Is India Repeating Its Biggest Historical Mistake🤯 | Vivek Raghavan, Maker of Aadhaar, CEO Sarvam AI

[00:00:00] Most people are already using AI every day, uploading data, asking questions, trusting the answers. But very few stop to ask, what does it actually know about us? If you go and figure out what ChatGPT knows about you, it's quite shocking. And that's not because of the model. It's because you're using ******. India is, after the US, the largest user of ChatGPT. So what are we doing? We're exporting our data and importing intelligence.

[00:00:25] A century ago, you know, when we exported cotton and imported cloth, that's exactly what's happening here. What is really free? That's something that we need to think about. And what are we paying for, for this free stuff? Today, the access is there. Tomorrow, access may not be there. The job of the model is to predict. Why does India need sovereign AI? And what happens if we don't build it?

[00:00:51] Firstly, this technology is going to pervade everything we do. And therefore, in a technology like this, it is important to have a sovereign effort. Today, we have Aabhas Malhadiyar, CEO of Sarvam AI. Sarvam AI is shaping India's AI, built for Indian languages and everyday needs. He helped build Aadhaar and one of the largest identity systems in the world. Hey Vivek, thank you so much for being on the podcast. So good to meet you today.

[00:01:19] Yeah, Roshan, likewise. Thank you for having me on the podcast and look forward to a stimulating conversation. Yeah, so today we're going to talk about a very important aspect. But let me begin first by asking you, why does India need sovereign AI? And what happens if we don't build it? See, I think first we have to talk a little bit about what the definition of sovereign AI really is.

[00:01:47] And I think that people look at things and, you know, kind of three different ways, right? One kind of a thing is that, for example, you have what the so-called, you know, kind of closed models are, such as, you know, Anthropic or OpenAI or Gemini, right? These models, most people access them through kind of APIs, right?

[00:02:15] Which run on infrastructure wherever they are. And, you know, that is, those things are kind of a black box. So those kinds of models, yeah. So when you say closed models, these are proprietary to the particular company. Yeah, they're proprietary. And they may selectively open up parts of it. Yes. That can be accessed by developers through querying. Yeah, it may be accessed by developers. And, you know, if today the access is there, tomorrow access may not be there, right? That is, and you can't do anything about it.

[00:02:45] You can turn it off whenever you want. The second class of models, really, which are today not as powerful as these closed models, are the open models, right? And today in an open model, what ends up happening is that actually you have the weights of the model are kind of published somewhere. For example, in, you know, Hugging Face and places like that. And then you can take that model and host it yourself, right?

[00:03:14] And then use that model. Now, the different, the issues with those things are two or three issues. The first issues, in fact, it has been shown that you can actually take a model and actually put a fairly small amount of poisoning data into a model to make sure that it actually doesn't work at a critical moment.

[00:03:35] So if you're looking at something, so even you have an open model, you have no idea about what data was used to indeed train that model, right? So therefore, especially for critical kind of infrastructure things, you, they could have, it could be poison. And today, almost all the best models in the world are from China, open models. Right. So I can't just, let's say, use a version of DeepSeq, run it locally on my machine or host it. So you can.

[00:04:05] And so the fact is that your data doesn't go to DeepSeq when you use it, when you host a local model. It's with you. So that part is clear. But what output comes could actually be poisoned. Okay. So that is the difference. Right. The difference in a, what do you call that, in a closed model is you, you know, your data is going there. Now they, obviously, everybody says that they don't use your data for various kinds of things. Right.

[00:04:34] But in the end, the fact is that your data goes there. Correct. So one way of defining a sovereign model is really a model where, you know, the sovereign has the ability to audit what data was used to train that model. And so therefore, and so therefore, I think for a country, I mean, firstly, this technology is going to pervade, right? Everything we do.

[00:04:56] And it's, and it's something that is going to be kind of what's the right word to, you know, it's going to be the importance for, for everything we do is going to be very high. And therefore, in a technology like this, it is important to have a sovereign effort. So therefore, that's something. Right. So how does this translate to in practice? I mean, does it mean building our own models? Does it mean having our own data or something else?

[00:05:25] So no, it translates to every, at every level. There are multiple different levels, right? It's interesting, right? When you look at something like, you know, you could train a model from scratch using some set of data. That data may be representative data that's important in the Indian cultural context, but also global data. And then you can actually train a model from scratch. That's one kind of a way that you could do it.

[00:05:53] The other kind of ways that you do it is, oh, okay. Maybe it's some kinds of applications. It's okay, I have an open model and I run the open model, which is also something, at least your data is not going out. The next level, actually, there are many levels of sovereignty. One level of sovereignty is that you actually use these closed models, but you use them in whatever as an API. Okay. Okay.

[00:06:22] So, and if you use them as an API, the models themselves are supposed to be stateless, right? And the interesting thing is because the models are supposed to be stateless. You give it a set of input tokens and it gives you some other output tokens and then you use that to actually build your application. But what really people end up doing is they end up using applications. Like, right, if you're using ChatGPT or you're using something, you're using an application. And there it's much more than just the model.

[00:06:52] They have memory, right? So, therefore, it knows your likes and dislikes. And I think, you know, if you go and figure out what ChatGPT knows about you, it's quite shocking. Okay. And that's not because of the model. Yeah. It's because you're using the application that uses the model. Right. So, the model itself is a piece of code. Yeah. And if I use it just as a piece of code. Yeah. Without my data actually moving to some other servers, it's fine. No, no, no, no, no, no. It's not true.

[00:07:22] Even when, so, of course, if it is an open model, you can host it yourself. Right. Then your data is not moving to, is not moving somewhere. That's true. But there is also the case where you're calling an API call. So, your data moves there. Right. But it is only that API call. So, that API call is limited. It's limited. Right. And at least what they say is that, you know, they are not using your data. Right.

[00:07:48] If you believe that, and I think, you know, certain cases may be okay to believe that. If you believe that, then at least you are giving them something and you are getting something out. And that's stateless. Right. Right. But when you're using an application. Right. Then all bets are off. Right. You understand? So, therefore, there is many, many levels of, you know, control. And when you say think of sovereignty, so don't think actually in sovereignty in terms of even, you know, there is sovereignty in terms of a country and that's probably where you're coming from.

