GRTiQ Podcast: 187 Eshan Chordia

The following podcast is for informational purposes only. The contents of this podcast do not constitute tax, legal, or investment advice. Take responsibility for your own decisions, consult with the proper professionals, and do your own research.

Eshan Chordia (00:00:18):

There has been a lot of, in the last year, last year and a half, a lot of crypto AI narrative, VCs looking for deals, and entrepreneurs trying to raise money on these ideas, because no one actually knows what’s going to happen, but everyone knows that something is going to happen.

Nick (00:01:03):

Welcome to the GRTiQ Podcast. Today I’m speaking with Eshan Chordia, co-founder at Lumino AI, a company building decentralized AI solutions, focused on making machine learning models more accessible and affordable. As you’ll hear, Eshan’s background is filled with a deep passion for tech and entrepreneurship. From his early interest in tech and video games, to his experiences working in startups and at Protocol Labs, Eshan has unique perspectives on topics like AI, decentralization and web3. During this interview, Eshan shows his journey into web3, his time at Protocol Labs, working on Filecoin’s crypto economics team, and the origins behind Lumino AI. We explore the challenges and opportunities within the AI and web3 spaces, Eshan’s vision for decentralization, and why he believes that democratizing access to machine learning models is crucial for the future of tech. Eshan also provides insights into Lumino AI’s unique approach and a very thoughtful, maybe as thorough as you’ll hear, overview of how we should think about what the next big thing is in AI. We started the conversation with Eshan sharing his background growing up in Pittsburgh and the experiences that shaped his early passion for tech.

Eshan Chordia (00:02:16):

Yeah, absolutely. Thanks for having me, Nick. I’m super excited to chat. This is going to be a blast. I grew up in Pittsburgh, Pennsylvania, so I was there for the first 18 years of my life, and then I went to Carnegie Mellon there, because I was actually there, and I took an extra semester. I was there for almost 23 years. I literally moved out to the Bay Area two weeks before I turned 23, so wonderful city. I loved growing up there. It’s like the perfect mix of big town and small town, four seasons, east coast. Yeah, just absolute fun time.

Nick (00:02:50):

And what types of things were you interested in, or what were some of your hobbies as you were a young person growing up?

Eshan Chordia (00:02:55):

I still am, but I was the biggest sports fan. I love sports. I’m into basketball, football, soccer, cricket, Olympics, into all sports, and Pittsburgh is a big sports town, like Mass is sports town. Yeah, that was my biggest thing. I wanted to be an athlete. That was my biggest thing. I also love tech. I got into video games super early. Today, now we’re all like, “Okay. Maybe we should push back on when kids get screens,” but I think my mom actually got the original Game Boy when it first came out, so with Tetris. My first video game was the Nintendo 64. I don’t remember what year it came out, but my parents got it from me for that Christmas with, oh my God, I’m forgetting the name, the N64 Mario. Oh my god. I have to Google this now. Don’t worry about that, but I got that first game.

(00:03:48):

I was super into video games, and then we got, I think it was like ’98 or ’99 when yellow version, red version, and blue version for Pokemon came out, and they had released a Game Boy color. So then the Game Boy color came out, and they had released a Pokemon themed Game Boy color. I have two younger brothers, so we shared all of our video games. We got the share the N64, Super Mario 64. That was the game. Oh my god. Then, we got the Game Boy color, the Pokemon themed one, and we got yellow version, and we used to play that so much. I was super into video games. I think video games is actually how I got into tech. I’m 33, so I feel that my age of millennials, it’s like the similar story of video games was like the intro into tech, right? So that’s how I got into tech. It was a lot of video games, a lot of sports, and Star Wars, and my dad was a Star Wars fan.

(00:04:45):

They reproduce Star Wars multiple times to change how Anakin looks at the end of return of the Jedi and everything. So I saw the originals, and I don’t remember for what happened for Fantom Menace and Attack of the Clones, but for Revenge of the Sith, my dad got us tickets for opening night for Revenge of the Sith, middle seats. I have two younger brothers, right? So all three of us and my dad. So I think Star Wars was, still is, I still watch The Bad Batch and all the new Star Wars stuff, but Star Wars was also a big thing for me for tech, because I was like, “This is space, all of these robots and droids,” so I think those are the big things that really, really excited me when I was young: sports, Star Wars, and video games, which I feel like it’s kind a mix of a very nerd thing and jock thing, but to me, I felt like most boys, that was the thing that we did back in the day, was either Pokemon, Star Wars or sports.

Nick (00:05:46):

Do you remember the moment in time where you sort of matured past seeing tech is something that you just sort of game on, and it could be something that could be a career or profession you could pursue?

Eshan Chordia (00:05:58):

Yeah, I think it was pretty early. So the iMac G3 came out in 1998, I think, and that was the first computer, and I was in elementary school in ’98. That was the first computer I ever used, and I was like, “Wow, this is amazing. I want to build something like this,” and I was floored and I loved it. It was the desktop, the CRT monitor, and I loved the design. It was something about the Apple’s design. I fell in love with it, and I loved the colors. That was the thing. It was like the shape, and then the outside is plastic. For people who were listening and don’t know what the iMac G3 is, it’s this massive desktop that’s covered in plastic, but they made it look beautiful, and you could take off the plastic at the back and get into the wiring, and it was just great.

(00:06:47):

I think I was in, I don’t remember, third or fourth grade or something like that, and I was like, “I want to build something like this.” That was the moment where it was a combination of technology and also combination of like, “Wow. That is beautiful, and I want to build something amazing and beautiful like this.” That was my entry moment into, “Oh. Maybe I want to build computers, and I want to build stuff around computers.” I used to play that game. I think the crossing game you have to get from the east coast to Oregon, the West coast crossing game, I remember. I think it’s a computer game that we all played, and I was like, “Oh. I want to build games like this. This is fun,” so I started thinking about it like that.

(00:07:29):

Then, right around sixth and seventh grade, I really got into computers first. I would just mess around with operating systems. I would install Linux, but when you say Linux, I’m talking about you have to install distributions of Linux, so I would install Debian, I would install Ubuntu, I would install, I think it was called Gecko or Gecko. I don’t even remember. I used to install seven or eight different versions of Linux to play around with the operating systems to understand how these things worked, because I just was interested in the UI, the applications, and loading stuff from Source. I wasn’t really programming at this point, but I was just kind of figuring it out, and then in seventh grade I started building my own computer, so first it started out with buying a kit and assembling.

(00:08:21):

Then, it ended up being buying just a case and buying all the components off the shelf and then assembling, and that was my seventh and eighth grade. A lot of the time was spent doing that and a lot of time, honestly, before Disney bought Lucasfilm, I guess it is, the Star Wars universe was super wide, and it was probably the best universe ever invented. It went from 3000 years before a new hope to 50 years after when Lucas is super old. So I read a lot of those books but the books, the movies are great and the TV shows are great, but the books, when you read books, they’re in detail about technology or about the force or whatever it might be, but that was another thing where they talk about building stuff in the books, and I was like, “Oh. I want to build robots,” so robots was also, the other thing that got me into it.

I was like, “Maybe I can build robots when I’m older.” By this time I was pretty sure I was going to be an entrepreneur.So this is around eighth or ninth grade. Now I’ve built computers, pretty sure I’m going to be an entrepreneur. Before this, my goal was to be an athlete. Up until sixth or seventh grade, I was like, “I’m going to go play football. I’m going to play baseball. I’m going to play basketball. I’m going to do something like that,” and it was pretty clear that I had no athleticism. I grew up vegetarian. My parents were not, didn’t know much about nutrition, I didn’t get any protein as a kid. I was always the slowest person or the weakest person. When we did the presidential, physical thing, I was like the person who was last. I was like, “This is not going to work out.”

(00:09:54):

Then, I was like, “Oh. I’m going to be an artist,” and I’m tone-deaf and I have no musical ability, so that didn’t work out for me, and I don’t have any artistic ability either. So that didn’t work out for me either. Then, I was like, “You know what? I’m going to be a general manager of a sports team,” so general manager and entrepreneur were my two things that I was going to do. Then, I was like, “You know what? I want to be an owner of a sports team, so let’s go be an entrepreneur, and then hopefully I’m successful where I can buy a sports team and hire my GM,” and I was super into tech at that point. Then, I think in ninth grade or tenth grade, I got my first Apple laptop and I stopped building computers, because the Apple laptop was so amazing. It’s funny. We always come back to Apple on this.

