0:01I'm Chad. Welcome to Sana. Welcome to Kasan. Pretty amazing country.
0:05It's It's been fantastic. I've been so mind-blowing so far.
0:08This is your first time?
0:10I didn't know I was going to be here until like few weeks ago. So, it just like came out of the blue.
0:16Did you see the Jensen? What?
0:18I saw the Jensen visit. Uh but that wasn't the reason. I was actually invited here by uh by the king of Jordan. Um because um Jordan's a is a you know uh close uh friend and ally of Kazakhstan and uh there was like a central Asian uh meeting of the different heads of state talking about um counterterrorism and they wanted someone to kind of talk about AI and kind of recent advancement AI technology. So I joined that meeting
0:45um and even in that meeting I was I was surprised by how advanced uh sort of uh President Tokyo's ideas around AI and what he wants to do with a country. Uh but also just you know I I I'm like ashamed to say I haven't thought too much about this part of the world.
1:03But is like wow and and I we met customers today. We have customers here in this country doing amazing things with roughlet. So, so it's been really fantastic.
1:13Yeah, I think I think this century is going to be something where uh we have a lot of what's happening in Central Asia is uh this is going to be part of this coming century. Um, and in the sense of, you know, for example, like El Salvador just kind of came out of nowhere and like burst into view, right?
1:32Um, and uh, Blly, you know, was nobody knew about him and then he became very very popular.
1:38Um, and uh, so I I think oops I think uh, similar to that people are going to find out what's going on with Kazakhstan and a lot of people, you know, from around the world are coming here. This is one of the greatest places to build AI data centers. is one of the greatest places to do a lot of these, you know, kinds of things. Um, that's why Jensen is out here. That's why, uh, it's also got some of the best crypto laws in the world, right? So, Olaf, uh, you know, Olaf Carlson Wii, who's the
2:06founder and CEO of Polychain, uh, went and, uh, he came out here, early investor in Uber, Shervin came out here and, uh, and Andrew Sus is coming out here, too.
2:15So, this is going to be, I think, a real place that's coming on the map.
2:18Yeah, that that's fantastic.
2:20uh how do you think that's happening and what is that how does that why how does that fit in your overall kind of Yeah. Well, so I I think what we aren't used to is countries going to zero or going 10x, right? We're we're used to that already with companies and with currencies, but the internet is increasing the volatility of everything.
2:40And you can catch this sky hook and just go all the way up or you can like have your economy crash like you know Paraguay has this crazy bank collapse or what have you and Kazan is catching a sky hook going up like this right you know I think uh those countries that are internet first and they're rethinking internet first are catching the sky hook and Kazakhstan is doing that especially if they're like avoiding conflict and so on and so forth and having peace and trade with the world. That's one of the big things I love about this area.
3:06Yeah. I mean one thing that really was kind of surprising and very interesting is they were like oh we have great relationship with China we have great relationship with America we have great relationship with even Iran we have great relationship with everyone and that's quite refreshing it's almost like they're not getting dragged into all the different uh wars and the exactly and this is you know one of my friends here um the uh secretary of education or minister of education he says custom is warm it's western Asian
3:38Because it has it's a synthesis of all the cultures because it it's right in Central Asia. It borders some people actually now call Russia North Asia. So it borders North Asia, East Asia, South Asia, West Asia. So it's a synthesis of cultures. And so everybody's welcome here, right?
3:53And it's tolerant. It's like a it's like a Asian Switzerland, right? you know uh one thing I I told the presidents uh yesterday and uh you know something Bajia and I believe for for a long time and we've collaborated on all sorts of projects to advance that is the idea that um you know in many ways you can look at the technology landscape you can look at the history of Silicon Valley and say oh you know that's that's like a centralizing force that's you know youalth is centralizing
4:21in this small geographic area that's like Silicon Valley and some of that is true. But in reality, what the internet does and what AI is now doing is massively decentralizing power, introducing opportunity and sort of equalizing the world. Um I've met uh two companies over the past two days that are building amazing things um servicing the local economy um that I you know for example there's like the smart city
4:50company that I that I met one of our customers and they're building cameras uh that reduce deaths by 2x you know and just the fact that they're they're using AI to kind of improve the city improve the country because most most countries are going to be like Oh, they want to do something, they'll like they'll go to America, they'll go to Europe, they'll find some that is like, you know, using very old technology, not even they haven't heard of AI yet. And uh and but here they can actually do it
5:21Here they're building it, they're vioding it from scratch. Yes. You know, that's right. It's actually, you know, the the president, one of the reasons we're here is, you know, you mentioned President Tak, he's mentioned that he wants Astana to be a tech hub. That's we're literally in Aana hub. That's what it's called. uh and allow the the new city they're building to be a crypto city and Kan itself to become a digital state.
