1 00:00:00,120 --> 00:00:03,400 I think AIX crypto by and large is a scam. 2 00:00:04,200 --> 00:00:08,400 The majority of businesses I see building an AIX crypto are doing 3 00:00:08,400 --> 00:00:11,000 nothing more than trying to launch and sell a token to 4 00:00:11,160 --> 00:00:13,760 unsophisticated retail market participants. 5 00:00:13,880 --> 00:00:17,880 When I started LaGrange, the cost of generating a proof for 6 00:00:17,880 --> 00:00:24,080 AZKEVM was like a dollar or in the range of 10s of cents per 7 00:00:24,080 --> 00:00:25,680 transaction. Ridiculous. 8 00:00:25,680 --> 00:00:30,440 Now it's about 100th of a cent. It allows you to ensure that the 9 00:00:30,440 --> 00:00:34,360 correct model is being used for the inference that a system is 10 00:00:34,360 --> 00:00:37,640 receiving, and it also lets you ensure that there are properties 11 00:00:37,640 --> 00:00:40,040 of privacy over the use of that aim. 12 00:00:40,720 --> 00:00:44,200 There's a subset of public market participants in crypto 13 00:00:44,600 --> 00:00:49,520 who trade charts behaving the same way when they're trading 14 00:00:49,560 --> 00:00:53,080 bonk versus, when they're trading pengu versus, when 15 00:00:53,080 --> 00:00:56,360 they're trading with versus when they're trading Doge versus when 16 00:00:56,360 --> 00:00:59,360 they're trading LA. All they care about is trading 17 00:00:59,360 --> 00:01:01,720 on price action and trying to catch a runner. 18 00:01:01,880 --> 00:01:06,200 And if the only participants in your market are trading on those 19 00:01:06,200 --> 00:01:10,400 characteristics, whether or not you are in for protocol or you 20 00:01:10,400 --> 00:01:13,560 are a meme coin, you effectively have converged the same market 21 00:01:13,560 --> 00:01:14,560 dynamics. The meme point. 22 00:01:14,880 --> 00:01:18,000 If you're looking to stake your crypto with confidence, look no 23 00:01:18,000 --> 00:01:21,840 further than Course 1. More than 150,000 delegators, 24 00:01:21,840 --> 00:01:24,960 including institutions like Bit Go, Pantera Capital and Ledger 25 00:01:24,960 --> 00:01:26,480 Trust Core is one with their assets. 26 00:01:26,880 --> 00:01:29,280 They support over 50 block chains and are leaders in 27 00:01:29,280 --> 00:01:32,440 governance or networks like Cosmos, ensuring your stake is 28 00:01:32,440 --> 00:01:35,400 responsibly managed. Thanks to the advanced MEV 29 00:01:35,480 --> 00:01:38,240 research, you can also enjoy the highest staking rewards. 30 00:01:38,720 --> 00:01:41,720 You can stake directly from your preferred wallet, set up a white 31 00:01:41,720 --> 00:01:45,520 label note, restake your assets on Eigenia or Symbiotic, or use 32 00:01:45,520 --> 00:01:47,760 their SDK for multi chain staking in your app. 33 00:01:48,320 --> 00:01:51,440 Learn more at Chorus .1 and start taking today. 34 00:01:52,800 --> 00:01:55,920 Hey guys, I want to tell you about Gnosis, a collective of 35 00:01:55,920 --> 00:01:58,760 builders creating real tools for real people on the open 36 00:01:58,760 --> 00:02:00,920 Internet. Gnosis has been around since 37 00:02:00,920 --> 00:02:03,160 2015. In fact, it started as one of 38 00:02:03,160 --> 00:02:06,520 Etherium's very first projects, and today it's grown to a whole 39 00:02:06,520 --> 00:02:09,280 ecosystem designed to make open finance actually work for 40 00:02:09,280 --> 00:02:11,560 everyday people. At the center of it all is 41 00:02:11,560 --> 00:02:13,680 Gnosis Chain. It's a low cost, highly 42 00:02:13,680 --> 00:02:16,560 decentralized layer, one that's compatible with Etherium and 43 00:02:16,560 --> 00:02:18,760 secured by over 300,000 validators. 44 00:02:19,160 --> 00:02:22,360 So whether you're building a DAP, experimenting with Defy, or 45 00:02:22,360 --> 00:02:25,720 working on autonomous agents, Nosys chain gives you a solid, 46 00:02:25,720 --> 00:02:29,320 neutral foundation to build on. But Nosys is more than just 47 00:02:29,320 --> 00:02:32,400 infrastructure, it's also tools that people can actually use, 48 00:02:32,840 --> 00:02:35,280 like Circles for example. Let's anyone issue their own 49 00:02:35,280 --> 00:02:38,280 digital currency through networks of trust, not banks. 50 00:02:38,720 --> 00:02:41,160 And then there's Metri. It's their smart contract wallet 51 00:02:41,160 --> 00:02:43,960 that makes it easy to access Circles, manage group 52 00:02:43,960 --> 00:02:46,400 currencies, and even spend anywhere. 53 00:02:46,520 --> 00:02:49,200 Visa is accepted. Thanks to their integration with 54 00:02:49,200 --> 00:02:53,280 Nosus Pay, all this is governed by Nosus DAO where anyone can 55 00:02:53,280 --> 00:02:55,880 propose, vote and help guide the network. 56 00:02:56,400 --> 00:02:59,120 And if you want to get involved, running a validator is super 57 00:02:59,120 --> 00:03:01,960 easy. All you need is 1 GNO and some 58 00:03:01,960 --> 00:03:04,680 basic hardware. To learn more and start building 59 00:03:04,680 --> 00:03:07,000 on the open Internet, head to nosus dot IO. 60 00:03:07,800 --> 00:03:10,760 Nosus building the open Internet one block at a time. 61 00:03:12,080 --> 00:03:13,920 Welcome to The Epicenter, the show which talks about the 62 00:03:13,920 --> 00:03:16,480 technologies, projects, and people driving decentralization 63 00:03:16,480 --> 00:03:19,600 and the blockchain revolution. I'm Sebastian Coutier, and today 64 00:03:19,600 --> 00:03:23,080 I'm here with Ismael from LaGrange Labs. 65 00:03:23,080 --> 00:03:25,520 How's it going, man? Doing well. 66 00:03:25,520 --> 00:03:26,920 Thanks so much for having me, Sebastian. 67 00:03:27,920 --> 00:03:29,360 Yeah, so. I mean, we've known each other 68 00:03:29,360 --> 00:03:31,720 for a while. This is actually your first time 69 00:03:31,720 --> 00:03:35,240 on Epicenter though. I think you were probably on the 70 00:03:35,240 --> 00:03:38,040 Interop at some point, like a while back. 71 00:03:38,400 --> 00:03:39,760 We we've done some podcasts before. 72 00:03:39,760 --> 00:03:40,920 I was this. Few times. 73 00:03:41,680 --> 00:03:45,800 First, first episode or episode disclaimer I'm an Angel investor 74 00:03:45,800 --> 00:03:48,800 in LaGrange Labs like to get that out of the way early on, 75 00:03:49,120 --> 00:03:51,640 but you know I'm going to grill you any anyway. 76 00:03:53,080 --> 00:03:55,720 But yeah, let's let's. I wouldn't. 77 00:03:55,720 --> 00:03:59,520 Expect anything different? Yeah, let's let's dive right 78 00:03:59,520 --> 00:04:01,800 into it. So like LaGrange has been a 79 00:04:01,800 --> 00:04:05,800 while, has been around for a while and like you guys like 80 00:04:06,440 --> 00:04:16,240 started as this highly research driven ZK proof ZK project that 81 00:04:16,279 --> 00:04:20,000 has now evolved into all sorts of verticals, including AI and 82 00:04:20,000 --> 00:04:25,120 defy and and scaling. Yeah. 83 00:04:25,120 --> 00:04:27,800 But first, like, yeah, let's talk a little bit about your 84 00:04:27,800 --> 00:04:34,040 journey and what sparked the idea for LaGrange and how does 85 00:04:34,040 --> 00:04:37,840 your background kind of fit in this verifiable computations 86 00:04:38,000 --> 00:04:41,680 narrative? Yeah, that's a fantastic 87 00:04:41,680 --> 00:04:44,360 question. So LaGrange has from day one 88 00:04:44,360 --> 00:04:47,800 been hyper focused as a zero knowledge proof company. 89 00:04:48,200 --> 00:04:51,920 And throughout our history, we have targeted a variety of 90 00:04:51,920 --> 00:04:54,760 problems that we solve with 0 knowledge proofs. 91 00:04:54,960 --> 00:04:58,240 In the very early days of LaGrange, this was things like 92 00:04:58,240 --> 00:05:04,520 Interop or Defy Co processing. And over time, we have scaled 93 00:05:04,520 --> 00:05:07,440 the business, scaled the go to market motion and tackled 94 00:05:07,440 --> 00:05:10,440 increasingly large problem spaces and increasingly large 95 00:05:10,440 --> 00:05:13,080 10's. The current version of LaGrange, 96 00:05:13,080 --> 00:05:16,720 the business that we are today has a very large part of our go 97 00:05:16,720 --> 00:05:19,360 to market focus oriented towards AI. 98 00:05:19,600 --> 00:05:22,680 Both the application of 0 knowledge proofs to improve the 99 00:05:22,680 --> 00:05:26,400 trust and safety of AI in crypto, as well as to improve 100 00:05:26,400 --> 00:05:29,080 the trust and safety of AI in traditional sectors using 101 00:05:29,360 --> 00:05:31,440 advanced cryptography and 0 knowledge proofs. 102 00:05:31,880 --> 00:05:36,840 But as a team, we've always been laser focused on ZK, and that's 103 00:05:36,840 --> 00:05:39,440 been in our DNA. Our chief scientist, Babis 104 00:05:39,440 --> 00:05:42,600 chairs the cryptography department at Yale and under him 105 00:05:42,600 --> 00:05:46,440 we have a very large research team in applied cryptography 106 00:05:46,440 --> 00:05:49,400 with a bunch of world class researchers, people like 107 00:05:49,400 --> 00:05:52,920 Demetrius Papadopoulos, who's a professor at HKUST, Shravan 108 00:05:52,920 --> 00:05:57,640 Trinvasin, Nicola Gaye, and a bunch of great, great people on 109 00:05:57,640 --> 00:06:00,840 the team. And so unfortunately I'm not a 110 00:06:00,840 --> 00:06:03,800 ZK researcher myself. I was a venture investor before 111 00:06:04,040 --> 00:06:06,520 and then I worked in financial services before then leading 112 00:06:06,520 --> 00:06:08,960 digital asset strategy for large insurance company. 113 00:06:09,440 --> 00:06:14,920 But you know, the bread and butter, the DNA of the business 114 00:06:15,040 --> 00:06:18,400 has always been ZK, and that's embodied with the research team 115 00:06:18,400 --> 00:06:23,600 that we've built at La Grange. And how much of the team now, 116 00:06:23,600 --> 00:06:27,680 like if you were to sort of, you know, look at the different 117 00:06:28,200 --> 00:06:31,240 people on the team, how much of the team comes from the ZK 118 00:06:31,480 --> 00:06:34,640 research background versus now like the AI component? 119 00:06:36,800 --> 00:06:39,840 Yeah. So I would say that everyone at 120 00:06:39,840 --> 00:06:43,120 the company is a ZK person. We're a cryptography company, so 121 00:06:43,120 --> 00:06:45,680 we don't build new foundation models. 122 00:06:45,920 --> 00:06:51,160 We don't build LMS or agents or, or really anything besides 0 123 00:06:51,160 --> 00:06:53,760 knowledge proofs. What we do is we apply advanced 124 00:06:53,760 --> 00:06:55,960 cryptography and 0 knowledge proofs to AI. 125 00:06:57,280 --> 00:07:00,520 And so we do have people on the team who have familiarity with 126 00:07:00,560 --> 00:07:01,600 AI. We have people in the team 127 00:07:01,600 --> 00:07:05,080 who've worked at companies doing AI engineering. 128 00:07:05,280 --> 00:07:08,600 But you know, as a business, what differentiates us isn't the 129 00:07:08,760 --> 00:07:11,280 AI talent and the AI skills, it's the cryptography. 