Worldbuilders: Building Digital Minds and Why Humans Still Matter | Dara Ladjevardian (Delphi)

0:00I want to invent something that was not possible before. I'm interested in creating something new. And no one wanted our product. It was really painful. But then we had our first big believer, Brandon Burchard, this coach with a million followers. And he was kind of our evangelist. And a lot of other people heard about Delphi. More people came. More people came. At every stage, you're refinding product market fit in new user groups. I'm glad I went through that because I'm almost like it's supposed to feel like that for every user group. There's one version of you that gives up and pivots. And the amount of companies that have tried Delphi's idea that pivot four months later, because it's really painful to make work, is so many.

0:37So I'm just like, you just have to continue and survive.

0:49Sumeet Singh:My guest today is Dara Lajavardian, the co-founder and CEO of Delphi. Delphi is redefining how knowledge is shared by creating a new medium for human communication, interactive digital minds that allow you to scale your presence and allow others to learn from the best. Dara and I first met when he was building his first company, and we later strategized on how to leverage human psychology to build Delphi. In this episode, we talk about why human connection is still so important in this age of AI, how Dara knew he had to go all in on Delphi, and what principles got him. Enjoy the show. Hey, Dara, how are you?

1:23Sumeet Singh:It's me. I'm great. How are you? I'm great. I'm great. Hoya Saxa. Great to be here in the same city. Hoya Saxa. Yes, sir. I want to bring it back to the beginning, which is our beginning. I literally remember being on the phone with you. It was the middle of COVID. You were tinkering on ideas. I think I had reached out to you because there's not many people from Georgetown that are in startups. And I remember being impressed by you and just wanted to get to know you. And you told me about the e-commerce idea that you were working on at the time, Friday. Friday. I'd actually love to hear a little bit about what was the decision to go after Friday, to stop, to join OpenStore, to then leave and start Delphi.

1:59Sumeet Singh:I'd love to understand what was going through your head in those really pivotal moments. It was a very nonlinear path. I was working at C3 AI as a forward deployed engineer. Before it was cool. Before it was cool. I mean, this was really unsexy AI. It was predictive maintenance for oil rigs or turning text on papers into PDFs. Not as sexy, but I got early access to the GPT-3 developer list. And there was a community of people on Twitter that were building experiments, Jordan Singer, a couple of people. And there was some hype. And at the time, everyone was obsessed with crypto. So crypto and American dynamism were the hot things in tech, and LLMs were kind of like the weird, like, what is that?

2:43Sumeet Singh:But I can tell you it was not a main topic of discussion in our firm at the time. But American Dynamism were. Yeah. I mean, I was tempted. I joined OnDeck Catalyst to kind of figure out what I wanted to do next. And I definitely had shiny object syndrome. I'm like, all my friends are making so much money in crypto. Right. Like, should I be joining Olympus DAO or any of these other things? Yes. Yeah. But I decided to just like start building things and tinkering. And one of those first things, I was on Uber Eats. And the UX of Uber Eats is not great. I was like, I wish I could just buy anything with a text.

3:16I wish I could just text something. So for the On Deck Hackathon, we built Friday, which was literally that, connected to Uber Eats, Amazon, and Walmart, where you could text a phone number and it would buy it for you. And it went kind of viral on Twitter. And at the time, I mistook virality for success, which is not a mistake people should make. But it kind of kick-started my journey.

3:37Sumeet Singh:I was like, I should start this company. Did it give you enough signal to say, hey, actually choosing LLMs instead of these other catalysts to build upon is interesting? Yeah, yeah. I think when you first get into startups, external validation has a role. And at the time, for me, I had quit my job. I had no money. I slept out of my car for a week. So I was mentally really anxious. And this got a lot of virality. And I was like, I'm going to start this company. So I was super focused on Friday and was talking to users. I was very lonely at the time. I didn't really know many people in the tech area.

4:10I didn't really have any mentors. Right. You were working at this big company, not this inner circle of startup. I wasn't in the scene. I wasn't the guy who was going to happy hours like a lot of people do these days. And my dad had been pushing me to read a book about my grandfather for so long because my grandfather was like this titan in Iran before the revolution, had 30 ,000 employees at his peak. And it was during this time I was like, you know what? I was super stressed. And I was like, maybe I can find like a little bit of wisdom in someone who's like my family. And so I read his book and I loved it, not just because he was like a legend, but it was very much about his principles.

