Alain Richardt: One of the— and how the Schrödinger equation is like, is very difficult in terms of exponential complexity. We don't have a computer that can do it, these sorts of things. So quantum computing kind of as a whole offers us the promise that maybe we could solve that equation in an acceptable amount of time. So, so yeah, what would— what that would mean is that largely all of these sort of shortcuts and all of these models that we're focusing on right now, we could just throw out the window and solve the kind of root physical law itself, which is like super, super exciting. Even when you sort of hear about what are the practical advances of quantum computing as a whole, you always hear breaking cryptography and chemistry. They always say those two examples, and they're absolutely right. And— Yeah. In both of those. There is going to be a time, hopefully in the next sort of 5 years or less, where quantum computers will be good enough to actually solve the Schrödinger equation, and then all of our tooling will just go out the window because we'll solve a fundamental law ourselves. 27:52 Georgie Healy: Wow. Is this good for you or bad for you? Would this like make your day or be like, oh my gosh, I kind of wanted another decade at like playing with the tools? 28:01 Alain Richardt: No, It's a really good thing. We're super on top of it. Like, we've got every couple of weeks we're kind of running experiments with other quantum computing partners. We've been running systems with quantum variational encoders and helping us like solve optimization problems and things like that because we want to be there to take advantage of it when that, when that, when that clickover period happens. 28:26 Georgie Healy: Yeah, you're like, I see it coming and we're ready. That's amazing. Okay, next headline. 'Hot new thermodynamic chips could trump classical computers.' So apparently there's this new chip to accelerate AI. This guy, Ghislaine, is building this new chip. What are your thoughts? Is this fake news? 28:49 Alain Richardt: No, no, no, no. These are— these— I think it's super important that we do experiment with these new forms of architectures. And also just anything that can speed up AI in general. So we, just an example of this, we do like femtosecond simulations and a femtosecond is like, one of the cool facts is like a femtosecond to a second is the same timescale as 1 second to 31 million years. 29:16 Georgie Healy: Stop it. That breaks my brain. Femtosecond. Did I say that properly? 29:23 Alain Richardt: Yeah. That's 10 to the power of negative 15 seconds. 29:27 Georgie Healy: Oh my gosh. 29:28 Alain Richardt: Really enormously short amount of time. And most chemical reactions happen kind of at the upwards. So another example is like the proteins in the back of your eyes, they take about 200 femtoseconds to react to light, right? So that's the speed at which this is happening. And we do, so we do simulations from femtoseconds all the way up to like hundreds of nanoseconds. You know, so there's quite a bit large period of time there, but that does actually mean that every day we're doing like 2 million, 5 million inferences through AI, right? So anything, any chip experts that can build a new chip that does AI faster, we're completely on board with because we're doing millions and millions of these inferences a day through AI, and even a tiny 10% speed up would help us a lot. 30:18 Georgie Healy: Wow. Wow. I mean, it's not the same thing, but I briefly talked about you before we started recording about my love of Formula 1. And because I watch it at home, I can't really understand— like, they're nanoseconds apart from like the first place to second place because they're just going at such a fast speed. And I'm trying to imagine it sitting at home, but like they show it on the TV in a way that it looks like they're ages apart, right? But femtosecond, I think that might be a of my compute level for my brain. 30:52 Alain Richardt: Yeah, not many people know, even know of that timescale that's so short, right? People hear of nanoseconds, but then down to picoseconds. 30:59 Georgie Healy: I mean, we worked with nanoseconds in engineering. I don't remember femtoseconds. 31:03 Alain Richardt: That's— Yeah, pretty crazy, huh? 31:06 Georgie Healy: Incredible. There's one more headline, Accelerating Scientific Breakthroughs with an AI Co-Scientist. Please tell me a little bit about this one, Elaine. 