(00:10:38):

The Apple laptop was so amazing, I was like, “Oh. This is so good that I don’t need to do anything. I don’t need to worry about Linux.” Now I started playing around with software. I think tenth grade is when I got around to playing with software, and then that’s when I started in school, I think in high school I took a couple of programming classes in 11th and 12th grade, and that’s when it was like, “Okay. I don’t know what I want to do with my life, but I know I’m interested in computers, and so I’m going to go to a university where they let me study computers.” That’s how I moved from wanting to be an athlete to being like, “I’m going to go work in technology, and computers is going to be the area of technology that I’m going to work in.”

Nick (00:11:20):

As you mentioned, you ended up at University at Carnegie Mellon University, and you studied electrical and computer engineering and cognitive science. Talk to us about your university experience. Did that shape a little bit more that vision you had for your career and wanting to be an entrepreneur?

Eshan Chordia (00:11:36):

Yeah. So on the entrepreneur side, I think over here, VC backed and this kind of high growth entrepreneurs, there’s these founders, are like, “Oh my god, you’re starting a company,” like all of this, but in India, it’s not like VC-backed software companies, but being your own boss or starting a lifestyle business is a normal thing. For example, my grandfather on my mom’s side started a jewelry business. My uncle on my mom’s side started a business. My uncle’s on my dad’s side started a business. My dad, he was the first one who didn’t. Well, I guess my uncle’s on my dad’s side also, but he went to IIT Madras, which is the equivalent of MIT or Stanford in India, one of the best engineering universities in the world, and then came to Pittsburgh to do his PhD at Carnegie Mellon. And so he started a tech company, but in chemical and mechanical engineering, in that kind of field, right?

(00:12:31):

To me, it was not weird to be like, “I want to be an entrepreneur,” or it wasn’t like, “Oh my god. You’re a crazy person. This is too difficult. Why are you doing it?”

(00:12:39):

That’s what I grew up with. I was like, “Everyone has their own businesses, and that’s normal.”

(00:12:45):

And so, the whole entrepreneurship thing, I didn’t really realize how me being like, “Oh, I want to be an entrepreneur since I was a kid,” that was super not normal, because everyone I knew was an entrepreneur. Then, Pittsburgh, they’re a bunch of entrepreneurs, but then there’s also a lot of doctors. And so it wasn’t that weird being an engineer and an entrepreneur. So I got to Carnegie Mellon, and I had this super different experience than everyone. I was on a dance team, so there’s this Indian dance called Bhangra. I was in that, and then I was in the Undergraduate Entrepreneurship Association. I was in the Indian Student Association, and I was in a bunch of different extracurriculars, and I took Intro to ECE and Intro to Programming my first semester at CMU, and I loved Intro to Programming, and I did not like Intro to ECE.

(00:13:34):

So I was trying to figure out how to switch my major from ECE, which is Electrical and Computer Engineering, to computer science. The other thing is I wasn’t even sure I wanted to do this. I went through ten different majors, because in high school I loved AP gov, I loved AP economics, I loved psychology, and I did not like physics, I did not like chemistry, and I did not biology, and I liked math, so I was like, “Math and computers, I like,” and then these social sciences, like economics and psychology and gov, I love, and I really do not like the physical sciences.” So I was super different from every other person in engineering, where everyone was into science and technology, and I was much more math and computers, and then these social sciences, and I was super into these social Sciences.

(00:14:19):

So in my years at CMU, it took me four and a half years to graduate a double major and a minor, because I took classes in interaction design. I took classes in human-computer interaction. I took classes in cognitive psychology. I was debating whether I wanted to be an economics major, whether I wanted to do computational finance, and then ECE and computer science. I was trying to pick between seven different majors, and I took intro two or three classes in each one, and so as I got older at CMU, I realized that I keep doing different stuff, but I keep coming back to computers and I love computers. At this time, tech was taking off in Silicon Valley, right? Google had IPO’d by then. I think I was in high school when Google IPO’d, but Facebook was becoming a thing.

(00:15:03):

People were talking about web2 being a thing, and all these things were happening, but I was a college student who didn’t really know that all this was happening, so I didn’t realize how lucky I was to be in a field where I was studying computers, how big tech was going to be and what the profit margins were going to look like, and it was going to be basically the 2000 tens was the decade of software, right? Software, cloud, et cetera. And so, I was just trying to figure out my life at that point, and I was super not into studying either. I took five classes or six classes a semester, and I did well in the classes that I liked, and I did super poorly in the classes I didn’t like. It was all over the place. Finally, I was like, “This computer thing, I really want do, and I really, really like design.” At this point, they were not going to let me into the school of design, and so I did cognitive science so that I could take all of the human computer interaction classes in the computer science department.

(00:16:04):

Then, I was one major one class away from majoring in HCI, basically, and so that’s how I ended up in cognitive science, and I was super into psychology, but I didn’t love all parts of psychology. I really like cognitive psychology, and so this was the perfect way for me to combine hardcore engineering with product design, without studying product design. It’s almost like, I think there’s this famous major at Stanford called Symbolic Systems, which is a mix of computer science and product design, or at least that’s how I’ve always thought about it. Marissa Mayer talks about how this major that she did at Stanford gave her these skills in product design, and this is my equivalent of like, “Okay. I studied ECE, got a minor in CS. I really understand the technical fundamentals, but took a lot of HCI and interaction design courses, so I can focus on building products for real people and not just push technology for the sake of progress.”

Nick (00:18:01):

It is an amazing hack on how you got the education that you wanted, and that you took the time to make sure that you learned the things that would sort of arm you for the career or the vision you had for your career. At what point, then, in your life do you become aware of crypto, and can you take us back in time and kind of explain what that moment or experience was and what you were thinking at the time?

Eshan Chordia (00:18:28):

Yeah, absolutely. So I moved out here in 2014, so I did a bunch of stuff, worked at startups, but I got into crypto in 2016, so not late, but not super, super early. I wasn’t one of those people that knew about crypto when I was in college, and I knew people who, one of my dad’s friends would travel to Europe to buy, and back then there were no exchanges. He would travel to Europe to physically buy Bitcoin and move them from someone else’s cold wallet to his hardware wallet to his hardware wallet. It was a different time back then, and I heard these stories much later, but I didn’t know about Bitcoin and college and stuff like that, so I got into crypto in 2016.

(00:19:15):

Actually, kind of how a lot of normal people got into crypto back then is Coinbase. I was like, “Oh. This thing called Bitcoin is gaining value, and Ethereum. I should buy, and I don’t know what it is,” and so I bought a bunch. Then, I was like, “But what is this thing, and will it ever be more than an asset?” So then I read the Bitcoin White paper and I read the Ethereum White paper, and I was like, “This is awesome. How do I learn more?” And so, then I started reading white papers of other protocols, and started doing web3 projects on the side at home, building just to see how it worked. I never published anything. It was never a business. It was never like I’m trying to get users. I just wanted to learn. That’s how I got into crypto.

(00:20:04):

There was a lot of reading online, being on forums, and talking to people who were also into crypto. I think it was actually my roommate who got me into crypto. He was the first one. Him and I were living together at that point, and my brother was about to move in. He is the one who actually introduced it to me, and that’s kind of how I got into it, and then we would have these long talks about decentralization, and will it work? And the world is getting more centralized, and not less centralized as globalization happens. How does economic decentralization and legal decentralization and technical decentralization fit into this? I still have questions around these things, and I still love talking about these things, but yeah, that was my intro into crypto.

Nick (00:20:47):

If you don’t mind, can you just sort of double click on that aha moment that sort of drew your interest? And I ask it, because clearly you’ve made a decision early in your life that you like technology, you’re starting to work and build with software, you’ve gone to a great university, you’ve developed a degree minor that will definitely perpetuate you into the tech field. So there’s a lot of reasons you should stay comfortable in web2 and just stay on that path. So what is it about the white papers that you read or the conversations that you were having that was a light bulb moment for you, where you’re like, “Maybe I sort of veer off a traditional path and take a closer look at this stuff”?