[00:08:18] But sovereignty is actually even an enterprise. Yeah. Or sovereignty is even as an individual, how much control are you giving to people? Yeah. So, for people watching this, I'm sure, you know, we've uploaded a doctor's report or some kind of an MRI or something onto ChatGPT and asked, hey, can you interpret this in layperson's language and so on? What could be the potential downside of doing something like that? See, I think this is a, this is, this, I think that, you know, the benefits of it. Right.

[00:08:46] I mean, I think that, that, you know, it can give you insights and understanding in these kinds of things that are, you know, much beyond really. So, people do that because they see value in it. But if you look at it at a larger scale, then there are questions. Right. If many, many people do this, then what does that mean?

[00:09:09] You know, does, you know, and, you know, people's, your actual, especially if you're using it in an app, for example, there are obviously that app has now a full, your understanding of your entire health history. And that can obviously be, you know, then the bar gets much higher, right? For how, what it needs to be protected and, and, and, and how it needs to be thought about. Right. So. Right. Yeah.

[00:09:36] So, at the very least, I mean, it could give pricing leverage to someone sitting out of cure. Yes. On how they price their drugs based on this mass health data that let's say we're uploading to this application. And that's the least harmful thing I can think of. That's, that's, yeah. And you can start from there and you can even go to, you know, you can go to much more. Depending, depending on the level of evil genius you want to apply to it. Right. But, but yeah, I mean, I think it's a, it's a real issue.

[00:10:06] It's a real problem. The, the benefits are so vast and so immense and, and so immediate that we kind of ignore that. But of course, that's a looming challenge with this. Right. How do we go about building sovereign AI? See, I think the questions is, there is really many levels to that question. Right.

[00:10:29] Firstly, you need to go and start figuring out, I mean, if you're building foundation models from scratch, it's, it's a complex process. You have to figure out how to get data, right? High quality data and very large amounts of data. That's one challenge. And then you need to have the compute, the GPU compute to actually train these models.

[00:10:53] And that takes money and engineering skill and, and, and, and, and time to do those kinds of things. And then, of course, and the data needs to be such that, you know, especially if we want to take build models that are more relevant to our country, then, you know, the data that we are having needs to be, you know, focused on Indian context, Indian languages.

[00:11:19] But on the other hand, the models need to also understand general skills, right? Which is like things like reasoning and ability to write code or, or to think, you know, and I, and I think, or call, call functions, call functions. So all of those things are general capabilities that a model needs to have. And so the process of building a model, you know, is a money, many month effort.

[00:11:43] And of course, you know, it depends upon both what is the size of the model you're training and how much data you're using to train that model and how you're actually creating that data, those pipelines. So it depends on all of these things. We know that, you know, you know, the frontier companies, for example, spend hundreds of billions of dollars training models.

[00:12:09] But it's interesting that if you want to say slightly behind the frontier, you can actually do these things at a much lower cost. Right. So, and, you know, which means like, like, you know, if you try to say that you remain one generation behind, right, you know, you remain two. For example, you know, I think that the interesting thing is this.

[00:12:36] When you train a model, especially you train a model from scratch, it's very, you know, kind of capital intensive thing to do. But, you know, but the interesting thing is that, you know, GPUs are getting better every, you know, every year.

[00:12:57] So if I try to train a model that, you know, costs me X million dollars or X hundred million dollars to train this year, and I want to train the same model next year. Right. Then actually you can do that at a cost that is, you know, a fraction of what it was.

[00:13:16] So I think that's something that, that, that, so therefore that's one area by which you can actually bring down the cost of training these kinds of things. Because, yeah. Right. So that's actually optimistic for us, right? So, which means that, you know, a lot of folks believe that, you know, we were too late to this and that, you know, now's not the time. I mean, we've already missed our bus and so on and so forth.

[00:13:42] But what you're essentially saying is that today it is easier to get closer to the frontier models for much cheaper than it was even, let's say, two years back. So, yeah, I'd like to, and I think you're, you're absolutely right in the, in the directionality of your thinking. Right. The interesting thing is that, you know, there's two things.

[00:14:08] What a foundational model today can do is improving all the time. Right. A foundational model of today is dramatically better than the foundational model of, say, you know, three years ago when, when chat GPT 3.5 came. And it is, it's very interesting that at some point you reach, you can make a model which is good for doing most things that people want to do. Right.

[00:14:37] At a fairly low cost. Okay. Now, if I'm aiming to go to AGI or something like that, maybe that's a different ball game. Right. But the intent, if the intent is to actually do things that allow most people, what most, both individuals and businesses and governments, the kinds of things that they want to do with AI.

[00:14:59] And you're willing, you can actually do like 95 to 99% of those things using a model that is trained, you know, you know, in a fairly, you know, cost efficient way. And that's really the game, right? Yes, there will be a difference. There'll be some more things that, that, that the frontier model will be able to do. Right. But if you take a view that I'll be a fast follower and I'll make sure that we can actually do, then the costs of these things are.

[00:15:28] And in the long run, see, it's your, you're, you, we are, you know, you can always worry about which model is the best or which model is the best at benchmarks or which kind of, you know, chip is the fastest or. I think you have to look at the sovereign, the, in the long run, right?

[00:15:46] If you take a sovereign approach, it's slightly harder, it's harder to do, but you will end up in the ability to be being able to control your own destiny in this technology, which is probably in my lifetime, the most fundamental technology that we have looked at. Yeah, certainly. Yeah. So in many ways, it's, it's like a ravaged PC, right? I mean, perhaps, I mean, I can imagine maybe 40, 50 years back, everyone needed a certain level, right?

[00:16:13] And that level was not very much different from, let's say, what some scientific organization or defense organization used. But today, most of us are fine with, you know, the standard version of a Mac, right? And the very, very high level of computer is restricted to a very narrow band of use cases, let's say in research or, you know, some sort of fundamentally, like, you know, fundamentally innovative things, right? Yeah, no, absolutely.

[00:16:43] And actually, the more, the interesting thing is that if you want to go deep into a particular domain, then actually even having a small model for that particular domain, you can actually do better than even the largest frontier models because I'm focused. I've made my domain narrow and I can do that at a cost.

[00:17:01] So the ability to train models, right, which are in my way, frontier minus one, are something which is a core capability that a nation should have because, you know, there are so many things and so many impacts that can have by, you know, having these kinds of models. You know, we jumped right into the thick of things with sovereignty and everything, but can I just take a step back and if I would have you explain to folks, you know, how AI works, right?