4:46And I wanted to be able to ask him, like, how could I apply these principles to my own? To your founding journey. To my own founding journey in AI. He was like in textiles and oils. He had had a stroke, so I couldn't ask him directly myself. But there was a book that kickstarted my career in tech, and it was called How to Create a Mind by Ray Kurzweil. And in this book, he says that the mind is a hierarchy of pattern recognizers, where a pattern recognizer takes an input and produces an output. And if you can stitch a bunch together, you can recreate the mind. Right. Like a token. Yeah, like a token.

5:16LOMs, prompt, output. And I was like, I can do this with my grandfather. Right. So I started doing a bunch of research. You know, there's people who created like a polygram bot. This was like while you were still doing it on Friday. While I was still working on Friday. Got it. Yeah, yeah, yeah. And I ended up creating a digital where my grandfather used it for advice. I actually was so excited about this idea that I almost almost ditched Friday and I built this thing called Helix at the next on deck hackathon. Okay. Got a lot. The hackathons were a good forcing function to like get to go from idea.

5:46Totally. This weekend I'm just going to build this. Yeah. Got a lot of hate on Twitter for that one. People were like this is immoral. This is stupid. It was also very expensive. So I was like I'm not going to start this company. I'm just going to continue working on Friday. A couple months into Friday I had like minor PMF with parents like people who are using it to buy like paper towels staplers it's so busy but i was like i don't care about this idea at all like i'm so not passionate and like i'm thinking about this other

6:11Sumeet Singh:idea well one had like real a real emotional connection and real emotional value for you versus oh this is an interesting technology catalyst i can build upon yeah which is of course important for startups but you know i i ask a lot about the why like when it drives you to the late Saturday nights, you're alone. You're thinking about the work. Right. Especially in this day and age when the cost of software is going to zero. To me, motivation matters more than anything. Why will you go through the near death moment? Why will you shatter through the glass ceilings? You need to have that emotional...

6:43Sumeet Singh:You need the fire. And so I had gotten an acquisition offer from a YC company, but they wanted me to come work for them. And I was like, I don't want to do that. So I shopped around. Was that because you knew you wanted to be an entrepreneur, that you didn't want a boss necessarily? I had a very specific idea of what I needed to do next. I wanted to go work somewhere where I would find a co-founder. So the culture needed to be a talent magnet. But the number one thing was actually I wanted to have a mentor. I wanted to learn from someone who was really respected and the best. You wanted to see what great looks like.

7:20I wanted to see what great for Friday who just wanted to buy the asset versus me. And I found OpenStore, Keith or Boy was the founder. Again, I didn't care much about e-commerce, but they were interested in me because I had worked on Friday. And I ended up moving to Miami to work for Keith so I could learn as much as I possibly could. And six months into working at OpenStore, I had built a bot of Tony Fidel because we were reading Build in our book club. And I was like, this is a lot cheaper than last time. And there's like a new paper on Twitter every week. And a learning lesson from Keith, where he says he only invests in companies where the majority of people laugh at him.

8:06I was like, maybe the fact that everyone was sending me death threats is actually why I should double click into this idea. It's a new category. Why is it so visceral? Yeah, it's a new category. Like Airbnb, Uber, like every category that was new, people thought was weird. By the time people think it's normal, by the time there's a market map, it's too late. So I was like, I got really motivated by Keith's mentality of looking at the world. I was like, oh, like this is something new. And I like, it's up to me to will it into existence. So a month before ChatGPT came out, left OpenStore, started Delphi.

8:37Sumeet Singh:Great timing. Great timing. When it comes to, I'd love to double click on that sort of, you know, the, what ends up becoming these mainstream consumer applications, marketplaces, et cetera. do get, you do get sort of weird looks in the beginning or... We still get weird looks. Yeah, you still get weird looks to this day. Like, why do you think that is? Like, why do people actually have that sort of visceral reaction at, you know, an adverse visceral reaction? And why do you think that is? I think all these things where people have an adverse visceral reaction involves some sort of social layer.

9:08Like you are being perceived in another way or you're interacting with someone like Uber, you're getting in a stranger's car, Instagram, posting photos online, like how cringe, like YouTube, posting videos online. You're creating a new kind of social contract, and people are very anxious about their reputation and self-conscious. But then if you just try to simulate what it must have been like to move from a society where we only shared knowledge by speaking to each other to like I'm reading someone's thoughts. Yeah, why wouldn't I just go talk to you? That's so inauthentic. Right. Like why would I read someone's thoughts?