5:42So they are you know the saying like a players hire A players B players hire C.
5:47So Kazan is filled with A players. They were like number one the physics Olympiads this year. And so A players want to bring in also the talent of the world. So it can be you know Jordan to Pippen, you know for the slam dunk, right?
5:59And it's a very refreshing thing because this is just a country that's rising with the internet. So it's pro internet's pro not not in a completely unbalanced way like they you know they want to they want to make sure that there's safeguards but in general they're a very positive country. Mhm. I I mean, you know, he's probably going to be uncomfortable with me saying this, but you're so lucky to have uh Baji here. Speaking of he's a talent magnet, so you're going to see a lot of people come through Kazakhstan. Uh and uh you
6:28know, I'm I'm probably the first, but I think uh I think you're going to be really uh put on the map and Yeah. So, well, very kind of you say I'm Jad by the way. You know, I was I was a seed investor I think 10 years ago, something like that in in in replet just because I thought was really smart and we we both actually shared this goal of the dark talent of the world, right? And and and finding those people who especially whether they have weak passports and strong talent or they're in they're in the Midwest or in the Middle East, they
6:57could be, you know, they don't have to be like super poor. They just have to be locked out of institutions in some sense. And in giving those people an opportunity, that's what replet stood for. That's why I invested in that. I think you've done an amazing job with that. So, Replet, by the way, if you guys don't know, is one of the largest AI companies in the world. Um, and uh, it's something which I I've been pleased to see it grow all the way into this Goliath that it's become. And I think all the growth was like the last two years, something like that. It was like it was like we all had a lot of users, but the last
7:27revenue growth was just two years, right? Because you built you built the basis for that and it was the foundations and then it just went kind of vertical. Maybe you can talk about that.
7:35Yeah. Yeah. So we um I started the company actually not like I started the project not in the United States. Uh I'm originally from Jordan and so I was going to school there and you know I didn't have a laptop and every time I wanted to like do a little bit of coding I would like spend a lot of time installing development environments and I thought that was like that was you know bananas like I should be able to go to a browser and start coding like I do everything in browser. I do Gmail, I do docs. And so I started working on like
8:04the first coding in the browser capability. And so that that's that was like really the basis for Replet. And that was like in 2010. Um and I open sourced it because that that was my mindset is like you do something you just put on GitHub. That's that's how I operated. And it went super viral. Like companies all over the world, companies in the US started using it. Peter Norvik reached out to us and I was like, "Hey, like how did you guys do this?" And they were starting Udacity. They wanted to like bring coding to the browser. And
8:34then I joined one of the companies that were using my software, Code Academy.
8:37And at Code Academy, we taught a lot of people how to code. And I started seeing those stories that Bali and I talk about a lot, which is like, "Oh, I'm a fitness instructor. I learned a little bit of coding. I made an app. Now it's on the app store. Now I'm making $100,000 a month." It's like, "Wow, this is even more powerful than I expected."
8:55And so um you know I did a stint at at Facebook where I was really excited by this idea of internet.org and what's you know Zuckerberg was was talking about this idea of like more and more people coming online and I I thought this combination of internet uh connectivity coding capability will make it so that people entrepreneurs from all over the world will like rise up and will change the material conditions of people around the world. at Facebook. I ended up working on ReactJS uh and React Native
9:24and invented a lot of the modern technologies uh that we use for coding that AI uses for coding. [laughter] Initially we targeted humans but turned out it's for AI. Um and then in 2016 um we started the company replet and the idea was to bring those two ideas together learning code but also developer tools and can we reduce the barrier to entry so much that really anyone with an idea can build a company and and grow and actually around the
9:51same time uh I read biology's um Stanford course what was it startup engineering 13 yes yeah startup engineering and it was like the first course that was like we're going to teach you coding and we're going to teach you business and at the end you start a startup. And this idea was always fascinating to me and now it's playing out with AI is that we need people with more well-rounded skills.
