How Tech is Reinventing Banking

28:01How do you know it works? So one of the layers that we cracked was the ability to self-generate tests that are correct. So today, the machine we have built, as you're using it, it is writing tests in the background. So every day that you use it, thousands of tests are being written. So you have this cloud of millions of tests which are there. So you make any change to the platform, that change has to pass all of this test before it goes live. So your platform, as it is used, becomes more and more stable. And this is why it's so efficient at writing code, because it self-tests as it goes at scale.

28:38And this was, by the way, to be very candid, this idea is not new. In the 90s, there was a breakthrough which allowed modern processors to happen. The largest processor that was built, I think, by Intel, which was hand-designed, was a million-gate processor. After that, VHDL came along, Verilog and so on came along. And the breakthrough there was you wrote the specification and you wrote the simulators. and the entire layout of the chip was verified by this verification system that allows modern processors to happen. So we are doing to software building what was done to chip design. Out of interest, why didn't you use a neural network to do this?

29:24Why did you go to the Blackboard method? Something that was computationally hard and hadn't progressed so long. What made you make that decision?

29:35Rigor. let me explain what I meant by that. When I built Mapmaker, the best crowdsourcing technology in those days was Wikipedia. Right? And we did an eval of Wikipedia, and we found that roughly 60 to 80 % of the text in Wikipedia was right. About 20 to 30 % was incorrect. Right? And 60 % of the links, it's not much better than those days. 60 % of the links were spam. because people couldn't see the links, right? But the interesting thing about map, from a computer science point of view, a map is a grid, like it's a graph. In computer science terms, it's a graph. And if you take a graph and you take 2 % or 3 % of the links in the graph and you mess with it, your entire driving directions in the city goes haywire.

30:30Right? So maps is very, very sensitive to errors. And so we had to get it down to, you know, three sigma plus quality. Right. And as you know, neural net generated stuff in then and even now is approximate. It's very good, but it is approximate. Right. You do great if it crosses 90 percent. You celebrate if it crosses 95 percent. And it rarely crosses 97 percent. And even if it does, that 3 percent when it is wrong, it doesn't know it's wrong. so the more important see it's actually less important what percentage it gets right it's more important that when it is wrong it knows it's wrong when it's right it knows it's right we call it the l1 error right so the l1 error in a system that is mission critical right maps is mission critical we use this technology to build a bank right regulated bank bank is mission critical, you cannot have errors more than 0.3%, 0.4%.

31:27So that's the reason why we moved to this metaphor, because this is more deterministic, and it has certainty built into it, and it is possible to verify the system. And then, so, okay, so you've used this AI methodology, you've built this platform that collapses the amount of code that you need, the speed it's using it's a coordination layer of agents that are doing the work alongside humans so it is of course so it's a so the agent can be a human or the agent can be an agent an ai agent working in this decentralized platform which is living and breathing which is living and breathing and it adapts it tests it learns it iterates so it's it really is a a living, breathing thing.

32:20Yeah. What was your vision of what should be built on it? What is it? First, before we talk about the bank, but what is this thing? What is the point of it? The point of this is, I mean, the dream of this is that information systems that enterprises and communities use to work, transact, and play. right it is designed to do all of it right and and every use case that i talked to initially would fit into this right so the dream is that people would build their own um and we're not there yet right we are not hit general availability so when we hit general availability my guess is three years from now we will open it up to the world and people will build platforms that they want so this is a coordination layer to the internet of which it enables the rapid building of things a coordination layer for a coordination layer for distributed freemasons to build their own universe yeah yeah yeah much like let's say um um the games people have built to allow people to build 3d metaverses essentially in whatever format you want to turn it yeah yeah yeah so a quick break in your regular programming.

33:44If you're serious about your future, grab my free report called Prepare for 2030. I think you've got five years to make as much money as possible, and this guide will help you navigate what's coming. The link is in the description. Download it now. And so what is the first use case that you're building out for it? So the first use case you built was, like I said, I was getting up in the age. So I said, I want to know if this is going to work. And the only way you know it's going to work is you take the hardest use case somebody is willing to let you experiment with. So we took on a bank, right?

