9. Shravan Nageswaran, Atman Labs
In this interview, Shravan Nageswaran, founder of Atman Labs (formerly involved with Worldcoin), discusses his "conscious flywheel"—the interplay of awareness, agency, and synthesis—as a framework for personal and organizational growth. He contrasts zero‑sum and positive‑sum energy flows, and shares practical habits (reflection, openness, action, comedy) for expanding awareness. Drawing on the Kenyan mobile‑money system M‑Pesa, he explains a depth‑first scaling strategy that targets deep penetration of peripheral subnetworks before expanding outward, and applies this to Worldcoin’s early growth. Shravan recounts why he left Worldcoin in 2022, citing the limits of large language models (LLMs) that excel at single‑shot deterministic tasks but falter on open‑ended, multi‑step reasoning. He outlines his vision for a new AI architecture that revives expert‑system ideas: building large, evolving knowledge graphs and using reinforcement learning to generate decision‑tree strategies on the fly. He critiques meta‑prompting as insufficient, demonstrates how their system outperforms LLMs on a country‑guessing game, and argues that such reasoning models can be instantiated across domains (medicine, sales, recommendation). Finally, Shravan connects his work to Vedanta philosophy, explaining the naming of Atman Labs and Othman as an attempt to bridge individual consciousness (Atman) with collective knowledge (Brahman), aiming to create systems that expand human awareness rather than merely automate tasks.
Chapters
Hello everybody, welcome back to Chaotica. Today I have Shravan. Shravan is the founder of Athman Labs. I was lucky enough to invest in Shrav a couple of years ago when he was just starting out. He's probably the founder that has influenced most my way of thinking and my way of understanding and assessing talent. So super grateful to have him here today. He previously did stints at Worldcoin and he built other projects before. He has a super interesting journey. And the conversation we have today is extremely conceptual. So please go on this conscious journey with us because it's going to be an interesting one. One place to maybe start is to talk about the first time and the interactions we had. In the sense that during the investment process, it was really quite unique because I met you on a, I think, Thursday, and then on a Friday, like I met you for 30 minutes on a Zoom call and that 30 minutes became an hour and a half. And then you were like, actually, let's— I have a 30-minute meeting, but can you hop back onto the call right after?
Yeah.
You told me and I was like, okay, sure, let's do that. So we hopped back onto the call and we did, I think, another hour after that. So we spoke 2 hours and a half. And then I asked you, can you come in person the day after?
Yeah.
So you came and we did, what, 4 hours in person at the whiteboard?
Got a whiteboard out.
Yes.
Yeah.
So you came in and you sketched out the whole plan for, for Athman and what it was going to be and the architecture that you had in mind. And it was interesting because it was super conceptual in the sense that, yeah, it, it really was built from a place of first principles in a way that you were deconstructing, okay, what are we trying to achieve here? And at the time, if I remember correctly, I mean, the core that you were trying to achieve is a system that is, yes, simulating human expertise, but the consciousness itself was, was important. You know what? What does AGI mean? How do you achieve that and consciousness within that? And you deconstructed how that is achieved in humanity, in human systems. And then you tried to formalize an architecture around that. So that was interesting. So my question for you maybe to start is—
Yeah.
What does it mean to be conscious in humanity for us in the first place?
Yeah, I mean, it's definitely a question, I think even to preface, right? I think it's a question that none of us in our lived experiences can potentially even truly understand. And so what we have at our disposal is the ability to make frameworks and think in abstractions and at least think that we know the answer, right? And if you asked me right now, I think that when I think about what is a framework that can be used to explain our existence and define what it means to live more consciously versus live less consciously, I came up with this framework called the conscious flywheel. And I guess if you think about the universe as a whole, it's just this massive composition of energy. And as humans, what do we do? We go through the world, we consume energy, we transform energy into different manifestations that we call different things, and we continue that cycle basically. That is our life. And so when we think about what it means and how we transform energy and why we transform energy, I think This idea of a conscious flywheel is broken down into 3 major pieces: awareness, agency, and synthesis. And so I think our conscious experiences, at least our individual conscious experiences, fundamentally start off with awareness. And awareness is, I think, what it means to be us. It's our mental model of the world. It's the abstractions that we use to define the things in front of us. It's the abstractions that we use to define know, things that we like or, or don't like, things that, you know, we think are good or, or bad. And we all have fundamentally a completely different awareness of how we perceive the world, how we perceive our existence. But it's that mental model that kind of like makes us human. And then using that awareness, we can decide to take action in the world, and we take action to basically transform energy around us. Taking the action to have this discussion, even a conversation, it's not a physical transformation of energy, but it's like a mental exchange of ideas, which in itself is exchanging energy. The idea of creating a piece of artwork or even the idea of any action that you take in the world that was guided from your awareness is what we call agency. And then the outputs of that action is synthesis, right? The output of taking an action ultimately transforms energy and creates a new reality. And ideally, the synthesis that you create, maybe you design a new piece of artwork or you decide to explore and learn about a new subject. And that synthesis, that process of actually completing the action and the action itself, The reason why it's a flywheel is that should ultimately expand your awareness. And you kind of have this flywheel where as your awareness expands, the actions that you take change, and then how you transform energy changes.
Why does it feed back into awareness? Is it because there is a reinforcement mechanism where I see the outcome of my work in synthesis, and as a consequence, it affects my awareness?
Well, let's take this conversation. we might introduce new ideas to each other. And just by committing to that process, the ideas that you introduce to me or the ideas that I introduce to you might change my mental model on certain things. And ideally, a highly conscious conversation versus shooting the breeze or talking in pluses and minuses. I think, as you said, the Italian expression was— Should lead to a situation where I leave the room and the actions that I am going to take would be different had I not interacted with you. And so I think there are individual conscious experiences, right? Maybe it's like learning about a new subject. And then fundamentally, because we don't exist in silos, right, there's also these joint conscious experiences. We come together, we generate a new outcome, it could be a conversation, it could be something physical. But ultimately, that then should expand our awareness and influence the next actions that we take. And I think the idea of what it means to live consciously versus what it means to kind of live subconsciously or unconsciously is, I think if your awareness of the world is quite deep and quite nuanced versus an awareness that is quite quite shallow. I think that's probably like an indication to where, you know, I think aware— you can have kind of deeper awareness on a, you know, given— for example, if I'm looking at this plant, right, if I'm able to kind of describe this plant in, you know, with so many different abstractions and connect that to other experiences that I've had in the past, versus if I just look at that plant and say, oh, it's a plant, and kind of don't really connect to it. you can have deeper awarenesses of the world. And I think when it comes to agency, if you're able to proactively take action into the world and take action with intent versus acting reactively, I think fundamentally, what's an example of an unconscious action? You throw something at me and I reflex. So I'm reacting. I'm not understanding deeply why I'm taking that action and how my unique awareness has brought me to that position. I'm just trying to survive, right? And fundamentally, I think how you kind of determine, you know, truly acting consciously in the world or not is, are— do the syntheses that I create, you know, does the way that I truly transform energy lead and generate output that not only expands my awareness but can expand the awareness of others that that interact with it. And it's just like, for example, why artists, for example, are considered like, or at least the best artists are considered highly conscious individuals because they see the world in a different way. They have motivation, they set an intention to translate that awareness into something physical.
Interesting.
And then the output. Think about the last piece of, not to digress, but like the Beatles, for example. Have you ever heard from the album Revolver, the song Tomorrow Never Knows?
I'm sure I have, but—
It's a lesser-known Beatles song. Basically, they took psychedelics for the first time and they wanted to generate an output that could show the average person what it would feel like to be on psychedelics. And that creative process led to an output that should expand the awareness of others. And so I think And most people, and I guess the less you think about it, it can be very easy to go through life relatively unconsciously. The content that you consume, the routines that you fall into, the jobs that you have, the people you interact with. It's just like what it means to live consciously is, are you existing in a world where your awareness is constantly expanding, or are you operating with the same mental model and just surviving day to day?
And your approach is that I mean, your perception is that it's net good or positive, or there is a positive quality to the increase of awareness because that enables better action and synthesis as well, I guess.
Yeah. I mean, I think it's hard to think that there is an absolute truth, like a good or bad, but I think about it as there's 2 mental models of existence. The first is surviving, right? And the second is thriving. And I think naturally, you can't exist 100% in one bucket, in the bucket of thriving versus surviving. But if you asked me, what is thriving and what is the purpose of humanity, or what it means to be human, I think self-expression is a huge part of that. And I think fundamentally what self-expression means is, generating syntheses that reflect your unique awareness of the world. And as a side note, if people say, oh, but actually reproduction is the purpose of humanity, reproduction is actually the most conscious of syntheses. Raising a child through the world with your unique mental model is an example of creating a synthesis that's a projection of your awareness. But I think fundamentally there's probably no absolute truth on what it means to to what's the ultimate, what's the goal of life, right? But if you asked me, I would say, actually being able to, we're on this world for a short amount of time, to be able to understand as much of this world as we can in a very nuanced way, and to actually be able to connect with people on a deeper level, AKA connect with people in ways that expand collective awareness.
Mm-hmm.
is what it means to thrive versus kind of just survive, which is you have your kind of fixed and shallow mental models of the world, and you're just kind of going through life reinforcing those mental models and surviving. And obviously, I think it's not 100 to 0. There's balances, right? If you really want to expand your awareness by doing something deeply dangerous, then if you have no more awareness, then there's, there's nothing left to expand.
What's the— what's the— not correct, but what's the, uh, direct— you know, how do you assimilate information in a way that expands awareness? How does that process work? Is it a, you know, a conscious sitting down where you try to assimilate and then it expands your awareness? Do you write? Is there like— I assume for everybody it's different, but it seems as though there is a preferential way of doing so, uh, to some extent that works better. Is—
I think, I think it's about reflection on, on the, on the things that you consume, right? It's like, for example, like someone, an artist or a poet that like is more conscious, right? And for example, one, one example of self-expression that, that I, I love and I actually have done in the past is, is stand-up comedy. And when we talk about awareness, standup comics, they can look at something, they can sit on the tube. And when most people have their heads down trying to not be awkward or talk to anyone, a standup comic is just observing everyone on the tube and they can see things about the tube and they can see connections and they can connect that to other abstract topics or ideas in ways that most people don't or wouldn't dare to do. And that leads to a synthesis that makes people go like, oh, wow, we are actually that way. We are kind of strange. And it's like, I think, one of the most purest examples of having a heightened state of awareness and generating synthesis from that. And so I think when you ask, how do you then expand your— how do you actually think about, or I guess, how do you— what is the process of expanding your awareness? I think it comes from actually reflecting on the syntheses and the actions that you've taken and the things that you've consumed and connecting that and putting that within the framework of your existing mental model and going, does this break it? Does this reinforce it? Or does this actually expand it in a certain way?
And do you practically take time for that?
I think it's not something that is, you know, requires like at the end of the day, you don't have to sit down and say like, these are all the things. So it's a way of life in the sense that it's just a way of life. It's just like, you know what it means. You're someone who's deeply curious, right? And like even in this conversation, you know, both of us, we can like end this conversation discussing a lot of new ideas and go down and actually like write down, this is interesting, this is what I agreed with, this is what I disagree with, this is what I take away with it. But it's just something that happens naturally because we're open, right? We're open to it. Whereas like, if you're not open to it, if you go in with kind of like a preconceived mental model, um, of a conversation, for example, and you, you can, you can both be talking, for example, you won't really be listening and, and digesting the conversation and your mental model won't really be expanding. So I would say it's more of like a way of life. Like you either go through life kind of open and just trust the fact that, you know, as I expect, as I ingest more energy and ingest more information, my awareness will, um, will reshape itself, right? Or you go into it kind of clinging to like what you believe is your mental model of the world, to where even when you interact with something, it won't really penetrate you.
Indeed.
And so yeah, I think it's just like we talk about the universe being, being energy, and we are just as humans, we're channels of this energy. And so I think it's really like, do you let this energy pass through you, or are there like certain blockages that, you know, prevent you from actually experiencing the situation in its purest form and allowing that to truly like change?
What kind of blockages, uh, could they be?
Blockages could be ego. I think fundamentally, right, it comes down to when we talk about surviving or thriving, right? It's like love versus fear. And so one example of an abstraction that gets in the way is our ego, which is, I think, just like an abstraction that this is a plant. I am Shravan, you are Michelangelo, is an abstraction. But we use these abstractions to protect us. And so I think ego and fear are examples of I either might not take an action because I'm afraid that it will end up this way, but it could also end up in a different way. And ending up in a different way expands my awareness. Ego, for example, I think certain cultures can be quite ageist. And so ego would be like, this person's 20, I'm 50. I have so much more lived experience. There's nothing I can gain from hearing some 20-year-old, even though their awareness might be shaped from a completely different thing. And so, yeah.