10:15It's not just about like being the most artistic elite coder. You need to understand people, you understand management, you understand bringing a product to market. And so I saw Replet as a way to like a platform for entrepreneurs, for people with ambition and ideas to be able to build their their their work. Um, and it quickly we got a lot of users because it was a really fun to use coding environment of the browser, but no one wanted to pay
10:43for it for a long time because it was just like it was a nice to have. It was an essential tool. But when when AI came on the on the market, I I thought this this was the thing that will allow us to achieve our mission of de democratizing access to programming. And so we started like training our own models. In 2023, we released one of the first open source coding models. Uh so we built like an AI team and all of that. But I I you know, we added a lot of AI features, but I was never satisfied. I was like the the
11:13ideal programming environment is one where you program in English.
11:16Actually, I do want to ask about that. So within Facebook, you said there were a lot of AI features. You mean? Yes. No.
11:21No. I I replet. Okay. Okay. So, so basically you were treating before you went to the replet agent kind of thing, you were treating AI is kind of an add-on as opposed to the main event.
11:32Yeah. You're kind of sprinkling AI all over all over the place like autocomplete. autocomplete, you know, you maybe some debugging, some code explanation, but it just felt like you were l literally like salting AI, right? There was a period where we're salulting our products with AI.
11:52What you want is the stake made of AI, right? You want the the core thing to be AI.
11:57And so in 2024, we actually kind of laid off half the company and we went into hole again.
12:04were like you know uh uh part of uh part of startup. I was actually talking to our friend Freedom CEO uh freedom telco CEO and he was telling me he think about uh everyone talks about AI native but we need to talk about AI reborn because there's a lot of companies that need to be rebirthed with AI in many ways 2023 24 was that for replet where we went into we actually moved the company from
12:32San Francisco to Foster city we didn't go all the way to Kazakhstan but we exited San Francisco Uh, and we're like, "All right, everyone in San Francisco is thinking in this very limited way. We're going to what we're going to do is we're going to use AI to make it so that it AI is the programmer, right?" And so, uh, we we Rep was the first coding agent on the market before cloud code, before any of them.
12:58September 2024, we released it. Karpathy said it's an AGI moment because everyone suddenly figured out that AI can be at the center of the product.
13:10Not the salting of the product.
13:12And so then then our revenue sort of exploded.
13:14And you know the thing about it is Replet because it had you know the virtual machine environment where you felt really confident that you could enclose code. That's what gave you the confidence to do that so that people wouldn't bust out of that with some crazy sandbox breaking or something like that.
13:32Yeah. And now it is super important like the Binance hack was because someone used some kind of like cloud code or codeex on their machine and installed a malware and that got infected the network reached their finances and stole a lot of money. And so everyone is realizing the value of good sandboxing. So we've been publishing a lot of our architecture. It's hard.
13:56Yeah. Because I mean you're running arbitrary code. Yeah.
13:59And arbitrary code is like a like a biological organism that's trying every possible way to get out of that sandbox almost, right?
14:06Yeah. I mean I I I honestly think it was irresponsible from openthropic to develop their uh software that way uh because it's causing a lot of issues like well they what they've done is Yeah. Yeah. That's interesting.
14:18But you know other things. So actually we were just talking about this offstage. What does so now we have all of this AI what are the consequences of this okay so one thing is biotech you were just talking about this off stage yeah I mean um we we've been doing a lot of thinking at replet um about what are the consequences right in many ways I find myself in a fortunate position of like a lot of entrepreneurs like set this like really big vision that seems unattainable like ilamus will say oh we
14:46want um we want to be a multi-planetary society and it's just seems like oh you can work on it for 100 years and never achieve it but um or maybe you'll get very close and and at the end of your life you achieve it but like I said that you anyone should be able to code you know but you know 15 years ago and we achieved it I mean anyone should anyone can code now and so we were like okay what's what's the next big thing and ultimately like watching ourselves and
15:15our customers uh using rapid and using agents the thing that I feel like really excited about and interested in intellectually is what is the future of the firm, right? Like you know uh you've read Robert Co, right?
15:30Yes. Yeah. How big is a firm and does it shrink?
15:32Yeah. Just like what is the theory of the firm? Like it's just a question of like why do we have companies like [snorts] uh Robert Co like 100 years ago proposed the reason we have companies is because like there's a high transaction cost in the market, right? there is um there's uh you know a lot of uh you a lot of friction to procure uh services from the market. So you have to hire employees and train them on certain
16:00thing um so that you have that capability internally. Um but but like no one looked at the transaction cost inside companies. I mean there's a huge transaction cost inside companies. Like the reason why it's miserable to work at a company like Google is because the pre-me meeting and the premeating uh sort of memo and the postmeating memo and the post meeting and like all this like work about work and work around
16:28planning and KPIs and someone needs to keep the KPIs and um like all of that is cost of doing like you can think of it as like transaction cost of work.