31:15 Alain Richardt: Yeah, this is, I feel like this is the biggest sort of most exciting one in general, and it's also an area that we're focusing on a lot. So one of the, you know, AI fundamentally is like the mechanization of thought, right? So if you can have like an AI co-scientist that can help you review a paper, form an alternative hypothesis, and even run some of the experiments, that really takes us back to that earlier ethos of what we wanted to do for a materials scientist. So if a materials scientist can walk up to their computer and say, hey, I just read this paper, can you reproduce the work in it? Then the AI co-scientist will actually run the simulations, will reproduce the work, will interpret the results, and then come back to you saying, I was able to rerun these simulations to create what was in this paper. And then the scientist can say, okay, rerun the whole thing again, but this time swap out all copper for aluminum, go, right? That's what we want to get to, right? We want to get to the point where we can just say to the machine, go and try this experiment, and it will go. So, AI scientists, yeah, a couple of months ago, I gave a talk in Riyadh at the Global AI Conference in Saudi Arabia, and I had our first version of this running. So, we have AutoAtomic running internally, which is able to review a paper and then set up and run simulations. And yeah, and while I gave the talk, I said it on the great big projector behind us, and the crowd was able to watch the AI agent actually build and run the sim— build a set of hypotheses, import the data structures, run all of the simulations, and then come to the result while I was telling about, you know, the future of AI and materials science. 32:58 Georgie Healy: And so much of this, I'm sure, is hard to imagine without seeing it, right? So being able to actually see the model, you know, explain what it's doing and go ahead and do it. That's incredible. So exciting. 33:14 Alain Richardt: Yeah, it's very visual. It's nice to be able to see summarize into the points of the AI, you know, and then that we have a system in the platform as well where the AI can write notes, right? So the AI writes a couple of little notes about what it's thinking and then it runs the experiment and then the results come out. So it's all live on the workbench. 33:31 Georgie Healy: Elaine, can you make an LLM that creates like lab-grown, you know, precious jewels and things. I know we can already make lab-grown diamonds and certain things, but there's certain inclusions that are in, say, an emerald, right? And if you do a lab-grown emerald, it just doesn't look the same as a real one. Not that I have real ones lying around in my house, but like, these are the things I want to simulate. Like, how can you make it really look like heritage, you know, Sri Lankan sapphire or something like that? That's That's what I would be doing the models on. 34:06 Alain Richardt: Yeah, definitely. Yeah. And this, and those kind of— so often reality in terms of like all of these like GEMS and how alloys perform and things like that, they are, it goes back to like just imperfections. So instead of being able to model very simple systems like a pure slab of copper and things like that, you need to be able to model all of the different types of defects and dislocations and impurities and or these sorts of things. So, um, in a material science platform as a whole, you need to be able to sort of say these are the parameters that you want to operate within, and then the system will generate lots and lots of different imperfections and then average the results towards the end. And that's how— that's how you go from like something that's just too much theory in terms of like a pure system that's really basic to something that really reflects reality. 35:01 Georgie Healy: Yes. Look, we've talked about your background as a, you know, as a very deep domain expert in the chemistry field. We've also talked about your technical background and working at Google. But as a founder as well, I'm curious to ask you a couple of questions. Like, you're breaking ground here with Atomic Testator. So how do you measure success? What are you looking looking for? Is it the number of new materials you discover, or are there other metrics that, that you're really focused on at this stage? 35:32 Alain Richardt: Yeah, so for the last, um, I would say 4 or 5 months, we've been working on what we call experimental parity. So what we— what basically what we do is we take a bunch of, uh, very high-quality lab data, um, that says that these particular materials with these particular ratios have these mechanical properties, have these electrical properties, have these thermal properties, these sorts of things. And we've been making sure that our simulations match reality. 36:02 Georgie Healy: Wow. 36:03 Alain Richardt: So there's always this trade-off between like accuracy and computing speed. There's two levers that you're pulling all the time. And what matters is that you can do stuff fast enough so that you're not costing a lot in terms of computing power and bills. And, but also accurately enough to reflect reality. So when we started, we were doing things, we have these parity graphs that show like reality versus predicted. And we had, and when we started, we were all over the place. 