And so that is a— but is it only about protection? Because to me, it's also a lot about saving energy. As you say, there is a— it's all about the energy flow. And so it seems as though we use these abstractions as proxies for concept, because I'm not going to recompute the idea of a plant on the fly every time I see one, right? Because right now I'm focusing on the conversation, so I don't want to from scratch rebuild a model of what this is going to be. And so we build this abstraction layer for that is a plant, this is shrub, etc. To— yeah, so, so I, I don't necessarily think that it's always someone's fault. Oftentimes it seems to me as though it's good actually to save resources on some things. It seems to me— maybe please challenge me if you disagree— but it seems to me as maybe the most interesting place in which to put energy is not to consciously compute the carpet and the plant, but to focus on it, right?
Yeah.
And so I kind of tune everything else out so I can consciously be present here. Do you see energy as finite or infinite? Like, is this a zero-sum game or positive sum?
I think that Most of us see the world as something that is zero-sum, right? Whereas I think that fundamentally energy can be positive-sum and can be compounding. And so it's a question of, I think your point, right? I'm not going to consciously reimagine the idea of a plant. or a carpet, but it's like, given the situation that's in front of me, what is the best place to channel my energy that would lead to something that expands my awareness in the highest derivative? And in our case, it's talking to each other. If I was sitting in a room by myself, I was locked in this room, and I was someone who thought that energy is compounding versus zero-sum, and you weren't in front of me, then I probably would really look at this plant and deeply ruminate, because what's my alternative? My alternative is just being bored and waiting.
On your phone.
Yeah. And so it's like, fundamentally, I think with everything, especially with consciousness and energy, we won't know in this experience if there is an absolute truth. truly unlock in our lived lives with the capacity, the cognitive capacities we have. There's a theory that we will never understand consciousness. We will truly never understand truths of, is energy zero-sum or is it infinite? And so I really think that the idea of believing in something, Pascal and this idea of Pascal's wager, right? Of like, I want to believe in God, not because I think God exists or I think God doesn't exist, but the power of believing in God is something that can make my lived experience feel more limitless. And so I'm choosing to believe that I think there is an infinite flow of energy that can exist. And then to your point of like, You know, it's— it becomes like the agency of where I direct my energy and how I want to transform energy. That becomes the, the challenge and the decision that each of us make every day.
Yeah.
The agency of like, I want to direct my energy towards you in this conversation versus tune you out and just look at this plant and like ruminate on that existence. That's an important decision that I make. Um, whereas like if you weren't in this room, Am I going to take agency to try to expand my awareness in some form, or am I going to just waste almost?
It's interesting. It sounds to me as though you are not 100% certain about whether it is zero-sum or positive-sum. And it sounds like you're choosing quite deliberately to lean one way because you feel as though that's the positive recursive loop. But what happens if you're building your frameworks onto a wrong assumption there?
I think you have to tell yourself, and I think you have to be— because a scientist will always say, if something isn't proven, then I'm not going to adopt it to my mental model.
But I think we don't have time.
Yeah, I think it's just this belief that if something fundamentally— if you think that something fundamentally can't be proven in this existence, in this limited existence, then you are the power of your beliefs.
Right.
And I truly believe that your awareness determines the syntheses that you create and the actions that you take. How can you frame your awareness in such a way that truly allows you to act in the way that you feel is most conducive for you? And to your point of maybe this awareness was founded off of a wrong assumption, A, I don't know if that assumption can be verifiably proven in our lifetime. And so I'm acting under the assumption that that isn't the case. So it's a choice and it's a choice of free will. And I think the choice of assuming that energy is infinite versus energy is zero-sum and the purpose of life should be to expand our awareness, I'm taking that choice because I believe that my lived experience will be better because of it. And then B, I think In terms of, I mean, going back to this idea of Pascal's wager, it's like, I believe in the power of believing. Because if it was true that energy is infinite, and I spent my life existing in a zero-sum way, the outcome of that would be far more negative than if, for example, energy was zero-sum, but I interacted in life as if it was was infinite. I think it would be a worst-case scenario to have been living in a very limited way if life was limitless versus living in a limitless way when actually life was limited, if that makes sense. It does.
But I feel that I'm not so sure that if we assume that there is a zero-sum game to be played, it's fundamentally necessarily a negative thing. It could very easily actually, you know, provide a lot of meaning, provide a lot of added— oh, this is temporary, it's an existence. It doesn't need to be a continuous energy flow that goes infinitely. It could very well be this, and it forces the sort of conscious function because of it.
Um, so you're saying the constraints almost breed creativity, or—
I, I'm not sure. Well, yes, in the sense that Interesting. Well, yes, because this is what we see in art as well, right? This is— I mean, this is McLuhan, the medium is the message, in the sense that if there is a specific tool used, that breeds a specific type of output rather than another. Um, so I'm, I'm not so sure if it's necessarily a good thing or not, whether it's zero-sum or not, uh, in terms of flow of energy. But, um, I'm— I just don't know whether it would drastically influence or, or, or drastically change the way you would live your life. You would still try to, uh, gather that energy. You would still try to process the information, build your awareness, process it maybe even more so because there is little of it.
Interesting.
It doesn't sound like a super harmful, uh, assumption in the sense that either way—
I think it's more of like, I guess when you think of zero-sum, right? It's like the idea that if I create energy or I create some feeling, someone has to lose it. And I guess we talk about these buckets of surviving and thriving. And I think surviving is built on the idea that resources are finite. And so it's like we have to collect these resources. And when I collect a resource, you don't. It's like there's— And I think Thriving is just this notion that we can all have positive energy. And so I actually think that there are certain times in life where, yes, we have to operate under a mentality that resources are finite because that is the society we live in. But for example, if you think about this conversation, it's like, I think this conversation can be a compounding exchange. of energy. It can be where ideally both of our— yeah.
But it is taken away from somewhere in the sense that someone listening to this will have to allocate their resources towards this rather than the plan that is in the room, right? Like, there is a flow of energy that is focusing on this that could have otherwise gone somewhere else. Like, the computational power is fundamentally limited, it seems to me.
Yeah. So, yeah, so you're saying that like Yeah, I guess for me, what I think about when I think about energy is the idea of your mental model changing, even from a chemical and neurological standpoint, is something that generates energy.
Right.
And so if I'm looking fundamentally only at this plot, and I'm just thinking this is a plant, right? And my mental model isn't changing. Actually, the energy which I'm— I'm not generating much energy because my awareness fundamentally isn't changing. But I think in our interactions, if we are having a discussion that is quite deep and expanding our mental models, I think about the energy that is generated. Right versus the energy that is like there's like physical energy that's like what it means to like actually have this discussion and like you know actually like expand energy expand energy towards having this discussion but then there's the output of that discussion which is you know if our awareness is like truly expanding if our our thoughts on the world are truly expanding and how we see our existence and how we see the world truly expands and like even like fundamentally, there is like this expansion of a mental model. There is this energy that is required, or that is generated, and even like processing that and expanding our mental model. And I think the point is, is that in this conversation, it's like we might— just by having a conversation, we might— that might require like X energy, but the resulting energy of both of our awareness expanding is greater than the energy that we put in.
And I wonder where we gather that additional energy from.
Yeah. I think it's like the process of a conscious synthesis generates a lot of energy to where— and I think it's this notion of you're not necessarily expending energy in a way that makes you tired. For example, you can go into a job that you hate where you kind of know the, the output every day, and you're, you're generating, you're, you're expending energy, um, but you're not receiving any energy back, right? You're putting that energy into the universe, but your awareness isn't fundamentally changing. And so that's why it feels zero-sum. It's like, I, I put all this energy in, right?
Right.
And, and now I have no energy, right? And so that feels zero-sum, but actually, like, energy and energy flowing through you is like a good thing. It's like being charged, right?
And so it's like, right, you know, so it seems to me like, okay, independently of whether there is a zero-sum game of energy or positive sum, it doesn't really matter. But in your daily life, uh, pursuing activities where there is a positive-sum exchange seems to be the path to, to, to do well.
Yeah. And, and, and pursuing activities and in this case, the positive energy and the compounding energy comes from an expansion of your awareness and your mental model where if I take this action, the next action that I would take would be different. If I take action A, then the next action that I would take, action B, would be different if I hadn't taken action A at all.
And so, I mean, it's interesting what you say about— so into someone living a conscious experience that is The thing is, a lot of people are stuck in jobs you don't like, which you just mentioned very, very simply, where there is a zero-sum exchange of energy. Yeah, but they still have to do it because, you know, it's the reality they live in. The challenge here is that you propose a world in which you do the exercise of thinking positive, something, and you need to gather the strength to get yourself out of the situation.
Yeah, it doesn't mean also like I don't want to be the the founder that comes on a podcast and says like quit your day job. Like exactly. But it's like and you can be in a day job like intentionally, right? And and and there's gather energy for something else, right? Gather even gathering energy in your day job as well because it's a frame of mind, right? If you go into it thinking this again, the nine to five and waiting for the clock to hit five, then It's just like if you're in this room with this plant and you're just thinking of like, someone let me out, right? It's like your energy is not, you know, and the frame of mind that you have is not something that like can, of course, cultivate energy. So it's really a frame of mind thing. And then, yeah, sure, if you're not able to expand your energy in your day job, like, do you have aspects in your life? Do you have relationships in your life that can generate compounding energy?
And so what are the ways you cultivate compounding energy in your life? I mean, You're a founder. Tell me a bit about the different paths you've taken to have this positive reinforcing mechanism around you of energy.
Yeah, it's a good question. I mean, I feel like for me, my awareness and what gives me compounding energy, I feel like was shaped— obviously everyone's awareness is shaped by my upbringing, but I think some of these frameworks I have on life as well as the actions I take to strive for compounding energy was shaped a lot by my upbringing. I mean, I think my dad and my mom are 2 examples of having really high awareness, but in completely orthogonal areas. So the example is my dad is the ultimate scientist, in the most respectful sense.
I met your dad.
You met my dad, right? He's like the ultimate nerd in the best way possible. He's the kind of guy that growing up, he left India when he was like 18 and wanted to come to America, and he does master's degrees for fun. He did like 3 master's degrees. He's deeply interested in science and understanding the universe through these physical phenomena. And he is also someone who is quite Quite expressive and likes to see the world differently. I think he's—I think at least twice he's like left corporate jobs once, like right after I was born as well too, which is not—not a time when people say you should leave a paycheck that's coming in. But he left corporate jobs to start his own businesses, and you know one of his first businesses was like, you know, he is like obviously a. A geek on data and infrastructure. And so how to help small businesses set up data systems in the right way. And he ran a consulting gig for it, and he started basically making his own courses and doing that all over the US where I grew up. And one of my earliest memories was he turned my childhood bedroom into a bookbinding station where he would print copies of like the, the books that he would, he would use in his courses. And I just remember like binding books with him when I was like, uh, like 4 or 5 years old. And so my dad always, always kind of saw the world in this, you know, ones and zeros way, but like at a very deep, at a very deep level. He was, he was a scientist. And but he also kind of gave me the courage that like sometimes the best way to express yourself and the ideas that you have is to like You know, work, work, start a business, right? And so that the idea of starting a business wasn't foreign. But then on the flip side, and kind of going back into like the interpersonal part of our conscious experience, my mom was someone who like she's worked in the same job for like the last 20 years, right? She's a school teacher. But I think the thing that I always notice about my mom is like where my dad is like quite like ones and zeros, quite intense and positively geeky. My mom is like someone who is deeply, deeply sociable in a way that isn't frivolous, but in a way that comes from strongly understanding people's characters and philosophies. And so my mom is like one of the most incredible interpersonal judges of characters that I know to where I remember being 10 years old and having friends come over. And after I would leave, my mom would be like, this guy has your back. This guy probably won't. If push comes to shove—
She was good.
She was good. And my dad's like, what are you doing? Don't be judgmental. These are just 10-year-olds playing video games. But the same people that she said would have my back or not are still in my life today. And she is someone who's quite magnetic, and people really feel safe around her. And at the same time, she understands group dynamics incredibly well to where, you know, she could, whereas like, you know, my dad, for example, hated corporate America because of like certain group dynamics. And also he found like he wasn't able to express himself.
Mm-hmm.