130 00:07:11,760 --> 00:07:15,000 And it is, how do we take cryptography and apply that 131 00:07:15,000 --> 00:07:18,360 cryptography to companies that have AI expertise? 132 00:07:19,480 --> 00:07:24,600 The same way that like you don't have to be a insurance expert or 133 00:07:25,160 --> 00:07:29,680 a, you know, consumer expert to build AI that's used for 134 00:07:29,680 --> 00:07:32,200 consumer or or financial service purposes. 135 00:07:32,400 --> 00:07:35,880 You don't, we don't think you have to be an AI expert or 136 00:07:35,920 --> 00:07:39,000 shouldn't have to be an AI expert to build cryptography. 137 00:07:39,000 --> 00:07:40,920 That can be very impactful to AI. 138 00:07:41,240 --> 00:07:44,880 It allows AI companies not to have to deal with cryptography 139 00:07:45,000 --> 00:07:47,400 when they work with us. All they have to do is use our 140 00:07:47,400 --> 00:07:49,320 technology. We don't necessarily they have 141 00:07:49,320 --> 00:07:52,080 to deal with AI either. They just have to be able to use 142 00:07:52,240 --> 00:07:56,000 very simply, a technology that adds a a real zero to 1 143 00:07:56,000 --> 00:07:58,600 improvement in the security and trust properties of what they 144 00:07:58,640 --> 00:08:03,760 built. So there's always been an AI 145 00:08:03,760 --> 00:08:09,720 crypto convergence narrative since as going as far back as 146 00:08:09,720 --> 00:08:13,600 20/15/2016. You know, guys like Trent 147 00:08:13,600 --> 00:08:17,560 McConaughey who've been on the podcast multiple times have been 148 00:08:17,800 --> 00:08:20,920 kind of pushing like different narratives around AI and crypto, 149 00:08:20,920 --> 00:08:27,000 whether it's for private AI or user own data or provable AI. 150 00:08:27,000 --> 00:08:29,240 And you guys are like now at the forefront of that. 151 00:08:30,480 --> 00:08:34,840 How much of this is sort of hype, right? 152 00:08:34,840 --> 00:08:39,280 And what's the genuine, the genuine signal versus noise sort 153 00:08:39,280 --> 00:08:44,520 of thing that people should look at when analyzing or sort of 154 00:08:44,520 --> 00:08:48,320 observing projects, the building in the AI space within crypto? 155 00:08:50,520 --> 00:08:53,360 Yeah. So you know, you started this by 156 00:08:53,360 --> 00:08:54,800 saying you're going to ask some tough questions. 157 00:08:54,800 --> 00:08:56,200 So I'm going to give some tough answers. 158 00:08:56,760 --> 00:09:00,040 I think AIX crypto by and large is a scam. 159 00:09:00,800 --> 00:09:05,040 The majority of businesses I see building an AIX crypto are doing 160 00:09:05,040 --> 00:09:07,640 nothing more than trying to launch and sell a token to 161 00:09:07,680 --> 00:09:10,400 unsophisticated retail market participants. 162 00:09:11,040 --> 00:09:14,400 The, the reality is that there are some things that have been 163 00:09:14,400 --> 00:09:17,880 financed and built in crypto that are actually very, very 164 00:09:17,880 --> 00:09:22,040 relevant for AI. And so one of those we believe 165 00:09:22,040 --> 00:09:23,600 is, is 0 knowledge proofs, right? 166 00:09:23,600 --> 00:09:27,480 It's a technology that comes from academia, but has been 167 00:09:27,480 --> 00:09:31,800 productionized in crypto because of the demand for scalable, 168 00:09:31,800 --> 00:09:35,080 provable block space, right? ZK rollers and because of that, 169 00:09:35,080 --> 00:09:39,160 you know, private capital has flowed into ZK in crypto and 170 00:09:39,240 --> 00:09:43,880 about a billion dollars of venture dollars have been spent 171 00:09:44,080 --> 00:09:47,000 on R&D 40 knowledge proofs in the space of crypto. 172 00:09:47,000 --> 00:09:50,800 Now great, we can scale blockchains more, but what are 173 00:09:50,800 --> 00:09:53,720 the other applications of this technology that we have as an 174 00:09:53,720 --> 00:09:58,440 industry plugged a billion dollars into and AI and adding 175 00:09:58,440 --> 00:10:01,480 trust and safety to AI, we would argue as one of the largest 176 00:10:01,480 --> 00:10:05,280 markets. So it's not that AIX crypto 177 00:10:05,280 --> 00:10:08,000 where we're doing the centralized agents to move your, 178 00:10:08,320 --> 00:10:11,640 you know, to rebalance your yield aggregator on chain. 179 00:10:11,880 --> 00:10:14,800 It is how do we take a fundamental piece of 180 00:10:14,800 --> 00:10:17,960 infrastructure that's useful in crypto and that is a technical 181 00:10:17,960 --> 00:10:22,000 breakthrough financed by crypto and apply that to other sectors 182 00:10:22,160 --> 00:10:24,720 and other areas. That's how we think of where our 183 00:10:24,720 --> 00:10:27,040 business is positioned. Crypto is a capital formation 184 00:10:27,040 --> 00:10:29,040 mechanism. Cryptography is just 185 00:10:29,040 --> 00:10:33,040 mathematics. It isn't a crypto, a crypto 186 00:10:33,040 --> 00:10:35,240 thing, right? Cryptography has existed well 187 00:10:35,240 --> 00:10:38,120 before crypto and it will exist for the entirety of crypto. 188 00:10:38,320 --> 00:10:40,560 It secures the Internet and now it secures AI. 189 00:10:41,280 --> 00:10:45,840 Now where I don't think crypto XAI is a scam is in some very 190 00:10:45,840 --> 00:10:49,920 specific applications such as, you know, I think sourcing of 191 00:10:49,920 --> 00:10:54,760 GPU's from, you know, very large subsets of users who may have 192 00:10:54,760 --> 00:10:56,960 latent compute sitting around. I think is a very interesting 193 00:10:56,960 --> 00:10:59,840 thing the the ethers of the world, the prime intellects of 194 00:10:59,840 --> 00:11:02,640 the world. I also think it's a fantastic 195 00:11:02,640 --> 00:11:06,520 market and AIX crypto for for certain types of agentic things, 196 00:11:06,520 --> 00:11:09,120 right where you want to have like natural language based 197 00:11:09,120 --> 00:11:10,680 wallets, I think is very interesting. 198 00:11:10,680 --> 00:11:15,440 It decreases the user experience or improved user experience of 199 00:11:15,440 --> 00:11:17,280 using crypto. I think those things are very 200 00:11:17,280 --> 00:11:20,600 cool. But by and large, I think the of 201 00:11:20,600 --> 00:11:24,600 the the 100 companies that you see that are announcing stuff 202 00:11:24,600 --> 00:11:27,360 every week building an AIX crypto, maybe two of them are 203 00:11:27,360 --> 00:11:30,320 not scams. Yeah, that I think that 204 00:11:30,320 --> 00:11:32,520 resonates with me. And I think one of the things 205 00:11:32,520 --> 00:11:35,840 that stands up from what you just said is that AI in crypto 206 00:11:35,840 --> 00:11:40,240 is, is not just one vertical, it's different types of problems 207 00:11:40,240 --> 00:11:43,720 that are trying to be solved. You know, whether that is, you 208 00:11:43,720 --> 00:11:49,960 know, scaling access to GP us applying LLMS to user 209 00:11:49,960 --> 00:11:54,720 interactions, or you know, in your case, providing provability 210 00:11:55,320 --> 00:11:58,520 and verifiability to to AI inference. 211 00:11:58,920 --> 00:12:01,360 Those are all like very different problems that use 212 00:12:01,600 --> 00:12:04,720 crypto and cryptography in very different ways. 213 00:12:05,440 --> 00:12:09,480 Let's like maybe dive into the CKML use case a little bit and 214 00:12:09,480 --> 00:12:12,480 like for people who are not familiar with this particular 215 00:12:12,480 --> 00:12:16,240 technology and how you guys are solving some really tough 216 00:12:16,240 --> 00:12:20,800 problems there, like what is what is CKML and how does 217 00:12:20,800 --> 00:12:24,560 LaGrange sort of fit in this, in this use case? 218 00:12:26,400 --> 00:12:29,400 Yeah. So what ZKML lets you do is, is 219 00:12:29,400 --> 00:12:33,320 effectively 2 things. It allows you to ensure that the 220 00:12:33,320 --> 00:12:37,240 correct model is being used for the inference that a system is 221 00:12:37,240 --> 00:12:39,800 receiving. And it also lets you ensure that 222 00:12:39,800 --> 00:12:43,040 there are properties of privacy over the use of that AI. 223 00:12:43,720 --> 00:12:48,040 Now, where is this valuable? Well, most places that use AI 224 00:12:48,040 --> 00:12:51,360 have a remote system where that model is running, that it's 225 00:12:51,360 --> 00:12:53,920 communicating with something else that is dependent on it. 226 00:12:54,160 --> 00:12:57,040 You can think of this in aerospace defense as, you know, 227 00:12:57,040 --> 00:12:59,840 command and control systems. You can think of this in 228 00:12:59,840 --> 00:13:03,680 healthcare as a user who's interacting with a diagnostic 229 00:13:03,680 --> 00:13:06,240 LLM, or a doctor that's interacting with a diagnostic 230 00:13:06,240 --> 00:13:08,600 LLM run by a third party company. 231 00:13:09,240 --> 00:13:13,720 Or you can think about this Encrypto as a user who wants to 232 00:13:13,720 --> 00:13:18,200 generate a bunch of transactions from a natural language prompt 233 00:13:18,560 --> 00:13:21,840 for, you know, buying an asset on Ethereum and bridging it onto 234 00:13:21,840 --> 00:13:24,280 BNB and then swapping it into something else, right? 235 00:13:24,520 --> 00:13:28,440 In all of these situations, you have someone with a materially 236 00:13:28,440 --> 00:13:32,520 amount, a material amount of financial value tied to the 237 00:13:32,520 --> 00:13:37,480 correctness of an AI output. What 0 knowledge proofs lets you 238 00:13:37,480 --> 00:13:42,120 do is to generate a proof that effectively with this model and 239 00:13:42,200 --> 00:13:46,240 this input, this is the output, and you can be 100% sure of 240 00:13:46,240 --> 00:13:48,320 that. That's the first property you 241 00:13:48,320 --> 00:13:50,920 get from ZKML. It's a very, very powerful one 242 00:13:51,240 --> 00:13:54,400 in the field of applying security to AI and safety to AI. 243 00:13:54,640 --> 00:13:58,520 Now, the second property is privacy, right? 244 00:13:58,640 --> 00:14:01,960 There's a lot of conversations about AI ethics, and privacy is 245 00:14:01,960 --> 00:14:04,320 generally central to all of those conversations. 246 00:14:04,840 --> 00:14:06,760 But what privacy is is, is twofold. 247 00:14:06,760 --> 00:14:11,360 It is how do I ensure that the model that's being used doesn't 248 00:14:11,440 --> 00:14:13,680 potentially, or the person who's running the model doesn't 249 00:14:13,680 --> 00:14:15,800 potentially have access to the underlying user data? 250 00:14:16,160 --> 00:14:20,720 So you say, hey, I have this, you know, this weird chest pain 251 00:14:20,920 --> 00:14:25,480 and I want to interact with diagnostic model, but I don't 252 00:14:25,480 --> 00:14:28,680 want the the mega Corp running this diagnostic model to be 253 00:14:28,680 --> 00:14:31,920 like, hey, I have chest pain. Let me serve, you know, 254 00:14:31,960 --> 00:14:35,400 Sebastian adds for chest pain medication that that's kind of a 255 00:14:35,400 --> 00:14:39,360 dystopian future of all of your health data becomes just, you 256 00:14:39,360 --> 00:14:43,040 know, dispersed across the Internet with whoever's running 257 00:14:43,040 --> 00:14:45,720 these models. And so that's where privacy is 258 00:14:45,720 --> 00:14:49,680 like very, very important in AI. And the second place where 259 00:14:49,680 --> 00:14:53,040 privacy is very important is in keeping the models private, 260 00:14:53,040 --> 00:14:55,440 right? So closed sourcing of models and 261 00:14:55,440 --> 00:14:59,560 closed sourcing of weights generally is, is significant in 262 00:14:59,560 --> 00:15:03,440 fields like healthcare and financial services, we're 263 00:15:03,440 --> 00:15:05,280 fine-tuned models. The weights of those will be 264 00:15:05,280 --> 00:15:07,200 considered PII or client information. 