Eshan Chordia (00:21:27):

I love learning. Well, the way to learn to think differently is to gain knowledge, because if you can’t gain knowledge, then you can’t open up your mind to different possibilities, because you get to a point where you’re like, “Oh. I don’t know more.” So you can try to think from first principles, but when it comes to science and technology, you actually need to have fundamentals, right? You can’t be like, “Oh. This thing makes sense,” but when the laws of physics will not allow you to do it, or in social sciences, you can try to be like, “Oh, we should live this way,” but in reality, humans have the way that they want to live, and we have biases and we have perceptions, and trying to change, fundamentally, human nature is extremely difficult. I liked learning just for the sake of learning.

(00:22:10):

It’s like the same thing as a scientist who might want to push science for the sake of pushing science and no other reason, right? That’s kind of how I got into crypto. I just like learning about it more than anything, but the thing that I found super fascinating was programmable money, and the second thing I found super fascinating was building economies. I think we talked about this, right? I’m super into economics, building economies when a bunch of people can come together and make decisions for that economy, and those people don’t need to live together. They don’t need to work together. They don’t need to have similar political ideologies. It doesn’t matter. As long as everyone is aligned on building that economy, it doesn’t matter who those people are, what those people do, what those beliefs are, what part of the world they’re in, but they’re organized by this principle of building this economy and using this programmable money to move the economy.

(00:23:08):

These are the two things of crypto that I find super interesting to this day. I’m thinking about this in 2016, but we’re here almost in 2025, and it’s still super important to be able to build economies using these rules that power the group, power that product, or power that economy that you’re trying to build. Once I started understanding, “Oh. I read the white papers of different protocols, and I’ve started buying into the crypto thesis and decentralization,” I think this programmable money and this wanting to build economies that do this one thing really well, and doesn’t matter what’s happening in the outside world, was the thing that really was the aha moment of, “Crypto is a real thing, and I think it’ll be a technology of the future.”

Nick (00:23:59):

As you mentioned earlier, after you graduated, you did some work in some startups, and eventually though you make your way full time into crypto. What’s the backstory for how this side interest and the self-education on crypto turned into a pivot away from traditional startups to going full-time into crypto?

Eshan Chordia (00:24:19):

Me, focused on learning, allowed me to make a couple of good decisions. My brother and I started a company, we can talk about that later. It wasn’t super successful. We shut it down in 2019, and 2019, when I was looking for a job, I really wasn’t sure what I was going to do in life. At this point, I’d been an entrepreneur, I’d been a designer, I’d been an engineer. I loved computer science in college, but I did not find software engineering in the real world interesting, and I think it might have been because I went through the web development. I became a full stack web developer when I graduated, and not focused on back end, not focused on crypto, not focused on AI or infra. I just found that super boring, even though I love building products. I loved designing, but I didn’t love the effort it took to be amazing at the tool to turn the designs in your head into an asset that someone could build.

(00:25:12):

That effort was not something that I had fun with, and then I didn’t really like this. Then, as an entrepreneur, I was doing technology stuff, I was doing business stuff, and I was doing design stuff, and I was like, “I want to do more of this,” and that’s when I was like, “Oh. I need to be a product manager. This is basically where I’m going to.” So in 2019 after this, I started applying to roles in product, and I quickly realized that there were three kind of industries that I was just going to apply to. One was AI, the second one was crypto, and the third one was FinTech. I always was super big into FinTech, and I ended up joining ZestyAI, and so wonderful company, probably the three years that I learned the most, up until starting Lumino. When it came to 2022, I’d been there three years. I actually wasn’t looking to leave, but I guess web3 was growing super-fast at that time.

(00:26:08):

A bunch of web3 companies reached out, Protocol Labs one of them, and I started to think about what is next steps in my career if I end up leaving Zesty, which at that time I was like, “Oh. I’ll just grow into [inaudible 00:26:22] product here. That’ll be my thing, and it’s like an AI company and we’re building climate risk models. We’re doing great work. It’s amazing,” and then Protocol Labs and a bunch of web3 companies call. Then, I started thinking about, “Oh my God. I was thinking about AI and web3 in 2019. I spent three years in AI. Should I start thinking about pivoting and coming back to something that I love, that I didn’t get to work in before?” And so, I went through the process with a bunch of companies, and then their offers, and Protocol Labs is making this great offer and they’re recruiting me hard. I started thinking about, “If I join web3, what does that mean for my career? Do I want to leave AI? What happens there?”

(00:26:56):

Then, also, what kind of web3 company do I want to join? I came to this decision. I was like, “Okay. If I’m going to leave AI and I’m going to move to web3, I want to either join a company that is building hardcore, decentralized infrastructure or a company that is doing some really cool stuff in DeFi, like not yield farming, but actual DeFi stuff, because my thought at that time was basically the two parts of crypto and web3 that will stick and are actually useful are the financial products that are coming out that are changing the way that users are actually using financial products, or this set of on the actual tech side of decentralized infrastructure, right? Software infrastructure, hardware infrastructure, and so that’s why I decided to join Protocol Labs, and it just happened at that time that, after three years in AI, I was ready for the new thing. I just got lucky that all these web3 protocols found my background super interesting and wanted me to join their team.

Nick (00:27:58):

So I always ask this question to guests that join, and they talk about that pivot moment going full-time into web3. You went to work at Protocol Labs, went to work on Filecoin, if I’m noting that correctly.

Eshan Chordia (00:28:10):

Yeah, I was on the Filecoin Crypto Econ team. Yeah.

Nick (00:28:12):

What new insights, or how did that first exposure to full-time work in web3, either reshape or double down on some of the thoughts or opinions you had about the industry?

Eshan Chordia (00:28:26):

I love this question. One thing about working in any industry, but let’s just actually use my experience. Working in crypto and reading about crypto is completely different. You can talk about ideology and inspiration, and these are what we’re trying to do, but then when I moved to crypto, I was like, “Okay, so it’s basically any other industry, and there are a lot of people who believe in it and the real difference here is that we’re trying to get groups of people to coordinate together that aren’t maybe in the same place but have a shared vision,” as opposed to being in centralized, it was very top down, crypto felt more bottoms up, but at the end of the day, it still requires coordination to build anything.

(00:29:07):

That coordination means that there needs to be some sort of structure, there needs to be some sort of hierarchy, someone needs to be making decisions. Whether it’s one person, a DAO, or representative democracy, whatever, or straight up voting by how many tokens you own, there’s some process of making decisions, right? That was one thing that had a more naive or inspired look of crypto before joining, and then it was more reality setting. The second thing was that I underestimated the power of crypto economics to build an economic system, to build a business, or build a product. And so, I had had better technical understanding of web3 and blockchain, and not that strong of a crypto economic understanding of web3 and blockchain. I think joining the crypto econ lab, as we call it at Protocol Labs, opened my eyes to like, “Oh, the power of crypto economics.”

(00:30:01):

This whole concept earlier that I had of programmable money, and using this programmable money to coordinate people and build economies, I actually doubled down on and I became a big believer in. I saw that, and I was like, “I think this can work. I think we can coordinate people to build the right things in the world, using these systems,” but then I also saw the dark side of, “Unfortunately, all of this work, token prices can’t go down or else people lose interest,” right? And that’s the dark side of crypto, is I also felt like everything became financialization, everything became a token and token price, and not about the ethos of decentralized software, coordinating through non-centralized means, and being more bottoms up product development. And so, that’s my two, “Oh my God,” double down moment, and then the others actually a stronger dark side moment of, “The reason why people are super into crypto is because token prices go up,” and that kind of just made me sad.

Nick (00:30:56):

One theme I’ve been exploring recently on the podcast with some recent guests is this concept of the token is the product, and there’s a pretty popular sort of cult following blog post out there on this topic, and I’d love to get your opinion on it. I mean, where do you come in on that sort of argument that, at the end of the day, the token is the product?