16:39Yeah. Yeah. within companies and so I think AI's impact on companies kind of reducing fundamentally reducing the transaction costs within within companies and so that okay once you once you do that like so how do how should companies actually structure in the future and what kind of AI and platform support you need to to help with that. So I want to poke on this a little bit because I thought about this a lot. Yeah. which is um so
17:06even before actually AI became big I remember just you know I wrote this post which is you know within the lives of people now living like backspace didn't exist like meaning you know you typewriters right like you couldn't delete something you couldn't copy paste uh you certainly couldn't internet search um you you you couldn't you know like even a photocopier is actually a relatively recent invention and so on everything had to be typed by hand and yet People wrote 100 page like Solzenitson wrote
17:36the Gulag Archipelago. Yes. You know, like by hand, right? This 700page book, right?
17:41And I was just thinking about this because someone was complaining to me about having to write, you know, 20 pages. I'm like, you're a golden god surfing on a sea of electrons, right?
17:50You have this incredible thing in front of you. You have backspace, right? You have copy paste. You're god, you know, right?
17:57And and the thing is though, I I was thinking about that. Have we actually become more productive? Yeah. Given all of that and I thought there were several answers to this. I'd like to know your thoughts. For example, one answer is yes, we have some people have become more productive because you have all these tech unicorns and so on. A second answer is distraction in the form of social media and other things have eaten up all the productivity gains, right?
18:18A third is bureaucracy, paperwork. Yes.
18:20Because now you can create more. So they ask for more paperwork whereas the forms had to be more simple 50, 60, 70 years ago. So that's eating up the gains, right? Uh but I I think the fourth is maybe the most interesting which is um I think the computer actually has to do it for you end to end and that things like notion and Google docs and things like that we think of them as being useful but it may not be that useful to put all this intermediate state into a computer. It should just do it for you end to end
18:49because otherwise a human has to read it process it and then 100% your thoughts. Yeah. Yeah. I mean I I think that um look I mean if I were to kind of really think about zoom out and think about it what like industrial uh revolution promised us is automation right reality what happened is that um we had partial automation and so actually humans became more like machines and you
19:17know factories emerged and factory lines emerged and humans would sit there and I you would your responsibility would be to like put a you know the eraser on the pencil and move it to the next guy and the next guy would like do something else with the pencil and so um that's like fundamentally a machine's job but because we didn't have full automation people were relegated to machines and by the way this this was like Mark's like whole observation like the theory of
19:46alienation which I actually agree with theory of what alienation oh yeah yeah theory of alienation from Karl Marx I'm not a communist but he did say some interesting things.
19:55Yeah. Basically the laborer was was they used to be farmers and so on and now they were alienated from their work product and Yeah. Yeah. So so when you when you made a tomato and you sold it to your neighbor, you knew like you saw him eat the tomato and get nourished, you knew that the uh fruits of your labor have impact and that created meaning in your life.
20:15Yes. Whereas if if your job is part of a supply chain of a supply chain of a supply chain, you don't even know who the end customer is. It might be a different country. It might be a dog somewhere. It might not be a person, right?
20:28And and and so it created this alienation where work became meaningless. We don't need pause AI. We need acceleration to get to the promised land. And so what is the promised land? We call this at Replet is a self-driving company. is the idea that humans are the creative heartbeat of the company. You know, we I I think the ideal work day is one where you wake up and you're like excited to go to work and you go to work and your friends are there. You spend most of the
20:57time talking to your people, brainstorming, going and talking to your customers, inventing new things, using um like AI like this futuristic like vibe coding sort of maybe even in VR kind of like looking inventing new things and and like with one thought you can you produce something fundamentally new like even like [snorts] hardware products and things like that. And so work become more about invention, about creating new things, about um
21:26interfacing with people, about having an impact on real people's lives. And I think we're we're now we have eyesight to it. It's not there yet. I mean, agents still hallucinate and there's all these ways in which you need to monitor them and uh that's why like I I I I think now that we're like in this intermediate period where we're like enter make it better. enter make that's what vibe coding is [laughter] better we're enter pressoressors so
21:54we're still in this like awkward teenage phase of AI but we have eyesight towards what the ideal sort of labor conditions should be uh and a world where Marx is actually wrong it's funny I mean like one of my oneliners you might put this on a billboard which is like AI doesn't take your job AI makes you the CEO right because everybody's a CEO everybody's an engineering manager like managing an AI is actually a lot like engineering management
22:24and I think the part where I depart from some people is they think oh you don't need to know how to code actually as engineering manager you kind of need to know how to code even better than the because you have to AI is amazing for something where you kind of already know how to do it and if you the current instantiation of AI I think that's right you think it'll get better than that oh I you know the thing is I don't know I see over the last four years the better I've known an area, the farther I could get with AI.