265 00:15:08,240 --> 00:15:10,680 And so being able to keep the model private actually allows 266 00:15:10,680 --> 00:15:12,040 you to use it in some interesting ways. 267 00:15:12,360 --> 00:15:15,360 And so the two things you get from ZKML is being able to keep 268 00:15:15,880 --> 00:15:18,280 a lot of information private that otherwise would have to be 269 00:15:18,280 --> 00:15:22,080 public and being able to add security on top of the use of 270 00:15:22,160 --> 00:15:24,760 AI. OK, right. 271 00:15:24,760 --> 00:15:27,560 So we have inference verifiability, which is like a 272 00:15:27,560 --> 00:15:34,400 very important use case in military and industrial settings 273 00:15:34,400 --> 00:15:40,760 where you, you want assurance that this query, this prompt has 274 00:15:40,760 --> 00:15:45,760 been sent to a particular model and that the inference comes 275 00:15:45,760 --> 00:15:49,520 from that particular model. And then you have privacy, I 276 00:15:49,520 --> 00:15:53,680 think. I think so I want to maybe just 277 00:15:54,080 --> 00:15:59,000 kind of zoom in on the inference verifiability part, because I 278 00:15:59,000 --> 00:16:02,440 think for, for most people who think about ZK and, and kind of 279 00:16:02,440 --> 00:16:06,880 0 knowledge circuits, what comes to mind is like a computation 280 00:16:06,880 --> 00:16:12,480 environment that is, that's very limited, right? 281 00:16:12,480 --> 00:16:16,000 So the, the, the types of computations that one can do 282 00:16:16,000 --> 00:16:19,800 inside AZK proof are like quite simple and rudimentary. 283 00:16:20,040 --> 00:16:23,920 While when thinking about AI inference, it's like this very 284 00:16:23,920 --> 00:16:28,480 complex compute problem. So can you maybe like clear, 285 00:16:28,800 --> 00:16:32,320 clear a little bit of that, that misconception and, and how we 286 00:16:32,320 --> 00:16:35,880 actually get to do inference in a ZK proof? 287 00:16:36,040 --> 00:16:37,360 Like how does that actually work? 288 00:16:39,240 --> 00:16:41,320 Yeah, this is, this is a, this is a really a very good 289 00:16:41,320 --> 00:16:43,720 question. So I, I would say that part of 290 00:16:43,760 --> 00:16:46,920 the hard thing about staying on top of ZK for, for the broader 291 00:16:46,920 --> 00:16:49,720 market is how fast ZK changes, right. 292 00:16:49,720 --> 00:16:54,600 So, you know, I, I, since I started LaGrange 4:00-ish years 293 00:16:54,600 --> 00:16:57,840 ago, about four years ago, we've seen an order of magnitude 294 00:16:57,840 --> 00:17:01,680 improvement per year in the performance of ZK. 295 00:17:02,240 --> 00:17:05,079 And that's consistently been every single year all the way 296 00:17:05,079 --> 00:17:08,280 from improvements in the core cryptography, improvements in 297 00:17:08,280 --> 00:17:11,160 tricks and circuit writing that make things faster, improvements 298 00:17:11,160 --> 00:17:15,200 in hardware acceleration. All of this has just drastically 299 00:17:15,200 --> 00:17:16,359 improved the performance of the space. 300 00:17:16,359 --> 00:17:20,400 When I started LaGrange, the cost of generating a proof for 301 00:17:20,400 --> 00:17:26,560 AZKEVM was like a dollar in the range of 10s of cents per 302 00:17:26,560 --> 00:17:28,160 transaction. Ridiculous. 303 00:17:28,160 --> 00:17:32,280 Now it's about 100th of a cent. I saw from from ZK Sync's newest 304 00:17:32,280 --> 00:17:35,360 benchmark for Bluejam too. Fantastic improvements in speed 305 00:17:35,960 --> 00:17:38,960 now. AI would have been a pipe dream 306 00:17:39,240 --> 00:17:43,760 to prove in ZK four years ago. Today it is actually quite 307 00:17:43,760 --> 00:17:46,400 performance. So our library deep proof that 308 00:17:46,400 --> 00:17:51,160 we built can actually generate proofs of GPT 2, Llama and 309 00:17:51,160 --> 00:17:58,200 Gemma, which are two, which are three open source models that 310 00:17:58,320 --> 00:18:01,600 are are open source LLMS and and we can do those. 311 00:18:01,600 --> 00:18:03,560 And obviously I'm not going to claim the performances like 312 00:18:03,560 --> 00:18:06,840 anywhere near real time, but we can do those with relatively 313 00:18:06,840 --> 00:18:10,240 reasonable performance. And for a lot of very from a lot 314 00:18:10,240 --> 00:18:12,280 of much smaller model architectures, we can generate 315 00:18:12,280 --> 00:18:16,880 proofs in the order of seconds. And that's without the 316 00:18:16,920 --> 00:18:19,840 specialized hardware that we expect to be available in the 317 00:18:19,840 --> 00:18:23,200 next year, which should add a one to two order of magnitude 318 00:18:23,200 --> 00:18:25,280 improvement in the proving times of these systems. 319 00:18:25,320 --> 00:18:30,480 Well, and so but. You're saying GPT 2 and Llama, 320 00:18:31,320 --> 00:18:33,160 those are fairly old models, right? 321 00:18:33,160 --> 00:18:36,200 I mean, yeah. So what what we like is it 322 00:18:36,200 --> 00:18:40,280 expected that you know, we'll be able to do verifiability on you 323 00:18:40,280 --> 00:18:44,560 know, large like very performant models like like Gemini 2.5 Pro 324 00:18:44,560 --> 00:18:49,280 or or like Grok 4 super heavy like is it, is it reasonable to 325 00:18:49,280 --> 00:18:53,000 think that ZK can also verify inference on on these very 326 00:18:53,000 --> 00:18:56,440 complex models? Right. 327 00:18:56,440 --> 00:19:02,400 So like GPT 2, Llama, Gemma, those are like let's say 10 328 00:19:02,400 --> 00:19:07,680 figure parameter models, right? So, you know, Gemma, I think is 329 00:19:08,000 --> 00:19:11,360 this version of Gemma that are 1 billion parameter, 5 billion 330 00:19:11,360 --> 00:19:13,880 parameter. There's versions of Llama that 331 00:19:13,880 --> 00:19:18,120 are like 6-7 billion parameter. GBT 2 I think is sub 1 billion 332 00:19:18,120 --> 00:19:21,080 parameter. It's like 600700K. 333 00:19:21,680 --> 00:19:25,040 But you, you you're, you're talking about, you know how many 334 00:19:25,040 --> 00:19:27,960 orders of magnitude you need to reach in a performance 335 00:19:27,960 --> 00:19:29,880 improvement to run those models efficiently. 336 00:19:30,160 --> 00:19:34,000 So getting from a, a 5 billion parameter model to a 50 billion 337 00:19:34,000 --> 00:19:35,840 parameter model. SO1 order of magnitude 338 00:19:35,840 --> 00:19:38,800 improvement in memory optimizations and proving time 339 00:19:39,240 --> 00:19:44,200 getting from a 10,000 parameter model to a 5 billion parameter 340 00:19:44,200 --> 00:19:48,760 model or a 10 million parameter to a 5 billion parameter is what 341 00:19:48,760 --> 00:19:51,600 it's four or three orders of magnitude. 342 00:19:52,080 --> 00:19:58,280 So we're closer to being able to run a frontier model with 5060, 343 00:19:58,280 --> 00:20:01,560 seventy, 100 billion parameters than we were to being able to 344 00:20:01,560 --> 00:20:08,480 run Gemma or LLAMA or GBT 2A year and a half ago. 345 00:20:08,840 --> 00:20:11,880 So we can run. In the current version of the 346 00:20:11,880 --> 00:20:17,200 proof a variety of LMS that are transparently smaller in size 347 00:20:17,200 --> 00:20:21,160 than what would be used for a lot of you know chat apps today. 348 00:20:21,440 --> 00:20:25,000 But we're probably about a year, 18 months from being able to run 349 00:20:25,360 --> 00:20:27,760 the, the frontier models that people are familiar with. 350 00:20:28,760 --> 00:20:32,000 Generally we're actually, So what I'm seeing, not seeing an 351 00:20:32,000 --> 00:20:34,640 increase in parameter count proportionate to the 352 00:20:34,640 --> 00:20:37,240 improvements we're seeing in ZK performance every year, right? 353 00:20:37,240 --> 00:20:40,960 We, we, we did not go from 50 billion parameter, 60 billion 354 00:20:40,960 --> 00:20:45,440 parameter models last year to, you know, 500 billion parameter 355 00:20:45,440 --> 00:20:47,680 models, trillion parameter models this year, right? 356 00:20:47,840 --> 00:20:53,400 And we did go from being able to prove, you know, 8 figure 357 00:20:53,400 --> 00:20:56,160 parameter models being able to prove low 10 figure parameter 358 00:20:56,160 --> 00:20:59,360 models in a year. So right the the rate of ZK 359 00:20:59,360 --> 00:21:01,960 improvements a lot faster than the rate with which models are 360 00:21:01,960 --> 00:21:04,920 on. Interesting. 361 00:21:04,960 --> 00:21:09,320 So, so you think that ZK will be able to continuously catch up 362 00:21:09,320 --> 00:21:16,200 with the speed at which AI models are are are also 363 00:21:16,200 --> 00:21:19,440 improving? At least inference, right. 364 00:21:19,680 --> 00:21:23,600 I think there's a question of whether or not, you know, if 365 00:21:23,600 --> 00:21:28,120 you're training a, a model on like, I think it was the newest 366 00:21:28,120 --> 00:21:29,520 Grok one. They're training on that, that, 367 00:21:29,520 --> 00:21:33,480 that giant data center that they did, you know, get financing for 368 00:21:33,480 --> 00:21:37,000 in the range of of, you know, several billion dollars. 369 00:21:37,240 --> 00:21:39,800 I, I don't think that you'll be able to generate efficiently a 370 00:21:39,800 --> 00:21:44,880 proof of the training of Brok 4 or Brok 5 in those types of 371 00:21:44,880 --> 00:21:46,480 environments for, for a very long time. 372 00:21:46,840 --> 00:21:50,360 But I, I do believe you will be able to generate proofs of 373 00:21:50,360 --> 00:21:54,360 inference in the next 1218 months for any model that you 374 00:21:54,360 --> 00:21:55,760 want to with reasonable performance. 375 00:21:57,120 --> 00:21:58,800 And I actually don't think it's a Volt prediction. 376 00:21:58,800 --> 00:22:00,480 I think it's a rather conservative prediction. 377 00:22:01,800 --> 00:22:02,960 So what? What is the? 378 00:22:03,240 --> 00:22:09,560 What is the incentive for closed source model providers to 379 00:22:10,840 --> 00:22:15,040 implement ZK proving of their models? 380 00:22:15,040 --> 00:22:18,960 So you think this is something that will, you know, can at some 381 00:22:18,960 --> 00:22:21,840 point be included in all models or does it, you know, will, will 382 00:22:21,840 --> 00:22:24,960 it remain some sort of like a premium feature that only sort 383 00:22:24,960 --> 00:22:29,560 of enterprise and governments and military clients would have 384 00:22:29,560 --> 00:22:33,680 access to? Yeah, I mean, I think it depends 385 00:22:33,680 --> 00:22:35,240 on who wants to pay for it, right. 