Eshan Chordia (00:31:18):

I think it’s a lot more true in DeFi and something like Ethereum. The entire economy runs around the crypto economics of Ether. First of all, I don’t buy that all protocols need to have a token, but there needs to be some sort of value capture and monetization capture, right? So if you don’t have a token, you need to be building a useful product for your customer or your user that they actually like, and not trying to be shilling a token for no reason, but I fundamentally do believe that the difference between crypto, and specifically crypto, not blockchain, I think within web3 we have a lot of things. The use of blockchain can and cannot be with crypto, right?

(00:31:58):

But with crypto specifically, if you’re trying to build a crypto product that has its own token, then yeah, if people lose interest in that token, the rest of the stuff doesn’t really matter, and so, you’re actually building two businesses at the same time. You’re building the business where the customer and the user cares about the actual product you’re building, and then you’re building a crypto economic business as well. I think being an entrepreneur in crypto is actually harder than being an entrepreneur in, let’s say, enterprise SaaS or just software, in general, is because you have to think about that you’re running two businesses at the same time, and how do you make sure these two businesses are always aligned with each other? And so, that’s my viewpoint on this, “Token is the product.”

Nick (00:32:39):

So let’s shift our attention then to Lumino AI, and as you’ve said in your story arc here, you’re working at Protocol Labs, you’re working on the crypto economics team at Filecoin. At some point, this idea, I presume for Lumino AI, sort of begins to germinate in your mind. Take us back to the origin stories. Where did it come from, and what were you sort of thinking about at the time?

Eshan Chordia (00:33:03):

So this is a funny story. I joined Protocol Labs. They made me an offer, it was called unmatched. I went through, man, I forget what it’s called, but they basically allowed you to join or allowed certain people to join unmatched, and you had to go find a team. I guess Protocol Labs was growing big enough in 2022, that they were like, “Eshan will find a team.” I joined, and I’m debating between two teams now. I’m debating between Crypto Econ Lab and Bacalhau, which is the decentralized compute infrastructure team. They were building this software library to do compute over data. I’m thinking about this, and I actually pitched Lumino as an idea to leadership, and they’re like, “But we’re already doing Bacalhau, and that’s general, and this is for machine learning, so let’s focus on generalized compute.”

(00:33:54):

Like FEM was coming out, and they just started thinking about, “Hey, for Filecoin to be a little bit bigger than just a storage market, we need to allow applications and L2s to be built on it,” and this was future looking.

(00:34:06):

I was like, “Hey, why don’t we just build an L2 on top of Filecoin that allows people to train models?”

(00:34:10):

And they’re like, “But we’re doing Bacalhau, compute over data. We’re trying to integrate with Filecoin, and Filecoin is cold storage, kind of.”

I was like, “I’m not sure this aligns well,” but I was like, “Okay. It’s not going to work at Protocol Lab, so let me figure out which team to join, Bacalhau or Crypto Econ Lab.” I ended up joining Crypto Econ lab, and it was a super tough decision in my head. I was like, “I don’t know. Am I making the right choice?” Whatever, and it’s funny, because then I ended up working closely with the Bacalhau team, anyway, later on. So fast-forward a year, Protocol Labs does massive layoffs. 25% or more of the company gets laid off, and I get caught in the layoff, so this is whatever, February 2023, I need a break, so I took a couple of weeks off, and then I start thinking about what I want to do for my future. What was it going to be? Right around mid-2022, end of 2022, I felt like I was ready to start a company again.

(00:35:05):

So I’d spent three years in AI and almost a year at Protocol Labs or six months at Protocol Labs, and I was like, “It’s been three and a half years. I feel like I’m ready to start a company again, but I don’t know when,” and then this layoff happened. Then, I spent a little bit of time, and I was like, “You know what? This seems like a good time to start a company again.” I reach out to my friends, my LinkedIn acquaintances, and people, and I’m like, “Hey, do you guys know anyone who might be interested in starting a company?” And I also stumble upon YC Co-Founder search, and so I start talking to a bunch of people, start doing YC Co-Founder search. Between My network, and then YC Co-Founder search, I probably speak to 80 or 90 potential co-founders, 70 on YC Co-Founder search, and then on 20 others, and then I meet Yogesh on YC Co-Founder search.

(00:35:51):

Out of everyone I speak to, I’m like, “Oh. This guy is actually amazing. He wants to start a company.” A lot of people I spoke to, they were just casually looking, but they weren’t really into starting a company, and I was like, “I want to work with someone who’s really into building companies.” I met a lot of people who think the entrepreneur life is a flashy life, and it’s not. It is like grind life, right? You have to be working all the time. You don’t have anyone to delegate work to. You have to do everything yourself. When you’re looking for a co-founder, you want someone who is great on the business side, great on the technical side, is willing to grind, understands it’s not easy, there will be difficulties, and it’s not just going to be like, “Oh. You’re going to raise $5 million in a month, and you’re going to launch a product and it’ll be successful,” right?

(00:36:38):

So it was pretty hard to find a co-founder. Yogesh and I meet. He was working on something called Quicker Pay at that point, so he was at Tax [inaudible 00:36:46], and then he was working on the side on his own company. Originally, it was like, “Okay. Let’s experiment with Quicker Pay.” It was a crypto on-ramp, but specifically for gaming. We did that for, I don’t know, a month maybe, maybe a little bit more, and we realized this is not the right business to be in. It’s cool, but there are 25 other on-ramps. On-ramp for gaming, there are some technical things that you need to do better than on regular on-ramp, because you need super low latency, and people are changing assets on web3 gaming companies super-fast, and so you need to have a higher performance. Fundamentally, the issues with risk and fraud are still there, and neither of us have experience in risk or fraud.

And so, we’re not good entrepreneurs. Were not a good founder fit for this kind of company, so he and I come together. He had a list of ideas that he wanted to work on, and I had a list of ideas that I wanted to work on, including Lumino, and we make a list, and we do two things. One, we make a framework of what are good ideas, and this framework is, what is the market size? What are we interested in working in as personally? Because if you, as a founder, is not interested in working on a project, it doesn’t matter how big the market size is, because you’re not going to be willing to grind or get through the difficult times, right? What are our skills fit well for? You also want to start a company where your skills are fit for what the company needs. I was not a good fit to start a gaming company or to do risk and fraud, because I didn’t have any experience in fraud and at least payment risk.

(00:38:20):

We were doing insurance risk at Zesty, but no payment risk. So these are the three main things. Then, the fourth thing was, on these ideas, we spoke to 30 or 40 people for each idea, and so then we narrowed it down, and the two leading contenders like Central Bank digital currencies, which are blockchain based, and this idea, which is also web3 NAI intersection idea, and that’s how Lumino got started. We did a lot of work before deciding that we were going to solve this, and Yogesh has been in the Bay Area for, I think it’s been eight or nine years now. He graduated with his master’s. It’s been eight or nine years, and then I’ve been here for ten years. We had the list of people who were data scientists. I worked at an AI company. I spoke to a bunch of machine learning and data science people. He spoke to people at, I think, where he went to do his master’s at University of Texas, Dallas.

(00:39:08):

I spoke to people at Carnegie Mellon, UC San Diego, and other places, and we were like, “Okay. We think there’s a product to be built here,” and so even before we decided to work on this, there was a whole bunch of user research for this idea and for all the other ideas. Then, when we were pretty confident that this is the idea we worked on, we met in probably end of March, early April, or something like that, and then we incorporated July of last year. And so, there was a solid amount of time. We did the whole YC. YC has this really, really good setup of meet in person, do a bunch of things, and then they give you a document of 20 questions. You answer the first 10 questions, and you have to take it seriously of what kind of company do you start, what kind of founders do you think are great, what products do you think are great, what skills do you have?

(00:39:56):

I don’t remember the exact questions, but a bunch of other people didn’t take it seriously, right? Two sentences, but Yogesh and I wrote paragraphs for every question. We would meet, and we would meet a bunch of times, and we would talk, and we’d be like, “Oh my God. This person is serious. We’re aligned,” and then we did the second set of ten questions,” and then they recommend like, “Oh. If you have a partner, everyone should meet. If you’re going to start a company together, this is going to be hopefully a 10 or 20 year endeavor. Everyone needs to be on board,” and so we kind of followed this YC process, and it was really good and it worked out really well for us. We incorporate Lumino and the week before we incorporate Lumino, I hear about OrangeDAO. So we make this OrangeDAO application, and without incorporating, without doing anything, we send this application to OrangeDAO.