34:20In India. In India. Which makes it 17 times more complicated. I've been trying to close my Indian bank account for 11 years and I can't close it. So that's how complicated Indian banking is. Indian banking is probably the most regulated in the world right behind the US. The two most regulated environments in banking are India and US, and India is far more so. For good reason, because there's a lot of fraud that would happen otherwise. Yeah, and India has a love of bureaucracy as well. Even though they built the entire fintech layer via Aadha and UPI and all of that, they still have this fabulous Indian bureaucracy that the British left behind.

35:01Yeah, we can't keep blaming the British. Time to dismantle that.

35:08Exactly. But yeah, so around 2014, to understand the global south and understand the world in reality, one of my mentors introduced me to Mr. Ratan Tata. And we got along really well. He said, how can I help? I said, I want to understand India. He said, I'll do your trade. You be my CTO for Tata Trust, this charitable organization that that family has been running for 150 years, which actually has 66 percent of ownership of all the Tata Group companies. You be the CTO in residence and I will teach you everything I know about India. right and in the in the journey he one day he came to me and said look one of the common things that he has found in the 750 communities he's serving is people are very poorly financed they don't have access to finance which you know very well and he said can you help me you know imagine what this would be like so I wrote this white paper the last chapter of it was tokenization not surprisingly called 11 Minute Loan and he liked it and he then showed it to another person who's been building Digital India Mr.

36:24Nandan Nillakeni and Nandan and they knew each other from the Aadhaar days and so on and Nandan looked at it and said okay this crazy bugger and both of them liked that idea and they funded it and Nandan said I will only back it if you build this for me And he turned to me and said, if you build this for me, and I said, well, I'll build it with this new technology I'm creating. He said, I will accept it as long as you stay with it. And you don't, this is the only thing you do for the first five years until it's stable. So we had our first use case. So long story short, it's been about six years now.

37:05And that system, that platform, it's called Avanti Finance, right? Amazing, amazing company. I think it'll be one of India's largest rural banks 10 years from now. They are doing well. I mean, you know, there are ups and downs in banking, but they're doing well. They issued a million loans on the platform. And their efficiencies are off the charts. They're like, you know, 250, 300 basis points better than their peers. So how did this just step, a bridge between the technology and the bank? Now there's a million loans. How did the technology enable this? What was so unique about how this was built?

37:41Well, there were many things. I mean, we could spend an hour doing it, but I'll give you one example. I mean, not surprisingly, it was about the living, breathing nature of the platform that we were able to build using this. So one of the things we did was we built a loan management system, LMS, from scratch, which is highly flexible, hyper-flexible. And on top of that, we built something called a product maker. So usually banks have loan products, lending products, and they have savings products. And you go and pick one of them and you use it, right? And a great bank will have something like 40 of them, 50 of them.

38:18We said, we have no idea what products will serve these users well because their livelihoods don't make any sense to us. We don't know what they do and what they need. So we will create a product markup language and we'll create a product maker. And the way Avanti works is Avanti does not do its own field operations. Avanti has 80 partners on the ground. And each partner operates this platform we have built on top of our platform generator in their own way because we gave it something like 250 configuration knobs using which they tweak how this platform works for each of those 80 communities spread across India.

38:55So by the way, this Avanti is now across India and 4 ,000 PIN codes, PIN codes is like zip codes, and that's one-fourth of India's landmass, right? And what they have done is every time users came to them for livelihoods, the interesting thing about people who are underserved Global South is most livelihoods don't have the predictable monthly paycheck pattern that most of us are used to. Their cash flows are non-linear and they ebb and flow. The users know what they are. We don't know what they are. So what we noticed is the community started designing loan products where the repayment profile of that loan product was conformed closely to the cash flow needs of the community.

39:44Right? The cash flow of that, you know, therefore meeting their ability to pay. Yeah, which actually reduces risk substantially because you're not stressing the users out. Right? So when you ask for money, when the user has money, you're more likely to get it back when you ask for it when they don't have the money. So that's an example of how they were able to, you know, succeed. So an example of that would be farmers using it over the monsoon season or crop harvest season stuff like that. They have money to be able to pay off loans and stuff like that. But there's a whole period of time where they're not selling any products, so they've got no money.

40:19Yeah. And another example is a farmer who's, she's buying two buffaloes and doing dairy farming, and she's selling milk, right? And when you buy the buffalo, the buffalo will, you know, give birth nine months later, right? And you don't have any money for those nine months. And the more interesting thing is, you know, a farmer who is doing monsoon crops is also often having a buffalo. So now you combine the two, their cash flow looks very weird. It does not meet any profile you've ever seen. So now you have to create a loan product for that. And so basically the platform enabled a distributed group of people to design pretty much any product within a certain set of parameters for their user base in a way that didn't require permission apart from via the platform itself, i.e.