386 00:22:35,520 --> 00:22:38,920 The number one, the, the deciding factor of what people 387 00:22:38,920 --> 00:22:41,760 will integrate I, I generally find is, is the economics of it, 388 00:22:41,920 --> 00:22:45,040 right. And so if you are a user who is 389 00:22:45,040 --> 00:22:48,640 very privacy concerned and verifiability concerned, there's 390 00:22:48,640 --> 00:22:51,680 obviously a subset of users who are, you know, there's always 391 00:22:51,680 --> 00:22:55,320 going to be open source models that you can use that you can 392 00:22:55,320 --> 00:22:58,280 run your ZK on on top of yourself, right? 393 00:22:58,560 --> 00:23:02,040 You can use DeepSeek with ZK proofs at some point in the 394 00:23:02,040 --> 00:23:06,040 reasonable future and you know, have privacy private guarantees 395 00:23:06,040 --> 00:23:08,080 of works. And that's great, that's 396 00:23:08,080 --> 00:23:10,400 exciting. And then there are applications 397 00:23:10,400 --> 00:23:13,320 where you're like, OK, I want Rock to be used for defense 398 00:23:13,320 --> 00:23:16,440 purposes. And how do I know that a remote 399 00:23:16,440 --> 00:23:20,440 system that's communicating with XAI servers, it hasn't been 400 00:23:20,440 --> 00:23:22,720 tampered with? How do I know that nobody in the 401 00:23:22,720 --> 00:23:25,560 back end at XAI has pushed to change the code that's going to 402 00:23:25,560 --> 00:23:28,560 take down, you know, an entire fleet of US defense rooms, 403 00:23:28,800 --> 00:23:30,560 right? That's a situation where you 404 00:23:30,560 --> 00:23:33,040 really, really do need verifiability. 405 00:23:33,400 --> 00:23:36,880 And there's no shortage of money that will be willing to be paid 406 00:23:36,920 --> 00:23:39,080 for that. The great thing about ZK, and we 407 00:23:39,080 --> 00:23:42,200 cushioned this into privacy earlier, it's actually very well 408 00:23:42,200 --> 00:23:45,640 suited for closed source models because you can keep the model 409 00:23:45,640 --> 00:23:48,440 private. So I can prove to you that a 410 00:23:48,440 --> 00:23:52,440 commitment to the correct model was used to generate this 411 00:23:52,440 --> 00:23:56,200 inference output for you without actually having to ever show you 412 00:23:56,200 --> 00:23:58,160 the model. So you can have a commitment to 413 00:23:58,160 --> 00:24:00,680 Grok that just says, hey, this is Grok and here's a proof it 414 00:24:00,680 --> 00:24:03,440 came from Brok. And you never have to actually 415 00:24:03,440 --> 00:24:06,440 see the weights to bias use the model architecture, anything of 416 00:24:06,440 --> 00:24:09,800 the closed source model. And so it's very, very relevant 417 00:24:09,800 --> 00:24:12,320 to use Zcane enterprise applications because you 418 00:24:12,400 --> 00:24:17,400 actually can have guarantees of correctness over AI output and 419 00:24:17,400 --> 00:24:19,960 privacy over the underlying models that are kept closed 420 00:24:19,960 --> 00:24:23,080 source. Yeah, this has made me think 421 00:24:23,080 --> 00:24:27,000 like I recently finished reading Nexus, the You've All Known 422 00:24:27,000 --> 00:24:30,520 Harare book. And you know, part of his thesis 423 00:24:30,520 --> 00:24:35,800 is that AI poses a risk to democracy in its current form. 424 00:24:35,840 --> 00:24:39,200 And the way that AI is being used like on social media, that 425 00:24:39,200 --> 00:24:46,560 could like create sort of like misinformation and can be used 426 00:24:46,560 --> 00:24:51,160 adversarially by, by our, by our, by our enemies to create 427 00:24:51,160 --> 00:24:54,760 social unrest. And, you know, I, I think ZK 428 00:24:55,160 --> 00:24:59,640 could be in, in, in this context, could be used to curb 429 00:24:59,640 --> 00:25:02,880 some of that, but it would have to be sort of a regulatory 430 00:25:02,880 --> 00:25:08,960 requirement for AI companies to also include ZK proofs for all 431 00:25:08,960 --> 00:25:10,800 of the inference. So that, you know, when you're 432 00:25:10,800 --> 00:25:13,120 looking at a social media post, you know, that this is like an 433 00:25:13,120 --> 00:25:15,040 AI generated thing versus something that's not. 434 00:25:16,200 --> 00:25:17,640 Have you guys given any thought to that? 435 00:25:17,640 --> 00:25:21,720 And what's your view on like having ZK being sort of part of 436 00:25:21,720 --> 00:25:26,560 the AI stack from a, from a sort of like a regulatory 437 00:25:26,560 --> 00:25:29,560 perspective? Yeah. 438 00:25:29,880 --> 00:25:32,080 I mean, I think it, I think there will be increasing 439 00:25:32,080 --> 00:25:36,200 regulation surrounding AI trust and safety all as well as the 440 00:25:36,200 --> 00:25:38,320 trust and safety of data used to train AI. 441 00:25:38,720 --> 00:25:41,920 Some of that, those problems can be addressed with ZK and I'd 442 00:25:41,920 --> 00:25:43,280 like to see them addressed with ZK. 443 00:25:43,560 --> 00:25:45,280 And some of those problems can't be right. 444 00:25:45,520 --> 00:25:48,400 Things like, you know, preventing AI providers from 445 00:25:48,400 --> 00:25:52,320 scraping private user data and using user chats to train next 446 00:25:52,320 --> 00:25:55,400 generation of models that, you know, has potential actually 447 00:25:55,400 --> 00:25:59,440 generative capacities that were predicated on non public 448 00:25:59,440 --> 00:26:01,600 information or sensitive information that they shouldn't 449 00:26:01,600 --> 00:26:04,320 have had access to a training. Those things are always going to 450 00:26:04,320 --> 00:26:07,440 be concerns and they require regulation and ZK proofs. 451 00:26:07,760 --> 00:26:10,040 You know, maybe in some architecture could solve it, but 452 00:26:10,040 --> 00:26:13,240 it would be very, very complex. And the simplest answer is just, 453 00:26:13,240 --> 00:26:17,440 you know, having somebody with a clipboard run after the 10 454 00:26:17,440 --> 00:26:19,360 companies that are actually doing this and pointing at them 455 00:26:19,360 --> 00:26:21,520 and saying stop doing that. That's probably the cheapest way 456 00:26:21,520 --> 00:26:24,080 to solve that. Maybe not the most durable long 457 00:26:24,080 --> 00:26:25,640 term, but probably in the short term the cheapest. 458 00:26:26,920 --> 00:26:30,600 Where I think the ZK is uniquely positioned is in applications 459 00:26:30,600 --> 00:26:33,760 that actually have an imperative that is not established by a 460 00:26:33,760 --> 00:26:38,880 government, but is established by an economic motivation to use 461 00:26:38,880 --> 00:26:41,320 ZK. And this is where I think the 462 00:26:41,320 --> 00:26:43,480 most value in technology comes from, right? 463 00:26:44,040 --> 00:26:47,400 You know, why do we have the centralized systems and block 464 00:26:47,400 --> 00:26:48,720 chains? It wasn't because, you know, 465 00:26:48,720 --> 00:26:51,160 some government bureaucrats said you have it, you have to have 466 00:26:51,160 --> 00:26:52,480 it. It's because there was an 467 00:26:52,480 --> 00:26:55,640 economic motivation to build the centralized block chains to 468 00:26:55,640 --> 00:26:59,880 protect, you know, non custodial user assets and that that was 469 00:26:59,880 --> 00:27:01,480 the entire basis of our industry. 470 00:27:01,640 --> 00:27:03,640 What was the basis of financing for ZK? 471 00:27:03,640 --> 00:27:07,440 Wasn't the, you know, government bureaucrats saying, hey, you 472 00:27:07,440 --> 00:27:12,080 know, you should build private and verifiable scalability. 473 00:27:12,080 --> 00:27:14,840 It's because there was massive hacks in crypto and then people 474 00:27:14,840 --> 00:27:17,000 go, hey, maybe we should scale block chains in a more secure 475 00:27:17,000 --> 00:27:19,320 way so we stopped losing our money, right? 476 00:27:19,520 --> 00:27:23,800 And so where there is a, where there's a market for ZK in AI is 477 00:27:23,800 --> 00:27:27,520 applications that cannot actually even use AI in the 478 00:27:27,520 --> 00:27:31,680 current form because there's lack of safety and there's lack 479 00:27:31,680 --> 00:27:34,440 of privacy over it, right? Healthcare is an example of 480 00:27:34,440 --> 00:27:36,840 that, Aerospace defense an example of that, institutional 481 00:27:36,840 --> 00:27:38,840 finances, an example of that, right? 482 00:27:39,080 --> 00:27:41,480 Like there's a bunch of companies that can't use grok 483 00:27:41,880 --> 00:27:45,240 because they can't just pass, you know, insurance participant 484 00:27:45,240 --> 00:27:48,520 data over to X AI. And there's, you know, a team of 485 00:27:48,520 --> 00:27:50,160 lawyers there say, no, you can't do that. 486 00:27:50,160 --> 00:27:51,720 We're going to go to jail or we're going to get sued to 487 00:27:51,720 --> 00:27:53,680 oblivion. So these are the places where 488 00:27:53,680 --> 00:27:58,120 there's an actually a very large market for, for ZK in AI as well 489 00:27:58,120 --> 00:27:59,600 as like actually in crypto, right? 490 00:27:59,800 --> 00:28:03,080 How do you ensure that the, you know, agentics LLM you're using 491 00:28:03,080 --> 00:28:04,960 to construct your transactions won't rug you? 492 00:28:05,880 --> 00:28:08,600 These are these are where it's very, very valuable in my view. 493 00:28:08,840 --> 00:28:11,160 And I hope there's regulation that also pushes things in our 494 00:28:11,160 --> 00:28:13,240 favor. But I don't think those are the 495 00:28:13,240 --> 00:28:16,160 driving motivations that that is going to transform this 496 00:28:16,160 --> 00:28:18,800 industry. Yeah, can can you talk a little 497 00:28:18,800 --> 00:28:21,200 bit about your your collaboration with NVIDIA? 498 00:28:22,920 --> 00:28:24,960 Yeah. So, yeah, we recently announced 499 00:28:24,960 --> 00:28:27,200 some really big collaborations, one of them with NVIDIA, one of 500 00:28:27,200 --> 00:28:31,600 them with Intel and one of them with a very large hyperscaler 501 00:28:31,600 --> 00:28:35,600 cloud provider. And so in all of these, the kind 502 00:28:35,600 --> 00:28:37,200 of the central point is very simple. 503 00:28:37,200 --> 00:28:42,240 It is there is a imperative on the use of AI and confidential 504 00:28:42,240 --> 00:28:44,920 AI within a bunch of sectors these companies sell to. 505 00:28:45,360 --> 00:28:49,080 And so there has not been a company before LaGrange that has 506 00:28:49,080 --> 00:28:54,120 had a commercially viable product that has the capacity to 507 00:28:54,240 --> 00:28:56,280 actually be able to start addressing these problems. 