[00:17:30] I mean, I think broadly people understand that you have some data, you run an algorithm on that and the algorithm is in such a way that it gets smarter over time, right? I mean, it gets better at it and at the same time it generates new data and this combination of data and algorithm is what results in what we see fundamentally, right?

[00:17:51] I mean, whether you are asking about, you know, some, this MRI report you uploaded or just saying, you know, what's the best restaurant to go to in, you know, in this place that you're visiting to and so on, right? It gets smarter over time purely because of the new data and the ability of its, its own ability to improve the algorithm. Let me see if I can explain this. You've brought up quite a few, you know, kind of.

[00:18:17] Sorry, I may have confused people more than clarified technical points, but I think, see, so I think first you need to understand that what an AI model is, right? And what a large language model is, right?

[00:18:36] At its very core, a large language model is basically something that, and I will use words, though strictly the word is tokens, right? Where given a kind of a input of a series of words, the job of the model is to predict what the next word would be, right?

[00:19:03] And that's fundamentally, that's what the model does. There's no, interestingly, you know, there is no, the algorithm inside a large language model is actually something very, very simple, okay? So what ends up happening is you create this model and these models, we have sizes, right? Then for, let's say that somebody says, oh, there is this model with a hundred billion parameters.

[00:19:29] Then I actually have hundred billion numbers, which I then use, I pass the input through this set of these hundred billion parameters. And then it says, this is the most likely output, okay? And that's the job of the model is to predict that output given this input.

[00:19:49] And so when I'm training the model, I'm actually saying, how do I make the values of those hundred billion parameters such that I can best predict the next output in a general sense? That's what I'm doing. Right. And then when I'm actually using the model, what I'm doing is I'm giving it a bunch of input. And this input could be context. It could be, you know, what you're trying to answer. It could be the last output that came from there.

[00:20:19] All of them are used to predict what the next word is. And that's really fundamentally what these models are. They are these next token prediction kind of engines. So it's kind of amazing to say that, you know, given some context and given a bunch of parameters, you are able to actually predict in a, predict actually in a probabilistic way what the next word should be. Right. So, and that's, that's really what it is.

[00:20:48] And how you, how given those input words and given those parameters, you calculate the next word. The mechanics of that is very simple. It's a large number of additions and multiplications and, and, and some kinds of, some, some kinds of, you know, kind of some other functions like, you know, but, but essentially it's something very, very simple that actually, that actually the mechanics of what the output is. Right. Yeah. So it's not an algorithm.

[00:21:20] Okay. Okay. Yeah. And when you say weights and everything, I mean, like I, I would assume that, you know, context is one thing, right? And you wait that a certain amount. And then let's say the previous output that it gave us one, another thing and you, and you waited a certain amount. And then my own inputs over time is another thing you waited and so on and so forth. You have different parameters.

[00:21:42] So the, so the, so the interesting thing is the, when we say that the model learns, I mean, now there are maybe, you know, there are certain, the model actually, you know, you can, so the interesting thing, the model by itself doesn't learn. Right. So what is happening is, so there are a bunch of inputs, right? One is the, you know, the context, right? So you have given like, here it is, here I've sent 3000 words of context of, you know, what is it that we are talking about?

[00:22:12] Maybe then you give, you're asking a question that is actually your, your, your immediate input. Right. And then you know the history of what output you have given and then what the, even the previous word that you have predicted. All of those things comes as inputs to the model. Normally when we use the word weights, the weights are actually just the weights of the parameters of the model itself, as opposed to the input. Right.

[00:22:38] That's not, and, and that is really, and therefore that entire thing goes into the network and then you are ending up predicting, you know, the next token, right? That you want. Right. And, but this prediction is, is actually not, what's the right way? You're not learning. Okay. When you're training the model, you give it many, many examples and say, this is a good, good answer versus that is not a good answer.

[00:23:07] And then you actually modify the weights. That's training a model. But when you're using the model for inferencing, you're not actually learning anything more from it. You're just taking the input and generating the output from it. Okay. That was useful context to everyone. You know, there is a thought that, you know, why do we need to build everything from scratch? We have the likes of, let's say, Google and so on. Why don't we Indianize these folks, right?

[00:23:35] So why don't we have data production laws like the DPDP, for instance? Why don't we get them to pay their fair share of taxes locally? Why don't we also, let's say, I mean, if I was thinking along those lines, stipulate that the likes of Google should have an Indian at its helm in India, not just globally. Right. And so on. Right. And can we just Indianize these MNCs?

[00:24:01] So I think that, I mean, this is a complex question. I don't think there's a simple answer to this. But I think the fact is that, you know, all of these companies consider India to be an important market, right? Right. Outside of the US, right? Which, you know, India is the biggest user of many of these models. And of course, the reason is because they're not allowed in China.

[00:24:29] That's a different, that is the reason why that is the case. But I think, you know, I think that certain restrictions about how people should, you know, even people have their models, the restrictions that there needs to be, there should be some of these things. The first thing is that the models, you know, whatever models are there, they're hosted inside the country.

[00:24:55] That's a basic thing that, you know, when you're using a model, you have to host the model inside the country. The second thing is that when you are using an application, the application is what is storing the memory. And those kinds of things need to also be in, you know, in the country or at least in the control of the individual as well as in the country. These are very core things that people need to do. Right.

[00:25:23] So and then, of course, you know, I think that those are some reasonable things that, you know, we could have. The one interesting thing, and I don't know the answer to that kind of question is that, you know, many of these companies, you know, kind of provide these services for free. Now, this is a good thing for the people, right? So therefore, that's a, but the question is that, you know, what is really free?

[00:25:52] That's something that we need to think about. And what are we paying for, for this free stuff? So therefore, some of those things are really questions, right? That if you want to build kind of whatever Indian companies and other companies in any other industry, right? If you actually spent hundreds of billions of dollars investing in something and gave for free, it would be called dumping, right? But in this industry, it's different, right?

[00:26:21] So that's something to think about, right? What are the answers to these kinds of questions? And I think that if we want to have sovereignty, one is, of course, training the models. But more than training the models is using the models. Because only when you use the models and you get access to, I mean, today, right? What are we doing, right? When we use India as the, after the US, the largest user of ChatGPT. And we are the last largest free user of ChatGPT.

[00:26:51] So what are we doing? We're exporting our data and importing intelligence. Right? A century ago, you know, when we exported cotton and imported cloth, that's exactly what's happening here. Right? So that's something to think about. And we do that because we like the answers. Right? We see value in what is happening.