My mom could actually thrive in those environments, whereas like she could like dodge her way through like a corporate drama minefield so easily because she deeply like understands people and, For me, I always thought my awareness growing up was shaped in how do I understand the physical world in a lot of depth? And obviously engineering is stuff that I learned directly from my dad, but also how do I deeply understand interpersonal dynamics and philosophy and my awareness of character and how relationships work? That I kind of gained from my mom. And I felt like for me, like that awareness that I gained from my parents was like multi, multi-dimensional. And I think it kind of manifested itself in like naturally, like being a founder, not only being able to have like an idea on, you know, these are types of products that I want to build and this is how I, how I build it. But also these are the types of— this is the type of culture that I want to run that really like creates an environment of like tremendous energy and brings out the best in people. For me, both of those are as important. And I'm sure we'll talk about this, but that's kind of how my awareness was shaped and kind of led to things that I find give me a lot of energy. And yeah, I think even just personally outside of that, things that give me energy are You know, we talked about stand-up, stand-up comedy. Like, I actually used to be into stand-up comedy. I would perform shows.
And is there any gigs out there that— or on YouTube or something that we can still—
There are actually not any gigs on YouTube, but my claim to fame was I did do a show where I opened for a Netflix— someone who has a Netflix special. No way. Her name is Cristela Alonso. And I actually I actually opened for her. I only did stand-up, performed a handful of times, but I love comedy because it's an example of like, you're projecting a deeply unique awareness on the world in a way that connects concepts that like wouldn't be connected before.
It's true.
And that synthesis, if you— first off, if people are like laughing and they're kind of like diffused, they'll naturally be more open to actually understand the underlying messages that come through standup. And so standup is such an art. Music is also an art for me, and it's something that is both deeply technical as well as deeply improvisational and creative. I dabbled at acting, but naturally the reason why I didn't love acting was because I was playing someone else's and I was projecting someone. I was projecting a predefined awareness or predefined character versus projecting myself.
I have a lot of thoughts about this, obviously.
Actually, yeah, that's true.
Yeah, because for me, I mean, I did a lot of acting in the past. For me, the experience was, uh, by analyzing someone else's awareness and thought processes and behaviors, that was the path for me to actually self-analyze. That was the path for me to start questioning, ah, someone is doing it differently. Why are they doing it differently? And that exercise was the first awakening for me of any sort of awakening that made me change from living my life in a zombie-like manner to actually starting to be a bit more conscious where it's like, okay, why am I doing this? Yeah, what are the reasons for me to do this? So that for me was actually the most formative, uh, conscious endeavor in— to use your framework. Yeah. Um, to play around with other people's awareness and motivators and so on made me reflect and develop my own almost as if I didn't have anything, which I did, but I didn't really Self-analyzed before that.
That's really interesting. So, I mean, I'm assuming the types of characters you played, for example, there were probably certain characters that you deeply resonated with, and there are probably other characters that you played which might be so foreign to your mental model. And I guess, are you saying that—
I think it's a yes. I mean, when you practice acting, you play all sorts of characters. That's part of the lessons, the, the process itself of having to almost go in a meditative way to detach yourself and, and lean into someone else's being. Yeah. It feels very transcendental when you're in it. You know, people that haven't done acting maybe don't, but the way I practice it personally, it's very, it's very meditative. So it brings you really in another dimension. It feels like everything is different and there is a Really, you're accessing another dimension almost. And that process of detaching yourself by itself allows you to look introspectively. And so there is the first, there is the first little bit. And then when you're entering the, the next part.
Yeah.
I think even more so because at that point you really, you go even a step further where you really give up everything. And but then at the same time, you lean into some aspects of yourself because it's not like you're reinventing anything. fully net new. Oftentimes you're tapping into something that you already had, etc. Yeah, I guess you're just spinning the diamond enough times and in different directions, and it allows you to understand how the diamond is actually shaped, right?
I can imagine, like, there are certain characters— this is really cool, by the way, because it's like expanding my mental model and like what I think acting is and how you can evolve as a person. But I can imagine, like, you talk about that meditative state. I mean, like, one One way to describe it is like a flow state, right? Like if you're doing something, yeah, like if your agency is like really pure, you're just so invested in the action that like you feel in a state of flow. And I mean, were there like roles that you played that put you in a— even though you completely detached your mental model and you just invested in this role, were you— did you feel like you were able to get into a flow state with like every role that you would play in? Or like what was the also the difference between The roles in which you got into a flow state versus the ones in which you didn't? Like, were those roles the ones that expanded your awareness?
I'm not so sure it's about the role itself. Yeah, I'm not so sure it's about the character itself. It could be. Yeah. Because some characters you just get and you understand and it's like very from the script, you understand them and you relate to them. Some of these characters are really far from you and then it really feels like, oh wow, like, I mean, I'm in trouble. And those are the characters that if you have to do first read, you don't know where to start because you don't know what voice to have. You don't know anything. So, and they all, I feel like they all change you in different ways. Some of them faster because you have a faster feedback loop. The other one, it takes you 3 weeks maybe to find anything at all that you can work with.
Hmm.
But is there any character that shaped me? I'm not, I'm not so sure. That's a, that's a difficult question. The— what I know for sure is that there is— I mean, there is some characters that directly shape the way you're thinking. So there is, for instance, Lancelot into— in The Merchant of Venice, which was an interesting exercise because it's this character which is in a constant state of should I stay and should I go?
Yeah.
And I— should I run away or should I stay here as a servant? And I found myself many times in life looking back and feeling aligned with that. It's like, oh, I'm just like him at the moment. It's like, should I stay or should I go? And, you know, this pull and backwards. So there is, there is that as well. It's a recursive, you know, 5 years later you find something that you'd talked about.
Yeah.
Or you had a little segment or something that floats back after 5 years in the most unexpected ways. But I see it much more as a fluid practice rather than, oh, this character taught me that, this character taught me that.
Yeah, because you don't see yourself like that character for you is not a temporary experience almost, right? It's not like you play this character, you finish the transaction and that character is gone. It's like you—
Yeah, that's not how I live it. I'm sure other actors are really excellent at doing that. How I used to live it is, yeah, it's really something you tend to carry with yourselves. The learnings have— you have adopted into your mental model because I speak a lot about this, like This is also how I look at founders, which, you know, really understanding the motivator sets of a character, it's fundamental. Like, why are they doing a specific action?
Yeah.
And you really go and try to analyze that and that layer of understanding. Then it's almost as if there are small seeds that are seeded into you that then a month, 2 months later slowly start to come back, which is like you see some similarities here and there.
Yeah.
And it's like, ah, fuck. And maybe you don't even reconnect it to the character anymore, but you know that it's from having experienced the deep analysis and empathic connection with another entity of any sort, which in this case is a character, but it's— or a person or historical person or whatever it may be. And I'm sure historians maybe have the same, right? Like someone that writes a biography you know, you really understand someone. And I like to think that this is maybe one of the reasons why I invest, because that is my way of exploring more. It's like it's my way of understanding and exploring more, which is—
Your awareness expands at a— you use characters as a framework to expand your awareness, right?
It seems as though that's the— yeah.
I've gotten definitely that sense from you because like you're— it's like there's tons of frameworks, right? I mean, we could talk about loads of frameworks that like expand your ownness, but it's just like, for you, it's like this character and how they live their life is a framework that you use to like organize, you know, your mental model. Or you can like pick and choose certain things of that character that like resonate with you or kind of manifest themselves like in another situation and cause you to remember that experience. And yeah, that's really interesting. It's like where most people Yeah, yeah.
I mean, it's also— I, I mean, I've, I've said this before, but, uh, I think, you know, the, the fact that I'm meeting such exceptional people for my job, which is founders that have deep, deep connections with their whys— yeah, for instance, you— yeah, that it's almost as if I'm stealing their ideas in the sense that I'm meeting people that have such developed frameworks, etc., about the world and the universe. And I, I get to— yeah, I get to learn about the world through them as an additional mean. And that's how I reinforce it, which is— and I think you above all, actually, in the sense that I said it many times, but since I think out of— you're the founder that influenced most my way of thinking in this proactive motivators versus acting more reactively on how the world. Yeah. how the world around you works. And but there's so many shared learnings from one discipline to another that is, is pretty incredible.
I think seeing your— like, one thing I noticed about you from interacting with you is that you are a deeply curious person, almost like a proactively curious person who wants to like ask all the questions. And, you know, it's like you see the world in multiple dimensions, right? Like you want to understand a founder's why as well as like you want to understand Every single detail about the business, about the technology. And I think like you have like this intense app that's like quite conscious to me, like because it comes from a place of kind of purely like expanding your awareness, right? And so I see actually you've expanded my awareness on like actually what it— how acting is also like quite a conscious endeavor because I can imagine that same level of curiosity and precision and desire to connect with a character is pretty much the same in terms of like when, when we interacted. I think maybe because we were on the same wavelength, or maybe because some of the things that I said really interested you, I felt like we connected quite quickly because of both of our kind of like proactive desire to understand like each other better and, and kind of like generate maybe some even shared frameworks from like the frameworks that we've that we've had individually. And so, yeah.
Yeah. Which is hopefully what should happen in a lot of founder-investor relationships. But not only— but this is ideal, right? Because we have this reinforcing feeding mechanism where there is a proactive exchange. And I know this was part of your decision-making when choosing who to work with. Like, you know, we've spoken about this, but what's interesting here is I think that this level of, of appreciation and, like, conscious activities can actually be found pretty much anywhere in any art, any craft. Yeah. Almost independently of the craft itself and much more closely linked to the degree of intensity in which you practice that art and that craft.
Yeah. What are the motivators for, like, actually, do you let the craft consume you or do you kind of like interact, engage with the craft? Quite a bit.
I feel like if you engage to a certain extent, it, it drains from other sides. Yeah, you're right. Where you don't have the energy to recompute a little plant, right?
Yeah. Yeah. I mean, I guess there's probably levels, right? But I think it's like with, with everything, do you— and then it's like, what is your motivation for pursuing the craft? Is the motivation to express yourself and expand your awareness, or is the motivation because there's something external that you want to reach. And I always found that if the motivation is external, then actually it limits the ability for your awareness to be expanded. Because if you do the craft and you do the deed, you take the action and you get the result you want, then it doesn't expand your awareness at all. It's like you have certain founders who aren't thinking of a business as self-expression. They want to just you know, build something, make money, um, you know, be on some, some famous lists. And then once they reach that and they realize that actually, like, hey, there's actually very little energy that comes out of it because the whole process has been working towards some, like, external phenomenon that is actually then realized in the worst case and doesn't actually expand your awareness, they'll ask themselves, oh, was that worth it? And it's just like, you know, I think you don't want to be someone who's like posting, was it worth it on LinkedIn? You want to be someone who really like appreciates whether it's acting, whether it's building a business. Let the craft consume you. Understand all the intricacies of the craft and focus. I guess the goal should be, can this craft serve as a mechanism to like deeply evolve myself and evolve my awareness? And if so, the output just by committing to the craft The output will naturally be something of kind of high degree of consciousness and high value in the world if you just kind of like trust the process of letting the craft consume you and getting into a flow state.
It's interesting to— I— it's interesting, like this, this entire conversation. Yeah. Because it's almost as if it's a given in the industry that the right way to to invest in founders is to back truly broken individuals, you know, that are really scarred by something or that are, you know, somewhat different, you know, completely unique. And partially, you know, there is— I mean, those are different things, right? But the— it's, it's a— this— the current status of things is that VCs really like, uh, broken people because that is fundamentally the That is fundamentally a very secure way of saying, "Oh, that person is broken. Look how obsessive they are about you know covering that up or running away from that or whatever it is." And it usually correlates, which is true. It usually correlates to intensity. It does happen because I mean not it's well I'm not trying to say that broken people have intensity, but I'm trying to say that a lot of the people that are intense and broken like they share this. Yeah. It's so the industry is kind of aligned that the way to back founders is to find these individuals that are really maybe have some deep insecurity, something really deeply that is affecting them, and they're building companies so intensely because they they have to, they're forced to. It seems as though you're proposing a different type of of founder that is much more proactive and grounded in in different. Different truths. Do you want to elaborate why this framework may not be the winning one, the ones that we have adopted as an industry for the past 15 years? And what else is there out there?