508 00:28:56,680 --> 00:28:59,200 Now, I wouldn't claim that the version of the proof we have now 509 00:28:59,200 --> 00:29:01,760 is the version of the proof that we'll have in 12 months, 18 510 00:29:01,760 --> 00:29:05,440 months, 24 months. But directionally, it is moving 511 00:29:05,440 --> 00:29:07,840 faster than anything has been able to move previously to 512 00:29:07,840 --> 00:29:10,840 address these problems. And that's opened up a lot of 513 00:29:10,840 --> 00:29:13,520 opportunities to us commercially to actually be able to work with 514 00:29:13,520 --> 00:29:17,200 some very, very large AI companies to start exploring 515 00:29:17,200 --> 00:29:23,360 what it looks like to use AI to improve trust and safety of of 516 00:29:23,480 --> 00:29:27,680 deployments that they have. All of these AI companies have 517 00:29:27,800 --> 00:29:31,000 healthcare, defense, institutional finance, relevant 518 00:29:31,000 --> 00:29:33,560 contracts, kind of services, relevant contracts. 519 00:29:33,720 --> 00:29:36,360 They have international contracts, you know, very 520 00:29:36,360 --> 00:29:39,240 complex legal requirements surrounding how data can be 521 00:29:39,240 --> 00:29:41,960 transited between countries and how AI can use between 522 00:29:41,960 --> 00:29:45,800 countries. And what we have is a technology 523 00:29:45,800 --> 00:29:49,120 that's uniquely positioned to address many of those problems. 524 00:29:50,880 --> 00:29:52,680 LaGrange has recently launched its token. 525 00:29:52,680 --> 00:29:59,720 It's the LAW token or the LA token and LA. 526 00:30:00,440 --> 00:30:01,960 So yeah, it's got the finance listing. 527 00:30:01,960 --> 00:30:08,240 Coinbase, what's the role of the LA token and what's planned here 528 00:30:08,240 --> 00:30:09,960 for like staking and governance, etcetera? 529 00:30:12,000 --> 00:30:15,400 Yeah, So, you know, we were very, very excited to finally be 530 00:30:15,400 --> 00:30:17,600 able to unveil and to launch the LA Token. 531 00:30:17,600 --> 00:30:21,560 The LaGrange Foundation did a fantastic job orchestrating and 532 00:30:21,560 --> 00:30:26,440 coordinating that whole process. And so, you know, we were very 533 00:30:26,440 --> 00:30:30,640 lucky as well to be listed on a variety of top liquidity venues, 534 00:30:30,640 --> 00:30:34,480 Binance, Coinbase, Upbeat, and many others. 535 00:30:35,800 --> 00:30:39,040 And we were very excited to see an overwhelming community 536 00:30:39,040 --> 00:30:41,160 support behind the launch of the token. 537 00:30:41,480 --> 00:30:44,400 The utility of a token as designed by the LaGrange 538 00:30:44,400 --> 00:30:49,000 Foundation is as a fuel for the cryptographic engine that 539 00:30:49,000 --> 00:30:54,360 LaGrange builds effectively. There is a network of provers 540 00:30:54,360 --> 00:30:58,440 that generate proofs for Deep proof, RCK, machine learning, as 541 00:30:58,440 --> 00:31:00,400 well as a bunch of other commercial applications we 542 00:31:00,400 --> 00:31:02,880 target as well, ranging from roll ups to Co processing to 543 00:31:02,880 --> 00:31:05,560 more as well as verifiable database infrastructure. 544 00:31:06,560 --> 00:31:10,600 And at the end of the day, the token is used and staked into 545 00:31:10,600 --> 00:31:13,920 individual provers in the network who have an economic 546 00:31:13,920 --> 00:31:15,960 motivation to generate proofs correctly. 547 00:31:16,680 --> 00:31:19,600 If, for example, they don't generate a proof on time, or 548 00:31:19,600 --> 00:31:21,720 they failed to participate in an auction the way they were 549 00:31:21,720 --> 00:31:25,880 supposed to, they can face a penalty in the form of slashing 550 00:31:25,880 --> 00:31:28,600 or non payment. In the current version, you know 551 00:31:28,600 --> 00:31:33,080 if it's possible to stake the LA token into provers and there is 552 00:31:33,400 --> 00:31:35,400 programs designed by the LaGrange Foundation to 553 00:31:35,400 --> 00:31:39,600 incentivize the staking of LA tokens based on fees that the 554 00:31:40,480 --> 00:31:44,800 network collects from being able to render inference or render 555 00:31:44,800 --> 00:31:48,080 proofs of inference to many of our counterparties. 556 00:31:49,520 --> 00:31:54,680 And you, you tweeted something a little while back, which which I 557 00:31:54,680 --> 00:31:56,160 thought was was kind of interesting. 558 00:31:56,160 --> 00:31:58,880 You said if your intra protocol has no revenue, it's just a meme 559 00:31:58,880 --> 00:32:03,080 coin. Can can you unpack this, this 560 00:32:03,080 --> 00:32:06,400 thought and you know why? Why do you think? 561 00:32:06,400 --> 00:32:10,040 I mean, I think it's it's it's obvious, right that the crypto 562 00:32:10,040 --> 00:32:13,480 needs to move to more towards a more revenue generating model 563 00:32:13,480 --> 00:32:15,640 than a simply like up only model. 564 00:32:17,520 --> 00:32:19,240 How? How will revenues flow back to 565 00:32:19,240 --> 00:32:21,960 token holders in the case of the LA token? 566 00:32:23,480 --> 00:32:26,320 Yeah, this is a great question and I'm glad you asked it 567 00:32:26,320 --> 00:32:27,720 because that was one of my favorite tweets. 568 00:32:28,000 --> 00:32:34,200 But there's a subset of of public market participants in 569 00:32:34,200 --> 00:32:39,040 crypto who trade charts. And all they do is they trade 570 00:32:39,040 --> 00:32:41,880 listings and charts. And those listing and chart 571 00:32:41,880 --> 00:32:47,120 traders are behaving the same way when they're trading banc 572 00:32:47,480 --> 00:32:50,240 versus, when they're trading Pengu versus, when they're 573 00:32:50,240 --> 00:32:53,240 trading with versus, when they're trading DOGE versus, 574 00:32:53,240 --> 00:32:57,320 when they're trading LA. All they care about is trading 575 00:32:57,320 --> 00:33:01,000 on price action and trying to catch a runner or momentum in 576 00:33:01,000 --> 00:33:04,480 the chart. And if the only participants in 577 00:33:04,480 --> 00:33:08,280 your market are trading on those characteristics, whether or not 578 00:33:08,280 --> 00:33:12,080 you are in for protocol or you are a meme coin, you effectively 579 00:33:12,080 --> 00:33:14,160 have converged the same market dynamics. 580 00:33:14,160 --> 00:33:17,760 The meme coin right people trade goes up, people sell goes down 581 00:33:18,680 --> 00:33:23,080 and that is really not an inspiring or long term durable 582 00:33:23,080 --> 00:33:26,440 way to build a infrastructure protocol. 583 00:33:26,800 --> 00:33:29,000 The objective of an infrastructure protocol should 584 00:33:29,000 --> 00:33:33,560 be to create net new value such that the economics broadly of 585 00:33:33,560 --> 00:33:37,920 that infrastructure protocol are creative to the network dynamics 586 00:33:38,160 --> 00:33:40,080 that include the underlying asset. 587 00:33:41,440 --> 00:33:44,640 And that is for example, why hype has done so well in market. 588 00:33:44,880 --> 00:33:49,320 That is for example why you know the many other L ones that have 589 00:33:49,320 --> 00:33:52,600 high demand Solana, Ethereum have done well by and large in 590 00:33:52,600 --> 00:33:54,760 market. It is the hope that many 591 00:33:54,760 --> 00:33:58,720 investors have brought to something like pump in the last, 592 00:33:59,360 --> 00:34:03,720 you know 30 days. But anyway, to get back to the 593 00:34:03,720 --> 00:34:06,200 point, if you do not have revenue and you do not have 594 00:34:06,200 --> 00:34:09,000 traction, your token is nothing more than the meeting point. 595 00:34:09,520 --> 00:34:12,120 And so LaGrange, as you know, we've talked a lot about today 596 00:34:12,280 --> 00:34:16,440 has material traction both outside of crypto and within 597 00:34:16,440 --> 00:34:20,440 crypto in the adoption of our technology in both enterprise, 598 00:34:20,440 --> 00:34:24,920 AI, financial services, aerospace, defense and crypto 599 00:34:24,920 --> 00:34:28,840 asset sectors. And because of that, we've tried 600 00:34:28,840 --> 00:34:32,400 to design our network in a way where the fees that accrue from 601 00:34:32,400 --> 00:34:35,560 the generation of proofs and from agreements that we have for 602 00:34:35,560 --> 00:34:38,120 the generation of proofs agreement, the foundation has 603 00:34:38,120 --> 00:34:41,840 for the generation of proofs accrue back to people who have 604 00:34:41,840 --> 00:34:46,800 staked and who are generating proofs within our network. 605 00:34:47,800 --> 00:34:51,400 And so This is why we, you know, we're very excited with with 606 00:34:51,400 --> 00:34:53,560 many of the traction numbers that we have right now that are 607 00:34:53,639 --> 00:34:57,880 are publicly verifiable on chain wherein you can see the movement 608 00:34:57,880 --> 00:35:03,480 of fees for the generation of proofs to prove as a network and 609 00:35:03,520 --> 00:35:06,120 very strong demand for the generation of proofs that's 610 00:35:06,120 --> 00:35:09,560 visible in the network. And so long term, we think that 611 00:35:09,680 --> 00:35:15,320 the majority of fees that accrue for staking the LA token will 612 00:35:15,320 --> 00:35:18,360 come from fees that are paid directly for the generation of 613 00:35:18,360 --> 00:35:22,560 proofs such that is a positive economic market wherein there is 614 00:35:22,560 --> 00:35:27,640 an incentive to hold and stake the LA token into the network 615 00:35:28,600 --> 00:35:32,880 that isn't simply just trading chart action and isn't simply 616 00:35:32,880 --> 00:35:38,040 just trading on a meme coin. What So when when operating in 617 00:35:38,040 --> 00:35:44,720 the enterprise space and selling LaGrange products to enterprise 618 00:35:44,720 --> 00:35:49,760 customers, how is the crypto component perceived and how do 619 00:35:49,760 --> 00:35:51,560 you get over some of the objections that people might 620 00:35:51,560 --> 00:35:55,520 have simply by virtue of like Larache having a crypto 621 00:35:55,520 --> 00:35:57,800 component? You know, some companies or like 622 00:35:57,800 --> 00:35:59,560 clients might see that as a risk. 623 00:36:00,040 --> 00:36:03,280 And, you know, I know that like working with an impressed 624 00:36:03,280 --> 00:36:06,520 clients, it could be complicated to disassociate, you know, the 625 00:36:06,520 --> 00:36:12,400 technology from a lot of the negative press that crypto gets. 626 00:36:13,120 --> 00:36:16,480 Yeah. Yeah, that's a great question. 627 00:36:16,480 --> 00:36:19,880 So Deep proof is a library. Our ZK machine learning 628 00:36:19,880 --> 00:36:22,920 technology is a library. You could run it on top of our 629 00:36:22,920 --> 00:36:25,640 Prover network with the same security guarantees as you 630 00:36:25,640 --> 00:36:29,280 running it on top of an edge device used in a battlefield. 631 00:36:29,600 --> 00:36:33,920 There is no difference in where you choose to operate that 632 00:36:33,920 --> 00:36:36,160 library. The library will operate with 633 00:36:36,160 --> 00:36:39,320 the same safety guarantees over proof generation anyway. 