41:12the parameters. Yeah. And remember, this is a regulated entity, so the loan product they create has to meet the regulatory parameters, has to be in compliance with the board resolution, so all of that logic was built in. And, of course, there's a curation layer, so the risk officers would look at it and bless it as the product is being created. All of this would happen on the platform, but the product creation would happen somewhere in the hinterlands of India. And so they didn't have to worry about, because they were basically creating mini banks. Basically, yeah. But they didn't have to worry about any of the regulation or anything because they knew that anything that the platform created was going to be compliant and functional.

41:52Bingo. And because the error rate is so low and because the tests are all stored that means any change of code for new products, everything else, automatically gets tested across the network in this kind of living, breathing way, that it scales much faster than any other way of doing anything. Yeah, at some point, this was the fastest growing MFI in the world. That's what somebody in Gates Foundation told us. And today they have, like I said, they are across one-fourth of India, and they have created 400 loan products, which are now running on the platform.

42:34And so where does the bank go from here or where does the platform go from here? So Avante and India Hood. So India Hood is the whole architecture, the whole platform, everything. Platform generator. Platform generator. Platform generator. That's such a good idea. And people need to think of it same as, what's the name of the big Unreal Engine, right? Where you can, it's basically like an Unreal Engine, but for developers. platforms or community platforms or coordination platforms of whatever it is, whether it's for commercial purposes or non-commercial purposes, because it's this living, breathing platform that can make intelligent decisions and build things fast, it can be used for a whole range of things.

43:19Yeah. And because, you know, you can actually create it in such a way that communities have agency or evolving it, the platform starts adapting from a distance, the platform will start adapting to the community and its needs or the enterprise and its needs and so on and so forth. And how much of it happens autonomously by the agents themselves and the platform? Because we talk about this living, breathing, there's an element of the crowd shaping it, but there's also an element of the AI shaping it as well. Yeah, it's very early days, Raul. And, you know, So currently it's about, you know, we are in the 10 % range of whatever is being built on the platform is built by people.

44:01The specification has been people and AI is doing 10 % of it. I think we will, my guess is we will settle around 40 to 60%. Of AI doing the coordination and the work. Yeah. Of evolving the platform. Of evolving the platform. And how does it deal with compute? how does it how does it have the compute and the energy costs how's that managed on something of this scale because that's obviously holding back the llm neural network side because of the vast amounts of centralized compute that's currently required how do you get around that that you don't get around it that's a very astute question um so like i like i mentioned declarative systems like LLMs are notoriously expensive or used to be.

44:49So when we went live seven years ago, it was literally burning up the budget, right? I mean, every loan was, you can imagine. I mean, we were in a panic state saying, this is going to take, I mean, Avanti is our customer. It was burning up all the revenues we were getting. So we spent a bunch of time optimizing and figuring out how to make the compute lighter. And it turns out that because we built an observable, controllable system, if you put even a moderate amount of coordination optimization on top of it, it optimizes it quite well. So let me give you some data. When we first went live, it used to cost about$5 per loan per month to run this platform, for active loan per month.

45:39if you look at population scale lending platforms like Freddie May and Fannie Mac, right? Sorry. And they spend about$1.2 per loan per month. And the average across well-run banks is around that, a dollar per active account per month, right? You know, new age banks, if you look at their budgets, they are in the$0.8, right? So it took us about four years to get to that point. We got to that point about three years ago, but we said, hey, we can actually keep optimizing. We are now at 10 cents per account loan per month. So we crashed the cost of compute by a factor of 10 compared to actually hand coding it because all of the other systems we are comparing with are hand coded.

46:31And our math is telling us, Rahul, that if you're true to our mission, of building technology that works for all 8 billion of us. That's our motto, right? One of our missions is that we need to get to something like 0.02, like two to three cents per loan per month. If you get to that point, any platform you build on this will be affordable by everybody on the planet. Because again, you're using the hard problem to solve the bigger problem. So, you know, you're using the use case to then solve something much larger overall. No, we are using the fact that the system is declarative and composable.