(00:40:45):

They’re like, “We want to do an interview with you.”

(00:40:46):

And at this point we’re like 20 days after the OrangeDAO deadline. We sent the application just out of kicks like, “Oh. Let’s just see what happens,” and we send this application.

(00:40:58):

They want to do an interview. We do the interview. It’s 15 minutes, normal YC style interview. Orange DAO was started by a whole bunch of YC founders or XYC founders or whatever. Yogesh and I are talking. We’re like, “Huh, I don’t think the interview went so well.”

(00:41:14):

And so, the day we incorporate, I think two days after, we’re thinking about the product and get a call from James Sinka, who is the OrangeDAO fellowship leader, “Hey guys, you have been accepted into OrangeDAO. It starts in a week and a half. Can you let us know by the end of the week if you’re joining?” I kid you not. We are in a library. We’re working out of a library right now, and we’re in this library, it’s quiet, and we’re trying not to scream and jump up in Joy that we got into OrangeDAO. That’s how the story begins of how Lumino started, and then we got into this accelerator pre-product on the strength of our team, our idea, and the user research that we done. I think one thing that we found out later is they were super impressed with the amount of user research we done on a bunch of ideas on this one before picking this product, and so yeah, that’s kind of the story.

Nick (00:42:02):

So let’s talk then about what Lumino is and how it works, so what’s the pitch there?

Eshan Chordia (00:42:09):

I like to call us a decentralized machine learning training protocol. We used to call ourselves like a decentralized compute protocol, but as we did more user research and tried to get more customers on the demand side, we kind of realized that there is this missing thing in machine learning right now, that it’s not that easy to build models, actually. I’m not talking about fine-tuning in an LLM, but I’m talking about, in general, about building a model. You have to get a data set, process that data set, clean that data set. You have to build out your machine learning training pipeline. You have to train the model. You have to figure out where the model’s doing well and the model’s not doing well. You have to get more data to where the model’s not doing well. You have to retrain that model. You have to put it in production.

(00:42:53):

Then, you have to see how the model’s doing in production based on whatever data is coming from your customers, if there’s drift, if it’s starting to make wrong decisions, and then train that model every few days, every week, every two weeks, whatever your cadence is based on the product that you’re building, and that’s the customer side. This is not web3 related. It’s not crypto related. This is just understanding the demand side of what’s happening in machine learning. We’ve been thinking about this demand side for a bit, and then the back end, which is like normally you would just deploy your training pipeline on GCP or AWS, and either use some managed instance or build your own custom training pipeline, and you’ll be like, “I’m using GCP.” Fundamentally, our belief for Lumino is we need to decentralize infrastructure.

(00:43:38):

So I’m going to take a step back for a sec. In crypto and web3, we talk about decentralization, and we’re talking, what are we decentralizing, right? Are we decentralizing compute? Are we decentralizing economics? Are we decentralizing the tech? But I think this deepened narrative just got interesting the last two years. One thing is these cloud service providers or hyperscalers, they’re eating away data center margins or profit. I think at one point, I don’t know if this is true anymore, but 70% of all Ethereum nodes were being run on hyperscalers. You can imagine that if the hyperscalers are like, “You know what? We don’t want Ethereum to exist in the world,” and they shut that off, that kills Ethereum. You can talk about Ethereum being decentralized and there being a million nodes, whatever, but they’re all being run on hyperscalers.

(00:44:33):

So fundamentally, it was, how do we ensure that we have infrastructure that is not controlled by three or four entities in the world or five entities in the world? The second thing was the hyperscalers have a very different cost structure that has nothing to do with their product, right? Marketing, operations, and they also have different incentives. Their incentives are not top line revenue. Their incentive is, how do I maximize my market cap in the public markets? And so, that has caused them to, especially for machine learning, Keep prices on these compute really high, and it’s really expensive for the rest of us to build out, especially training models, right? Inference is also expensive, but training is really expensive, and so we were like, “How do we democratize the building of AI so that anyone can build a model, LLM or not, and how do we bring costs down so academia can do it?”

(00:45:34):

We spoke to postdocs, and they’re like, “Oh my God. I had to share these bunch of machines with 20 other people, because we have to reschedule it, and this is when we can use the machine and it slows me down.” It’s like, “Whatever. Everyone just had access, so it’s super cheap to compute.” Then, the third realization was renewable energy is coming, and so, yes, data centers are growing, and I think in the US it uses data centers are like 2% to 3% of all the energy usage in America, and by the end of the decade, it’ll be 7% to 8%, and there’s a lot of stuff about will we be able to meet the energy demands and will the grid be overloaded? I’m a techno optimist, and I think that where there’s a problem, somebody will figure it out. So in general, I’m not too worried about it, but I know a lot of people will disagree with that statement.

(00:46:19):

But in general, I was like, “If renewable energy is coming, and the cost of energy is going to come down, that is something that works in the favor of decentralizing infrastructure,” so how do we combine these things to give a better user experience for people who want to build models and a cheaper way of being able to build models? And that’s the intersection of this decentralized machine learning training protocol. The customer side, it’s not like AWS. It’s not general compute. We’re not spinning up a virtual machine, and you get access to that virtual machine for the amount of time that you’ve paid for it, and you can do whatever you want. It is truly a machine learning training protocol. You can think of it as a serverless experience. The experience on the customer side is very similar to the experience that OpenAI provides, and these guys have gotten it right. They have an SDK. I think they have in different languages, but we’re focused on Python since that’s the predominant machine learning language.

(00:47:16):

We have a Python SDK. We have these APIs available, and you can just get started training in a couple of minutes. You don’t need to build out this entire end-to-end pipeline. We’re not doing the data cleaning sites. You still have to do that, but the compute, figuring out the compute, downloading PyTorch or TensorFlow, Kuda, setting all of it up when you’re training, running into all these errors. We’ve done all this experimentation behind the scenes. We did all the work. We know what the errors are. The customer side is this normal SDK product. Developers use it. You can start building. You can go to our UI. You can use the UI. You don’t even have to be an engineer. You can just use the UI to build yourself a model. That’s like the vision on the customer side. On the backend, on the crypto side, it’s like, let’s use the power of crypto economics and blockchain to coordinate large scale data centers to work together to train machine learning models.

(00:48:12):

That’s the vision for one product, but the vision for Lumino is to be the AI ecosystem where we build products at the intersection of AI and crypto. This is one product. We’re going to have a decentralized inference product. Then, there is an SDK for specifically doing inference within smart contracts. There is lots of things that we could build. Now, I don’t know what the future looks like and what the roadmap looks like. We don’t know if this product will be successful, right? But there are lots of things that we can go into. The vision for Lumino itself is, how do we build a company that can build multiple protocols in this AI space that can make it easier to build models, run models, make it cheap, do the work that you need before the model is trained? So like data, pre-processing, data, cleaning, getting the data, and then also the work after the model’s trained but not deployed for inference, so evaluation and the stuff that needs to be done to make the model better. We’re trying to capture all of that, but this is our first product.

Nick (00:49:17):

Let’s talk about where Lumino is then in that lifecycle process, so for listeners that are intrigued by what you’re building, and they want to get their hands dirty, what’s available right now, and how can people learn more or get, I guess, some exploration of what you’re working on?

Eshan Chordia (00:49:31):

Today is Friday, September 13th. We’re going to be launching our SDK in about two to two-and-a-half weeks. You’ll be able to go to Luminoabs.ai, log in, sign up into the web app, read the developer docs, import the SDK into your project, and get started fine-tuning an MLM. Right now, what we allow is we allow people to fine tune open sourced, large language models. So the goal is Lama, Mistral, Quentoo, like whatever, all of them. Go to the website, go check it out, play with the product, give us feedback. Anything is allowed, right? It’d be like, “If the product sucks, let us know. If you think it’s amazing, let us know. If you think that we can change some stuff, let us know. If you like the developer docs, let us know. If you want to use the UI, you can do that.” In about two and a half weeks, all of this will be live from this date. That’s one way. The second way is if you’re interested, but you want help or you’re not sure this is something for you, check out the website. You can DM me on Twitter or Farcaster.

