WEBVTT 1 00:00:00.040 --> 00:00:03.879 I think what's really powerful about crypto is its ability 2 00:00:03.960 --> 00:00:07.879 to like mobilize a massive amount of resources. Right, Like, 3 00:00:07.879 --> 00:00:11.720 if you look at just the number of GPUs that 4 00:00:11.800 --> 00:00:14.640 were in the proof of workdays of ethereum set up 5 00:00:14.720 --> 00:00:17.760 to mine ethereum, it's like fifty times as much compute 6 00:00:17.760 --> 00:00:21.320 power as what was used to train chatchipt. Right, So 7 00:00:21.359 --> 00:00:24.559 you have like a huge amount of resources. And AI 8 00:00:24.640 --> 00:00:27.480 model required huge amount of resources to be trained, right, 9 00:00:27.480 --> 00:00:32.759 it can cost like ten million dollars of train a model. 10 00:00:36.920 --> 00:00:39.240 This content is brought to you by Bitco, which is 11 00:00:39.240 --> 00:00:42.520 one of the top crypto custodians in the crypto industry. 12 00:00:42.840 --> 00:00:46.159 Bitco works with many big companies and brands such as 13 00:00:46.159 --> 00:00:51.280 Pantera Capital, Bitstamp, and bitcoin Ira. Nike also selected Bitco 14 00:00:51.479 --> 00:00:55.399 to power its wallets for its NFTs, and Bitco has 15 00:00:55.479 --> 00:00:59.079 many great services such as hot wallets, custodial wallets, self 16 00:00:59.119 --> 00:01:04.040 managed cold walls, and NFT wallets. Many institutions trust bitgo 17 00:01:04.159 --> 00:01:08.239 with its top level security and incredible services such as 18 00:01:08.239 --> 00:01:11.000 being able to deploy your capital while it's in custody, 19 00:01:11.200 --> 00:01:15.120 which includes lending, borrowing, trading, staking, DeFi access and more. 20 00:01:15.359 --> 00:01:17.879 If you'd like to learn more, about bitco. Please visit 21 00:01:17.920 --> 00:01:20.959 bitgo dot com link in a description. Welcome into the 22 00:01:20.959 --> 00:01:23.879 Thinking Crypto Podcast. I'm your host, Tony Edward, and with 23 00:01:23.920 --> 00:01:27.040 me is Anna Keslawskis, who's the co founder and CEO 24 00:01:27.280 --> 00:01:29.359 of VANA and a great to have you on. 25 00:01:30.040 --> 00:01:30.959 Yeah, great to be here. 26 00:01:31.760 --> 00:01:35.359 Anna. You're doing some great things and very interesting things 27 00:01:35.359 --> 00:01:39.799 at VANNA with the convergence of AI and blockchain, and 28 00:01:39.879 --> 00:01:43.159 I'm really excited to speak with you because I've been 29 00:01:43.400 --> 00:01:46.640 noodling and thinking a lot about these things and data 30 00:01:46.799 --> 00:01:49.040 and how you can monetize your data and own your 31 00:01:49.120 --> 00:01:52.120 data in the web three world and token i should 32 00:01:52.159 --> 00:01:55.640 say token iss world. We're headed to. So lots of questions, 33 00:01:55.680 --> 00:01:58.760 but let's start with your background. Where are you from 34 00:01:58.799 --> 00:02:00.519 and where'd you grow up? Yeah? 35 00:02:00.599 --> 00:02:04.560 Yeah, so I was born in Montreal, lived there for 36 00:02:04.599 --> 00:02:06.920 the first five years, moved to Sweden, and then moved 37 00:02:06.959 --> 00:02:12.240 to Minnesota, so mostly grew up there. I yeah, half Lithuanian, 38 00:02:12.400 --> 00:02:16.280 hence the last name, half Filipino, and then I ended up. 39 00:02:16.199 --> 00:02:16.879 Moving to. 40 00:02:18.960 --> 00:02:23.199 Boston for school at MIT, and then yeah, left school, 41 00:02:23.319 --> 00:02:25.759 moved to San Francisco and have pretty much been here since. 42 00:02:26.680 --> 00:02:31.719 And before you started Vana, what was your professional background. 43 00:02:31.719 --> 00:02:34.360 I noticed you were at crypto company Celo. If I'm 44 00:02:34.360 --> 00:02:37.199 not mistaken, tell us a bit about that. Yeah. 45 00:02:37.280 --> 00:02:41.000 Yeah, So I got into crypto through, like I guess, 46 00:02:41.039 --> 00:02:44.639 an interest in traditional currency. So I had worked at 47 00:02:44.639 --> 00:02:46.439 the FED and high school. I actually had a picture 48 00:02:46.439 --> 00:02:48.439 of Janet Yellen, chair of a FED and in my 49 00:02:48.520 --> 00:02:51.759 high school bedroom, which is like totally like who does that, right, 50 00:02:52.520 --> 00:02:57.000 And then got to MIT and learned about these decentralized currencies. 