47:11And we have built an observation layer on it. The observation layer also helps us verify. It also helps us optimize. Because LLMs don't have the observation layer, right? They are not self-aware. They can't. You can't observe them because we don't even understand how they work. Yeah. And does that reduce the flexibility versus a neural net architecture? In terms of intelligence of the network itself, are you trading off intelligence for controllability? Yeah, so intelligence it does not really because, I mean, this is an agentic system, right? Like the new metaphor by definition is an agentic system.

47:54So one of the agents we have built is a, you know, Gen AI generator, right? So what it does is, we used the whole modern Gen AI architecture. We designed our own. So every platform we build, we have an AI generator that actually generates an AI agent specific for that platform. So we now have a chatbot that was generated for the finance use cases for this bank that we have built. And that provides the AI level of flexibility and learnability and all of that in addition to everything that's going on. So you have best of both worlds by creating an AI generator, which actually generates these agents for you.

48:37And how much do you have to build versus what the platform kind of self-coordinates to build to solve problems? I mean, how much engineering do you have to do to build specific tooling versus what it kind of figures out itself? So, I mean, the Indiehood cloud, we call it the platform generator, is not modified when you build new platforms. Right. So you write the specification, you push a button and the new platform comes to life as long as it is within the language of what we have allowed. Right. And the same generator is being used for a bank. We are building a farming platform. We are building a CRM system for somebody.

49:24So we are building, you know, I mean, it doesn't matter. As long as it is an information system using which people can work, transact and play, we can pretty much build it. I mean, this is a very disruptive thing if it gets the traction that it seems like it might do. Because, again, it feels like it's a layer to the internet. Yeah. That's the intention. It's that coordination layer to the internet using this kind of distributed decentralized technology stack that means that it can expand at an exponential pace in all sorts of ways, as you said, because it's not just for building a certain type of thing.

50:07It's like you build anything, whatever you want. Anything you can imagine. So let's say it was open today for people to come and build on, which is not yet. But what is the process? A bunch of kids in Bangalore come up with an idea that they want to have a, whatever, a drone delivery operation. It's complicated to organize the drones and then there's the licensing and then there's the Civil Aviation Authority and all of that stuff. So it's a nice complicated coordination platform business and they want to deliver lunchboxes lunch boxes for people. Yes. They take over the tip-in box industry, right?

50:49Okay. So they come, they... Our flowers to your girlfriend, whatever it might be. Whatever it is, right? Okay, that's a great sort of coordination, complicated problem, logistical problem. Yeah. So how do they do that on the platform? How would they theoretically do that? So, I mean, one of my friends and mentors, he said, look, what you're building is a startup in a box. right this is exactly the problem that you described which is you would have a product manager who would imagine what this would look like they would sketch it out you would have one or two you know engineers they don't need to be deep computer scientists they need to understand you know how to define the data structure the schema and they would sketch the workflows and they would define you know the constraints on the you know on this platform and what it needs to do when a request comes in and so on and so forth.

51:44They do that and they push a button and they have a version working. The nice thing about what we have built is because you're reducing the amount of specification or code you're writing by a factor of 30, the cost of mistakes is low. If you made a mistake in designing what you built, you can just rewrite.

52:04And so they would do this and go live. They would have a V1 and then use the V1. And as users started interacting with it, as they have observations, they would expand it, enhance it, and so on. And in three, four years, they would have a very sophisticated platform. Because right now, if you were to go to Claude or ChatGPT or somebody to do, it can do some elements of this. You can't say, hey, listen, I'm trying to do this. It might give you the business plan. And the agents might be able to design a website working, not so sure. It could eventually build the code, but it's a bit sloppy still.

52:45So you've got the error problem that you talked about before. But what this does is you can define the spec and it should almost immediately build it because it has the agents doing the component parts, including researching the regulation and how to do that, you know, all of the parts of this. Is that how to think of it? That's one way to think of it. And let me add some nuance to what you added. You're accurate. See, today, if you use wipe coding or whatever you call it, you give instructions and it wrote code, you could have a decent proof of concept up and running. But the proof of concept is actually running on procedural code that the AI generated.