Yeah, I mean, it's a really interesting thesis, actually. I think, I think investing in a broken founder, I think it's like a similar degree of like You know, if you're entering a relationship and you're entering a relationship when my mom— this is like my mom talking, right? She's like, she hates— she's been happily married for what, 26 years? You will never hear her say the phrase my better half because that implies that she's not a whole individual, right? And she's like, if I feel broken or feel empty and I turn to another person or I turn to Another, another, you know, or I turn to like a craft or like a vision, or I turn to work to kind of like make me feel complete. Then the second that I feel like it's kind of like a drug almost, right? It's like kind of I need to once you kind of take a drug. over and over and over again, you first might feel like a baseline high, right? But then once you become dependent on that drug, taking that drug only gets you to a certain level of dopamine versus when you took that drug the first time around. And so I actually think if you're a founder because you're running away from something, or if you're a founder because You have some deep insecurity that one day when you show up on the Forbes 30 Under 30 list, you'll be able to tell all your friends that made fun of you in high school, look at me now. I mean, I guess one of two things can happen. One, the best case scenario is that you realize that you were never broken to begin with, and then it's just like, what happens now? do you still follow the path that you've been traversing, or do you one day say, oh, now I've decided to retreat to a beach in Jamaica and make reggae music?
Right. I don't know because the thesis— fundamentally, it feels as though you're betting on people that are so broken that they need to build truly gigantic businesses to get fixed or whatever close to it. So it's, um, obviously you're, you know, you're betting on drinking people, but not a little broken, but like really, really problematic. And so the belief is that carries you all the way. Um, the problem I have with that is that besides the fact that you get whole, which by the way, I'm not sure happens, I mean, it usually happens less times than, than it does.
Yeah.
You know, people don't really get whole wealth-building company, But the problem for me with that is that it's like you're intruding a lot of personal problems and issues into building something that should be pure, right?
Yeah.
So you're optimizing for things. Someone that is married, maybe extrinsically driven, nothing wrong with wanting to build for— for instance, I saw a tweet the other day, which is retiring your parents. If you've come across a founder that wants to retire their parents, back them immediately. It's like, I see it. Like, okay, it's nice.
I also don't love like shallow quotes like that.
Yeah, of course, of course.
I think we live in an industry that tries to like compress and like oversimplify.
Right.
Versus like—
Of course. Yeah, yeah. But the core thesis there is like, oh, there is this incredible drive. But then it's, yes, but is that the right framework to build $10 billion, $100 billion businesses? Because at the end of the day, that's the goal of a startup, which is why it's coming together. But fundamentally, it's— it feels very short-sighted to me.
And I think it's like deeply like— I think the second point as well is like you then go through life. I get like there's kind of 2 buckets, right? The thriving, the loving bucket, the exploration bucket, and then the surviving bucket and the fear bucket. And you go through building a business with a lot of fear fundamentally. Right. And so it's like The people that are broken because they feel like this is my life raft, right? Their every action will be like, how do I stay on this life raft as long as possible versus how do I swim as far as I can? And I think that leads to— I mean, in a startup, you can tell cultures that are fundamentally based around fear, right? You can tell cultures that are fundamentally based around love and curiosity. The people feel quite emotionally secure versus emotionally charged and emotionally insecure. The types of risks that people are willing to take in a culture that's based around love and excitement and exploration are fundamentally different than the types of risks that people are going to take in cultures that are quite insecure. Cultures that are quite insecure have so much more overhead of people trying to show that they have worked and clocked the hours versus—
True.
And so I think it's just like short-term validation cycle, short-term validation. It's like, I got to text this. I have to— if I'm broken, maybe also part of my— the reasons why I'm broken is like I have this obsessive need for like, for validation. And so round after another, one, you know, get the round, you know, get the fast— get that dopamine hit. Tell that investor I hired you.
Secure on my path whilst I'm doing this.
Exactly. So it's like, you know, it's like Do you think of a startup as a life raft, or do you think of a startup as just like, naturally my DNA is a swimmer, so I'm going to just swim and just see where the journey takes me? And I could probably come up with a better analogy for that, but I think fundamentally the 2 points of being broken versus being pure and letting the energy go through you in the case of a startup is like, A, these are the types of people that you'll wake up one day and you'll actually see them sitting on a beach in Jamaica because they realized that they weren't broken or that the startups don't complete them. And B, if you're lucky, these are the people that will deliver what they promise, but the cultures that they create will be ones that are deeply insecure, focused on a lot of short-term validation because they're operating from a place of insecurity and lacking versus operating from a place of being deeply complete and feeling confident in being able to navigate the ebbs and flows of the journey but stand on the surface.
So intensity at the cost of volatility and at the cost of operating from a place of strength. So I guess then the second question is, how can you operate with intensity whilst not having the broken part?
You don't say that you're optimizing for intensity. You treat that as a byproduct of being in a culture that is energy. Think about compounding energy, right? Like, this is an intense discussion, but like, none of us come in thinking, oh yeah, let's, let's make a podcast that's going to be number one on the chart. And like, you ready, Shraav? You know, we're just having—
The audience here is pretty small. We're going to have a very niche audience for this type of podcast.
Yeah, but yeah, exactly. It's like it comes from this place of just, we're coming together on this jointly conscious effort. We're first off focusing on deeply understanding each other's awareness such that the actions that we can take are going to be net new and represent both of our personal investment. We're both just so eager to see where this conversation goes. And I think the resulting synthesis is going to be something that creates a lot of energy. And that exchange of ideas and that exchange of energy is deeply intense. You optimize for intensity as the byproduct. Like, I see these like 9-12-7, right? In order to catch up, you know, sleep with a mattress under, under the, under, under the office. And the reason why people are doing this and also find the need to post about doing this is really just to optimize optics. And it actually limits the potential energy that you can, you can create.
Um, And so, I mean, it's also to feed into narratives, and there is a very utilitarian point of view, and it's like you're creating— and yeah, it's a pretty precarious situation of trying to cram as much resource as you can because you're actually desperate. That's the truth.
Yeah. You think that it's finite. You think that it comes from a place of zero-sum versus infinite mindset. It's like, oh, if I— this is why founders that don't have deeply creative hobbies, I think there's a limit on the level of success that they can have because at a minimum, the ideas that they create or the ideas that they think of will be capped. And at a minimum, they'll deliver what they— or at a maximum, they'll deliver what they say and nothing more. And so this idea of like, oh, if I'm working 9 to 12, if I take an hour off to go play music or read this orthogonal piece of literature or learn about this subject. It's zero-sum. It'll actually take away from the amount of energy I can invest in my business. That's a zero-sum mindset. But a compounding mindset is being able to actually ingest a lot of diverse information enhances my creative ability, which then feeds back into the business at a much higher rate and actually changes the potential and the trajectory of the business by being more well-rounded, by being more—
Yeah. You know, I think there's also different, different types of businesses and different types of companies that need different things, probably in the sense that, you know, when building something that is really, you know, you're really ambitious and really out there and really pushing the boundaries, then you probably want to acquire some sort of level of ground truth where you're gathering some fundamental information about the universe from acting or whatever, whatever one's practices. And you're introducing that back into the business and that allows you to do lateral sort of synthesis where you come up with something that is really interesting and additive. There is also the YC actual, you know, playbook, which is much more iterative. And rather than net new, it's much more, you know, what do we know today that is true? Yeah, very scientific. What do we know today that is true and how can we iterate towards a you know, our success.
Yeah.
I'm not sure, you know, I'm not sure if there is necessarily one company. I think every company needs a slightly different approach, but it seems like to do something that is really, really ambitious.
Yeah.
You should lean towards the one that allows you to work from a place of net new ideas and creativity rather than iterative, which has been the playbook for quite a lot.
Yeah, exactly. And I think it's heavily dependent on the type of company that you want to run and the types of outputs that you want to create. If you have a very clear output that requires minimal invention, but you just need to ship it and grow business, which I think is definitely kind of— if you look at the distribution of YC companies now, it might be that distribution was closer towards the invention in the early days, but now it's more focused on like playbooks and generating value from playbooks. And so that requires a different type of energy. And then, yeah, for sure, like in that case, I think invest in, in, in the broken founder who you can give a playbook to and just say like, deliver it, execute it, execute it. But yeah, I think in the most ambitious of ideas, which I'm lucky enough to have even before starting Authman, have worked on projects that are like moonshots and require a level of creativity from openness. And you can't let things like fear, you can't let things like ego, you can't let things like this get in the way because you don't even know. There's no playbook for building moonshot companies ideally. And for me, that's kind of for me in the idea of expanding awareness and doing things that are focused on expanding my awareness as much as possible. The type of companies and the types of businesses I want to run are the moonshots of the world. And I'm telling you right now, being in a moonshot versus being in a company with a clear ceiling requires so much more energy and so much more work. But no moonshot of a company has 9-12-7 on the board and has people that are comparing the scores of their Oura Rings and competing to see who has the least readiness score. They're all just people like this conversation that come in without things that get in the way of their awareness and their desire to learn more about the world and follow this moonshot. They come together and they just act in unison and they just are following a trajectory. They don't know exactly you know, there might be certain things along the way that like deeply surprise them, but they're confident that just like by following this trajectory together with this group of people, you know, the path is just going to be, going to be incredible.
And so, so since you mentioned, since you mentioned your previous endeavors, why don't we speak a little bit about the learnings and the journey from, from your different adventures?
Yeah. Yeah.
You know, not everybody gets to work closely to, to Sam and and other people that were involved and are involved at Worldcoin and also previous companies that you've worked through. Yeah. So why don't you, you know, why don't we speak about Worldcoin where you had the chance to lead growth for, for a little while? Yeah. So curious to learn more about that experience and how it went and what you took away from it.
Yeah. I mean, Worldcoin was a, was a crazy ride. I think it was kind of the experience that made me comfortable working on moonshots and seeing what that required, reorganize my mental model that we just talked about on how to frame your motivations, how to frame the types, how to frame, how to understand what is a truly high-performing culture that can invent versus one that's following a playbook. And I think Worldcoin also just gave me so many stories. But I think if we even want to start from the beginning, how I got involved in Worldcoin. So I have an applied AI background, actually. So when you mentioned running growth, I have— I had zero operations experience before I started. And so how did I get to Worldcoin? I did my undergrad and my master's focused on applied AI. And in my undergrad, actually, as part of my master's thesis, I kind of worked on my first, like, what I thought was a moonshot idea, which was, this was before the crypto bubble of like 2021. It was like 2019 and stablecoins were kind of all the were up and coming. I mean, the 2 kind of most popular stablecoins were like Tether and Gemini Dollar. And I thought from like a store of value and from like a security perspective and from an ease of transaction, they represented something that could be valuable, like if widely distributed. But the problem with stablecoins was that these, like the Gemini Dollar, for example, like the Winklevoss brothers will basically send out an audit once a month that said, these are the tokens in circulation and this is the amount of money in the reserves, just to show that they lined up one to one. And it was literally just like a table that they would send every month. And so I figured, what if there was a way to build something that could audit both tokens in circulation and reserves in real time to give true confidence that the stablecoin truly can be exchanged one-to-one for whatever currency. And so I was living in the UK at the time doing my undergrad. So I built a stablecoin, or at least a proof of concept of a stablecoin that was tied one-to-one to the pound. I called it DigiPound. And I set up kind of like this auditing service where you can audit both the tokens in circulation, which was the easy part, and the reserves in real time. And I actually ended up— the project was a huge success. And actually my supervisor was like, I want to present this to the Bank of England and see if they'd sponsor you actually working on this.
And they hated it.