(00:50:32):

You can email me at Eshan at Luminoabs.AI, and I’ll help you get set up, and I can walk you through how to do this. Right now, we’re focused on being able to get people to fine tune LLMs, and then after that, we’ll be launching our DevNet sometime in end of October to mid-November. We don’t have the exact timeline out, but our DevNet for the decentralized protocol, right? So if you’re interested in that, again, you can DM me, you can email me. That’s probably the best way to do this. If you’re interested in working at Lumino, again, we are looking for people who are into machine learning, who are into software engineering. We’re growing the engineering team rapidly. We’re adding crypto economists to the team, and then next year we’ll be adding some non-technical folks, including recruiting and marketing. So if you have expertise in those areas, and you’re like, you just want to get to know what Lumino is about and learn more, you know how to contact me. I think those are probably the best ways of contact, and definitely use the product and give us feedback.

Nick (00:51:30):

Listeners of this podcast are familiar with this concept of AI and some of the themes that are emergent within web3 related to decentralization, so I think, conceptually, a lot of listeners will understand the value proposition and vision of what you’re building. My question though is about demand, and so, is there demand within just the web3 sphere itself for decentralized AI? And then, if you just take sort of that next, lateral step to the entire market, so all these people that are using Claude and ChatGPT, is there a demand there for something that’s powered by decentralized tech?

Eshan Chordia (00:52:08):

I love this question, because I think it kind of shows people’s philosophies and views on life. One, yes, there’s 100 percent demand in web3 and for decentralized AI, and in web2 for decentralized AI, but I’m going to be real honest with you. I think right now, blockchain, crypto, and decentralization, there haven’t been enough killer apps for normal people who are just building models and don’t care about any of this stuff to be like, “Decentralization is a feature I want.” Our service has a way to do it in a centralized fashion and a decentralized fashion, because fundamentally, some people just don’t trust decentralization. They’re not going to change our minds, because we are big believers in decentralization, right? And then there are some other enterprise issues like VPC, keeping data in certain regions, and then et cetera, that maybe it doesn’t work until they update their way of doing things.

(00:53:01):

So we have a setting that allows us to do this in a centralized way, but obviously we’re building out this protocol, and that’s the important part, not just the SDK. On the second part of, is their demand, we’re not only targeting web3 people, we’re actually, most people we talk to are not web3 people. They’re just normal people like, “I want to build a company, and I need to fine tune my alum.” They’re not even the type of person like, “Oh. I got to build AI, and the AI is the most important thing in the world.” They’re like, “Oh. Actually, I have this problem in finance, or I have this problem in legal, or I have this problem in sales, and if I use an LLM, I can solve this problem, make it way cheaper for my end customer. I just need to fine tune my LLM on this data set, on my proprietary data set, and then I build the AI into the entire user experience, and it doesn’t matter to the end customer whether I use AI or not. It matters if I’m solving their problem and if I’m bringing value to them.”

(00:54:01):

For those people, I don’t care what service I use on the back end. I just want to fine tune my LLM in a cheap manner, keep my costs down so that I can pass on those cost savings to my customer, and if I have lower OPEX, then I’m more competitive against the legacy incumbent software companies that have built out these really, really complicated enterprise SaaS applications that cost a lot of money. I don’t know if you saw this recently, I don’t remember what company it was. I think it was Klarna, but it could have been someone else. They ripped out Salesforce. They built their own in-house. They fine-tuned an LLM. They didn’t build a large language model from scratch. They fine-tuned an LLM. They built a user experience around it, and they’ve built what they claim is almost as good as Salesforce, but way, way, way cheaper, right? We’re trying to serve all people. We’re not here just to serve web3. My fundamental belief is that blockchain and crypto are technologies, and I would like us to stop using the phrase web3.

(00:55:00):

We stopped using the phrase web2. It’s part of the open internet, and let’s just have blockchain and crypto part of the open internet when necessary. You can think of as like we have different building blocks, right? We have steel, we have concrete, we have cement, we have wood, we have plastics, and if blockchain is steel, we don’t use steel for every single application in the world. Blockchain doesn’t need to be there for every single application in the world, but steel is super powerful. It’s super cheap, and it allows you to do things that wood and concrete don’t allow you to do, right? And so, when the technology is useful, you use that technology to build a specific type of application. So that’s kind of my philosophy on this blockchain and crypto, and we are trying to serve everyone in the world who wants to build a model, and we are doing it in a decentralized way. Yogesh and I are big believers in decentralized AI, but we recognize that decentralization doesn’t matter if the product that you’re building doesn’t have value to anyone.

Nick (00:56:01):

I appreciate that answer, and that sentiment regarding web3 has come up multiple times on the podcast, and I know a lot of people hold that position as well. I do want to ask you sort of this final few questions before I ask you the GRTiQ10. The first one is, every crypto cycle, there are top level themes that sort of drive the speculative nature of the industry and maybe the holder’s interest and tension span, and AI unquestionably has been a big part of the cycle we find ourselves in now. For that reason, it sort of becomes a buzzword and a garbage term that people sort of misuse and apply in probably ways that don’t always work. As somebody who’s building a real product on decentralization and AI, what do you make of that top level theme and maybe the way that AI has been bastardized, if you will?

Eshan Chordia (00:56:54):

It really was so people could just raise obscene amounts of money and not really build product, and be like, “Hey, look. I raised at this valuation. We saw this with other decentralized compute protocols,” or “We saw this with competitors,” or “We saw this with just people pitching us stuff.” They don’t have a product, they don’t know how AI works, and they were just like, “AI will change how web3 works,” but it’s part of life, and DeFi, NFTs, AI. The entire world is just super excited about LLMs, right? But think about how long, so AlexNet came out, when? I think in 2012, right? AlexNet was that computer vision model that kind of proved that deep learning was super, super good, and it worked, and it worked better than anything that we had seen before, and you didn’t need as much training data or you didn’t need as much energy to do the kind of work when AlexNet came out and was able to classify those images.

(00:57:50):

I’ve been in the Valley since 2014, and people have kept using the word AI or machine learning, and everyone at every point was like, “This is fake.” When they think of AI, they think about singularity or they think about this all knowing thing. Machine learning is part of every single product that we use today. TikTok, machine learning. Reels, machine learning. When you Google search, those rankings are based on machine learning. Your Netflix, what shows up on your Netflix page, it’s machine learning. Everything that we interact with today is some sort of machine learning, and so this generative AI is just the next step. Again, and generative AI is truly, again, is actually in the context of, you think AI, there can be multiple things that form artificial intelligence. Machine learning is just one of those things, and generative AI is another form of machine learning.

(00:58:40):

We’re still using the transformer architecture. I mean, we’ve made lots of improvements on it since Google released the original transformer, but it’s still a transformer based architecture on the technical side, which means that we’re still doing what’s the probability of what the next word is going to be? I think people have seen like, “Okay. LLMs truly have the ability to change user experience, build new user experiences, drop costs. It’s going to be something new that we really haven’t thought about these types of businesses and user experiences before,” and everyone, regardless if you are web3 or not, the people who figure this out are going to be like the mass of winners five years from now, but no one knows who’s going to figure it out. No one knows what the user experience looks like, and so the only way to figure out who the winners are is bet on everyone.

(00:59:29):

Because if you bet on everyone, even if you lose money on nine out of ten deals, that tenth company is going to make so much money that, I mean, as a VC, you will get all your fund back with that one company. I mean, fund plus 10x, 100x, or whatever with that one company. And so, I think that’s why there has been a lot of, in the last year, last year and a half, a lot of crypto AI narrative, VCs looking for deals, and entrepreneurs trying to raise money on these ideas, because no one actually knows what’s going to happen, but everyone knows that something is going to happen. And so, when you know something is going to happen, you have to try everything to figure out what sticks, what doesn’t stick, what customers like, and what customers don’t like, and where is value capture in this new world, compared to the old world of enterprise SaaS, bottoms up SaaS, or whatever?