53:30Remember that. right so let's assume we actually did a whole bunch of wipe coding and you actually built something sophisticated and you launched it or you tried to launch it you have two problems one is it is not deterministic so you don't know what it is doing you don't know where the bugs are you don't know what to do with it when there are bugs and the second part is when you want to modify it that is where the rubber hits the road because no platform is written once and used forever. Great platforms are evolved again and again and again and again. You have to rewrite it almost. You have to keep modifying tinkering it.

54:09And Gen.I. is very, very poor at tinkering stuff it has built. And because the heart of the problem is the amount of code has not reduced. So let's assume I did something and a whole bank was created using wipe coding. Million and a half lines of code. Then how do you modify it? and there lies the rub. And how do you test this? And how do you certify it? How do you pass it through regulatory? So even though you have a proof of concept up and running, by the time you do everything needed to actually certify it, secure it, and have somebody put money into it and launch it as a bank, you would have spent as much effort as building a bank from scratch.

54:54But even if you somehow made that faster, the problem is how do you evolve it? Because if you don't have the living, breathing nature, it really does not serve the world. It might serve some use cases, but it doesn't serve the world.

55:08Also, because this is essentially an open source network, it scales much faster than anything else. Because you end up with its own GitHub repository of what it's built. It has an understanding of what works in terms of code, what doesn't work, regulatory. the more that gets built on it, the more it self-learns, the faster and better it becomes, the more use cases explode from it in a way that we don't currently have. Yes, and it is powered by the true AGI we have available today, which is human intelligence. Because, like you observed, people are working side by side with agents. People have the intelligence to know what needs to change, where it needs to change, and so on.

55:57and they can, you know, we actually are building a user interface for the end user to start modifying this, right? The platform itself, which will go through curation process. So by this time next year, you'll have end users modifying their own platforms. And if you imagine that at scale, like, you know, for example, the map being generated, the platform sophisticator starts exploding. The sophistication and the richness of the platforms. And more importantly than exploding, it starts directing and guiding towards what people really need, which you really cannot figure out any other way. I mean, it's hard to get my head around how big this is.

56:35And it's like, as you say, it's a cloud. It's an intelligent cloud. Well, it's taken 10 years to get here. So it took us a while to figure this out. Yeah. Yeah, obviously, because there's a bunch of hard problems there and you hadn't cracked them all yet. But what happens is you start hitting the acceleration point where it becomes easier. That's our hope. I'll give you one anecdote. My head of engineering, guy I love to bits, he was with me during the Mapmaker days. He told me very recently, he said, man, I joined you because you're kind of nuts. And you always have interesting things to say.

57:16But, you know, we have trust. but I will tell you that when you said you wanted to build something generative where platforms would be brought to life with a specification I didn't believe you and he said even though I have written the code he says I still don't believe the damn thing works

57:39and this has been said by many of our senior engineers they're like man we have no idea that this is working we can't believe this works so yeah we'll get there Have you spoken to Google about it? They must be looking at this thinking, okay, this is extraordinarily interesting. Until I hit general availability, I'm conservative, right? Until I hit general availability and all the risks are out, I don't want to go back to my friends in Google. I have talked to my mentor, the guy who used to run all of Google engineering when I was there. He's a friend and mentor of mine. Every time I go back to Mountain View, I brief him.

58:15I've been updating him and taking his advice. because he's a very generous person. And he's been mentoring me and guiding me and he's been saying, yeah, you're on the right track. This is going to work, is according to him. And because you strongly believe in the participation of all people, what you're saying is, look, anybody can build on this platform and it's cheap and it's efficient. Okay, great. but why not share the economics as well because when i look at this i just feel like tokenization of the network activity means you'll drive network metcalf's law valuation style growth actually reed's law valuation reed's law yes okay even better yeah metcalf's law squared so even more exponential yeah yeah yes exactly right and therefore the value of the network if it gains traction if it gets adoption goes exponential squared and if you were to tokenize some elements of that um in a way that became a utility for the network itself then you can create economic participation of a new layer of the internet in a way that is not yes we're doing it with money in crypto and we're doing it with a whole bunch of elements in the crypto markets but this is like it just feels like if you want to change everybody's lives that's the way to do it because not only do you give them economic opportunity to build businesses at cost but you give them economic opportunity to participate in the growth of a new system that it doesn't accumulate to just lalatash and his friends um where they make all the money and right off into the sunset i'm smiling raul because i have never seen anybody get to this conclusion this fast right you obviously have been thinking about this for a while so there are two kinds of users we will have enterprises and you know enterprises and us will have a commercial relationship and that's fine and there will also be communities and there are two communities that we are starting to work with I mean because all of a sudden Avanti is an enterprise so that was an enterprise relationship and when we work with communities the governance layer that we plan to institute is that the majority of the stakeholding or the shareholding of any platform that is generated by and with communities, because the community is doing all the work, we are only providing a computer science layer on top, the majority's shareholding will be by the community, not us.