[00:27:12] But I think the most, I think things like data, you know, the models running in the country, some rules about how memory is stored and what kinds of companies. I think some of those things. And then the understanding of not using these things in sensitive and regulated industries. So some of these things, I mean, these are policy questions. And definitely, you know, beyond my pay grid. But the point is that these are important things that the country must think about.

[00:27:42] So typically, when we talk about, let's say, tech sovereignty, people bring up the China example. Right? Right? And say that, look, they don't need Google. They have their own, let's say, Baidu. They don't need Uber. I mean, they have their own Didi or whatever. Right? And similarly, you know, they are not using your ChatGPT or Anthropik or whatever. Yes. They have their own, right? I mean, DeepSeek and maybe like hundreds of others.

[00:28:10] So what has China done correctly that we can learn from them? See, I think China has done some things, whether it is correct or not, that, you know, see, China has for almost two decades, maybe two to three decades, has taken the view that, you know, that many of these platforms are going to be, you know, companies that are Chinese companies.

[00:28:35] That's fundamentally, and they have not allowed the kind of Western platforms to operate in those countries. And as a result of that, they've built strength, right, in some of these technologies. And that's really what has, you know, what has happened. And India typically has not been that way. We have welcomed and we have, you know, kind of leveraged these technologies.

[00:29:03] But as a result, we have probably stunted a little bit the growth of these more fundamental technologies from within our country. Right? And is that a question that, you know, you know, so that's something that, and, you know, when I think,

[00:29:24] I believe that when ChatGPT first came out three years ago, I think the Chinese, even though they were actually extensive users of AI, you know, in other domains, they were actually surprised by the capability of some of these large language models. But what did they do? They really said that, on top of that, they also had export restrictions and things like that.

[00:29:48] And through a combination of many, many things, they have, you know, in two years, they have come to a place where, you know, there's, you started with DeepSeek, but now there are almost, you know, 10 Chinese companies which have amazing models.

[00:30:04] And I think that, and that's because they have had both the, you know, the investment and the technical strength at a very different level to actually make some of those things, make some of those things possible. And more important than even those things, they also created that basically there is a market, right? Basically, if the other models are not allowed, right?

[00:30:32] If, if, if, then, then, then it becomes very easy that you say, okay, here are all of these Chinese companies competing for the Chinese market, which is a huge market. Right. Now, that's a, that's something that in India, typically we won't do. Yeah. Yeah. But, but, but, but it is to be, it is, it is though it is, it is something for us to introspect and see. Absolutely. Really what these things mean. And at least there must be some ways that you have to restrict and you have to say that we need, you know,

[00:31:02] Indian companies building fundamental technologies to be used at large scale. Right. Yeah. I can quite imagine as you're even saying this, I mean, a few folks are already saying, you know, quote unquote protectionism. Yeah. Right. Because there is this vein of free market maximalism that people exhibit. Right. But one thing that very important, what you mentioned is that China may lag at certain points, right? In these innovations. Right.

[00:31:29] So I'm sure the first version of Baidu was not as good as Google. The first version of BYD was not as good as Tesla, but they eventually catch up. And in some specific fields, they actually accelerate beyond further. So no, no, it's very important.

[00:31:44] I think it's very important that in all fundamental technologies, all deep technologies, we must have efforts and we must have Indian companies that are actually kind of nurtured and protected to actually build world class solutions.

[00:32:02] In the end, see, in the end, protectionism or actually tolerating technologies that are substandard and not working well enough is not a good thing, I think, in the long run. But to actually allow competition, right? And if you have multiple players, right? And you actually make, you know, and those players are actually competing for the market, then things will get better. Right.

[00:32:30] If there are monopolies, then, yeah, then kinds of that, of course, causes problems. But, yeah, I think that is one of the challenges that we have, that we need to actually have these, you know, I would say, you know, five, 10 companies, right? Which are doing these kind of important things. And then they actually have certain, you know, market space to play and improve, right? Okay. For the next couple of questions, I want you to wear the policy hat, right?

[00:33:00] And think large scale impact, right? Think government, policymaker, what are some two or three things that you would do? Now, these could be, let's say, pick 100 or 200 of the smartest, brightest people in India and create incentives in such a way that they focus on, let's say, deep research. It could be that I set aside a few tens of thousands of crores every year to spend on this kind of stuff, patient capital that can see these innovations through.

[00:33:28] What else can you think of, you know, from a large scale impact perspective? No, no, I think, see, I think one of the things, especially in AI being as horizontal a technology as it is, I think there are, I think all of these things are important. But I think that we have to, you know, when I, when I, when I, you know, look at things, I think for the first time, I think we have the potential, right?

[00:33:55] Of AI allowing every child to have access to, access to good education, right? Every person having access to at least good health, health care or health information and things like that. These are things or, you know, I think that the ability of every Indian to access services in a simple way.

[00:34:22] I think these are all fundamental things that can actually impact the governance of our country. And if, if the government takes a view that these are things that we want to be able to achieve, right? Of course, basic fundamental things like, you know, kind of self-reliance and defense and strategic sides also include. You look at some of these core areas and you say, how can, you know, AI actually transform these things?

[00:34:50] And then you look at it, how do we actually kind of create a group of, you know, you know, some Indian companies which can actually make some of these things happen? So therefore, in some sense, how do you, one way to strengthen it is that if in some ways the government becomes a buyer of first resort of these technologies.

[00:35:13] And, and it, I mean, and I said that there should be a, there should be a, you know, honestly, there should be a, there should be a few companies, not specifically any one. But, but I think if you did something like that, then there would actually be a way by which these technologies could develop and could actually have a real impact of in the governance of the country.

[00:35:35] Because I think the ability of this technology to change everything is, is actually, you know, change things in a very significant way is there. And so therefore, if you actually say that that is where we need to, it's the usage of these models which are there. And, and which is, you know, and, and obviously China has taken the maximalist position that only Chinese models be used, whether it is, whether it is, you know,

[00:36:04] whether it is a, you know, government application or whether it's a private, you know, kind of, you know, application. And now we don't know whether we need to go that far. But, but I think that some kind of a view that certain things, you know, actually would, you'd look at Indian players to actually do something like that. Right. Yeah. I mean, the other thing is, AI has become so geopolitical right now, right?