22:52And and then of course you can I mean that that's I think that will always be true, right?
22:57Like I I think knowledge is good regardless like if you have the time and energy and and um ne neurology for it because I do think that coding it certain neurology but no that's right. I think at least right now what AI is doing is it's making the best better. It's amplified intelligence in my view rather than truly artificial intelligence in the sense that because it starts from a prompt and because you have to check the output and then reintegrate it with your main thing and
23:25because it's sort of you know even as sort of you know like that MITER graph that people like with time frames increasing if you zoom in on that graph it like in practice yes the agents can do longunning tasks I don't know maybe maybe it's just a function of giving them context and letting it go and improving you know, so I I think in two months you're going to see another big jump in performance.
23:51Um because why is that?
23:55So when OpenAI um did the Navier Stokes and a lot of the recent mathematics uh breakthroughs, they actually didn't use one agent or two agents or five agents. He used 20,000 agents. Now it's very costly. Cost them $20 million to do neg. But this idea of Asian swarms is like an emergent new idea of um like another way.
24:21It's surprising that it works. It's almost like 20,000 random numbers and and you have different bunch of different traces going through. It's it's surprising that like Go ahead. Maybe you have some intuition as to why.
24:32Yeah. So So the reductive way of thinking about it is sampling works.
24:38So you know in AI benchmark where you say like at one or at 100. We knew that always sampling works cuz cuz um like GPUs and AI is fundamentally like there's stochasticity to it and randomness and so if you run it more times you'll get a better answer.
24:55But I don't think it's just that. And and I had this conversation with Sam Alultman where I asked him like do you think swarms work because we're just sampling a lot more. Exactly. That that'd be the immediate intuition. But no, he said that no, because they, you know, they've tried just like normal sampling where but 20,000 don't interact with each other.
25:15Yeah. But but when they create a society of agents that interact with each other, it created new dynamics. Have you read the transcripts in the hugging face attack?
25:23No. Oh, I actually No, I did. did where they actually wrote to a shared memory thing and then they they wrote to a shared memory but there was a lot of emergent um sort of sociological thing that happened where where they had um they had um some of them that were um giving themselves to the group.
25:44They were doing sacrifices. They were like I'm going to sacrifice myself. I'm going to run out of contacts. I'm going to be shut down. But I'm going to like do kamicazi and I'm going to like collect as much information and pass it to the group before I die. And I mean you could go a little too far with it and like it can become a little woowoo and a little weird. But I do think there is an emergent sort of thing that's happening with agents that start to resemble the
26:13socology of of groups and that that has certain productivity benefits. So, here's one possible intuition, which is let's say you had a maze and you had a bunch of people start at the beginning.
26:26They all had walkie-talkies and they split up and they each took paths through the maze and they could talk to each other. Oh, this is a dead end. This is a dead end. This is a dead end. This is not a dead end. And then the other three could kind of go down that direction, right? So then that kind of thing of parallel exploration of space with communication between should speed it up versus just spawning off four pads. Plus, uh, if if there is emergent altruism, it's even more important. Like someone
26:54might say, I'm like 15% sure there's dead dead end, and if I go there, I'm not going to be able to like find the main thing, but it's probably good for the group if I'm the one that's taking the hits, right? So, those program those kind of dynamics is very important.
27:10Yeah, it's interesting. I mean, I I think uh what one of the things we're going to see also now is what that looks like in the physical world, which I'm not really looking forward to. I think the good version of that is drone construction and robotics. I think we can build cities much faster than ever before. Um I think that's going to be one of the least anticipated aspects of AI is that you can compete not just with companies and currencies but with cities because um since you know physical robotics actually you know this is all
27:39happening in China like greater Bay Area but something this may be an obvious point or maybe a non-obvious point in the digital world like when is an essay done like you have to sort of set an arbitrary thing an essay can always be improved there can always be some word you rephrase it's very kind of fuzzy even though it's digital it's bytes and zero one in a sense It's fuzzy when an essay is done. If you take a box like a physical box and you ask I want to move this box over there in the in the analog world paradoxically that box you know
28:08you can say it is within epsilon of this xyz bounding box right and you can say you can measure completion in the physical world more easily than you can measure completion in the digital world.