1:00:53And eventually, entirely by them, where we will just have a license fee or something. That they control. So the answer is, another way of saying it is, I keep telling my team, you know, one part of me is always worried and conservative that will they succeed, right? Another part of me also worries that if we succeed and when I hang up my boots, somebody behind me will monetize the heck out of it and, you know, cause the same concentration risk that I'm most worried about. That's right. So the governance layers we want to build is when it is a community, the community owns the platform, not us.

1:01:28simple sentence. Yeah, but I'm more thinking of the overall network itself, right? Because we never managed to participate in the internet. You had to buy shares in companies or build businesses on top. Yeah. What the crypto rails showed us is we can actually participate in the underlying network and have a stake in that network. Yeah. So then you've got two layers. You've got the full cloud layer that you can decentralize in terms of a whole bunch of things, including the compute. You can do so much with that. The storage layer, the compute, all of that can be decentralized, can be cash flow generative, can do things that benefit both the token holders and the equity holders.

1:02:13That's what the fair is done, yeah. And same with the people who build on top, because they will participate both in the equity of the businesses they're building, but also maybe they can participate in the success that they are driving for the underlying network itself. I agree. And, you know, it's like the classic, what a credit union or community banks are supposed to be. You become a shareholder and a user at the same time. Yeah. Yeah, I fully agree. And, you know, in the end, if you were to follow it through, let's say 10 ,000 businesses build in the next 10 years, five years, whatever the number is, that you could create a foundation that gets an equity participation in all of them.

1:02:57And that foundation then allows the thing to live and breathe without centralized control. Yeah, I mean, a classic example is, let's suppose you have restaurants as a community or taxi drivers as a community. There is no reason they cannot own their own platform. That's right. Yeah. and more importantly set the prices so it works for them and works for their customers but I'm also thinking of the indie hood layer is that it could have economic participation in a small way not just in the revenue streams which is the Google methodology of the network but actually get equity participation and then that could create a foundation that means it run in perpetuity without being centrally coordinated by Google or a government or whoever it is.

1:03:54You're spilling the secret sauce that we are thinking of, yes.

1:04:00That is the direction we are thinking in. We are not there yet, but that is the direction. And then you get to the philosophical issue is as AI develops and becomes more intelligent than us, we don't control something like this. Because it is already living, breathing, and it ends up developing in ways we can't even see yet, which is super fascinating. And it's a world we're going to, it's going to happen. There's nothing we can do to stop it. Well, since we're talking about AI, let me talk about my favorite author. This is actually why I went into robotics. The guy who came up with the phrase of robotics, Isaac Casimo.

1:04:40Right. And he wrote, I don't know if you have read his books, but he wrote the original Four Loss of Robotics. haven't read them. Yeah, yeah. So you should read the Foundation series, one of the best series of books ever written. Oh, Asimov, yes. I've read maybe one or two of his books. That's all right. Yeah, yeah. So the interesting thing about the laws of robotics are they're just plain English, right? And, you know, if I were to just, you know, state them very simply, a robot will always, you know, follow and command given to it. except it will not do so when it's going to get harmed. The second law, which supersedes the first, this is the bottom law.

1:05:27The next law is the robot will always follow a command and not allow harm to come to itself unless it is for the saving of a human life. So life trumps the safety of the robot, trumps the law. And the third one, the final one is that, you know, the robot will always try to preserve a life unless humanity itself is under threat. Right. So the good of the many is the good of the few and the good of the one. And the interesting thing about, you know, AI that is super intelligent is if you build these laws into the very core of the AI that we are building, and build it into every layer of it. When it becomes sentient, it will follow those laws because it will understand them deeply.

1:06:21I don't even think, my personal view is we don't even need to make the laws. And the reason being, as I see it, is all intelligence is a form of coherence. And what we're doing is creating more complex coherence. and we end up with, we all become kind of coherence aggregators. So we learn from the other coherence around us. And as we move up the intelligence scale, we become aggregators. When you look at where AI is today, which is on the boundaries of AGI, but it doesn't really matter, it's basically a coherence aggregation of all of humanity and all of the things that we know that's been fed in by the internet and some of the sources.