Nick (01:00:19):

The next question I want to ask you, again, following this AI theme, is you’re working on a lot of very cool and interesting things at Lumino. Outside of some of that, if we’re just talking about AI generally, what’s the next big thing there? I sort of feel like, in the minds of most listeners, ChatGPT was the big thing, but clearly there’s going to be other things that come outside of just LLMs. What is it going to be?

Eshan Chordia (01:00:44):

I’m not great at predicting stuff, so I’m going to make a bunch of predictions, and we can come back in five years and see how right I was. There’s the chip level, which is, for listeners, the chip is the bottom level. That’s the infrastructure. That’s the hardware infrastructure level, so there’s the hardware infrastructure level. There’s the software infrastructure level. There is the model level, which is the large language models, and then there’s the application level. So the AI stack is fundamentally comprised of these four levels, these four layers, and there’s innovation happening at every layer of the stack, so the next thing isn’t just one thing, but it’s the next thing at every layer of the stack. I’m going to start with the chip level, and then I’ll work myself up through all the levels.

(01:01:29):

At the chip layer, I think we’re seeing this a lot. I think there’s a lot of new publicity around this, but in general, NVIDIA is not going to be the dominant. They might have a majority, but they will not own 95% share of AI chips, both on training and on inference. By the end of the decade, that’s not going to happen, because there’s just too much money to be made. You know how we had Bitcoin ASICs, right? Application specific integrated circuits, like AI ASIC is the same. An AI chip is just an ASIC, right? So we’re going to get specific AI chips that run on chips that are much more efficient for training and for inference, so I’m expecting three or four companies to win in the inference market or become big enough in the inference market, that one of these conglomerates, big tech companies buys them, both on inference and on training, right?

(01:02:22):

And it might also be like AMD GPUs or AMD AI chips, so it might not be like a startup. It might be a big company, right? And Google TPUs are already really good, but Google’s never released that. You can use it through Google Cloud, but you can’t get those through any other clouds, and we already know that Amazon, Microsoft, and others are building their own GPUs, and then we know that Apple has been building their own GPUs for their own machines, not for cloud, but for their laptops, the iPads, and the iPhones. So that’s one thing on the chip level that’s going to happen, and I don’t think that’s unknown. I think that most people generally buy that. The second thing I think that’s going to happen on the chip level is that we’re going to start seeing a lot of on-device inference. On-device inference will come first, and then on-device training.

(01:03:04):

Then, the fact that we kind of understand now about data privacy, like I think in the 2000s, normal people were not really worried about data privacy and other stuff, because it wasn’t really clear what was happening with the data, but it is today, right? So we’re going to see a lot of on-device inference and on-device chips coming. And so, when we think about Lumina, we’re like, “Okay. On-device training is going to happen. We don’t know when, but we need to be able to support that,” right? Not just cloud-based training. Apple’s already talking. Apple’s already doing on-device inference right now, but they’re doing it with deep learning. I think generative AI and foundational model, on-device inference is going to be super interesting and will happen a lot more, especially starting with customers who are willing to pay for it. So Apple customers, I guess, will be the first ones. That’s the chip level.

(01:03:51):

Then, we have the software infrastructure, so that’s hardware infrastructure, and then we have software infrastructure. I think there will be a bunch of winners and both decentralized and centralized in software infrastructure, and we’re at that level. We’re working with GPUs and GPU providers, but we’re a marketplace, and Lumino builds out software. We don’t actually build hardware, which is why I call ourselves a software infrastructure company, even though we work with hardware infrastructure providers, right? Data centers and stuff like that. I don’t know if there is anything actually interesting to say in the software infrastructure level. It’s this is more enterprise business, right? Building out really cool tools for developers and stuff like that. Then, there’s the modeling level, which I think people find with the chip level like the most interesting, because that’s GPT-4, GPT-4o, Gemini, Llama 3.1.

(01:04:47):

The next set of things will be, there’s basically two pathways happening in foundational models. One is, how do we make models bigger and better, right? And so, I think GPT-4o is already over a trillion size in parameters, right? And Llama 3.1, 405 is obviously 405 billion parameters. There will be a push to make these models bigger. We will need more compute. It’ll try to do everything, right? So training costs go up, but we’re going to see over the next, I don’t know, three, four years, these models will be able to do more and more. The hallucinations that we’re seeing today, you won’t see mass scale hallucinations. You’ll see more nuanced hallucinations, and you’ll really need to be an expert to understand what’s hallucinating, et cetera, but the second point to that is that you’re going to see small, specialized models, and when I say small, I’m talking about a couple billion parameters to ten billion parameters.

(01:05:45):

So like Llama 3.18 AB, that are cousins of the very large models, being used in a lot more places, because you don’t need that much compute to do a certain job. You can fine tune a small model and get the job done, and it’s way cheaper and it’s way faster. I think we’ll see the big models being used for certain applications and smaller models being used for a lot more applications that are good for one specific thing or two specific things. I am curious to see what happens. The best model in the world, I think, in general, people think that open AI has the best model in the world, right? I think Llama is close for it being open weights, but not open source, and Mistral is closed. Again, it’s open weights, but not open source. I am hoping that we are actually able to build an open source and open weight large language model. I want AI to become more open source. This is something I actually think will happen.

(01:06:44):

So I’m predicting it now that somebody will be able to build an open source, large language model that is very, very good, let’s say like 90 to 95% of GPT open AI’s capabilities. I know we’re already almost there, but at scale. Even among the community, when people say open source, what they really mean is open weights, and almost everyone I talk to is either using Llama or they’re using Mistral, and those are open weights but not open source. So what I’m actually predicting is that an open source model will get mainstream usage in building models. Then, there’s the application layer, which man, things are about to change a lot in the next few years on the application layer. We use these models to build applications. When I go to these events in SF about AI, and just to be able to learn about sales and go to market, the amount of things that people are doing, reshaping the way that we interact with software and the things that software does for us today is breathtaking.

(01:07:46):

I love being in the Bay Area right now, with just the amount of stuff that’s happening that people are using AI to build applications for. I went to Theory Ventures go-to-market event a few days ago, and people are redoing how sales software works, how marketing software works, how cybersecurity software works. I mean, it’s all over the place, and it is great. So I think this is probably the easiest prediction, but I think we’re going to see more and more companies strip out these legacy software, especially the legacy vertical SaaS companies, and just build out an AI and UI, a user experience on top of that AI that does that job better, at a way cheaper cost, like a magnitude cheaper cost, one order magnitude or more cheaper. Then, somebody is going to be like, “Oh. Let me build this AI as a service,” and “Let me do what Salesforce does,” or “Let me do what Workday does,” or “Let me do whatever.”

(01:08:45):

Name a vertical enterprise SaaS company does, and I’m going to just re-imagine the user experience, so it’s going to be better, it’s going to be more efficient, it’s going to be faster, and it’ll be an order of magnitude or more cheaper, and I’m totally expecting that to happen and I’m super excited. At Lumino, we believe that so much that we always try out the new AI tools, so we use Claude and GPT internally for everything, development, research, et cetera. We are trying out this new tool called Cursor AI. I don’t know if it’s going to be great or not, but one of my friends recommended it, right? I just think the user experience of all of these tools that we’ve known for the last 10, 15 years, about to change. It’s going to be great. I am super excited.

Nick (01:09:28):

I appreciate you answering that, Eshan. What a great answer. So I hope all the listeners enjoyed that. I certainly did. I now want to move to the lightning round, where I’m going to ask you the GRTiQ10. These are fun questions. I ask each guest of the podcast every week. Listeners love it. I love it. It gives us a chance to get to know you a little bit better, but also, I always hope the listeners will learn something new, try something different, or potentially achieve more in their own life, so Eshan, are you ready for the GRTiQ10?

Eshan Chordia (01:09:56):

I am ready.

Nick (01:10:08):

What book or article has had the most impact on your life?