[00:36:30] I mean, I, you know, you look at any of these panel discussions, think tanks and so on. Everyone seems to publish a paper on AI of some sort, right? I don't, I don't remember any of these folks talking about SaaS or multi-tenant or cloud or whatever, right? So this is fundamental and, and they do recognize that this can impact humanity very tangibly and change things, right? So how can India negotiate better for its companies?

[00:37:00] No, no. See, I don't know about negotiating better for the companies, but if you look at it, I think, yes, it is a geopolitical game. And the closest geopolitical game I can think about it is, is actually the nuclear NPT, right? Right. So there are nuclear nations and there are not nuclear nations and there is a two, you know, and, and their rights are different. And I think AI is going to be in that category. Wow. Okay. And there are going to be two, it's very clear.

[00:37:28] There are two AI powers which are there in the world. There is, you know, de facto powers, which is the United States and China. Will any other country join them or not is the question. And if you don't join them, you will have lower rights. This is, this is a, I think if you look at the way geopolitics is going, there is no doubt that it is going to head in that direction. It is literally that important.

[00:37:51] So, so I think that, you know, I think that that's, that's the question that, and the way you actually do that is, is, is, is you have strong enough companies and good enough technology. Then you are a player with a seat at the table. I think, you know, it's not a question. Yes, we have certain built-in advantages that we have a large market. Everybody wants to come to our market. Maybe that's one leverage we have.

[00:38:21] But beyond that, really, you know, I think the, the goals is that, you know, how, how good is, how good are you? And, and that's going to determine if you have a seat at the table. Yeah. No, it really puts things in perspective, I would say. Okay. Um, also given, you know, how complex the supply chain itself is, right? Um, from chips all the way to let's say applications and so on. Uh, there are things I feel that are out of like an individual company's, uh, control, right?

[00:38:50] I mean, uh, the access to GPUs for instance, or, uh, uh, let's say having our own, uh, chip manufacturing companies, uh, within, uh, India and so on. So, so these are things that I feel, uh, you know, the country can negotiate at a, at, at a government level, right?

[00:39:05] So wherein, uh, whether it is like negotiating with, you know, TSMC or whether it is, uh, uh, incentivizing through PLIs and whatnot, uh, some of our larger conglomerates to invest in R&D and manufacturing, uh, more, uh, you know, amenable to AI. Uh, are there things that, you know, you think the government should do and is doing enough? No, no, see, I think firstly, I think we ourselves were the beneficiaries of this India AI mission, right?

[00:39:32] So I think that government is certainly realizing that AI is different. And I think that, uh, uh, we, uh, got basically the ability access to GPUs to actually train this model without which we would not have been able to do that. Right. And I think that's a great start, but I think that, uh, we need to do, we need to figure out more ways by which, uh, uh, some of these things are done. And you're right.

[00:39:57] It is beyond any individual company or even individual conglomerate for that matter. Uh, it is actually has to be a whole of nation effort to figure out the strategy of how we are going to do this and how we can become a player. Now there are certain things such as, you know, chips and semiconductors where it'll be a longer road, right? It's not going to be, you can't do this in one or two years. It's a five to 10 year game. But if you play consistently, it has been shown.

[00:40:24] And if you actually have that policy guidance and, and, and with a very deliberate view to becoming a power, there is nothing that, that you can't say that in 10 years we can't do that. Right. And I think that, uh, you know, I, and that's, that's, but that's that, uh, in, firstly, there needs to be an intent that we must, we want to do it. And then the incentives need to be aligned and we have to, and the, you know, and the incentives need to be aligned.

[00:40:52] And if that happens, uh, and I see that some early green shoots of that, we are going in the right direction there, but, uh, you know, it's still, we are at the very beginning and there's a long road ahead. Yeah. Yeah. There's usually daylight between intent and action, right? I mean, when it, when it comes to, um, let's say anything governmental, right?

[00:41:12] And intent is of course a very useful starting point, but translating, translating that into actual tangible action, uh, you know, oftentimes, I mean, it's left much to be desired. Right. Uh, but do you, do you see that happening? I mean, are we doing enough? Are we doing things fast enough? Uh, what are like one or two things that we should absolutely prioritize right now? See, I think, uh, you know, we are, this technology is moving faster than any technology that I have seen in the past.

[00:41:42] So are we moving fast enough at every level? You know, the answer is you can always move faster. Right. I don't want to say that, that, so there is, there's no thing that we, you know, I think even if you are moving faster than yesterday, tomorrow you must move faster. Right. That is, that is really the, uh, that's really the way this, uh, this is at. And, uh, you know, um, I think, uh, you know, intent is the, is where you start.

[00:42:11] But, uh, I think it's, uh, you know, we have to continuously move, aim to move faster because the world is moving faster. Right. And, uh, things are, and partly it is this technology itself, which is putting it into kind of this feedback loop that makes things move faster and faster.

[00:42:29] And so therefore we have to, uh, uh, kind of, uh, you know, build, uh, ways and build structures that enable us to move faster. Right. You mentioned, uh, the India AI mission. Could you talk a little bit about, you know, the work that, uh, uh, they've done and also in general about the ecosystem right now?

[00:42:53] See, I think, uh, the India AI mission, I think, uh, you know, was the goal was to how do we actually get, you know, move India forward in this AI technology. Uh, I think they did, they've done a number of different things that they've been focusing on. Uh, they've been focusing on, they've, you know, supported us and 11 other model companies, right. And to actually train models, right.

[00:43:19] Some of them are, uh, models, which are, you know, general models like we have. There are other people who are training models that are in specific domains. And so therefore those are the kinds of things that, that, uh, and, and then they have created these, you know, trying to collect some of these data sets. Uh, they are doing, how do you actually kind of train people in AI? So they're doing a many, many different things. Uh, and, and so the intent, uh, is right. Right.

[00:43:48] And, and I think that they have probably reacted faster to this technology than they have, uh, in the prior compared to, you know, any other technology that was there. But having said that, uh, we need to do more. And it's, by the way, it's not just the interesting thing that I see is it's not just government, right. It has to be industry also. Right.

[00:44:11] And, and, and I think that sometimes, uh, you know, we have lacked ambition to actually, uh, go and come be in this, uh, you know, to, to play in these games.

[00:44:22] Uh, I mean, we were looking for more kind of guaranteed returns, uh, kind of a view, uh, rather than, than actually kind of, uh, you know, investing and seeing how some things, you know, which can take three to five years to actually effectify, to build those kinds of things. Yeah. That, uh, that, uh, because if you're looking at it from a one quarter perspective, you're not going to be able to compete in these technologies. Yeah.