9:37Why would I listen to someone's thoughts? And you almost can imagine that people had the same reaction, and then it became normalized. Mm-hmm. Mm-hmm.

9:45Sumeet Singh:Yeah. That's interesting. Something else that you mentioned when you're even starting Friday and focusing on GPT-3, how did you know not to go into those other areas you mentioned, whether it was crypto or American dynamism? Was it that, hey, there's just way too many people focused on this, and so there's no secrets to be earned? No. Was it the emotional connection to realizing, wait, actually, these language models, I can actually do something that has an emotional tie to something I've been thinking about, like with your grandfather? I'm curious how you knew to make that decision. It was kind of by accident.

10:20When I left C3, I actually applied to YC with a co-founder at the time. We were building anomaly detection systems for satellites. So it was literally American dynamism. He moved to Hawaii. So I was like, okay, this idea isn't going to work. And I also know nothing about satellites. I need to find experts. And then I then built a token tracker, a bot that joined every single Discord channel and would track how often a token was being mentioned. I wanted to make some money. I lost a ton of money. And then I was like, you know what? Crypto is not my thing. I'm going to lose a lot of money in this space.

10:56American Dynamism, I just don't have the network. LLMs, I can build. Like I'm building this pretty fast. Yeah, and you were a Ford template engineer. Ford template engineer. I built like a bot that allowed me to buy crypto through techs. I built a bot that connected my Neurosity device to order Uber Eats whenever I felt the sentiment of Hungary. So I was like, LLMs allow you to just build all these really cool things.

11:18Sumeet Singh:Yeah. In a way that's actually useful and meaningful. Yeah. Interesting. And when it came to, I guess like even leaving C3, right? You're serving these enterprise customers as a forward deployed engineer, working with some of those industries that you mentioned. Why build a consumer product? I just think it's more fun. I think there's like such thing as founder market fit and like i'm kind of a creative in a way i also like like to move very fast i think the thing at c3 that i didn't love was just like the slow cycles um bureaucracy you don't really get to be creative you're solving someone else's problem and i remember in on deck uh you write your goals at the beginning and one of the goals that i had is i want to invent something that was not possible before like i literally wrote that down like i want i'm interested in creating something new And I think that's like a double-edged sword of mine where I've always wanted to do something that was different, which puts me down the path of most resistance and hardship.

12:18But that's just, I think, kind of how I am.

12:20Sumeet Singh:Yeah. No, that idea of building something new is something that I think a lot about here at WorldBuild, which is I think a lot of the applications in AI are very skeuomorphic. It's very much the beer-drinking app version of the iPhone apps. What I think is more compelling is post-skeuomorphic. You start first with the nuances of these models. Like, what can we uniquely do today with models that was literally not possible before to create these new experiences and workflows? How has that evolved since, you know, sort of day zero at Delphi to today, you know, within the acceleration is what I call it that we're in now.

12:55I mean, it's funny because when I reflect on day zero at Delphi, it feels like we're in a different reality. Like, I don't know about you. It just feels like we've jumped dimensions.

13:03Sumeet Singh:and like... Well, I remember literally talking to you the first time you're getting started, jamming on the memo. And to get to today, yeah, it totally feels like a different time. Different time. And a lot of things have changed. There have been a lot of distractions, a lot of shiny objects. I think with a company like Delphi, especially when you're building during a time when things are changing so fast, you want to anchor on the things that are not going to change. That has kind of been my north compass. People trying to pull us into building an agent assistant or people trying to pull us into like building Clodbot.

13:34Right, right. It's a hot thing today. It's like the hot thing. It's like, okay, what is not going to change? Humans are going to exist. They're going to need representation. Humans care about what other humans have to say. Let's anchor on those things and see how AI can help us solve this problem versus anchor our entire company around the technology.

13:52Sumeet Singh:Yeah. When did that become clear to you? Because in the beginning, right, it started as, hey, let's tinker on this technology to then actually wait a second. here's the company that we want to build and this technology will propel us forward not vice versa well it was kind of around the problem of access you know i didn't want to start a company around like death tech and legacy so i think that thing like monetizing grief is kind of dark but my grandfather was like how do we access other human beings i've like never had mentors my whole life and now i can learn from these people but then character i came out and i was like hmm character i is gonna have issues because it's like all fake it's a bunch of characters So like, okay, for Delphi to win, we actually need to build a product that incentivizes people to create digital versions of themselves.