Yeah, they hated it. I remember. Really? Yeah. I spoke to these 60-year-olds, and I think they were just looking at me like, this guy is just some— one of them in the middle of the presentation pulled out a wad of cash from his wallet. held it up and he's like, if I have this, why would I need blockchain? And I'm like, fair enough. And I remember it sat in the back of my mind, obviously not agreeing with them, but it's just like, if you're lucky, these are the types of people that control your financial infrastructure, if you're lucky. Right. And then fast forward to 2 years later, I just finished my master's. I was working on a a couple of startups. The one startup that I was working on was Airbnb for self-storage. We don't have to talk too much about that because I don't think it's that much of a moonshot. It was a fun project. And actually, one of my friends who I did my undergrad with, and we kept in close contact, we would stay in touch. This is 2020. And at the beginning of 2020, we were on the phone and he's like, by the way, I just found some of these people. He was studying at Caltech, doing his master's at Caltech. We both did our undergrad together. And he's like, I found these people who are working on this crazy project in San Francisco. And he was flying to San Francisco every weekend to work on this project. And he's like, yeah, the guy from Y Combinator is involved. And I'm like, oh, okay, that's cool. And he's like, I really think you should come and join us. And naturally, every founder who's working on something, it could literally be like, I don't know, some automated CRM email sender, or it could be OpenAI, whatever. Every founder who's really drinking their own Kool-Aid and is in their bubble would scoff at the idea of dropping their business to do something else. And I'm like, ah, no, dude, I'm democratizing self-storage. Call me another time. But then I was talking to my friend again, And he's just like, yeah, it's been about a month working with these guys, and I think they're working on something really cool. Can I at least send you the white paper? And by then I was like, all right, send over the white paper. And the white paper talked about this idea, and I'd even forgotten about Sam and his involvement, but the white paper talked about this idea of creating a global financial network and a world currency. And it kind of piqued my interest. And it said Bitcoin is obviously successful from an infrastructure standpoint, but from a distribution of wealth standpoint and as a currency, it's not successful because 1,000 people hold like 40% of Bitcoin. And so I think the white paper talked about what if we could start Bitcoin in 2020, how would that work? And I kind of was really captivated by the mission of if we could launch a new currency and give an equal share of it to everyone in the world. Obviously, there was like this redistribution of wealth angle, which was cool and kind of like Silicon Valley-esque. But fundamentally, we could create this global financial infrastructure where people who previously didn't have a mechanism to like store value or they were in a country of high inflation or people that required remittance or, you know, who— where remittance was a big deal. There's so many potential benefits of having a single global financial network and Also, it's a moonshot, right? And it's an idea of whether or not this works or not, just following this trajectory is just going to be so cool. So I talked to the founders and basically joined the founding team. And I thought I was going to book a plane to San Francisco because that's where they all were. But then they told me, yeah, it's like, how quickly can you get to Germany? And I'm like, what? And then apparently what happened was The CEO, Alex, a good friend of mine, he actually wasn't allowed back into the country. They left San Francisco for a week to see their family, and they all went to school, and most of the founding team was from the small town in Germany. And President Trump, he closed the borders, right? And so they couldn't actually get back into the country. And so they're like, yeah, come to Germany. We're working out of a little office space here, but I promise in a couple of months we'll be back in San Francisco. go. And I'm like, sounds good. Where are you guys? In Berlin? Uh, actually they were in a town of like 100,000 people in, in Bavaria. And so in the middle of COVID I hopped on a plane just like kind of following this, this, this adventure and following this energy. Uh, then I took like a 3-hour train and was kind of in the middle of, at least to me, the middle of nowhere. I mean, I think it was still 100,000. It's not that small of a place, but still to me it felt like the middle of nowhere. living in London and the States. And so, yeah, I remember joining the team and how did I end up leading growth with no operational experience? Well, one of the things that kind of struck me when I joined the team was, first off, the mission of Worldcoin and the technology that we were building was if we were going to launch a new world currency and give a share of it to everyone in the world, we had to verify that everyone who claimed a piece of the currency had not yet signed up before. And so essentially we had to build these devices, in this case using biometrics, to verify that if you sign up and claim a piece of Worldcoin, you haven't signed up before. And so we were building these essentially like proof of humanity devices. And essentially what they were were really like high-tech iris scanners. And so When I joined the team, it was incredible to see, first off, how interdisciplinary everyone was. It wasn't just like a bunch of generalists, but we had people with deep hardware experience that were working on how to design a device that could withstand the heat of Ghana to the freezing cold of Norway. We had experts and PhDs in AI that would focus on if we had a bunch of cameras set up internally and someone was looking at the device, how could you figure out where someone's eye was and position the cameras in such a way that could get a good picture and then convert that to an iris code? And then we had people that were working on the blockchain part of it, which was then— and the crypto part of it and the security part of it, which was then how do you issue someone a wallet for this token without connecting the iris code to it? And really, we had experts in each of these domains, which, as we'll talk about later, kind of gave me the my first notion that if you are going to build a truly incredible founding team, having some level of complementary awareness, especially for an interdisciplinary mission and technology, is much more powerful than having a bunch of generalists and telling them to figure things out. But yeah, that was my first impression of joining them. And they actually wanted me, given my background, of course, which makes sense, to help them with the Orb in particular, the the AI and the iris code generation and the algorithms that would turn and position the cameras when someone was looking at it to get the optimal picture of the iris to run the biometric validation. And so I think I did that for all about half a day when we realized that in order to prove that and in order to ensure that our systems are robust, That means that they could work in a number of geographies with a number of different facial complexions and eye complexions. And also we wanted to prove that if this could work at a billion-person scale, then the hardware could also be robust in a number of climates. And so this was like the middle of COVID This was like August 2020, the kind of peak hysteria. But we had like 6 people in a garage in Germany. And no one was kind of willing to leave like a 1-mile radius to start seeing if these orbs and these biometric devices would work on different types of people. And I just remember like, hey, I'm the new guy here, may as well, may as well do this. And so booked a flight to Turkey because that was like the quickest and easiest place that, that you can get into that was open during COVID And we ran our first field tests in Turkey to prove that the orb would actually work in different geographies.
Does that mean that you had to show up to the airport with an orb?
That did mean I showed up to the airport with an orb.
How did that go with the police at the border?
Oh, man, so much. I think anytime we would enter a new country, going in with an orb would be just— it would be so funny to see how people would react. They would just pick it up, They would look at it, they would like sit and question you for a while. And the project was in stealth, right? And the project was in stealth for many years. Um, and so, yeah, like it— that was always a fun experience going through an airport. And we would never check in the orbs, we would always take them through carry-on. Um, how come? Well, because they were— it's quite expensive, quite a, quite a—
Too fragile.
Too fragile.
We didn't want to Um, so in your backpack you have this orb, which probably at the beginning was larger and bulkier.
Very large, very bulky. Yeah. And it was, uh, it was, it was an experience, put it that way, like taking it through.
So you arrived in Turkey at the beginning, and, uh, what did you do with the orb? What was the—
We, you know, we, we basically like tried to sign up, or we basically tried to get as many people to like look at the orb as possible and see if it could verify that like they hadn't signed up before or if they have. And for the most part, it like worked. It worked really well. But it was really cool because, you know, even though we went to Turkey with the guise of like seeing that this orb could work on, on different people, we started like pitching to them the project and the ideas of the project. And obviously the lira is like incredibly— like Turkey has like insane inflation, right? And The amount of resonance people had for the project was really, really exciting. And so ended up coming back from Turkey, and I think over the course of 5 days, we signed up hundreds, if not close to 1,000 people. And it was—
I mean, you were paying them, right?
We were.
So it's a pretty straightforward exchange.
It was a pretty straightforward exchange, right? But the point is we came back, And after working on the Orb for a while, it was kind of cool for people to see, like, hey, there's actually people in the world that have used the Orb. And then Sam always says the biggest problem of Worldcoin is that we don't sign up enough people fast enough and people just don't care. We had come back from Turkey and there was almost like an adrenaline rush about seeing users interact with the Orb to where we're just like, okay, actually just like, you know, when we— let's start like running operations and let's start, you know, actually like seeing if we can take the Orb to different geographies and actually start by, by giving away, away Worldcoin.
Interesting.
We obviously had like Europe, flew to like Ghana in the middle of COVID as well, Mexico, which were all like fun, fun experiences. And then we just, you know, one of the immediate things that was like really crazy to us is The idea of Worldcoin and the idea of how Worldcoin would grow originally was we would find local operators. And our profile of what a local operator was, was like a hustler. It was someone who would just be very passionate about the project and also just want to hold the orb and make lots of money for signing people up. And Sam would always say operators would do things that we wouldn't even have the creativity to think of from you know, where we're sitting. So we should just like find the best operators in the world, give them Orbs, and have them figure it out. And yeah, ultimately, like when we, when we started like running operations and actually like giving away tokens, giving away the token itself, again, like we had no one working on it. So it was kind of natural for me to just like work on it. Right. And what had happened was the first thing we wanted to do was just like find operators in any country that we could. We didn't have a sophisticated model for which countries we could go into, really. There was 3 criteria. The first is, is the country open during COVID or are its borders shut? The second criteria was, could there be some— whether it's inflation, for example, or remittance as a huge use case, some kind of driving factor for why What would make the pitch easier in this country? And it was from that financial perspective and the infrastructure perspective. And then the third thing was, can we find really cool operators that we trust? And that combination of ideas, we ran a number of strategies to try to find as many people as we could, given the constraints that we were in stealth, given the constraints that when we talked to the average person, they're like, you guys are crazy. What are you doing? So there was a number of things that we did to really find our original operators and build trust. And then once we built trust in an operator, we would fly to whatever country that they were in. We would teach them the pitch, give them the Orb, and then see what they would do. And it was kind of mind-blowing because I think 6 months in, I think we had launched like 10 countries and we would launch the country, we would see the operators kind of struggle a bit at the start trying to pitch the project by themselves and recruit their friends to help them. And then we would come back to Germany and then a week later we would start to see hundreds of signups each day and we're like, wow, they're actually kind of figuring it out. And we actually started launching, we had a testnet for Worldcoin and so we were actually just giving away test tokens and a very simple wallet interface. But people were quite excited about the project, the idea, the chance to work on something in Silicon Valley that be associated with something from Silicon Valley that wasn't even present in America at the time. And that was the early days of running really scrappy growth operations and ended up just following that energy as much as possible. And that was kind of the cadence which we would use to launch early countries, find great operators, teach them the pitch, give them the orbs, come back to Germany, see what they would need to sustain their operations. And it was quite decentralized and relatively unstructured and super scrappy at the start. But yeah, operations 1.0 was hey, this stuff is kind of, kind of working and the project is actually quite magnetic in the right communities. I mean, there's so much like we, we learned and we can talk about, but that was like, like early.
I mean, what I'm interested about specifically maybe within that is, yeah, you mentioned about, you know, you said scale is all that matters. Yeah, that, that was kind of the ethos, the, the modus operandi that you had, just scale and then scale will bring data and it will bring energy, entropy, et cetera.
It'll bring use cases.
Use cases. Do you still feel that's the right modus operandi? Is that a— was that a— what do you still align with that approach?
In the case of Worldcoin, yes. I think what we could see is even now when you sign up certain geographies like Argentina, for example, right? Or I think if you onboard and not necessarily scale arbitrarily, there's breadth and then there's depth. And so it's building as many networks as possible, finding closed networks where there can be a huge velocity of interactions even between a small group of people and finding as many of those networks as possible. I do think for a project like Worldcoin, in order to move faster than the haters, for example, who would—
In this case.
Who would hold the project back, that was necessary. It's like we needed to get escape velocity and we wanted to make Worldcoin such an inevitable part of certain communities that was providing value either as a store of value or as a mode of transaction or as a mode of remittance. One of the things that I was deeply inspired by was, you know, and it's why kind of like Kenya was one of our original first markets, was M-Pesa and mobile money in Kenya and a lot of countries in Africa. And I think this is what people don't realize. I mean, through Worldcoin, I spent many months in Kenya and I didn't see a single— I can't even tell you what the shilling looks like as a cash, from a cash perspective, because No one there uses cash, even in highly rural areas. Kenya has this form of mobile money called M-Pesa that I think over 97% of the population uses. And actually how M-Pesa scaled, which is counterintuitive to what people might think as well too. They didn't go into the heart of Nairobi and try to sign up everyone in Nairobi. They actually went to increasingly rural areas, what people would call slums, right? Not to make it sound derogatory, but they would go to incredibly rural areas where cash was a burden for multiple reasons. One, it was quite unsafe to walk around with cash. Two, there was a huge need for microloans to set up a small store, for example. And the inconvenience of taking cash to a bank in the middle of Nairobi was just too high. And 3, there was just a lot of microtransactions as well. And they basically onboarded these peripheral areas to M-Pesa. And then what happens is what people don't realize is where do these peripheral areas get all their money from? Well, it probably comes from you know, people that are working in the city that are sending money back to these areas. And so by like onboarding these peripheral areas, these rural areas, by generating like lots of value in those areas, it meant that those, those people can say, hey, send me money in M-Pesa versus, versus give me cash. And then it was actually easy to onboard more urban areas. And, you know, M-Pesa caught on like wildfire in that regard and became incredibly useful incredibly quickly because You would find atomic units or atomic networks. You would find as many of those atomic networks as possible. You would generate some form of value, and then that would kind of catch fire and reach escape velocity. And then the utility would just become incredibly widespread. And so I think scale does solve all problems, but it has to be done in a depth versus breadth mentality, right? Where it's like, It's not about signing up 1,000 people around the world, all in different countries. It's about signing up, going in, getting insane penetration in this even subnetwork and onboarding as many of those subnetworks as possible. And then that's how you can reach escape velocity. And that was point one that I agree with, or the nuance. And then point two is scale will probably come in non-consensus ways, ways that come from a different— seeing the world in a different way than the average person would, like a different awareness, whether that's M-Pesa starting in the slums, whether that's— I think that scale can, like, and those networks can be non-consensus and the strategy of scale can be non-consensus and start in different places that might otherwise seem intuitive. And so that's my 2 nuances in Worldcoin's case. Like me agreeing with that thesis.