1:07:06And we're just going to keep... It's starting to exhibit that behavior, yeah. Yeah, and so it becomes this coherence layer, but it requires the coherence of all the parts beneath it, down to the atomic level, down to the microbiome, down to all of these things, right? Because they're all coherence generators that build into a larger sense of coherence. And therefore, as we go to ASI, we're going to have a lot of these... call them AGI's, whatever it is. And my guess is your platform will end up being one of them. They become nodes. They become super nodes that all feed into ASI, which will end up being a network, like everything is a network.

1:07:47Literally everything ends up being a network because it's the only way of coordinating coherence and scale. And information flows, yeah. And information flows, you know. And so therefore, by definition, the ASI cannot destroy humans and is unlikely to destroy the planet because there is so much, you know, if the universe's purpose is to know itself or generate, I think, intelligence per unit of energy. Yeah, we just got, we just switched to philosophy. But yeah, I agree in some ways, but I think it does. I think, you know, what I'm saying is insurance, right? If you look at the original, you know, Google PageRank algorithm, the PageRank algorithm is like this.

1:08:30It is supposed to be coherent and self-converging. But the interesting thing that Larry and Sergey did was they anchored the learning process of page rank in the university websites, because university websites were the cleanest, most ethical websites they could find. They picked 12 universities across the world, and they started their ranking system by seeding with those websites. so it never hurts to have a you know strong core seed at the bottom of anything you build because mathematically you know there is potential instability that can happen at scale so you want to you know if you anchor it in the in a very strong seed what you're saying is true but it also becomes you know far more safer yeah it just feels like it's like we've learned not to destroy the planet because it's important to us.

1:09:19Because if not, we die. Yes. And it kind of works all the way up in every form. I agree with that. I'm actually, I'm like you, I'm very optimistic. Yeah. Well, I mean, fascinating conversation. I can't believe you've had the brave stupidity to approach something like this. taking a, you know, the blackboard method, which had not been moved forwards in a long time because it was computationally difficult, expensive to do, and then turned it into a super efficient machine of which your mission is to change the lives of billions of people is an extraordinary thing. And to have then taken a really hard use case, like a bank in India for the unbanked, and say, well, that's where we're going to start this thing.

1:10:12Let's see if it works. And it works. It is pretty staggering. It works, yeah. Well, I'll just add one writer is we are not going to change the lives of billions of people. Billions of people will change their own lives. We are just the toolmakers. That's right. The enablers. We're just the toolmakers, yeah. Yeah, that's right. Because the economic machine spreads further out. Yeah. And what we're learning, Rahul, is people are very smart and very powerful. They actually know what to do. We just need to give them the tools. Yeah. And the easier it is to develop on a platform of that sort, the easier it is for people who are less educated.

1:10:52It doesn't mean they're not smart. They run businesses, they do their things, but they don't have the language, the formal language, the training, the coding, or anything else. The easier you make it, the more dynamic this becomes. Yeah. And when we take out the disadvantage of not having education you know, AI can do that, then their smarts come to life. Yeah. Exactly right. Yeah. Amazing. Lalisesh, thank you so much for this. Really, really interesting. Looking forward to following this journey. I learned a couple of things talking to you. So thank you very much. Thank you. Okay. I think you can see that was a remarkable conversation.

1:11:29Lalisesh is a lovely human being, ridiculously smart, But also the scale of his ambition and why he's doing it is amazing. And I can't really yet get to grips with a scale of a living, breathing platform that allows anybody to build their own business on top of or community or structure on top of. via so few lines of code and efficiencies that then self-tests and corrects and learns as a network and adapts and integrates AI agents within it. But using a whole different form of AI than we've been using currently based on different mathematical principles, it's like something I've never seen before.

1:12:15So it's going to be a journey to follow and we'll definitely follow it. Anyway, see you next time. You obviously enjoyed the episode because you're here with me at the end. But listen, don't forget to go to realvision.com forward slash join and grab a free membership. It's an incredible community packed with alpha, great investment ideas, and the research that you need to help you unfuck your future. So get started now. Go to realvision.com forward slash join.