[00:44:52] And, and there are obviously, and, and I, and I hope that some of those things will change. There are also obviously companies with, you know, tremendous balance sheets, right. Which can afford to do those kinds of things. And I think, uh, and, uh, you know, uh, and they just need to have the confidence that they can win in the world if they make that investment. And that's, that's really that, that confidence is what we need to have, right.

[00:45:16] That, you know, uh, and, and, and, you know, uh, they must believe that, that, you know, that the risk that they take here is, is worth it. Right. And that, that's, that's, that's something that I think we are still kind of struggling with in, in, in, in industry. Right. I mean, in some sense. Yeah. Yeah. I think, is that a, like a key mindset shift that needs to happen?

[00:45:39] Because, you know, we often hear about, uh, the private sector, private sector contribution to R&D being almost less than half of, you know, what the average, uh, standard is across the world and so on. And that, you know, a comfortable, uh, you know, operating in steel, cement, um, et cetera, et cetera. But we don't want to do anything that's really out there. We don't want to do anything really innovative. Now at one level that is perfectly understandable given that, you know, where we are, right.

[00:46:07] I mean, truth is that, you know, we're still two and a half thousand dollars per capita at this point of time. There are more basic needs and there are easier ways to create wealth and value. But do you think that, that inflection point is right now where that mindset is shifting? No, it has to shift.

[00:46:22] I think because I think, uh, you know, uh, I think the reason it has to shift is there as some of these things, the easier ways of, uh, you know, very successful business models and to, to what people do, to make money. Some of these things are changing.

[00:46:39] And so hopefully as these things change, people will realize that they need to do new things to actually, uh, you know, uh, and, and have take a little bit of a longer term view, uh, to build fundamental technologies and believe that we can build fundamental technologies. Right. It's, it's right now, the view has always been, and maybe it will continue that always been is let someone else develop the technology. Yeah.

[00:47:07] I will use the technology and I will basically, you know, uh, uh, use that technology and make money off of that technology. And I'll play, you know, I'll, you know, kind of, uh, focus on profitability from day one rather than, you know, and, and, and, and, and actually, uh, instead of, uh, you know, uh, and because that has been always possible. Now, now the question is that, you know, uh, you know, should that change in the, in this area, right?

[00:47:37] That's the question. Do we, if we don't play the fundamental games, we'll be left so far behind that we can never catch up. Right. So therefore we have to do this is my view. And I think it's everybody. And, and I think everybody, whether it is, uh, government or whether it is, you know, uh, uh, uh, the private sector, everybody needs to be, needs to, needs to, needs to play on this. Yeah. I think when the old moats collapse, as it will with AI, right?

[00:48:04] Um, it will become an imperative for these folks to act now. Yes. Uh, than to, you know, wait it out. Um, a shiny example of how the private sector technology and the government came together was the India Stack project. Right. And you've been closely associated with UID AI as well. Um, in fact, you're an advisor still, if I'm not wrong. Uh, I'm not an advisor, but I've been on their vision committee. But I, until fairly recently, I was an advisor. Yes.

[00:48:32] What can we learn from that whole experience that we can apply to AI? See, I think the core thing, which I think is one of the core things, two things. One is that we built India Stack as digital public infrastructure. So the focus of it was to actually help people, not necessarily, you know, kind of, and that, that is really something which was different from both the American model and the Chinese model. Right.

[00:49:00] We are saying that we're building certain things as digital public infrastructure. And, uh, that's really one of the big successes, right? Whether UPI is certainly for the first time, we feel that something is better in our country than, than is there in other parts of the world. Right. But having said that many things in our country are extremely hard, right? Still. And we need to get to a place where we actually make everything easy for all citizens. Right.

[00:49:26] And I think that's something that AI actually combined with this digital public infrastructure can actually do something that is, that, that's the other thing that I want to say is this, that this digital public infrastructure, we designed it to be sovereign. Right. Right.

[00:49:42] We designed it to be sovereign because we allowed, we said that it is so critical that at the wrong point, you know, these things can't be, and we didn't worry about whether, uh, you know, uh, you know, at the, in the short run, uh, whether some, uh, what's the right, uh, in short run, whether what the latest technology was or whatever it is there. We said sovereignty is a core principle that we need to solve for here.

[00:50:11] And I think that's the other part. Otherwise this India stack would not be what it is. Right. So, and I think that that thing at, at a much, I would say more fundamental and deeper level, we need to think of in the, in the case of AI, that, uh, we need to actually combine and make AI such that it makes the lives of all people better. Right. And, and I think that, that, that, that's something to be learned, uh, uh, learned. That's, that's something to be learned. Right.

[00:50:39] So therefore, I mean, you could, I mean, we could, we could have asked Google to make an identity system. Right. Why didn't we do that? I'm sure they can. Right. So, so, so, so I think that those are the kinds of questions that we need to answer. Yeah. No, I think the scale brings with its, with itself, the, the complexity of things. Right.

[00:51:04] And, and also given how diverse India is, I think there is scope for us to solve this, uh, ourselves. Right. And rather than, you know, adapt things from outside. Um, how do you envision AI in the India stack? I mean, do you think that it'll make existing services better? Or do you think that you see it as a stack in itself? No, I think, see, I think it's going to be a, uh, uh, uh, you know, it is certainly going to make existing things better. You should always think like that, right?

[00:51:31] What systems you have, how do you make those systems better? But in some times when that the, the impact is so large, you have to reimagine how these things are, are working. So it, it, I think it'll be, it'll be a bit of both. Right. So therefore there are many kinds of things you could never have imagined before, uh, which is, which, you know, when you have the, the access to a new technology, right? Before mobiles, how you would think about something and after mobiles, how you thought about things was very different. Right.

[00:52:00] So, and I think those kinds of things we have to, um, so I think it will be part of existing services as, as well as new things. Um, and I think one important thing that you kind of alluded to was the fact that, you know, India has certain unique advantages. Firstly, it is a very large market and the whole world wants it. It is a very cost sensitive market. AI by definition is very expensive and we need companies which are focused on bringing

[00:52:28] the cost of AI, of the valuable AI down. And I think that is one of the things that, that an Indian company can do is that. And the third thing is India is diversity, diversity of life. Language, diversity of culture, diversity of these things. And that actually makes, uh, an AI for India be different, uh, from the AI that, you know, somewhat different from, from what it is in the rest of the world. So I think that's something, these are some unique advantages that we have, uh, which can

[00:52:57] allow us to actually not just build AI. Let's build AI for, for all Indians and then see for the rest, you know, five billion people in the world who are, uh, you know, uh, not in the first world. Yeah. Um, I want to talk a little bit about your personal journey as well, right? I mean, you had a storied career in the U S built and sold companies, uh, came back, was associated with the UIDA mission and built something that's so meaningful for all of us to use and benefit from.