36:32 Georgie Healy: Scatterplot. Okay. 36:34 Alain Richardt: Yes. And then we went, so we've been refining and refining and refining, and now we're well past the point where we can be confident in our own predictions and for not just simple systems, but complex systems. Yeah. Yeah. 36:46 Georgie Healy: So you've got to have a new goal now. You've nailed that one. Well done. 36:51 Alain Richardt: So now that we're confident in our predictions, we actually have a bunch of new ultra-strong materials. We have some radiation-resistant materials. We have a bunch of materials that we've actually predicted, and we're working with the labs now who are synthesizing them and like proving that they'll work in reality. But I'm not actually— I'm not actually at all worried about that because our like previous predictions are so accurate compared to all of the other lab results that I know it's going to be close enough. So— 37:19 Georgie Healy: I'm asking really trivial questions here, but do you get to name any of the new materials you come up with? You do? 37:26 Alain Richardt: Yep. 37:29 Georgie Healy: Oh my gosh. Okay. You've got my email. I want to know all the cool new names. I can drop them in sentences. I'll be like, I met the guy who, who invented or created, or I don't know what it is. Whatever you call it. That must be fun. 37:43 Alain Richardt: Yes. 37:45 Georgie Healy: My grandfather was a town planner in Brisbane back when there were like hardly any houses in certain areas. He had so much fun naming stuff after like family members and stuff like that. I bet you have a ball. 37:58 Alain Richardt: Yeah, yeah. We have like all of our systems, all of our internal things, they're all named after various video games and various— and actually, Atomic Tesillator itself is named after an object in a video game. All right, so— 38:11 Georgie Healy: Oh my goodness, I love this thing. You deserve it. You should call it whatever you want. That's very well earned. 38:18 Alain Richardt: It's cool because like we were also inspired by— there are other companies who have done this, like US Robotics has been named after the Isaac Asimov novels. Oracle was a sci-fi name. Like there's a lot of history of great companies being inspired by sci-fi, and that's what we want to we want to try and do it well. 38:36 Georgie Healy: Just anyone that says there isn't a lot of emotion and heart behind it is wrong. Wrong. Look, I would love to hear how you hire, Elaine, because, you know, the deep knowledge and understanding of AI is one thing, and then there's the material science aspect. How do you, how do you find people in the space? 38:56 Alain Richardt: Yeah, it is really, it's super difficult. So We're a New Zealand startup in the sense that the company is here, but I'm actually the only employee in New Zealand. 39:07 Georgie Healy: Good for time zones, I'm sure. 39:10 Alain Richardt: But it's cool being a remote company because we get to choose the best from everywhere kind of around the world. What we have found is that it's certainly easier to teach a programmer physics than teach a physicist programming. So we tend to go heavier on the programmer side first. And the science can be picked up. But also one of the other things that we really, really value internally is like a growth mindset. So, anybody in the company can challenge any idea. I expect a certain number of experiments to fail within all of our staff, right? So, if somebody is working here at Atomic Test Lab and everything they do is successful, we encourage them to be more ambitious because That means that there's great, even greater greatness in them. So it can, we try and bring that out. And I remember sort of from my careers, you know, being at large organizations like Google where I managed, you know, 20 people, 30 developers, that kind of thing in my history. I kind of lost a little bit of that throughout my career as a senior programmer as well. And now I'm rediscovering the joy of managing incredible people. Like it's just so amazing. To see all of the staff members come back to me and be so excited at the things they're discovering and the new sort of frontiers they're pushing and— 40:29 Georgie Healy: It does sound rewarding. It really does. 40:32 Alain Richardt: And also with everybody in the company, everybody has their own internal speech goals, right? So, you have the amount of work that we have to do, and then you have some crazy ambitious plan that's aligned with the person's career, right? But also as like a crazy stretch goal. And when the staff, like, when you can, the staff can see that the manager is interested in the person and growing them as well, they kind of get excited about what crazy thing can I invent or come up with. So. 41:01 Georgie Healy: I have a sense that the environment's fantastic also, Elaine, because you're clearly so on the tools yourself. You're not kind of standing back and just like, you know, tell me what you guys come up with. Like, it's so clear that like before we even started recording, I was like, oh, there's a bit of background noise and you had all these servers running. Like, clearly you're like in it, you're doing it yourself, and you're just as passionate and excited to discover and create yourself. I really get that sense. 