And then you decided to move on, though, from Worldcoin at a certain point. What was the thought process? Because at the time it seems like it was the, you know, Worldcoin became what Bitcoin is known as today and there was a real inflection point. And I'm sure there was also some sort of, you know, you know, monetary incentives attached to that. And there was a lot of energy around it. So what made you switch?
Yeah, I mean, maybe I can. I mean, I'll speed. I know I spent most of the time talking about the first 5 months of Worldcoin, so I'll speedrun talking about 5 months until when I decided to leave and then can tell you why I left. But I think one amazing thing I learned from Worldcoin is within the first 5 months, it was like scrappily having a network of operators and having kind of very little or not an insane amount of visibility into how they were signing up lots of users over time. And so it was really incredible to kind of see. But actually, the goal was always like, how do you build an arm, an operations arm that is as sophisticated as how Uber launches markets and how Airbnb launches markets? And so this idea of just throwing orbs into the wild and seeing you know, what people do to sign up other people can only get you so far. And so for me, the tremendous amount of learning was basically like building teams, like both locally and globally, learning from like Uber and like Airbnb's playbooks on how they would like launch markets. And ultimately, like, you know, the strategies of how you would go into a market the types of operators that you would find, you know, they didn't necessarily have to be like these scrappy, like, college kids anymore, especially as Worldcoin gained more like press and exposure. You could find like quite sophisticated operators who already had kind of local businesses and kind of gig workers for hire already set up. Like, that's how Uber would, would launch certain markets as well. They wouldn't just go and find, you know, individuals who had their own cars. They would find like fleet management companies who had a bunch of gig workers for hire and employ them to accept the first rides. And so there was just a tremendous amount of learning that we did over time to make the operations much more sustainable, much more sophisticated, and much more effective, whether that's also setting incentives in the right way that optimize for not just signing up a user, but to show that they really understand what they're signing up for, ensuring delaying kind of the reward of their signup for from an operator's perspective until you prove that a user retains in the app a week later and uses their Worldcoin for something or uses the tokens for something. And so there was just like a tremendous amount of learnings and playbooks and infrastructure that we had set up. And so basically, over the course of like a year and a half, 2 years, we built our own operations arm from scratch. And it was truly incredible. And I think we had a team that was ready to go and just like print operations at scale, kind of like we talked about playbooks, right? And, you know, it was an incredible adventure for me. I spent 3 years kind of living out of a suitcase as well, mostly between like our top 4 markets, which was Lisbon, Buenos Aires, Nairobi, and India, both like Delhi and Bangalore. And it was incredible. We had launched like 1.5 million users in the course of like 1.5, 2 years, and I had stories, stories of a lifetime. And I know if I had like continued down that trajectory, I would probably still have like a number of stories. But I felt like for me, what was important at the time was, you know, the thing that was kind of lacking when we talk about like, is this truly like expanding my awareness now? Like, am I at a position where, you know, running the playbooks in different markets like really expands my awareness? You know, I felt like I had set up the engine in such a way where I, in a good way, like abstracted myself out of it. And I think that's the best thing in business. You kind of always want to like fire yourself by like setting up the right team and the right infrastructure. You want to fire yourself at certain functions. And the beauty is then either you can reinvent yourself within the business based on what it needs, or you might realize that there's other opportunities. that can better serve to expand your awareness. And I think why I left Worldcoin was going back to my technical roots as well. We talked about this idea of scale solving all problems, and that was the case in the case of Worldcoin. But around the time I left Worldcoin was 2022, late 2022, which is when ChatGPT launched. And when ChatGPT launched, everyone was so obsessed with the underlying models that ChatGPT was built upon, which are LLMs or large language models. And when you see the types of output that ChatGPT would generate, there was this moment where not only was it incredible from a product perspective, it was the fastest product to a million users, but from an AI perspective, there was a lot of discourse around the idea that these language models, by just scaling them up with more and more data and more and more compute and making them bigger and better, could lead to solving some of the hardest problems in humanity. And fundamentally, from an infrastructure standpoint, that's something that I actually didn't agree with. And I think for And I think Othman actually, and we can talk about that now, but Othman was actually started by this idea of LLMs are built on basically the most fundamental unit of AI and of deep learning, which is the neural network. Right. And the idea behind traditional deep learning is I give a system a ton of input and The system, the neural network should recognize patterns in input, and it should generate output that is accurate in line with what it should be predicting. Right. It's like the simplest example is I give a system enough images of a cat, and I give a system enough images of a dog, and I tell it, these are images of a cat, these are images of a dog. Now, if I give you another image, you should be able to predict with high accuracy, is this a cat or is this a dog? And LLMs are like that on steroids, right? Except You're not predicting the species of an animal. You're predicting what are the words that should come after the words in this input. And that's incredible because trained on lots of human-like text, right, you can actually generate things that sound quite intelligent and sound quite smart. But fundamentally, what neural networks and what kind of these traditional deep learning architectures enable are only single-shot Q&A-like experiences, right? It's like, I give you a question and give me the answer, and then that's the interaction and that's done. And so naturally you can solve harder and harder single-shot Q&A-like problems just by scale. Scale solves all problems. Scale doesn't solve all problems. Scale solves all single-shot deterministic problems. And even today, actually, these were insights that I had a couple of years ago. But even today, if you look at the launch of models like GPT-5, for example, and you see these AGI benchmarks, they're all on math problems or code completion problems, right? Which are problems that are single-shot deterministic problems with verifiable solutions. And so naturally, I can learn harder and harder math problems through just kind of more and more data and becoming much more adept at recognizing patterns. But if you take an LLM and you drop it in an unknown environment and you give it a goal and you say, what is the next action that I should take to reach this goal? And you try to have it form a strategy to reach a goal and adapt to its environment. Those types of tasks, these more open-ended exploratory tasks that require forming strategies, LLMs struggle with. And those are actually fundamentally the tasks that we promised our kids that AI would solve. Everything from drug discovery and emulating scientists in a research lab to medical diagnosis. How can you build AIs that ephemeralize nurses or doctors that are asking the right questions and running the right tests and working towards a diagnosis and treatment? Even things like Sales, which are, you come to me looking for inspiration and how do I ask you the right questions to narrow you in on the products that are the best for you and give you confidence in buying them? All of these are not single-shot problems. And fundamentally, all of these are problems that humans, particularly human experts, can solve really well and have the reasoning abilities to solve. But LLMs in 2023 and in 2025 aren't able to solve. And all of these fundamentally are problems that if you're able to interact with a machine or an expert, the interaction with an expert to solve those problems, problems in science, problems in sales, problems in product discovery, are all experiences that should fundamentally expand your awareness or the collective awareness of of mankind. And so if you look at the types of experiences that you can have with LLMs, solving, being able to solve lots of math problems doesn't expand the collective awareness of humanity. LLMs are amazing, but really what their function is, is increasing productivity or workflow automation of highly deterministic tasks. But LLMs alone can't ephemeralize expertise, which is, I think, what the universe is in short supply of and what our world is in short supply of. And so the idea behind starting Othman was if we could build new types of AI models that could, just like humans, form strategies towards a goal and take actions in ambiguity better than LLMs can, we could ephemeralize and actually supersede most expertise and most experts across multiple domains and actually build AI that can coexist with us to not only solve the types of problems that we promised our kids that AI would solve, but actually serve to expand our awareness and the collective awareness of mankind. And so that was the mission behind Othman. And that's why it was the right time for me to leave Worldcoin because I actually had an intuition on how we would design those systems and I wanted to follow that.
You had an intuition about what the new architecture to actually not only go at one-shotting problem would be. And what was your intuition?
Yeah. So I think it stemmed from this belief that First off, it stemmed from my undergraduate background where I discovered disciplines in AI that even predated the neural network and predated traditional deep learning. And so I think most of our industry thinks AI started in the 1990s with the invention of deep learning and with the invention of neural networks and with the cat-dog trainings, classification trainings. But actually, there's a rich world of AI that even predated the neural network. And one of the first disciplines in AI was called expert systems. And the goal was, how do you actually simulate the reasoning or the decision-making of human experts to look at an environment, ask the right questions to learn more about the environment, and narrow a solution space until you reach a certain outcome. And this was developed in Stanford in the 1960s. And one of the first implementations of it was in sales and product recommendation. So if someone's like, I'm going skiing next week, versus something single shot that just shows you products, it asks you the right questions, like what's the type of snow that you're going on? So if I'm going to a beach holiday, it'll basically create the decision tree that determines the questions you should ask to narrow in on what are the products that you should recommend. Right. And so you can simulate how experts interact with the world and experts' reasoning abilities through these decision trees. And in the case of the expert systems in the 1960s, the decision trees are fundamentally hardcoded. You had a system with a hardcoded explicit set of decision trees and facts. And so while they could do well in certain situations, they fundamentally were incredibly brittle. Right. But the intuition on how do you find or how do you generate the most optimal decision tree in a given scenario that will optimize a reward, that intuition was spot on. But the execution wasn't about— yeah.
Why was that spot on? Why do you think decision tree is the path to— why is that architecture spot on?
Because if you think about why does a human take an action, we take an action with 7 other actions in mind to reach a goal. We take an action because there's some reward. It's like in the case of playing chess, you're not just playing one move in advance. And then when you, you know, get the next move, you're like, oh my God, I'm in a completely new situation. The best grandmasters play a move because they're thinking 7 moves in advance, right? Like, if I play this move, he's most likely to play this move, and then I'll play this move, and then ideally we'll get to this state of the board. But also, if they don't get to that state of the board, they also, you know, have like 7 scenarios in advance of like how—
Okay.
The next 7 moves can, can unfold. And that's what separates things that are deeply strategic from things that are deeply arbitrary. And unstrategic. And so an LLM is only thinking kind of one move in advance if you give it kind of a long-term, more open-ended task. But the key is, is how can you build systems that can generate decision trees on the fly versus retrieving a hardcoded decision tree and breaking once the environment changes things that it's had experience on? And so for me, when I was an undergrad at Imperial, I was really lucky to meet the team at DeepMind. Who was working on AlphaGo. And that moment, I think, for me, taught me, showed me the power of reinforcement learning, which was kind of the discipline where you can have a system that had an understanding of an environment, had an understanding of a goal, and could essentially form strategies, form these decision trees to optimize a certain reward and adapt as their environment changed, aka as their opponent made a move. And for me, this kind of felt like the kernel of a truly superintelligent system. But the problem with these systems, with AlphaGo, is it could only operate over a fixed kind of game board. Right. And so our thought was, what if you could take the power of reinforcement learning and the ability to kind of explore an environment, but actually, instead of exploring a fixed game board, you could actually explore a conceptual map of knowledge. And so what we realized is that the best way to actually represent rules and facts about a domain, in the case of expert systems, this was kind of a list of decision trees and a list of facts. But as humans, our awareness, basically the way kind of our mental model of a domain works is We think in terms of abstractions and connections of abstractions. When I look at this, I see separate entities like the plant and the table, and I have names for them, and I have a relation that says this plant is on that table. And so I think just like in any domain, you can represent the conceptual knowledge of an environment that can be used to make a decision as a kind of set or as a massive map between concepts and relationships. Or in computer science, graphs, right? And so the idea is if you can represent the kind of conceptual knowledge and model that conceptual knowledge with a graph using reinforcement learning, you can explore different paths in that knowledge, kind of forming decision trees on the fly to reach a goal. And you can also model that goal via the exploration. And so I know it's quite abstract, and so I'll describe some examples that ground basically this intuition. But the idea is our research at Othman, or our reasoning models, aren't based on— are fundamentally focused on how do you solve and generate strategies for open-ended problems that adapt as you get more information from an environment, and versus an LLM where you're prompting them at each instance with here's the environment, what's the next action I should take? And the actions are quite stochastically generated and not trained to optimize reward. How do you model the knowledge of a domain in a graph? And how do you then build systems that can explore different paths in this graph to reach a reward? And that connects the power of AlphaGo and the explorative power of reinforcement learning but not to explore fixed game boards, but actually these really complex structures of knowledge that represent human understanding of certain rules in a domain.