[00:53:25] Um, what was that moment where you decided, okay, now's the right time for us to do this? No, it's, it's, you know, some of these things happen, right? So therefore, uh, you know, I, I am, I am accident. Yeah. I, I think that, uh, when I came back to India, I was kind of, you know, looking for something interesting to do, but I never thought that I would be a full-time volunteer in government for almost, uh, 15, uh, 15 years.

[00:53:54] Uh, but, and, and I thought once I had done that, I said, you know, the rest of my life, I'm going to be either doing open source stuff or, uh, you know, helping government to do various things. That was at least the, the, my default thinking of what I wanted to do. But I think something when, when the three years ago, when the original, uh, you know, Chad GPT moment happened, I said, this is something very important and we need to figure out how to do this in India.

[00:54:21] And my initial, uh, kind of, uh, you know, instinct was, uh, let's go, uh, ask the government, let's go and ask philanthropy, let's go and do these, these are kinds of ways to do something like this. But then I realized this is moving way too fast. And, and I think that this is too important and either I can give advice to everybody, to everybody, how to do this, or I can say, okay, let's try to do something. Right.

[00:54:47] And I think, uh, and, and then it became, you know, I mean, I'd, obviously I'd been an entrepreneur in the past in the U S not, not, not, not actually in India ever, but, uh, we said, okay, then we need to kind of, uh, uh, figure out how to do this. Right. And that's what we ended up, uh, uh, you know, we said, we're going to start server and we are going to see that, uh, you know, we think sovereign AI is something important. And I think it's interesting how it's played itself out in the last two years.

[00:55:16] And both the impact of the technology is, I always thought it was big and it is actually more than even I imagined. And I think the geopolitical games related to them have also been, uh, you know, uh, have been, uh, now it's much more clear what could happen, uh, in, in those kinds of things. So therefore that's, it's interesting that both those theses, uh, have played out, but I think we are still at the very beginning of this journey.

[00:55:43] We want to get to a place where, you know, uh, uh, uh, where this technology is actually being used in everything, you know, and making people's lives, ordinary people's lives, uh, better. And I think that's something that, you know, at least the opportunity to have been involved with Aadhaar to actually see such a cross section of people in the country and, and see what their

[00:56:08] lives really are, uh, I think, you know, we look at this technology as a way to maybe, you know, kind of make their lives better. Yeah. Yeah. Um, you know, for those few who haven't heard of Sarvam, uh, how would you describe Sarvam as it exists today and where it's moving towards, like what the vision is? So we conceived of Sarvam as a full stack, uh, you know, uh, AI company, which is working

[00:56:35] both at the foundational level and moving all the way to the applications where people are experiencing AI. And we believe that, you know, I think we need to build AI that is, you know, for India. And it is also something that we believe voice, for example, is something extremely key. And we believe that how Indians communicate, we love to talk, maybe too much sometimes. The argumentative Indian. Yeah.

[00:57:03] We love, and, and so, and everything, how we interact and how we do that. So I think that, uh, AI that is very voice first is something that is, uh, extremely important. We've built certain technologies. We've shown, if you look at the models that we have launched in the last month, uh, we've launched models that, you know, uh, and I don't know how many people understand is compared to the deep seek model launched a year ago, it is actually better.

[00:57:29] So therefore you are, you are actually behind only in terms of months. So therefore I think that's what we show that it can be built. Yeah. Which is a huge deal. And India can do it. Considering the fraction of the cost and time. Yeah. Yeah. And, and I think, I think, I think the point is this, it's the point is, uh, you know, we've got to be able to have that confidence that we can do it. And also the imperative that it needs to be done.

[00:57:55] And I think, I'm hoping that what we have shown thus far is that, you know, this is not impossible. We can do it. But having said that, we are at the very beginning at the pace of the mountain and we have a long road ahead. Right. Yeah. What are some of the hardest decisions you've had to make over the last like two, three years? See, I think, uh, you know, I, I think that, uh, it's, it's actually been, uh, you know,

[00:58:22] people work hard, but I, I don't think that it's been super hard, frankly, uh, frankly so far, but I think that the road ahead is hard. So, because we have to get to a place where, you know, in five years, a significant portion of the AI consumed in the country is from Sarbam, right? That's, that's the intent and that's the, and we are not even, we are just barely scratching the surface.

[00:58:47] So the question is, how do we build a company that is ready to take that thing on? Right. We've done a few things, right? What, what we are calling as, uh, Sarvam 1.0 is to actually prove that you can do something like this and, and, and, and, and, you know, and, and, but then to really take it to meaningful things, we are at the very beginning of this journey. So it's kind of interesting that, you know, uh, that, you know, we're, we're barely a startup

[00:59:17] at the very beginning of our journey. Right. And I think that, uh, but it's an important, it's a responsibility because it's a responsibility to show that we can build a deep tech company out of India that has meaningful outcomes, uh, for the country. Right. That's something that, and that's our responsibility. And, and that's really where, uh, we are, you know, we finished the first innings, you mark your guard and then you start again. Right. Yeah. That's a nice analogy.

[00:59:46] Um, what are some popular use cases that are emerging? And do you see Indians use AI, uh, uniquely compared to let's say the global, global, uh, See, I think, uh, we are obviously seeing that voice calling and, and using voice. Uh, and these are typically businesses. They are actually, people are talking, uh, to AIs to do various kinds of things. And, uh, and I think this is something that Sarvam is doing a million calls a day, right? Yeah. Calls a day today.

[01:00:17] And, uh, and, and we think this is a very, uh, it's a very beginning as far as these things are concerned. We think that, you know, um, in the last few months, I don't know whether you've been hearing about things like, uh, you know, open claw and stuff like that is really exciting things happening. Right. I mean, and I think that the ways things can be used to do interesting things is, uh, there's whole new things that are beginning to happen.