41:30 Alain Richardt: Yeah, yeah, I've got server rack here in my house. There's a big sort of full rack unit right beside us. And, you know, I mean, we put my electricity bill, uh, I think is about 6% the national average. Yeah. 41:42 Georgie Healy: Look, there's a laptop behind you. You're on a computer speaking to me. I can only imagine what would happen if I zoomed out. It would It would be a war room in there, I bet. Look, this is my favorite part of the interview. This is where we go into hot takes, spicy questions. Are you ready for the last part of the show? 42:01 Alain Richardt: Yep, sure. Yeah. 42:03 Georgie Healy: Periodic table mugs, you know, those mugs with the periodic table printed on them. Are these cute or lame? 42:10 Alain Richardt: Cute. 42:11 Georgie Healy: Thank you for saying, 'cause I actually had one for, I think, a decade there for a while. So you nearly, you nearly made me cry. Do you have one right now? 42:22 Alain Richardt: Uh, no, not a periodic table mug, but we've got periodic table everything. A towel, you know, like different stuff. Yeah. And we, and we, we've actually open source modeled a whole bunch of materials for open periodic table as well. So— 42:38 Georgie Healy: Hot tip! 42:40 Alain Richardt: When we interview scientists, they actually comment that our peri— they keep going back to our periodic table and using it because it's the most— it's like has 3D models of each of the elements. Crystals spinning and cool stuff like that. 42:52 Georgie Healy: So— Can the listeners see this if we put a link in the show notes? 42:56 Alain Richardt: Yep, sure. 42:57 Georgie Healy: That would be amazing. Would love— okay guys, we're going to have that in the show notes. If you were to make an atomic tessellator cocktail, what would go into it? 43:08 Alain Richardt: Uh, yeah, that's quite an interesting question. I think probably, um, probably something like gin, um, because it's kind of sort of sharp and a very distinct taste. Mm. And then also just for the chemistry, maybe some lemon for acidity. 43:25 Georgie Healy: Lemon and gin. It sounds fresh. It sounds unapologetic. It sounds honest. It's an honest cocktail. You know what you're getting. No hidden, you know, agenda in there. I love it. I love it. I feel like this is something that everyone could get on board with. Very on brand. I have a feeling you would be snapped up, Elaine, if I'm honest, by any of the tech companies. They would be desperate to have you in one of their executive roles. You could do that. So why go this hard path of being a founder? 43:59 Alain Richardt: It's a really nice question because it kind of touches at what you are as a person. So I was always, even since I was young, I was a bit rebellious and a wee bit sort of against the grain. And working at these large corporations taught me a lot about people management and skills and large-scale engineering, but really I like having the freedom to do our own thing and push at the boundaries a wee bit more. So having, like you say, having your own startup is like definitely on hard mode, taking the hard route, but it's just so rewarding seeing the staff grow, seeing the company grow, making new discoveries at the rate we are as well. So. And every week we're finding some new angle or partnering with somebody new, and they have a set of challenges that are just really wild. One of the— if I may, just one very cool example was just 2 weeks ago, I was at the NVIDIA conference over in San Jose, and I just serendipitously met a professor from Stanford who told me about this very special structure inside the shell of mollusks. Mm-hmm. Called mother of pearl, and it's a very special material, right? But what it does is it's actually built in such a way that stops crack propagation. So you have these little wee microscopic discs that are really strong in one direction. They get cracked or hit from a strong impact and they break. But because due to the way that they're set in this matrix of material, that crack doesn't actually propagate. Like when a crack is in glass, it propagates. When a crack is in metal, it propagates. But this kind of— nature's kind of figured it— figured out a way of preventing crack propagation. And these kind of things that you have as a founder, just going to all of these new meetings and saying, what are you doing? And these amazing, you know, sort of ideas that come together. We just— we immediately went home, jumped onto Atomic Tesla, started running all of these simulations, found even better ways that we can improve it. And we think that there's this sort of bio-inspired anti-crack propagation material that we can even start experimenting with. 46:08 Georgie Healy: So. That is so fascinating. And you do hear about this happening, right? Like people studying like even airplane flight by being inspired by birds and things like that and shapes and, and, and taking things from nature. 46:22 Alain Richardt: Yeah. Nature's had so long to figure out these kinds of problems. It's amazing how we can get inspired by them. 46:28 Georgie Healy: And mother of pearl's really pretty too. So. 46:31 Alain Richardt: Yes. 46:31 Georgie Healy: Might make everything look really nice and shiny. 