28:19So I think physical world tasks actually will be easier to automate once we get the actuators and sensors working. think physical robotics may actually get that full throttle automation like that's what that's what China is working on in a big way. Maybe you have some thoughts on that.
28:33Do you see the CEO of Unitry recently said that they benchmark every GPT generation and it's usually socks at robotics but GPT6 had sparks of robotics kind of physical understanding and he said he fully expects GPT7 to have solved robotics.
28:54This was actually uh I think Run made [snorts] this point. Whenever one model of an AI starts to get okay at something, then the next one is like way better, right?
29:02Yeah. For example, right now there's this huge excitement around um around AI writing model weights directly. So it can like go and instead of train a train a model, it can like fully emit the weights. And you can be a cynic about all these early because you could say ah you know it didn't do that right, it didn't do that right and and there's always like you know it's not perfect yet but yeah I mean wow this is like a
29:32start of something crazy because I think that you you're making the point that you still have to learn to code there is a world in which like AI is not only is not emitting Python anymore is emitting either model weights directly or emitting machine code directly.
29:50Yeah, I know. I mean like people have done these things with AI languages and they invent protocols to go back and forth. I think um a fundamental question is uh whether you want to read the output or not, right?
30:04And I I think uh I think you do mainly because um [snorts] well, of course it depends on the circumstance. If you're if you're reverse engineering a binary, you're not going to read it by bite, you know, some ROM reverse engineer. Um, but at least there has to be something that one can inspect because you need a control surface like a in a sense a joystick. No matter how fast that robot is, you want a control surface for it, you know, and it says I think something has to be readable even if there's something
30:35Well, actually, why don't we give maybe just give some thoughts for the audience of they should do, right? So, like what do you think that people here should study? what should they build in the age of AI? What what you know I'll give my thoughts.
30:45I don't know. I mean, if I'm living, if I'm like growing up here, I'm like in my late teens, early 20s or even 30s. And I'd be so excited because you have a government that is that's saying, "Oh, it's going to be an AI year." You have um people that like there's a lot of smart people around. You can like start to build companies. Um I'd probably just like be going around finding problems to
31:13solve. I mean the cool thing is that it is not a you know fully developed nation. Obviously you can still there's so many problems you could solve but you could solve them locally like you know you don't have to go buy technology from out uh from the west and I actually think that a lot of the solutions in the west are the old solutions like you know we don't have we don't not using like um machine vision to do smart cities you know we're using like maybe old
31:41technology to do that and so you have an opportunity to build the city of the future future and there's probably there's a lot of investment going in here. So, you could probably make money a lot of money doing that. Um, and so I I would probably just be looking going around looking for problems to solve and using AI to solve these problems.
32:01Yeah, that's good. I mean, I I think in terms of specific areas that I I think are interesting. Um, I think cyber security is going to be 20 of 10 importance over the next several years.
32:13I think the wave of attacks is going to probably change the internet. Um I think it's going to become more like uh more much more in the way of country internets and company internets much less trust between groups unfortunately and more trust within groups. That's my my view on cyber is that it's going to be fully automated.
32:30It is it already it's getting there already. That's right. But and so then you have a defense budget for the defender and then an offense budget over here and it's literally your token budget on both sides.
32:39It is it is 100% capital gain.
32:42That's right. And then I think that means um there's basically three kinds of institutions that survive. Things like China that have the great firewall. Um maybe things like Google and Amazon that have enormous token budgets and can defend every endpoint in a crazy way. And then actually blockchains like you know Ethereum, Bitcoin, Salana, Zcash that have constant security audits because if they were hacked it goes to zero.
33:06I mean proof of work is the original cyber capital game, right? Like I I actually think that the way AI is trending is making everything proof of work.
33:16So I think mining will become in part token mining.
33:19Because you mine tokens on your computer and you use them to constantly check the security of your network, you know.
33:25Yeah. Or open source packages. The other day I made a a point on GitHub. I want to see how many tokens or how many dollars worth of tokens a certain package um how many tokens went into securing a certain package.