14:35Why would someone do that? They want to represent themselves the same way someone creates a LinkedIn profile or a resume or writes a book. They want to scale their knowledge. So that was kind of the idea maze where it's like, okay, first let's go consumer. But actually, if you're thinking long term, we need to create the incentives first to collect the data and get people bought in.

14:51Sumeet Singh:Yeah. And interestingly, maybe, you know, this will get a little specific, but we had talked about, hey, relationships are at the top of Maslow's hierarchy of needs. Yes. This is something that we spoke about when you were working on your early memo. And, you know, we jammed on that creatively. But interestingly, like, you know, there is almost like two paths of Delphi, right? One was the more characterized companion, digital clones, relationships path versus the more utilitarian path around helping me scale my audience, my reach, but then also helping someone learn something new. Are those two different paths?

15:26Sumeet Singh:How did you think about that? The way I make decisions usually is what leads to the biggest outcome and usually what is the hardest, unfortunately. The hardest, yes. The first path probably would have been easier, but I think commoditized. You can make self-help bots. And when knowledge is commoditized, what you can't really compete with is trust and owning the identity layer. And so by creating the utility of having a Delphi, we actually end up doing the first thing in the long term. Because we end up getting all the best people you could possibly want to learn from the world. And it's verified, so you know it's them.

16:01You trust it more, and then we can recommend you the best people. And it can help you that way.

16:05Sumeet Singh:Yeah. So is it an active bed of Delphi that you think IP, in this case, individual IP, will still continue to be very important even in the age of AI? Yeah. What can't get commoditized? Real things, real estate, people. And people have... Delphi is a verified version of you when all these social media platforms have deep fakes and bots. We want Delphi to be the place where you know who said what, and I know it's true. But do you think like AI breaks IP or do you think IP will always, you know, have importance? And in this case for Delphi, right, IP is the individual's IP. Yeah. Right. In other cases, it's, you know, we've seen, you know, Disney go after folks, right?

16:44Sumeet Singh:Like just curious to get your thoughts on this because I think you do touch, you know, again, individual IP, maybe not, you know, a creative studio, but perhaps you could scale that one day. Yeah, I mean, I think about IP. Right now, IP is something that is like a rare, where I think IP grows into perspective. You care about what certain people have to face. You follow people on Twitter, even though you could probably learn everything you possibly could learn on Chachi. We're having this podcast because there's another human. So it's almost not that I'm owning my IP and then I'm going to sell it, but this is just a digital version of me that you know is true.

17:18This is how I want to be perceived in the world. And so the humanity of it, the human connection still matters. That's our bet. Yeah. That is kind of the bare case of Delphi is we don't care about humans anymore. That's when people ask you, like, why does Delphi fail? And it's like, Delphi fails if we end up in a world where we're just talking to Chachi BT and we're just talking to Claude and we actually don't care what anyone else has to say. We don't meet each other. We're not curious about each other. I don't think that's going to happen if you just look at, like, human psychology and habits over time.

17:49But I'm an optimist. So we shall see. Yeah.

17:53Sumeet Singh:Well, it's interesting, right? Like even if relationships are at the top of Maslow's hierarchy of needs, the fidelity of those relationships are still important. So, you know, I think you're, you know, from what it sounds like, it's, you know, maybe a similar bet to some of the companion apps, but you're much higher fidelity than a made up character in that, hey, this is someone that I look up to. This is someone that I respect. This is someone that I want to learn from. Or meet. Or meet. Delphi is how humans represent themselves, but also how people meet. Like eventually, you know, Delphi talking to Delphi.

18:18Sure. This is the person who thinks most similarly to you. But I think the insight with my grandfather is an important one. It's not that his clone said anything to me that was so profound that helped me. It was the fact that it came from someone that I trusted and I was more likely to absorb it and adopt it. Because I could have read that information anywhere. It would have been like in a self-help book. But like hearing information from someone you know and trust, I don't know, it had a bigger impact on me. And so I think similarly mentors or people have impact on each other.

18:49Sumeet Singh:We talked a little bit about this sort of like adverse visceral reaction to new consumer platforms in the early days. What's like the positive version of that when someone uses Delphi for the first time, whether it's someone who's trying to create a digital mind of themselves or, you know, a customer? What's the visceral reaction that you want to induce? I think for the person who is creating a Delphi, it's almost like never repeat yourself again and only spend time on things that matter. So examples is you talk to someone's Delphi before you meet with them, and they already kind of know everything about you, and you can just talk about new things.