46:34 Alain Richardt: Yeah, that's true. 46:36 Georgie Healy: Yeah. 46:36 Alain Richardt: And that same color, that like that iridescent rainbow color that's produced is actually an artifact of this really crazy structure that's also super hard and stops crack propagation. So yeah, it's really cool. 46:51 Georgie Healy: Yeah. Thank you for sharing that. I'll never look at it the same way again. That's incredible. So what do people get wrong about you? You know, you're from New Zealand, you've got this startup in chemistry. What, what don't they get right about you, Elaine, do you think? 47:06 Alain Richardt: Um, yeah, I think because we're kind of not in Silicon Valley, we kind of tend to get dismissed a bit more. But that's okay though, because also it does mean that we— it motivates us a lot more. So I definitely think like we're kind of the ragtag team of weird, kind of weird outsiders. 47:27 Georgie Healy: Yeah, there's power in that, right? There's a lot of power in that. I love it. 47:32 Alain Richardt: Yeah. And also the other thing is like, we don't, we don't really worry about doing large amounts of marketing and things like that because we let our results speak for themselves. So yeah, we're just kind of like the misfits on the, on the outside of materials science. 47:45 Georgie Healy: I love the honesty of your answer. And it's also kind of like, it doesn't really matter if they don't get it. You know what I mean? Because the proof will prove itself. 47:57 Alain Richardt: Like— Yeah. And one of the things that I found really kind of inspiring was I heard the story about like Larry Page and Sergey Brin when they started Google. What they would do is they would go to an investor, they would explain Google to them. And if they didn't get it, they would just move on. And they would say, if the investor doesn't get it, 10 more minutes of me explaining it, is it going to make you get it? Right? So. 48:20 Georgie Healy: Words to live by. 48:21 Alain Richardt: You'll find people, you naturally attract people and investors and staff and things like that, that you vibe with. So it's about not forcing the universe and just letting it go where it will. 48:32 Georgie Healy: We have a lot of founders that listen to the show. I think that's brilliant advice. Thank you for sharing it. We've got one last question. What's one headline that we might see in 2025? 48:46 Alain Richardt: One of the cool things that's coming up is declarative materials. So basically, one of the holy grails that material science and maybe us is coming to soon is being able to say, instead of being able to like say, here's a bunch of materials with all these different properties that we want to simulate, we actually flip it completely over and do it what's called inverse materials design, where we say, we need a material that has this radiation resistance profile, this conductivity, this thermal stuff, and then the AI will just produce it. So we declare what we want, and then, and then the AI fills that declaration of, of properties. So that's something that I think was coming within the next 2 years, certainly. 49:29 Georgie Healy: Declarative materials. You declare what you want, the AI will just generate that. That's incredible. Heard it here first. What a way to end the show. Elaine, you've been Such an amazing guest. Seriously, I feel— I genuinely feel privileged to have been able to speak to you. Thank you so much for coming on the show. Before I let you go, is there anything you can shout out to the listeners? Anything they need to know? How can we follow you? How can we see what's coming next for Atomic Test later? 50:01 Alain Richardt: Uh, yep, sure. Thank you. So, um, at the moment we're approaching general availability for the platform. So, uh, that means that, uh, if a researcher or if you're a university student or you're even just an enthusiast looking into material science, then please reach out. We'll bring you part of our, like our beta user program. We'll give you some credits, we'll give you platform access, we'll give you sort of primary support and help so that we can grow it. And then also on the investor side, so we are raising a seed round at the moment and we're looking for investors who are looking to be, who are deep tech, and also an interest in materials science because this sort of field is growing a lot this year. So we're super excited to hear from anybody in any of those areas. 50:46 Georgie Healy: Amazing. I think there will be a lot coming up, so definitely we will all be following you on LinkedIn and can't wait to see what's next for the company. Thanks so much for being on the show. 50:58 Alain Richardt: Thank you as well. Thank you for having me. 51:01 Georgie Healy: Thank you for listening to In the Blink of AI. You can check out the show notes for anything discussed in this week's episode, and we will be back next week. This podcast was produced by Day One with music by Dan Hansen and visual artwork by Sophie Tyrell. If you loved the episode, please tell your mates, and I love AI news. Please share your thoughts and suggestions to [email protected] . No transcript lines match that phrase. Try fewer words.
▶ 
▶ 
▶ 
▶ 
▶ 
▶ 
▶
▶
▶