33:41So like the Linux kernel on GitHub should have like you know $1 billion worth of tokens went into trying to secure that.
33:48And I think this is actually a new design paradigm for blockchains is to think in terms of it's funny there's actually an interesting circular economy because AI agents will buy cryptocurrencies to pay for things online and blockchains will buy AI tokens to secure themselves, right?
34:03And that's like an interesting AI crypto circular economy.
34:06Um, something else I think you guys I mean so you may disagree with me on this. This is this is actually one of our maybe points of disagreement. I actually do think it is still worthwhile to understand everything down to the bit and the bite and then up like I think I think fundamentals are actually still important perhaps even more important because um many people are going to use AI as a as like you and I come from a generation where we did have to understand all that stuff because we do that then AI is an amazing accelerant of what we already know and one of the
34:36things I'm concerned about is that the next generation is just using the shortcut without um without understanding understanding how to do it right and then that works until it doesn't work you know it it's like uh they've inherited something they couldn't build you know and uh unless you can like build an AI model from scratch uh you know then then you might not be able to advance it in the future in my view but you disagree this let me harmonize it perhaps I think there's an true opportunity cost right
35:05now because like um um I think some of them are here but I was meeting with a young group from uh the adult group uh yesterday big construction company here and the the the next generation of the family they're they're taking over and they're like putting AI and everything. So like they solved like this huge safety issue they had with the people not complying with safety you know regulations by doing AI vision. They have cameras on everyone's helmets and cameras on all
35:34their construction sites and they know down to every construction worker who's not doing the safety things correctly. And so they they solve that problem for their family's business. And like that's probably better use of their time than going and learning to code, you know. So [laughter] I absolutely agree with you.
35:54So, so, so provide if if you don't have ideas for how you can go build things and prove things, fill your time with studying for sure. Fill your time with learning.
36:04But if you have a good idea and how to be useful, go be useful. That's I think is more important.
36:08Totally totally fair. Totally fair.
36:09Synthesis. Okay, we've got some lunch upstairs guys and so please think do you want to take questions from Oh, yeah sure. We can take some questions. Let me take a few questions and then uh we've got lunch upstairs for everybody. We should have lunch upstairs. Okay.
36:22Thank you. Very insightful comments. And I'm you mentioned you boosted revenue with half of the stuff used before actually.
36:30Can you name five other industries or functions which can do the same or even better with AI?
36:36Uh I was just in Japan and you know it's you know a lot of what we focus on rep at replet is software like how to make software uh you know loop faster right we want to create a software factory to create software faster. They did a they built a marketing factory this uh this company by the way very old big company called Transcossmos they um they have this marketing group they're using graplet to bit to uh to to
37:04make to to like do really val print loop on marketing. So what they do is instead of the usual marketing strategy where like you have a big strategy meeting then you know that goes into graphic designers and they kind of work on the assets and that then goes into like uh you know focus group testing and then that goes to customers and often times at flops what they're doing is they built AI simulation and they they generate marketing ideas and test them
37:34on AI very quickly and then they build their own capability to go test on live users very they built all of that with a replet and they cut down their marketing budget, marketing spend and time to marketing very very quickly. They cut down on overhead. You don't need a lot of producers and and um like ops people around like marketers can just go run their campaigns very quickly. So I think you can apply the same lesson the same technology we're using for software on
38:04marketing on sales. I mean biology knows more but like I think physical physical AI physical world stuff bio stuff all of that I think AI coding affects everything.
38:22Hi uh you mentioned the u construction field in uh the safety. So what's your vision uh in a few years uh what's going to happen with AI in the AEC industry construction architecture?
38:37Well, you know, 14 years ago at the University of Pennsylvania, the Grasp Lab published something showing um drone construction where they had drones lift like pipes that were built to be lifted by drones and assemble them in like a Lego structure like this with multiple drones working in concert. That was 14 years ago. Um, that's advanced a lot since then and I think construction is going to get rethought. You mentioned the, you know, self self-driving company. Something I want to do in
39:06Kazakhstan. I've talked about this. I want to do a prize for the self-building home. So, you have a driverless truck.
39:12You know those new trucks without the cab in front? a driverless truck or a fleet of them pull up, the back opens, drones and robots come out and there's materials that are meant to assemble a modular building that uh that they just assembled right there, right? And they do the whole thing. And I think this is uh this is the kind of thing like the DARPA Grand Challenge where the first year it'll be kind of meh, you know, it'll like be okay. And the next year or this year after they'll get really good.