634 00:36:40,800 --> 00:36:44,280 And so some people really like the centralized proof 635 00:36:44,280 --> 00:36:46,840 generation, but they go and they seek that out. 636 00:36:47,280 --> 00:36:50,120 So when we work with enterprise clients, we don't sell them on 637 00:36:50,120 --> 00:36:53,120 the centralized proof generation, we sell them on core 638 00:36:53,120 --> 00:36:57,280 cryptography. The entirety of the Internet has 639 00:36:57,280 --> 00:36:59,880 been secured with cryptography, right? 640 00:37:00,280 --> 00:37:05,400 The, the, so TLS on top of HTTP is what enables online banking. 641 00:37:05,400 --> 00:37:07,320 It's what enables payments infrastructure. 642 00:37:07,560 --> 00:37:11,800 It's what enables, you know, everything you do on your phone. 643 00:37:11,800 --> 00:37:13,360 That's what enables social media. 644 00:37:13,360 --> 00:37:15,520 It's one enables online dating at what it is. 645 00:37:15,720 --> 00:37:19,840 It's the modern society that we have today is predicated on the 646 00:37:19,880 --> 00:37:23,360 use of cryptography to add safety and privacy on top of web 647 00:37:23,360 --> 00:37:28,320 connections. What LaGrange does with ZKML is 648 00:37:28,320 --> 00:37:32,680 adding those same 2 properties, safety and privacy on top of AI 649 00:37:33,720 --> 00:37:37,240 and that is what we sell when we interact with enterprise 650 00:37:37,240 --> 00:37:38,760 clients, Web 2 customers, etcetera. 651 00:37:39,200 --> 00:37:43,720 It is 2 properties that unambiguously need to be 652 00:37:43,720 --> 00:37:47,480 included on top of AI for us to have a robust and functioning 653 00:37:47,480 --> 00:37:50,320 and safe economy that predicates itself on top of AI. 654 00:37:50,560 --> 00:37:52,600 The same way those two properties had to be added on 655 00:37:52,600 --> 00:37:56,560 top of, you know, ICT Internet connectivity technology to be 656 00:37:56,560 --> 00:37:59,880 able to add those properties on top of the web. 657 00:38:00,600 --> 00:38:05,360 And so that is what Deep Proof and RZKML work is sold as and 658 00:38:05,360 --> 00:38:07,920 what it sells. Now there is a subset of 659 00:38:07,920 --> 00:38:10,120 customers, right? Wallet providers, for example, 660 00:38:10,760 --> 00:38:13,560 or people who really like the centralized proof generation 661 00:38:13,560 --> 00:38:16,000 because they think the properties you get over liveness 662 00:38:16,040 --> 00:38:19,360 guarantees, remove dependencies on cloud providers who might 663 00:38:19,720 --> 00:38:23,160 shut off, right? And in that case, you know, we 664 00:38:23,160 --> 00:38:25,600 have a prover network that's fantastic and it can be used for 665 00:38:25,600 --> 00:38:30,200 that. But when we sell to web two, we 666 00:38:30,200 --> 00:38:33,360 don't sell the crypto token. We don't sell, you know, people 667 00:38:33,360 --> 00:38:36,080 having to use or interact in any way with the crypto token. 668 00:38:36,240 --> 00:38:39,000 We sell core technology and impactful technology. 669 00:38:39,520 --> 00:38:44,040 And as a business, we have also our crypto network, which we 670 00:38:44,040 --> 00:38:46,720 think probably is the best way to generate proofs long term. 671 00:38:46,720 --> 00:38:49,120 We think the whole world's going to use the centralized proof 672 00:38:49,120 --> 00:38:52,480 generation long term, but we want to see people using proof 673 00:38:52,480 --> 00:38:55,760 generation 1st and then they'll eventually, in our view, start 674 00:38:55,760 --> 00:38:57,120 moving to the centralized deployments. 675 00:38:59,000 --> 00:39:02,200 So, so far we've talked a lot about AI, but you guys are also 676 00:39:02,200 --> 00:39:06,120 doing a lot of interesting work on on the scaling side, 677 00:39:06,880 --> 00:39:08,760 particularly like those recently announced. 678 00:39:08,760 --> 00:39:11,400 So you guys are working with Matter Labs to handle a lot of 679 00:39:11,400 --> 00:39:16,040 the proofs on on, on the on ZK sync. 680 00:39:16,040 --> 00:39:19,160 Can you talk a little bit more about like the Co processor and 681 00:39:19,160 --> 00:39:22,640 and some of the other products that are in the LaGrange product 682 00:39:22,640 --> 00:39:24,680 line? Yeah. 683 00:39:25,040 --> 00:39:27,760 So it's a really good question. So as I started with a little 684 00:39:27,760 --> 00:39:31,240 bit today we're AZK company and where we see the largest Tam for 685 00:39:31,240 --> 00:39:35,040 ZK today is on adding trust and safety on top of the use of AI. 686 00:39:35,920 --> 00:39:39,000 But that's not the only thing that effectively we sell that 687 00:39:39,000 --> 00:39:42,040 uses ZK, right. So we we sell verifiable 688 00:39:42,040 --> 00:39:45,280 database infrastructure where you effectively are able to have 689 00:39:45,280 --> 00:39:49,240 a database that is represented by a commitment to that data. 690 00:39:49,240 --> 00:39:52,080 It's like a hash of all of that data that we can prove the 691 00:39:52,080 --> 00:39:55,280 correctness of queries on top of which is very useful for a lot 692 00:39:55,280 --> 00:39:58,880 of contexts where you want correct provenance over data. 693 00:39:59,120 --> 00:40:01,560 It's very useful if you want to introspect into a chain and 694 00:40:01,560 --> 00:40:04,920 query over the history of chain. And we have a very large market 695 00:40:04,920 --> 00:40:08,200 for that that we we sell to within D Phi and NFT protocols. 696 00:40:08,200 --> 00:40:11,320 Many of those we've announced like Gearbox, Azuki, etcetera. 697 00:40:11,920 --> 00:40:15,520 We also have work that we've done on using our Prover network 698 00:40:15,520 --> 00:40:17,960 to generate proofs for roll ups. Right. 699 00:40:17,960 --> 00:40:20,120 And so we have a very large deal that we've signed and we've 700 00:40:20,120 --> 00:40:23,800 announced Matter Labs, we're up to 75% of Matter Labs proof 701 00:40:23,800 --> 00:40:25,880 generation for the next two years to be done on La Branche. 702 00:40:26,560 --> 00:40:29,160 And for us that's a very exciting market opportunity. 703 00:40:29,160 --> 00:40:32,000 We we think the Matter Labs team and the ZK sync ecosystem is, 704 00:40:32,280 --> 00:40:35,640 you know, one of the, the, the, the largest and one of the most 705 00:40:36,040 --> 00:40:40,280 important ZK roll up ecosystems in crypto and we love being a 706 00:40:40,280 --> 00:40:42,560 part of supporting them in their growth ambitions. 707 00:40:44,600 --> 00:40:48,080 So just switching gears a little bit, I want to, I want to ask 708 00:40:48,080 --> 00:40:51,960 you some questions about, about your personal journey as a, as a 709 00:40:51,960 --> 00:40:57,560 founder and you know, what's the, what's the thing that 710 00:40:57,920 --> 00:41:01,240 you're the most proud of at LaGrange, but that most people 711 00:41:01,240 --> 00:41:02,760 either don't know or don't care about? 712 00:41:05,200 --> 00:41:09,440 Yeah, I think that there is a fallacy in founding companies 713 00:41:09,440 --> 00:41:12,880 that the journey to be successful is linear and that 714 00:41:12,880 --> 00:41:15,840 you catch lightning in a bottle, you become successful and then 715 00:41:15,840 --> 00:41:17,840 all of a sudden you're off to the races and everything goes 716 00:41:17,840 --> 00:41:21,000 great. The truth is that at LaGrange, 717 00:41:21,000 --> 00:41:24,960 we've had very many periods where things were going our way 718 00:41:24,960 --> 00:41:27,360 and very many periods with things weren't going our way. 719 00:41:27,800 --> 00:41:30,840 And the resilience of the company and the team is the 720 00:41:30,840 --> 00:41:33,640 thing that has allowed us to continue to XLS a business. 721 00:41:34,600 --> 00:41:36,880 And that's the thing that we're that I'm the most proud of about 722 00:41:36,880 --> 00:41:40,800 our business. You see a lot of companies in 723 00:41:40,800 --> 00:41:44,800 crypto that they come up with a cool idea, they raise a big 724 00:41:44,800 --> 00:41:47,840 round, they launch a token, things don't go their way and 725 00:41:47,840 --> 00:41:50,560 they go to zero and then the team goes on to the next thing. 726 00:41:50,920 --> 00:41:52,920 Or you see companies that you know, they had come up with a 727 00:41:52,920 --> 00:41:55,240 great idea. They raise a first round. 728 00:41:55,640 --> 00:41:58,880 They, you know, everything's very exciting for them. 729 00:41:59,040 --> 00:42:01,320 They never end up catching that momentum again. 730 00:42:01,320 --> 00:42:03,320 They never raise a second round. Nothing ever happens. 731 00:42:04,000 --> 00:42:07,080 We are. I have to spend significant time 732 00:42:07,080 --> 00:42:10,120 raising my first round. People don't know this. 733 00:42:10,360 --> 00:42:14,040 I actually failed to raise my first seed round twice before 734 00:42:14,040 --> 00:42:15,720 the third time that I succeeded on it. 735 00:42:16,600 --> 00:42:19,600 We had many periods in the history of LaGrange where you 736 00:42:19,600 --> 00:42:21,800 know the market was swinging away from ZK. 737 00:42:22,040 --> 00:42:24,680 People weren't excited about Co processing, people weren't 738 00:42:24,680 --> 00:42:28,320 excited about roll up proving and consistently what the 739 00:42:28,320 --> 00:42:30,720 research team at LaGrange has done, the engineering team, the 740 00:42:30,720 --> 00:42:34,800 business team, everyone was stick to the fundamentals that 741 00:42:34,800 --> 00:42:37,600 we believe works, which is building technology that our 742 00:42:37,600 --> 00:42:40,600 customers love and then aggressively commercializing 743 00:42:40,600 --> 00:42:44,600 those into large Tam verticals. And through that strategy, we 744 00:42:44,600 --> 00:42:48,640 have been able to weather very many bad periods and get to very 745 00:42:48,640 --> 00:42:52,200 many very positive periods. And that's a resilience that I 746 00:42:52,200 --> 00:42:55,040 think too few companies in crypto prioritize. 747 00:42:55,040 --> 00:42:58,160 They prioritize the fast exit, the hot trade, the cool 748 00:42:58,160 --> 00:43:01,400 narrative, and they don't build aggressively on a fundamental 749 00:43:01,600 --> 00:43:05,040 that carries them through both bear and in the bull markets. 750 00:43:06,400 --> 00:43:11,080 Yeah, I think fundamentals are highly, highly underrated in 751 00:43:11,080 --> 00:43:13,920 crypto. I mean it, it's, it seems so 752 00:43:13,920 --> 00:43:18,280 obvious, right? By like more and more I'm, I'm 753 00:43:18,280 --> 00:43:24,280 finding that the, the, the thing that sets high performing teams 754 00:43:24,760 --> 00:43:27,360 and successful teams apart from the rest is just, you know, 755 00:43:27,360 --> 00:43:30,200 fundamentals and, and 1st principles thinking. 