The graphs are static themselves. So the decision tree is made on the fly, but the graphs themselves are static.
The graphs can evolve as you get more information. For example, if I read a new article and I learn different concepts, I can add that to my— if I learn a new concept, I can add it to my graph, and that can represent an additional path of exploration in the future.
So eventually, is it just one big graph, or—
Eventually, yeah. You can represent really most of our conceptual knowledge as one big graph because all of these concepts are interconnected. And I know it's a bit abstract now, so I can give you an example of what is a baseline open-ended task. that benefits from this architecture, this algorithm. I was in the task actually to— this task actually came funnily enough from being in line in a tent in Oktoberfest with my friends. But we were waiting in line in Oktoberfest and we just decided to play a game. And the game was, if one of us was thinking of a mystery country, can the others guess the mystery country using the fewest amount of yes/no questions? And that would be able to allow us to see who was an expert on the world and who wasn't. And so people obviously who had deep knowledge of not only countries themselves, but geopolitics, history knowledge, they kind of had essentially the concepts that they could use to organize this decision space of countries. And so the idea was these people that were able to not only have a clear mental model of the world and how different countries connected to different concepts, like was the country colonized by the UK or does the country border the Atlantic Ocean? but could also explore that decision space the fastest and adapt as they got new information were able to solve the game much faster than people who, A, didn't have this sophisticated model of the world, and B, didn't know how to explore this model efficiently. And so through playing this game, you're able to see who is an expert on the world and who isn't. And naturally, we would play this game with ChatGPT because that was That was, we wanted to see how well it would perform and it floundered quite a bit. And so an example is the first question that GPT would always ask was, is the country in the Northern Hemisphere or in the Southern Hemisphere? And it's a question that sounds smart, kind of stochastically.
Half.
Yeah.
It's half of the, like, geographically.
Half. Yeah. But actually from a, from a standpoint of does it eliminate this decision space? How much does it eliminate the decision space? Only 32 countries are in the Southern Hemisphere. So a real expert who's forming strategies to eliminate as many questions as possible and has that as a reward that's kind of formulated would never ask that question. Right. And so what we found is also the strategies that these systems would generate were heavily biased based on the distribution of data they had, which meant that for questions like the US or China, or countries like the US or China, GPT would be fine. But you give it an island in the Pacific and it goes on for 30 questions and it gets completely lost and it starts to hallucinate. And maybe it'll fall into a rabbit hole where it just starts guessing countries one by one by one by one. It even gets frustrated with itself.
Is that because data pertaining those specific islands is too remote and not as obvious, and so it wouldn't get picked.
That. And then it's also heavily influenced by its previous tokens as well. So if it starts guessing, is the country Tonga, the previous tokens will probably lead to, okay, is the mystery country Samoa or is the next country? And so it's heavily influenced by its previous tokens as well.
To what extent is this shortcoming a fault of the architecture versus something that gets fixed over time with more data and power?
Well, the point is, even if you give a system more data, if it can't formulate strategies, then it struggles to solve tasks that are out of the distribution of its data where it doesn't have the strategies. I mean, you could say if I train the system on enough country guessing games and I gave it enough examples of really positive gameplays, then it would naturally learn to parrot those gameplays back to me. But then how do you generate those gameplays in the first place? And then as you get more knowledge, for example, say I discover that there's a new alliance that's been created and that there's 50 countries that are in this alliance. That's information that isn't directly tied to data I've seen on how to play the country guessing game. If I'm a human, I add this as a concept. I can connect like 50% of countries to this concept. And naturally that would probably be the first question that I would ask in a next game, hypothetically. And so the truth is the architecture of LLMs, because they're single shot and because they're stochastic, isn't set up in a way to— most open-ended tasks are about formulating a decision space and then being able to organize that decision space. With the knowledge that you have, and then figuring out how to systematically eliminate that decision space as fast as possible and adapt to your environment. It's like a doctor, for example. They've organized all possible diseases in their head with all possible symptoms and genetic predispositions and treatments, and they can run tests and they can get insight from users that can ultimately eliminate certain diseases or certain ailments. And they'll do that As efficiently as possible until they can narrow in with confidence on a disease that a patient has.
And to what extent can they become drastically better if they utilize some of the one-shot for a chain of thought? So you come up with some sort of strategy at the beginning and then the other tokens are for specific questions to execute on that strategy. Is that the path forward for them?
So I actually think chain of thought is a myth in the sense that it's not reasoning, but it's meta-prompting. In a way. And so where I think LLMs have a— and so actually we tested LLMs on the country guessing game and actually successive versions of LLMs. And naturally, I think GPT-4o took like 13.1 moves on average to guess.
On average?
On average to guess. 01 took about 12, 03 took about 10. So you do see that the chain of thought maybe helps it improve a little bit. We tested on GPT-5. Actually, we got the results back this morning, and GPT-5 took 12 moves. Right. And so it's actually not like scale solves all problems because fundamentally, it's—
Wait, 12 and 4.0, how much? Or 3?
4.0 took 13, 0.1 took 12, 0.3 took 11. And then GPT-5 took 12. And so it was actually rather arbitrary at these types of games. But of course, naturally it would get better at these single-shot prediction problems that are deterministic. We always see the math benchmarks.
Why doesn't GPT-5 come up with a better strategy to kickstart the prompting game to itself?
There's just not much data on playing these open-ended Tasks, right?
But why do you need data about playing the tasks? Can you execute a good strategy without having the data about playing?
You would think, right? But because the formulation— but the architecture doesn't enable that, right? There's some clear plateau where even if it has all this knowledge, like a human who reads basically the news and can form these concepts rather explicitly, All of this knowledge is fundamentally implicitly represented, right? So GPT, in the case of country guessing games, in the case of any task, doesn't have these explicit notions of countries and concepts that humans do. And they're not trained to optimize a long-term reward or plan multiple moves in advance. This idea of decision trees, LLMs aren't generating an output as an intermediary step. of a decision tree. LLMs are generating an output thinking this is the final output to this input.
Interesting. Yeah, because the fundamental goal which we were trained on is one shot.
Exactly.
And as a consequence, the interesting— and as a consequence, they're not particularly good at figuring out the meta prompting tasking.
Exactly.
Do you think over time OpenAI and all the others will go towards really doubling down on meta prompting?
Yeah, because that's the cash cow, right? LLMs serve a purpose. They fundamentally— and I'm not here to say LLMs stink, they're great. But the purpose of LLMs are things like text synthesis or workflow automation or tasks that can deeply enhance our productivity. And actually where LLMs serve a purpose in our systems as well is that prompted with the right ontology, an LLM can look at a piece of unstructured data and extract structure from it. So we actually prompt LLMs to read and kind of leading into how we built a system that solves the country guessing game way better than LLMs can. We use LLMs to actually help us construct these graphs because what LLMs are great at is they're a presentation layer of information. And they're a great way to interpret unstructured information. But the kind of strategy or the central nervous system part of the brain that's determining the next action to take, that's not their purpose in a superintelligent system. And so we've prompted LLMs with an ontology that says, hey, these are the concepts that you can use to describe things that help you play the country guessing game, like countries, oceans, flag features, organizations, read through these Wikipedia articles and extract mentions of things that you think fit into these buckets. And so it can read an article about Chad and say Chad is in Africa. Chad has red, yellow, and blue in its flag.
So it can process the information.
It can process this information and help you create the graph. And it can also help you present this information. The cognitive abilities to explore these graphs are where it lacks. And so actually, we've built these massive graphs using LLMs, but we use reinforcement learning and we frame the country guessing game in a way that says, hey, right now, your information about a country, the way to represent that is a node that's not connected to anything in the graph at all. And the actions that you can take are to probe edges in your environment between this country and other concepts. And so asking, for example, does the country border the Atlantic Ocean, forms an edge between your mystery node and the Atlantic Ocean. And that moves based on how we structure the graph and how we set up our algorithms. That moves the position of your knowledge closer to countries that border the Atlantic Ocean and further away from those that don't.
Wow.
And it continues to do that in succession, essentially forming a path through this graph by probing for edges until it gets closer and closer to just one country. And that's fundamentally how we model what it means to form a path through a knowledge graph to reach a reward. And in this case, the reward is how can you get as close as possible to one country in the fewest amount of steps? We trained the system and on average, where the best performing reasoning model took like 11, or I think it was 10.95 questions to guess the mystery country. Our first version of a system built on this architecture took 8.38 questions on average over 1,000 turns. And so that gave us a lot of confidence that kind of our 2025 implementation of an expert system could solve these multi-step open-ended problems in ways that LLMs can't, even with scale and data, and especially open-ended tasks where There wouldn't be much example gameplays that you would see in distribution of the training data.
Interesting. And do you think most value in the approach is found in the way you construct the graphs or is found in the way that the reinforcement learning layer Both. Equally valuable?
Equally valuable. Equally valuable. Because, yeah, both are equally valuable and both pose their own challenges. In the case of constructing knowledge graphs, if we want to use these reasoning models and extend them to other tasks, medicine, sales, et cetera, it starts with how do we define the right ontology, which represents the types of concepts in a domain. And how then can you, using foundational models, prompt them with an ontology to be able to then look at a piece of unstructured data and analyze it in the same way as a human does? That object is a plant. The plant rests on top of the table. The cups rest on top of the table. The cup has color brown. And that's a perception challenge that we're trying to emulate and essentially like a mental model forming challenge. And then the task of actually forming strategies to explore this conceptual knowledge, that's like cognition as well. That's a core part of cognition that requires challenges on how do we first off represent this graph in a way that can be efficiently explored by a reinforcement learning agent? Because this goes way beyond the environment or the state sizes that are used in AlphaGo, right? Like chess and Go. And then how do you formulate the reward in the right way? And so we've actually tested the models that we've built to structure knowledge and to explore knowledge. We've actually proven are also quite versatile and quite context independent to other situations as well, which means that if I swap out the knowledge graph, the system can reason about this knowledge. Just as well. So we've actually abstracted the meaning of the knowledge itself from the kind of cognitive or explorative parts of the system. And so we have confidence that these types of reasoning models can be used to emulate any sort of expert task as long as we're able to very clearly define the rewards and give the systems the right infrastructure to train in. And if we're able to very explicitly define how we should construct knowledge in a domain, we can construct these conceptual world models that represent the knowledge that experts have. And we can also train our models to explore this knowledge that represents the reasoning of an expert to act in an unknown environment, reduce basically the decision space of solutions until it reaches a goal.
So the knowledge graphs are themselves switchable. The reward itself is switchable.
Yeah.
And so this can become a very fluid architecture which really adapts and over time scales.
Exactly, yeah.
So what's the end state from there? Maybe, you know, 5 years, what does the Athman reasoning engine look like?
What can it perform? The idea is versus thinking of these domains as separate instances of the reasoning engine, right? At the start, going from country guessing to another task, we basically just created a separate instance of the reasoning engine. We replaced the graph, we changed the reward slightly, and actually we went from country guessing to building a movie recommendation. assistant. The reward was slightly different because it's not about guessing a movie that's known, but it's about understanding a user's preferences, which can be modeled as concepts in a graph. Like, what are the genres you like? What are the types of actors you like? What mood are you feeling today? And obviously—
Not only that, but also what questions are indicative usually to a person's preference.
Exactly.
What do I know about a specific, you know, different people seek different things. Yeah, I guess so. There is an element of also adapting the system for the— I wonder whether that's feasible at all.
Yeah, I mean, that's the idea, right? It's like, you know, through the right training, you will be able to like make sense of people that have like different preferences because you're focused on optimizing the rewards. So if I ask you this question, and I get this answer, I have to then explore this part of the graph versus if I see this question and I get a different answer, I can explore a different part of the graph. But I'm still able to zoom out if, for example, your exploration looks different than previous explorations that I've had. And so the idea is we've actually built this movie recommendation expert on top of this reasoning engine that basically solves the problem of what do I watch on a Friday night? It'll ask you the right questions and it'll It knows basically every movie on the internet. And this is a separate instantiation of our reasoning system because it's a separate graph. There's obviously no information on countries and it's a slightly separate reward function. But over time, in this idea of marching towards superintelligence, right through all the context that we train on, the goal is if we can construct a system or we can construct a complete mental model of the world that consists of all domains wrapped and fundamentally connected together. And we can train a system that can take any task as an input and implicitly determine how to kind of structure it as an exploration task, aka say, this is the reward that I'm optimizing for, and then form actions to to reach that reward and form a strategy to reach that reward. The goal is it won't be dependent on— we won't silo the knowledge and we won't hardcode the reward. Right. The idea is if we can, through training on lots of different contexts and lots of different domains, if we can append all of this knowledge together and the system can intelligently determine at this task that I've been presented with, this is how to formulate the reward as a graph exploration task. That to us represents the components of a truly superintelligent system that can adapt across a number of domains and a number of tasks.