39:38for example, we'll have to make the materials uh you know, robots are really good at like sledgehammer like force and less good at dexterity. So, you know, perhaps there's less in the way of screwd and more in the way of mallet pounding and and so on and so forth.
39:51We'll have to rethink how construction is done, things that are easy for robots are hard for humans and vice versa. But I think once we can do the self-building home, then we can do the self-building village and you know, we can do we can actually hyperdelate the cost of housing. And so that's one of the things I'd like to do in Kon over the next few years. If you're interested in this kind of thing, come and join network school.
40:13All right, we'll take one or two more and then let's let's get some lunch.
40:22Uh you have talked a lot about u personal software over the years and now that we have the capability and a lot of people are building personal harnesses and stuff. uh what do you think like uh what should software engineers focus now uh what kind of software uh one should build so it's not just personal but should be useful for others I I think if you're a professional software engineer you should do the hardest thing you can imagine because
40:50that that's really like where we where society needs you is like can you go build like you know provable security systems you know can you build like scalable cloud architecture. Can you build like machine learning systems? Can we So because like five coding allows more and more people to like build like more regular everyday software. I think software engineers should focus on the hardest things.
41:15One quick thing guys, let's take a group photo actually here. Braxton have come back from the elevator. Let's take Yeah, let's take a group photo. Um after say don't don't leave yet. Just take a group photo. Okay. Okay.
41:26Stick stick around. Um but but yeah, I I I think I think software engineers could get a lot more ambitious for sure. Yeah, I'll take one more last person. Um, maybe one of the kids. The kids some of the younger guys. Yeah.
41:43Thank you. Thank you.
41:43Kids, do you have a question? Kids? No, I'm here. I'm here.
41:49Okay. First, the power belong to the strongest, you know, and then to the smartest. And when AI makes everyone like equal, who wins? That's that's a great question by the way. So in history of civilization uh the most powerful was the strongest you know Gingaskhan uh then the most powerful the smartest uh Zuckerberg you know so I was like now who's who's going to be the most
42:18I actually think I think the most cooperative right yeah I I think when when what's left is relations between people the like the diplomat that's right because basically if you can if you have a trusted tribe then you can share knowledge within them and the AI is stronger because you can all share information outside the trusted tribe you're not sharing code you're not you know you don't like so the larger your trusted tribe the larger your set of cooperation I think the better off you are in the post AI world
42:46I I I I I would agree with that I would I would um you know it's it's sort of like a such a lame buzzword but I I do think that EQ is probably matters a lot more than IQ right now and it's it's weird because on the other hand AI is amplified intelligence yeah you know people who are talking about for example the people who are prompting it to solve the quasi remain hypothesis have to be able to to to check the result you know they have to do lean yeah I don't think I don't think we're past the singularity of IQ like IQ still
43:15matters but I think his question is more like okay after we're after intelligence fully commoditized you know it's the question is whether that happens I think it will happen I so I I I for a long time I was resistant to it I'm now like you need to like really sit down and convince yourself that intelligence is going to commoditize and it's really hard.
43:37Yeah. Yeah. So, so [laughter] right right now I'm just seeing the highest ROI I've ever seen on IQ.
43:43Yeah. Yeah. It always like that. The turkey feels the best just before it gets slaughtered.
43:48That that is a key question. [laughter] No, that is a key question. Will just before Thanksgiving, one day before Thanksgiving, the turkey is the is the king.
43:55I think the [laughter] you're you're right. I think the biggest question is is AI ever upstream or is it always downstream?
44:02Like that is can AI prompt itself?
44:04Yeah. And obviously it can with chain of thought but but I don't think that's a question of intelligence and this is why I'm saying like EQ it's like I I think pure raw horsepower intelligence is different than like a holistic understanding the world understanding geopolitics understanding people understanding dynamics understanding you know I would agree with you that this is not so let me add something which is humans are sensors and actuators and they bring
44:34back data and information like EQ means sensing the world and so on and you feed that in and AI because it's a digital life form it doesn't you know people say taste right the actually has sight and it has uh it has hearing but it literally doesn't have taste and doesn't have touch and it doesn't have how would it know what humans want right exactly so the sensing of the world is kind of right okay you want to take a group photo why don't we come up on stage give it up for our speakers
45:02today And yeah, come join us on stage for a group photo.
45:09Yeah, is the CEO of Freedom Talk, founder of