756 00:43:31,400 --> 00:43:34,840 You know, you, you talked about the team and, and how it's grown 757 00:43:34,840 --> 00:43:38,440 and everything. And what what is a a piece of 758 00:43:38,440 --> 00:43:41,000 advice that you would give to aspiring crypto founders that 759 00:43:41,000 --> 00:43:46,120 are building a great team, that want to build a great team for 760 00:43:46,120 --> 00:43:48,720 the long term? Yeah. 761 00:43:49,040 --> 00:43:51,920 I mean, the only way I think to be successful is to hire the 762 00:43:51,920 --> 00:43:55,560 best people, especially in a very research oriented sector, 763 00:43:55,840 --> 00:44:00,920 you need to go out preemptively and find the best people to work 764 00:44:00,920 --> 00:44:05,560 with, hire them and then be able to retain them. 765 00:44:06,280 --> 00:44:10,240 And so a lot of the early hiring at LaGrange, not even early 766 00:44:10,240 --> 00:44:14,480 hiring, a lot of the hiring until today is done by me for a 767 00:44:14,480 --> 00:44:16,320 lot of the research sectors, right? 768 00:44:16,400 --> 00:44:19,480 I've run all of the, the interview processes for anyone 769 00:44:19,480 --> 00:44:23,080 who's interviewed with LaGrange for the first three years of our 770 00:44:23,080 --> 00:44:24,440 history, the first person I met was me. 771 00:44:26,200 --> 00:44:29,320 I, I took a very high amount of ownership in trying to run the 772 00:44:29,320 --> 00:44:32,680 interview process the way that I thought had to be run to attract 773 00:44:32,680 --> 00:44:34,760 the best talent. And because of that, we were 774 00:44:34,760 --> 00:44:37,520 able to get a lot of very, very, very good talent. 775 00:44:39,480 --> 00:44:41,080 And now as we've grown, we've changed processes. 776 00:44:41,080 --> 00:44:42,720 There's some roles I interview for, some roles I don't 777 00:44:42,720 --> 00:44:45,120 interview for. But for anyone who's starting 778 00:44:45,120 --> 00:44:48,280 out, I, I, I would recommend that they take as much ownership 779 00:44:48,280 --> 00:44:51,720 and as possible on trying to run their interview and hiring 780 00:44:51,720 --> 00:44:55,160 process and then do as much work as possible in one on ones. 781 00:44:55,160 --> 00:44:57,560 I, I have one on ones with everyone on the team at LeBron 782 00:44:57,560 --> 00:45:01,720 still at a very regular cadence. We like to keep our team small. 783 00:45:01,720 --> 00:45:06,440 We like to make sure that we we have offer packages that are 784 00:45:06,440 --> 00:45:08,400 competitive with the best companies in the space and we've 785 00:45:08,400 --> 00:45:11,920 done that since day one. And we make sure that people who 786 00:45:11,920 --> 00:45:15,560 join LaGrange have a very, very, very high retention rate when 787 00:45:15,560 --> 00:45:19,200 they're at the company as well. What, what stands out in your 788 00:45:19,200 --> 00:45:22,520 interview process do you think from other teams? 789 00:45:22,520 --> 00:45:25,360 Like what's the, what's the one thing in your interview process 790 00:45:25,360 --> 00:45:31,200 that, that, that, that stands it out as a a great way to find the 791 00:45:31,200 --> 00:45:35,320 best talent? So I'll give these secrets 792 00:45:35,320 --> 00:45:38,160 because because obviously we, you know, I think, I think, I 793 00:45:38,160 --> 00:45:41,600 think founders should know this, but early on I wouldn't have 794 00:45:41,600 --> 00:45:45,560 shared these secrets. But the one that really was 795 00:45:45,560 --> 00:45:48,400 helpful was I, I was the first person who everyone would talk 796 00:45:48,400 --> 00:45:50,280 to. So when someone is interviewing 797 00:45:50,280 --> 00:45:53,360 with the company and the founder comes on for the first interview 798 00:45:53,360 --> 00:45:55,640 and says this is a one-on-one interview with me. 799 00:45:55,760 --> 00:45:58,040 And this role is so important to us and the role that you're 800 00:45:58,040 --> 00:45:59,520 interviewing for is so important to us. 801 00:45:59,760 --> 00:46:01,800 You will directly interact with me throughout this entire 802 00:46:01,800 --> 00:46:03,040 process and I'll guide you through it. 803 00:46:03,720 --> 00:46:07,080 Generally, people who are top of line and are trying to take a 804 00:46:07,080 --> 00:46:10,640 bet on an earlier stage company enjoy that level and appreciate 805 00:46:10,640 --> 00:46:14,360 that level of attention. Secondly, when we were competing 806 00:46:14,360 --> 00:46:17,840 against larger companies for very, very good talent, I would 807 00:46:17,840 --> 00:46:20,520 fly out to the city of that person who we made the offer to, 808 00:46:20,520 --> 00:46:23,240 to meet them in person in as part of making that offer. 809 00:46:24,120 --> 00:46:26,360 And we would spend time, we would take them to dinner. 810 00:46:26,360 --> 00:46:28,680 We would get to know them. We would get to know them 811 00:46:28,680 --> 00:46:30,160 personally. We make it very clear that if 812 00:46:30,160 --> 00:46:32,200 they were to join the bronze, they were joining a company that 813 00:46:32,200 --> 00:46:36,280 prioritized them and prioritized winning and that very few 814 00:46:36,280 --> 00:46:40,040 founders even today I see are willing to fly out and meet a 815 00:46:40,040 --> 00:46:41,800 compass someone in person who they make an offer to. 816 00:46:42,600 --> 00:46:45,120 This is one of the best ways I used to try to close deals when 817 00:46:45,120 --> 00:46:47,960 I was a venture investor, right? It's a hot company and a hot 818 00:46:48,280 --> 00:46:49,400 founder we're trying to invest in. 819 00:46:49,800 --> 00:46:51,480 Then I would fly out to meet you in person. 820 00:46:51,800 --> 00:46:54,240 And I see no difference where if you're, you know, one of the 821 00:46:54,240 --> 00:46:56,960 main differentiators you have as a founder is your talent that 822 00:46:56,960 --> 00:46:59,360 you're able to hire, that you shouldn't be doing the same 823 00:46:59,360 --> 00:47:01,280 thing. And I've told this to dozens of 824 00:47:01,280 --> 00:47:04,080 founders and secrets and none of them have done it. 825 00:47:04,440 --> 00:47:06,280 And so maybe I'll say it publicly and people will start 826 00:47:06,280 --> 00:47:09,920 doing it, but after all this time, I rarely see any founders 827 00:47:09,920 --> 00:47:14,360 doing it still. Yeah, it seems like such a 828 00:47:14,800 --> 00:47:16,480 simple thing, right? It's all about relationship 829 00:47:16,480 --> 00:47:19,200 building. And if you're competing against 830 00:47:19,280 --> 00:47:22,400 whether it's, you know, other VCs for a deal or other 831 00:47:22,400 --> 00:47:27,840 companies for talent having having that that like building 832 00:47:27,840 --> 00:47:29,960 that sort of personal relationship with that person 833 00:47:30,520 --> 00:47:34,760 early on can can make it sort of make or break the deal and like 834 00:47:34,760 --> 00:47:38,400 flying out to meet that person. It's definitely, you know, it's 835 00:47:38,440 --> 00:47:39,800 yeah, it's an effort thing. Totally. 836 00:47:39,800 --> 00:47:44,200 Yeah. And then another question like 837 00:47:45,160 --> 00:47:47,920 starting, starting a company, you know, and scaling that 838 00:47:47,920 --> 00:47:52,160 company to, you know, 10s of people can be challenging for 839 00:47:52,160 --> 00:47:53,560 some people. And there's a lot of founders, I 840 00:47:53,560 --> 00:47:58,240 think that have a hard time getting over that, that that 841 00:47:58,240 --> 00:48:05,000 sort of getting over that scale. So they may be able to run their 842 00:48:05,000 --> 00:48:07,840 company when there's a handful of people, but then it gets 843 00:48:07,840 --> 00:48:11,680 harder and then they might, they might kind of cap out and then 844 00:48:11,680 --> 00:48:15,400 you have another, another CEO sort of like come in and take 845 00:48:15,400 --> 00:48:18,320 that company to the next level. Like how do you as a founder 846 00:48:18,320 --> 00:48:21,920 think about operating at all of those different levels? 847 00:48:21,920 --> 00:48:24,800 Like if LaGrange, you know, scales now to, you know, over 848 00:48:24,800 --> 00:48:29,120 100 people you know in sometime in the future, how, how do you 849 00:48:29,120 --> 00:48:32,520 think about your role as ACEO operating at those different 850 00:48:32,520 --> 00:48:35,200 levels of scale? Yeah. 851 00:48:35,600 --> 00:48:37,600 So I, I have two answers for this. 852 00:48:37,680 --> 00:48:40,120 Firstly, I think it ties into what we talked about before. 853 00:48:40,120 --> 00:48:44,520 It's an effort thing, right? Nobody wants to spend in the 854 00:48:44,520 --> 00:48:48,520 first year of their company 6 hours to 8 hours a day 855 00:48:48,520 --> 00:48:50,600 interviewing candidates, right? Nobody wants to. 856 00:48:50,600 --> 00:48:53,680 It's a lot of work and you know, if you, if you have to also run 857 00:48:53,680 --> 00:48:55,840 your business and you want to win the best talents, you're 858 00:48:55,840 --> 00:48:57,400 interviewing everyone personally and you're flying out 859 00:48:57,400 --> 00:48:59,080 everywhere. It's a lot of work. 860 00:48:59,520 --> 00:49:02,200 And it's like, it's, it's, it's kind of a pain in the butt. 861 00:49:02,720 --> 00:49:05,680 But if you want to win, it's a decision you have to make if 862 00:49:05,680 --> 00:49:08,360 you're going to do it. And so I think there's a subset 863 00:49:08,360 --> 00:49:10,880 of founders who accept what being a founder is. 864 00:49:11,120 --> 00:49:13,480 Which is doing what's required. At any stage in the business, 865 00:49:13,480 --> 00:49:16,040 that business succeed, right, Even if you don't like it, 866 00:49:16,720 --> 00:49:18,640 right. Your, your job as a founder 867 00:49:18,640 --> 00:49:21,200 isn't to be an engineer. It isn't to be a salesperson. 868 00:49:21,200 --> 00:49:24,480 It isn't to be a tweeter, it isn't to be a head of HR. 869 00:49:24,480 --> 00:49:27,040 It's to it's to win. And every point in the business 870 00:49:27,040 --> 00:49:28,560 or something you have to do to win. 871 00:49:28,760 --> 00:49:30,840 The most important thing is going to be different at every 872 00:49:30,840 --> 00:49:34,000 stage. And early on it requires a lot 873 00:49:34,000 --> 00:49:36,040 of effort along the things we just talked about in my view, 874 00:49:36,360 --> 00:49:40,480 and later on it requires a lot of effort along different axis. 875 00:49:41,160 --> 00:49:44,800 And so, you know, I, I think as you scale your business, you 876 00:49:44,800 --> 00:49:46,560 have to accept that, that, that things change. 877 00:49:46,560 --> 00:49:50,280 And, you know, there's been a bunch of bumpy roads in my 878 00:49:50,280 --> 00:49:52,920 journey as a founder getting to the company scale that we are 879 00:49:52,920 --> 00:49:53,800 now. And there's going to be a bunch 880 00:49:53,800 --> 00:49:55,600 of other bumpy ones getting to the next scale as well. 881 00:49:55,600 --> 00:49:59,040 I'm, I'm, I'm sure of, and it just, you know, it will require 882 00:49:59,040 --> 00:50:01,440 a significant amount of effort for me from the management team, 883 00:50:01,440 --> 00:50:04,920 from the, from everyone at the company to continually hit the 884 00:50:04,920 --> 00:50:06,680 milestones we need to grow these scales. 