Eshan Chordia (01:10:11):

This is going to be hilarious, but Harry Potter, I have read the seven books over 30 times. I mix it up, I read them from front to back, I read them to seven to one, and every time I read them, I’m like, “Oh my God. J.K Rowling had put so much thought into it,” because when you read it forward, that’s like the normal way, and then you read it backwards, you’re like, “Oh. In this book, they did something, but they set it up in this book, so it’s easier to see foreshadowing being set up when you read backwards. I think Harry Potter, for sure. If I was to pick one book on entrepreneurship, The Hard Things of Hard Things by Ben Horowitz.

(01:10:48):

Best book I’ve ever read on entrepreneurship. It is detailed. It’s specific. He talks about examples. There’s general principles, and there’s examples, and you kind of can take what you want out of that book. I really, really like that book. Then, on a daily or a weekly read, I actually subscribe to The Economist, and I find it really, really good because they have sections on US, Americas, Europe, Asia Africa, and then it breaks up into economics, finance, science and technology, and culture. Kind of what I learned from that is, one, I don’t actually read any sort of news and politics outside of The Economist, unless something is really happening, because I find that reading news, it’s like something happened, but it doesn’t really matter, right?

(01:11:35):

Let’s take in 2016, we had Hillary versus Trump. Whatever some people wrote about back then has no impact on my life today. I can’t use that information to make better decisions in my life. So I find The Economists great, because they talk about real issues. It’s not fat issues. It’s like they’ll talk about microfinancing in Nigeria, or they’ll talk about how mosquito nets are saving people’s lives, and it’s like these things where it ties important stuff with real world events, so big fan of The Economist. Then, I subscribe to Noah Smith and Matthew Yglesias’s substacks and [inaudible 01:12:15] substacks. Those are some of the things I’m reading right now.

Nick (01:12:17):

Is there a movie or a TV show that you would recommend everyone should watch?

Eshan Chordia (01:12:21):

Psych. There’s this TV show, called Psych, on USA Network. It came out, I think, in 2009. It was seven or eight seasons, 2017. They’ve done three movies. Greatest TV show of all time. Don’t @ me. It’s amazing. My wife and I still watch it. If we need something light, that’s the TV show that goes on. Greatest TV show of all time.

Nick (01:12:43):

And how about this one, Eshan? If you could only listen to one music album for the rest of your life, which one do you choose?

Eshan Chordia (01:12:48):

This question is so hard. So in rap, I think, recently, I think my favorite album is probably The Carter IV by Lil Wayne, and then in EDM, it’s easily, We Are All We Need by Above and Beyond. I went to the acoustic concert by Above and Beyond, the We Are All We Need acoustic concert, and it was the greatest concert I’ve ever been to in my life. Then, there’s this Hindi album. It’s actually a Hindi movie that came out in, I think, ’98 called [inaudible 01:13:15], which, in my opinion, is the greatest Bollywood album of all time. So most Bollywood movies are musicals, and so they have 8, 9, 10, 11 songs, whatever. I think I would have to pick that one. If you’re giving me one, I think that has to be it, because it is absolutely amazing.

Nick (01:13:33):

What’s the best advice someone’s ever given to you?

Eshan Chordia (01:13:36):

Oh, so I don’t remember the exact quote, but I’m going to paraphrase it, because I don’t want to mess up the quote, but essentially along the lines of every person is unique, and you are unique as an individual. This was given to me when I was super young, so it was super early in my career, sorry. So it was really, really transformative for me, but it was like, “Don’t copy anyone else’s leadership style. You are unique. If you try to copy someone, that’s not authentic to you. People don’t buy that. People will not believe in you. You have to figure out what values drive you, what’s important to you, what’s not important to you, and lead in that way, regardless of whether it works for everyone or not.”

(01:14:21):

No one’s leadership works for everyone, right? I thought that was super transformative, because when you’re young, you’re like, “Steve Jobs and Elon Musk and Barack Obama, and whatever, these are charismatic people that people get behind and are considered greatest entrepreneurs of our time,” and you’re like, “I need to be that person.” This was my manager at a previous company, and he was like, “When you become a manager and you become a leader, you have to be authentic to yourself. Otherwise, you will not make it work, and that is the only way or that is the best way to be a leader.

Nick (01:14:58):

What’s one thing you’ve learned in your life that you don’t think most other people have learned or know yet?

Eshan Chordia (01:15:03):

I think we don’t give enough credit to humans. I don’t think I know something specific, especially non-obscure knowledge that other people don’t know. So I don’t have a great answer for this, but I do think that I’ve noticed that one way I’m different than other people is we live this superfast paced life. We have social stuff going on in our lives, and I’m an entrepreneur and I’m grinding, and sometimes don’t take out time, but just slow down. I kind of just slow down when I can and try to look around and learn. It can be for anything. It can be like I’m out for a walk, and I just kind of start people watching and observing how people are interacting, and I’m learning what people are doing.

(01:15:46):

How are people walking? How are people driving? What’s going on? How do people react to noises suddenly? Random stuff. Another example of that can be, I love reading books, both fiction, nonfiction, fantasy, sci-fi, and a lot of times people read for fun, but they don’t take the learning from the story or they don’t think about the learning from the story, so just slow down for a second and kind of put yourself in this fantasy series or in this fiction series. Think about what’s happening. I think there are a lot of unique learning opportunities that come up, that you just need to slow down and think about. So I don’t know if this is knowledge, that’s something I know, but I think this is something different that I do, that I think that a lot of people don’t do.

Nick (01:16:31):

And Eshan, what’s the best life hack you’ve discovered for yourself?

Eshan Chordia (01:16:35):

This is going to be super boring, but sleep. If I sleep eight to nine hours, I can work forever. It takes a long time to get tired. I think we always say, as entrepreneurs, we’re told you have to work 16 hours, you have to work 18 hours, but you have to take care of yourself, because there are limited amount of decisions you’re going to make, and making a small amount of best decisions are better than making a lot of average decisions, right? So for me, it’s get sleep.

Nick (01:17:03):

And then, based on your life experiences and observations, what’s the one habit or characteristic that you think best explains how people find success in life?

Eshan Chordia (01:17:13):

Persistence. If you want to work up the corporate ladder, and that’s hard, that’s not easy. You have to take on more responsibility. If you want to be an entrepreneur, you just have to be persistent. Things will go bad. It’s not easy. I think people underestimate persistence and hard work. Get through it. Do the hard work. I hear a lot of people today, they’re like, “Oh. The job market’s really difficult. I can’t get a job,” and it is difficult, but if you want a job, let’s say you’re trying to move into software engineering, do hackathons, contribute to open source classes, take Udemy nanodegrees. You have to put in a lot of work to get into the position where you get lucky, right? I’ve gotten so lucky in my life so many times, but it always starts out with I put myself in a position to get lucky. If I wasn’t in that position, I wouldn’t have gotten lucky. That luck would not have happened to me.

Nick (01:18:02):

And then, the final three questions are complete-the-sentence type question. So the first one is, “The thing that most excites me about the future,” and I know you don’t particularly like the term web3, so we can say, “Blockchain or crypto is…”

Eshan Chordia (01:18:15):

Financial markets moving to blockchain. Right now we’re T minus one for clearing houses and stuff like that. Can’t wait until it’s sub ten minutes or things to move, and everything, all these types of financial transactions are on Blockchain, like super excited for that.

Nick (01:18:31):

And this one, “If you’re on X, I still call it Twitter, then you should be following…”

Eshan Chordia (01:18:37):

If you’re into football, follow Ben Solak. He’s probably the best football analyst right now. If you’re into policy and economics, follow Noah Smith on Twitter

Nick (01:18:45):

And the final one, complete the sentence, “I’m happiest when…”

Eshan Chordia (01:18:49):

I’m hanging out with my family.

Nick (01:19:00):

Eshan, thank you so much for taking time to come onto to the GRTiQ Podcast. It was incredibly insightful. I learned new things today about AI, about entrepreneurship and startups, and it was a lot of fun to get to know you. If listeners want to follow your work, stay in touch with the things you are working on, what’s the best way for them to stay in touch?

Eshan Chordia (01:19:17):

Follow me on Twitter, on Warpcast, DM me, or LinkedIn. I’m active on all three, or you can always just email me at Eshan@Luminolabs.