So it moves beyond replicating human expertise because you apply expertise on any topic. And what does that become? It's really what you would define as potentially super intelligent at the point. So it's someone that has complete awareness of pretty much everything, has knowledge graphs built across everything.
It's probably the deepest conceptual knowledge of the world, and it can even use— and I think at the start, it is fundamentally something that is reactive. You can give it a task, The task can be diagnose this patient, or this task can be generate this kind of creative synthesis, or the task can be guess a country, or the task can be discover this new drug. And it knows intuitively from the input that it's been given and from the environment that's in front of it what parts of the graph to navigate to At that moment in time to solve a reward that it thinks is representative of solving this task successfully.
And that is a better architecture than just throwing a ton of data and scaling the data in neural networks.
That's the thesis. Because one is fundamentally single shot and stochastic and the other is actually trained to optimize rewards and is grounded in conceptual knowledge and facts of a domain. So one hallucinates. For example, GPT-5 tried to guess a country that wasn't actually even a country. virtual country that was created by some people that had launched a satellite in space.
No.
And so there's tons of hallucinations. Sometimes GPT-5 will take the same country across different turns, guess it one time in 29 questions, the other time in 12 questions, because it's not fundamentally trained to reach a reward. And every output is trained to optimize that output sounding smart to answer that question versus thinking 7, 8, 9 moves in advance.
Yeah, because it's not thinking in meta prompting terms, right? It's not thinking, let me develop the right strategy knowing that I'm an LLM.
But even if it's metaprompting, it's only metaprompting to get to that one answer, not to get to—
It's metaprompting to get to that one answer.
To get to the answer that represents what it thinks sounds best as the next action versus—
You say sounds best and it's optimized to sound best. Why do you say that that's what's been optimized on?
Because why do I say that? That's because why it's been optimized on. I mean, because it fundamentally hasn't. The goal of LLMs is how do you maximize the predictive capabilities of when I see these words, these are the most likely next words to follow, right? If I see the words, like, what is 2+ 2? Optimizing that is from all the text that I've read. Every time I see the words 2 and 2 together, the word 4 comes next. And so by saying that it's optimized to sound the best, what I mean is It's optimized to, from a predictive standpoint, generate an output which best correlates to the distribution of tokens that it sees in its data that sounds like what it's read before. Right. But what it's read before, just word for word and kind of parroting that word for word, isn't necessarily doesn't necessarily mean that it's strategic or trying to optimize a reward, if that makes sense.
I understand. And I mean, what's interesting here is that I remember in 2022 when I met you the first time, 2022 or 2023, maybe 2022, I think that you were saying at the time that LLMs are going to hit a wall. And you're like, it's going to happen. There is just so much you can squeeze out of there. They are going to get better, but eventually we'll just hit a ceiling and that's as much as you can squeeze out of them. And we were really uncomfortable with making that bet at the time because it seemed unstoppable, right?
Yeah.
LLMs were like, do you really want to bet against the industry consensus? And it just seems like it's going up and to the right and you throw much more data. Everybody's going to throw money at it. Yeah, you don't really want to bet against that. Um, we decided to do it because we felt incredibly compelled by you and your co-founder, um, and we believed in the, in the thesis and the mission, obviously. Uh, but it was a pretty big swinger to take. And now looking back, that was a pretty decent guess. Now we have much more data to actually look at. Yeah, it's actually, you know, it seems like it's plateauing very clearly at the moment.
The dialogue is increasing in the industry too. I mean, obviously Yann LeCun from Facebook or Meta is kind of one of the biggest or most loudest voices of LLMs kind of plateauing. But what I think no one has articulated in the way that we've articulated is that what are LLMs good at and what are they not good at and what does humanity need? LLMs are really great at re-synthesizing knowledge and kind of increasing our productivity for perceiving knowledge. Being able to summarize hundreds of thousands of emails is super useful, right? And what LLMs and ChatGPT is one of the products that we're all incredibly dependent on and I think incredible for productivity. But again, the benchmarks And this is why I hate terms like AGI and superintelligent. But the benchmarks of truly superintelligent systems shouldn't just be on kind of these single-shot predictive tasks. And no one has yet articulated what is a benchmark of a superintelligent system. And I'm here to say being able to play the country guessing game is actually a benchmark that correlates to being able to discover drugs or being able to diagnose patients or being able to Because it's ambiguous by nature. It's ambiguous by nature and requires a sequence of actions and adaptations. And that's how humans exist in the world.
And I mean, the bet is that the LLMs are fundamentally not optimized for the sequence of action. Yeah. And yeah, so interesting. What's in the back of your mind? What keeps you up at night in terms of, ah, this could go wrong, we could be wrong on this, and we don't really know what is going to be on that front?
I mean, I think structuring the internet's knowledge, and even this doesn't have to— I think these reasoning models that we're building can go beyond just reasoning on conceptual knowledge as well, as a side note. You can structure the internet's knowledge, which kind of can serve as a proxy for the collective conceptual knowledge of the world. But also in robotics and actually taking actions in the physical world, I think you can also map out, even just as the examples that I did, any kind of scene as a structured world model or knowledge graph as well. Train a system not to form strategies on solving a cognitive task, but to form strategies on how to perform a task in the physical world. And so as a side note, that's where I see kind of the extension of these reasoning models just beyond conceptual tasks in the digital world. But fundamentally, one of the hardest things is actually generating these models to reason on, these graphs to reason on. It's obviously quite compute heavy. It requires having the right ontology. It requires being able to adapt that ontology as you get more and more information. And so I think that is a big challenge. I think for us, we've definitely cracked both at a small scale, and we look forward to not only using harder tasks, but we're not just a research lab, we're actually going to be launching a number of products. Right. And using kind of the products that we build and kind of the base camps of intelligence that we define to figure out more scalable ways to structure knowledge effectively and then to explore larger knowledge graphs using reinforcement learning. But I feel like we've already proven that our reinforcement learning systems and the way our algorithms can can explore really large knowledge graphs efficiently. So I'm pretty confident that once we have these really large knowledge graphs for a domain and for a task and we train our systems on it, they'll still be able to explore it quite efficiently. So for me, I think the biggest challenge is how can we structure knowledge efficiently at scale? And actually, if we're able to do that as well, this also realizes previous internet visions in the past, like the semantic web, for example, actually making sense of the knowledge of information, like humans, or the knowledge on the internet, like humans can. So I think that is fundamentally the hardest challenge, not just in a domain, but at scale, structuring the world's knowledge.
We're slowly coming to the end, so I want to leave you with a question, um, which I'm curious to hear your thoughts on. Um, but it's actually a self-question, so I'm asking you your own reinforcement algorithm. What would be the most— you know, is there something you think is really important to speak about that we didn't get a chance to, um, in the context of your life and Athman Labs and, and your story?
Actually, we could talk about why Othman is called Othman. Why is it? We actually never never discussed this, but we have. We we have. You and I have. I mean, we actually didn't discuss it. No, not on the podcast.
This is news.
It could be a cool thing to end with, which is we talked about kind of the purpose. One of the purposes of life to is to kind of expand your. awareness, right? And to take actions in the world and generate output that serves to expand your awareness. That's what it means to live consciously. If you can channel your energy to expanding your mental models, you feel more limitless in your position with respect to the universe. The universe feels smaller. You feel like anything is possible. And actually, one of the kind of frameworks or the underpinnings for like thinking about this is actually a religious or spiritual framework. I grew up in a quite spiritual household. We weren't ritualistically religious, but we were academically religious, which meant that we actually, especially my grandma and my mom, deeply tried to Um, understand like scriptures and teachings from almost more of an academic standpoint versus like following it blindly. And so one of the, obviously the schools of philosophy in Hinduism, like we grew up in a Hindu family that, that's, that's quite inspiring and quite ancient, is, is the school of Vedanta, the school of philosophy called Vedanta. And Vedanta basically postulates the relationship between Atman, which is our individual sense of self and our individual conscious experience, which we perceive is, is limited, and Brahman, which is the collective conscious existence of the universe, the collective power of the universe. And actually, like, what Vedanta proposes and really questions is, what is the relationship between Atman and Brahman? What is our relationship to the universe? And Yeah, I think there's two schools of thought even within Vedanta. There's like Advaita Vedanta, which is that fundamentally the Atman and Brahman are one, which is quite a loaded, uh, loaded sentence. But fundamentally it means that your conscious experience and my conscious experience are actually all connected to like a singular power, right? And then there's, you know, that's called non-duality. But then duality is of course is like we have completely separate conscious experiences. And so for me, I guess, and kind of going back to Pascal's wager, saying that there's probably no way to absolutely prove, at least in this physical world that we live in, if our conscious experiences are connected or if they're fundamentally separate. So you are the power of belief, and you can subscribe to a framework to Organize your mental model in a way that you think allows you to live, live better. And for me, this notion of like, you know, if the Atman and Brahman truly converged, the reason why the company is called Atman was, first off, why is the company called Atman? If we build products that can emulate expertise, like they can emulate teachers, they can emulate scientists, they can emulate stylists, they can emulate— what that does is build counterparts that serve to expand our knowledge of the world versus showing us more of the same thing. They guide us through knowledge in a unique and individualized way that help us discover more of it and also help us kind of expand the nature of our existence. The first products that we're building actually are like personal shopping concierges that have deep knowledge of the world and deep knowledge of culture, as well as deep knowledge of products. And by getting to know us, they can actually introduce us in the right way to new types of products. And even as simple as the fact that I'm wearing a new style, for example, increases my consciousness and increases my confidence and increases my capacity to feel good about myself and opens me up more to the rest of the world and changes the circles that I can interact with. And so discovering new knowledge and existing, coexisting with systems that introduce us to knowledge, even in the physical world, can open us up more as a species and expand our lived experiences. And so how can we build systems? First off, by building a platform that structures all the knowledge in the world, how can we build systems that can help us discover more of that knowledge faster? If we represent Brahman as the total knowledge in the world, how can we build systems that can basically expand our Atman to explore as much of that world as possible? That's fulfilling the realization of bringing the Atman as close to Brahman as possible. And so from a company standpoint and from a how do these reasoning systems How should these reasoning systems interact with humans and what is the purpose they should solve? The purpose they should solve is how do you build machines that expand our individual perception of the world and our understanding of the world versus show us more of the same thing? And so that's why the company is called Othman. And I think how Othman influences kind of, or Vedanta influences my lived experience is that if you were to subscribe to the school of thought, which is that, um, our conscious experiences are all fundamentally the same and that Atman and Brahman are one, you know, there's this quote that I, that I love that says, you know, give me the, give me the courage to control the things I can, the strength to accept the things that I cannot, And the wisdom to know the difference. And I feel like if we had the mental model that we're all fundamentally different conscious experiences and we have a significant amount of control on how we can affect the universe, um, you know, you feel like you can control more of the world that you can. Whereas I think if Atman and Brahman were really one, one of the things that you realize is that The only, you know, our, our unique awarenesses were brought together at this point in time for a reason and are going to interact in, in the world in a way that is fundamentally outside of your control. And so the only thing that you can truly control is not letting your ego or your fear get in the way for how, um, this moment in life is meant to unfold. And all you have to do is just be open and aware of how your awareness expands as you interact with the world, and just be grateful for that as your existence, because there is no I, there is no you, there's just awareness. And I guess another quote that I love is the idea of operating reactively versus proactively and figuring out you know, what someone's true motivation is, is are you someone that lives as if no one is watching? Or are you someone that lives, you know, because you want to live in a way that, um, you feel like others are watching? And if you believe that the Atman and the Brahman are actually one, it doesn't matter. It doesn't matter. There's fundamentally no one that can or will or is watching your story. You're just letting, you know, that existence pass through you, and you're just following the energy, um, and you're following your innate DNA and your innate awareness, and you're just seeing, you know, how much you can expand it. And yeah, that's, that's kind of— I know it's very loaded, but it's how Vedanta, which is something that I was really lucky that my grandma was able to introduce to me at a young age from more of like an academic sense versus like a ritualistic sense, It influences my mental model for how I interact with the world and really is the underpinning for building systems that can expand our collective atman and our collective awareness.
It really feels like there is nothing else you could be working on. It has to be this.
Yeah.