885 00:50:08,320 --> 00:50:11,880 And, you know, I think teams should be cognizant of that. 886 00:50:12,400 --> 00:50:15,040 They should accept the reality ahead of time so that they're 887 00:50:15,080 --> 00:50:17,480 equipped to be able to tackle it when they face it in the moment. 888 00:50:19,560 --> 00:50:25,240 How do you juggle with sort of remote versus in person? 889 00:50:26,080 --> 00:50:29,440 Are are you guys mostly a remote team or do do most people sort 890 00:50:29,440 --> 00:50:32,520 of come to an office? Yeah, we, we, we're a fully 891 00:50:32,520 --> 00:50:37,840 remote team, which is a decision that we made because of our 892 00:50:37,840 --> 00:50:39,960 requirement to optimize for talent quality. 893 00:50:40,120 --> 00:50:42,680 Because we're a research organization, You know, a lot of 894 00:50:42,680 --> 00:50:44,280 our researchers are based all over the world. 895 00:50:44,280 --> 00:50:47,200 They're some of the top people who've authored and published 896 00:50:47,200 --> 00:50:49,800 papers and, and applied cryptography and computer 897 00:50:49,800 --> 00:50:52,360 science, specifically in things like ZKML and verifiable 898 00:50:52,360 --> 00:50:54,640 database design. Our chief scientist, Babis 899 00:50:54,640 --> 00:50:57,120 Papamanthu, who chairs the Cryptography department of Yale, 900 00:50:57,320 --> 00:50:59,800 Dimitrius Papadopoulos, one of our distinguished researchers, 901 00:51:00,040 --> 00:51:02,080 is a professor at HKUST in Hong Kong. 902 00:51:02,360 --> 00:51:04,840 Obviously it would be very hard based the whole company in Yale, 903 00:51:04,840 --> 00:51:10,400 in Hong Kong, you know, Nikola Gayi and Franklin the Leahy, two 904 00:51:10,400 --> 00:51:15,400 of our fantastic, Nikola Gayi is a fantastic senior researcher on 905 00:51:15,400 --> 00:51:17,640 the team and and Franklin's our head of engineering. 906 00:51:17,640 --> 00:51:20,080 They're both based in Paris. So we have clusters of people 907 00:51:20,080 --> 00:51:22,360 all over the world, but it would be very, very hard to force 908 00:51:22,360 --> 00:51:24,880 everyone to move to one city. There's personal things that 909 00:51:24,880 --> 00:51:28,120 just would be very hard. So we are a remote team. 910 00:51:28,360 --> 00:51:30,800 I think there's something, you know, very special of being in 911 00:51:30,800 --> 00:51:32,680 person team. It is very special in person 912 00:51:32,680 --> 00:51:35,640 time you get as a team, especially if your remote team 913 00:51:35,640 --> 00:51:39,680 is very little. In person time you get, but you 914 00:51:39,680 --> 00:51:41,000 know, just something you have to work around. 915 00:51:41,200 --> 00:51:45,080 We've always hired very, very, very high agency people. 916 00:51:45,920 --> 00:51:48,680 Everyone in the team has a tremendous amount of autonomy 917 00:51:50,200 --> 00:51:54,240 and that has just, you know, been very positive for some 918 00:51:54,240 --> 00:51:57,080 people who really, really enjoy that and some people don't. 919 00:51:57,080 --> 00:52:00,120 But we've been very lucky at the ones we've hired really do enjoy 920 00:52:00,120 --> 00:52:02,480 level of autonomy. They like to be able to not have 921 00:52:02,480 --> 00:52:04,800 someone, you know, breathing over their neck in an office. 922 00:52:05,360 --> 00:52:07,560 They like to be able to execute at the highest level on their 923 00:52:07,560 --> 00:52:10,480 work on their own time and then be able to contribute to a team 924 00:52:10,480 --> 00:52:13,920 also doing the same thing. And for us, we've been able to 925 00:52:13,920 --> 00:52:17,200 make it work. How often do you guys get 926 00:52:17,200 --> 00:52:19,800 together as a team? You know, do you have sort of 927 00:52:20,440 --> 00:52:24,080 quarterly retreats and how do you structure those so that you 928 00:52:24,080 --> 00:52:29,120 guys can get the most kind of out of that FaceTime during 929 00:52:29,120 --> 00:52:31,280 those moments when you see each other in person? 930 00:52:32,880 --> 00:52:36,000 Yeah, I, I think we always try to do off sites. 931 00:52:36,000 --> 00:52:37,480 I think I think companies should do off sites. 932 00:52:37,480 --> 00:52:40,720 It's very important. We also have smaller team off 933 00:52:40,720 --> 00:52:43,840 sites where people, you know, coalesce or on a conference, a 934 00:52:43,840 --> 00:52:46,320 subset of team. Generally the business facing 935 00:52:46,320 --> 00:52:49,000 people and the more go to market facing people are are often a 936 00:52:49,000 --> 00:52:50,600 more of the commercial crypto conferences. 937 00:52:50,600 --> 00:52:52,880 The research people are generally more at research 938 00:52:52,880 --> 00:52:55,760 conferences and have, you know, kind of smaller meetings there 939 00:52:55,760 --> 00:52:59,200 around publications. The the engineering team has, 940 00:52:59,480 --> 00:53:03,920 you know, kind of meet ups that they they sometimes do to do in 941 00:53:03,920 --> 00:53:06,880 person hacking together. And yeah, I mean, just very 942 00:53:06,880 --> 00:53:09,160 broadly, we also do off sites as as team as well. 943 00:53:09,400 --> 00:53:13,080 I think meeting people in person and you know, having the team 944 00:53:13,080 --> 00:53:15,160 meet in person in a regular cadence, there's no replacement 945 00:53:15,160 --> 00:53:16,400 for that. Yeah. 946 00:53:17,200 --> 00:53:19,080 So before we wrap up, I wanted to ask. 947 00:53:19,080 --> 00:53:21,080 You about some of the cryptography research that you 948 00:53:21,080 --> 00:53:24,560 guys are working on. You mentioned that that there's 949 00:53:24,560 --> 00:53:29,920 a paper coming out this year. Yeah, the paper came out last 950 00:53:29,920 --> 00:53:33,200 year Dynamic Starks. It's a fantastic work that was 951 00:53:33,200 --> 00:53:35,760 authored by a research team on a new paradigm for zero knowledge 952 00:53:35,760 --> 00:53:39,440 proofs, fully updatable 0 knowledge proofs and this work 953 00:53:39,600 --> 00:53:43,360 was published or accepted into SBC Science and blockchain 954 00:53:43,360 --> 00:53:47,120 conference, one of the top academic conferences in crypto 955 00:53:47,120 --> 00:53:52,960 for as K and consensus and generally science blockchain 956 00:53:52,960 --> 00:53:55,080 designs, hence the name of the conference. 957 00:53:55,360 --> 00:53:58,560 This is hosted generally either at Stanford for a bunch of 958 00:53:58,560 --> 00:54:01,720 years, last year's at Columbia, this year's it's at Berkeley. 959 00:54:03,320 --> 00:54:05,600 And so we're actually the only team in crypto that's had work 960 00:54:05,600 --> 00:54:07,280 accepted two of the last three years. 961 00:54:07,280 --> 00:54:09,680 We were wait listed unfortunately on the third year. 962 00:54:09,840 --> 00:54:13,960 But the the, the, the dynamic stock work is is going to be 963 00:54:13,960 --> 00:54:16,600 presented next week at Science and blockchain conference. 964 00:54:16,600 --> 00:54:20,800 Wei J Wong, one of our our our PhD interns last summer, is the 965 00:54:20,800 --> 00:54:25,800 lead author and then Travon, Dimitris and Babis are three of 966 00:54:25,800 --> 00:54:29,120 the other authors on the team. And so, yeah, we're anyone who's 967 00:54:29,120 --> 00:54:31,080 going to be at Berkeley next week for science and blockchain 968 00:54:31,080 --> 00:54:34,200 or or whenever this airs. If you're at Berkeley for 969 00:54:34,200 --> 00:54:36,080 science and blockchain, hopefully you saw the talk. 970 00:54:36,840 --> 00:54:39,440 And there's a bunch of other fantastic work coming from our 971 00:54:39,440 --> 00:54:42,520 research team this summer as well all the way like from 972 00:54:42,520 --> 00:54:44,920 things like new start constructions, the things like 973 00:54:44,920 --> 00:54:49,120 privacy preserving inference and privacy preserving MPC based 974 00:54:49,120 --> 00:54:51,760 inference, like Coast arc work. And then some other stuff I 975 00:54:51,760 --> 00:54:54,680 can't talk about that are kind of more foundational to ML and 976 00:54:54,680 --> 00:54:58,080 applications of of of ZK within kind of some more foundational 977 00:54:58,080 --> 00:55:01,400 constructs of ML. And so across the board, you 978 00:55:01,400 --> 00:55:03,880 know, I think one of the things we prioritize the company is to 979 00:55:03,880 --> 00:55:07,800 actually do fundamental ZK research alongside our 980 00:55:07,800 --> 00:55:10,640 commercialization aims for deep proven other ZK technologies. 981 00:55:11,240 --> 00:55:15,080 You know, I'd like to think of, you know what, what deep mind or 982 00:55:15,160 --> 00:55:18,440 open AI were to AI research. We are to ZK research. 983 00:55:18,680 --> 00:55:21,360 We hire the best people, we retain the best people. 984 00:55:21,600 --> 00:55:26,160 We have fantastic groups of people who are, you know, active 985 00:55:26,160 --> 00:55:29,240 professors or are active PhD students who are joining the 986 00:55:29,240 --> 00:55:32,920 company for different periods of time on sabbatical or on 987 00:55:32,920 --> 00:55:37,320 continuous, you know, part time basis to be able to construct 988 00:55:37,640 --> 00:55:41,880 systems and research into improvements in systems that 989 00:55:41,880 --> 00:55:44,480 are, you know, a standard deviation more advanced than 990 00:55:44,480 --> 00:55:47,560 what you would get from purely commercially minded team. 991 00:55:48,920 --> 00:55:51,200 Cool. Well, where can people go to 992 00:55:51,200 --> 00:55:54,080 learn more about LaGrange? What's your what's your? 993 00:55:54,080 --> 00:55:56,600 What's your CTA? Your call to action for the 994 00:55:56,640 --> 00:56:00,200 audience. Yeah, so I would, I would 995 00:56:00,200 --> 00:56:03,040 suggest anyone who wants to build a deep proof, go on 996 00:56:03,040 --> 00:56:05,040 GitHub, go to LaGrange and look at deep proof. 997 00:56:05,040 --> 00:56:07,920 It's, it's up there. Anyone who wants to follow us 998 00:56:08,200 --> 00:56:11,640 and you know, learn about some exciting partnership updates and 999 00:56:11,640 --> 00:56:14,960 research updates, follow us on Twitter at you know, LeBron dev 1000 00:56:15,520 --> 00:56:18,440 or if you are interested in kind of having a more personal 1001 00:56:18,440 --> 00:56:21,520 relationship with a team, I'd recommend you join our telegram, 1002 00:56:21,800 --> 00:56:24,440 I'm sorry, our Discord channel, which is linked on the website 1003 00:56:25,000 --> 00:56:27,400 and is a great way to interact with our community team and to 1004 00:56:27,400 --> 00:56:30,720 interact with the founders at a team as well as a whole. 1005 00:56:31,800 --> 00:56:34,840 And so I also would say anyone who really wants to, you're 1006 00:56:34,840 --> 00:56:38,960 welcome to reach out to me on, on Twitter or on Telegram and 1007 00:56:38,960 --> 00:56:40,960 anyone who's really, really motivated will be able to find 1008 00:56:40,960 --> 00:56:42,680 me. Cool. 1009 00:56:42,760 --> 00:56:44,000 Ismael, thank you so much for coming on. 1010 00:56:45,360 --> 00:56:47,160 Likewise, thank you so much for having me.