[01:00:45] And as, as a, both as, as, as a company and as well as a country, we'll be looking at how to actually figure out these new ways by which people can leverage AI to do interesting things. And I think, uh, it's, uh, as I said, we're a full stack company. We want to get to where people can actually do useful things, uh, with, with the AI. Right. Right. Yeah. I mean, um, the Indian consumer is extremely cost conscious, but also value conscious as well.

[01:01:14] They want the best things for the cheapest. Yes. Uh, so it's not an easy task to please this customer. Yes. Uh, but we do have, I mean, we do make a habit of things. Right. And for many folks, I assume that Sarvam will be the first interaction with AI. Right. Uh, so, so that could be like a moat in itself once it develops. So, yeah, we have to see right now. I mean, I think that, uh, you know, I think that, uh, while we are not, uh, while, while

[01:01:42] we do have an app, by the way, which, which people can, uh, we've been using the Indus app. Yeah. But the intent is the intent of that is not, uh, you know, the, you know, uh, our focus is actually to build this AI into existing systems so that whatever people do, they can do more effectively while we have the app. That's not the primary, uh, kind of surface that we are thinking about as, as, as, as, okay.

[01:02:09] So, and it, that's AI needs to quietly kind of permeate, uh, things that people do. Right. Okay. Uh, moving on to the final section, um, how do you see AI impact humanity and more specifically India? When we think about India, we think of a large demographic, uh, large numbers of people, uh, who need to upskill themselves, uh, need to get jobs and so on and so forth. And, uh, there are all these doomsday scenarios, right?

[01:02:39] That AI will destroy jobs and that a lot of people will be found wanting and so on. Um, but how do you see AI impact humanity and more closely, how do you see AI impact Indians? No. So AI will dramatically, uh, change how things people do in the world. I think there's no two ways about it. Right. So the one thing that you need to understand is the one guarantee that I can give you is that there will be change. Right.

[01:03:07] And the, I think the, how you be stronger in that thing is how is your adaptability to change is really the question that, that, uh, that, that you have to look at it. Everybody will be doing their work differently. Right. In a few years. And that having that, uh, mental, uh, what do you call that flexibility to say that just because I have done it this way in the past doesn't mean that I have to do it the same way. So that ability to learn, right.

[01:03:37] That plasticity of, you know, of doing, doing different things is one of the core things to be able to react to these kinds of things. Every job is going to change. I think that's, I don't think that I'm, uh, every job is going to change. Has changed. But I'm saying we're just, you ain't seen anything yet. So, so therefore I'm saying that, so I think that, that ability and that willingness to,

[01:04:05] to, to actually use this, embrace this technology, use it to actually make yourself more efficient, uh, to be able to be able to do things that you never could have thought or dreamt of doing. And we are in an exciting place. We are literally in a place that if you imagine that I want to do something, you can actually do it. We have never been in that place in the world before. So that's a very exciting, exciting place to be. Right.

[01:04:32] So, I mean, I mean, I'm, I'm, I'm going to see that, you know, at one level you say, Hey, what's going to happen to software, for example, maybe the other way of looking it is that every Kirana shops owner is going to be a, is going to be a developer. So, so, so I, so I think the, the, and so of course what it means to be a developer will change. Right. Right. So, so, and, and, and I think that is really the, uh, uh, and the, the, there are certain

[01:05:00] things which I think are, uh, you know, constant, right? Right. If you understand what is it, the problem you are trying to solve and who is benefiting from that problem and, and having a deep understanding of that, then all these things are tools. Right. And make, elevate yourself to think like that. What is the value that, that, that you are providing if you're actually interacting

[01:05:27] with someone and how to make that value greater or more efficient or more effective. That is the game that you need to think about what, what, how to do things. Right. So, and, and I think that that is, that's, uh, uh, really the challenge that I would give a young person today, right? That, uh, please learn continuously because now that is, you know, and it's not like you

[01:05:57] went to college and then something, everything is done. You learned engineering and you were an engineer for about 30 years. It doesn't, it's not going to work like that. You have to learn continuously. Right. Uh, you have to be continuously curious. You have to continuously figure out who your stakeholders are and how you provide them value. Right. And, and, and, and, and that's, and if you take those principles, right, you will find these tools so exciting and so empowering to do more than what you could do today. Right.

[01:06:27] Um, if someone wants to contribute to your mission, um, how should they, uh, go about it? How do we build, how do we build in India? Right. I mean, let's say, let's assume you're addressing a 21, 22 year old, uh, right. Who's got some degree, let's say an engineering degree and wants to make a difference. What is the best way for them to get involved? See, I, I think there are a few things.

[01:06:51] I mean, and I, I, and I would say this, uh, you know, uh, is that become a user of AI become, and it doesn't necessarily mean that what, what we are doing in Sarvam or wherever. I think that is, that certainly please do. But I'm saying that I think figure out in your lives, how do you make things better? Right. And I think better for you, better for the people around you. Uh, and I think that if you take that mindset, right. Then that's something that, and then you not, don't be satisfied.

[01:07:20] If something is the way it was, it's like this only, right. You, you shouldn't think like that. Yeah. You say, how do I make it better? And if you look at it and you will find a million ways that, that AI can actually help it make better. So I think that at a larger scale, I would say that people should think about those kinds of things and whatever tools, right. Enable you to make that better. Those are the things that, that you should use. Those are the things that you should use.

[01:07:48] I mean, I think that, um, and, and I, and I think that, uh, the interesting things are, we obviously have such diversity and so many things in our country that, that people do and you should think about whatever you do and how can you do it better. Right. And I think that's, that's, that's, there's no dearth of problems. Yes. There's no dearth of problems. And, and I think that we can move to a place where, you know, things are, uh, you know, easier for people.

[01:08:17] People can do things better. And I think that's that problem solving mentality. And you have now so many more tools to solve the problems. Right. All right. On that very positive, optimistic note, we'll end this podcast. Um, thank you so much, uh, Vivek for giving us your precious time on a Sunday morning. Uh, right. And this was really inspiring. I think people, um, tend to hear about these things and think about the possibilities on AI

[01:08:46] and whatnot, but can't quite figure out, you know, how to sort of conceive of it. Like what do you make of it? And so on. This was a helpful concept, uh, conversation to sort of situate them in terms of what is happening and how they could, uh, sort of contribute to this. So thank you so much again. Um, really, really happy to have this conversation. Thank you, Roshan. Glad to be in this conversation. And, uh, and I think that, uh, we are, we live in exciting times and, and look forward to all the new things that keep happening.

[01:09:16] Awesome. Thanks.