51 00:02:57.039 --> 00:02:59.879 This was back in twenty fifteen, and then it was 52 00:02:59.879 --> 00:03:03.919 like the five person MIT Bitcoin Club. I started mining 53 00:03:04.039 --> 00:03:07.360 ethereum from my dorm room and just got really fascinated 54 00:03:07.400 --> 00:03:12.240 in kind of the world of decentralization. And I also 55 00:03:12.560 --> 00:03:16.680 sort of came across the world of AI through MIT 56 00:03:16.840 --> 00:03:19.599 as well. I was doing some research at c Sale. 57 00:03:19.919 --> 00:03:23.039 Basically this was when the attention is all you need paper, 58 00:03:23.080 --> 00:03:25.680 which is kind of the foundations of chat. GPT had 59 00:03:25.680 --> 00:03:28.039 come out, but people hadn't figured out how to use 60 00:03:28.080 --> 00:03:30.560 it to generate new texts. They had just figured out 61 00:03:30.599 --> 00:03:32.560 how to use it to kind of model some text 62 00:03:33.639 --> 00:03:36.120 and so I use some of that to automate my 63 00:03:36.240 --> 00:03:39.479 job sorting documents at the World Bank, and then ended 64 00:03:39.560 --> 00:03:42.800 up leaving school, going through I Combinator, discovering this world 65 00:03:42.840 --> 00:03:46.199 of like Silicon Valley, my life's purpose is not to 66 00:03:46.240 --> 00:03:48.680 sort the world's documents, And so I actually ended up 67 00:03:48.719 --> 00:03:53.520 joining Cello as an early engineer there focused on kind 68 00:03:53.520 --> 00:03:57.080 of building a mobile first cryptocurrency and kind of having 69 00:03:57.120 --> 00:03:59.680 this stable store of value that you can use from 70 00:04:00.120 --> 00:04:02.000 any mobile phone around the world. 71 00:04:03.520 --> 00:04:08.560 Now, and you're at the forefront of two major emerging 72 00:04:08.639 --> 00:04:12.319 tech sectors AI blockchain and crypto. What do you love 73 00:04:13.400 --> 00:04:15.479 the best or do you love them both equally? 74 00:04:17.160 --> 00:04:20.759 Yeah, I would say ideologically crypto like, and I think 75 00:04:21.079 --> 00:04:25.759 because with crypto, it's really it's a very foundational shift 76 00:04:25.839 --> 00:04:29.240 in terms of how power structures in the world could work. 77 00:04:29.279 --> 00:04:30.959 And it is a sort of like I would say 78 00:04:30.959 --> 00:04:32.920 most crypto outcomes, it's sort of the sort of it's 79 00:04:32.920 --> 00:04:35.480 the thing where if it works, it really really works 80 00:04:35.480 --> 00:04:38.480 in this massive way like bitcoiner ethereum. And I think 81 00:04:38.560 --> 00:04:42.879 that that is just a really interesting problem space to 82 00:04:42.879 --> 00:04:46.160 be in. And it feels like the frontier right of 83 00:04:46.279 --> 00:04:50.079 like this is really still the very beginnings of the 84 00:04:50.639 --> 00:04:56.199 crypto kind of decentralized industry. But I guess there's some 85 00:04:56.199 --> 00:04:58.720 really cool math with a so like maybe, yeah, there's 86 00:04:58.759 --> 00:05:02.480 like definitely some fun math of AI, and yeah, I 87 00:05:02.560 --> 00:05:04.879 mean today a lot of the AI work is like 88 00:05:05.000 --> 00:05:09.199 training AI models, so it's almost like DevOps side stuff. 89 00:05:09.560 --> 00:05:13.160 It depends, so yeah, hard choice. I think that you 90 00:05:13.160 --> 00:05:17.319 can apply a lot of the principles of crypto, which 91 00:05:17.360 --> 00:05:21.560 are kind of like censorship resistance, true like self sovereignty 92 00:05:21.839 --> 00:05:24.399 towards AI, both in kind of an open source AI 93 00:05:24.480 --> 00:05:30.360 context and a decentralized AI context. And yeah, then that's 94 00:05:30.399 --> 00:05:33.680 a lot of fun. So sort of depends if I 95 00:05:33.680 --> 00:05:38.600 had to choose one. Yeah, crypto's ideology and like AI's 96 00:05:38.639 --> 00:05:42.439 core technology. But yeah, I mean I kind of like 97 00:05:42.480 --> 00:05:44.639 a little bit of both, so for sure. 98 00:05:44.959 --> 00:05:47.560 So what do you think about the I guess what 99 00:05:47.560 --> 00:05:51.240 you would call symbiotic relationship with crypto, excuse me, blockchain 100 00:05:51.279 --> 00:05:54.600 and AI, because AI we're seeing is being used to 101 00:05:54.759 --> 00:05:58.800 enhance certain blockchain attributes and crypto attributes, while blockchain is 102 00:05:58.800 --> 00:06:01.720 being used to leave AI a bit with deep fakes 103 00:06:01.720 --> 00:06:04.480 and things like that, and there's different technologies being built. 104 00:06:04.720 --> 00:06:07.000 So do you see them working together as the future 105 00:06:07.920 --> 00:06:08.759 as things progress. 106 00:06:09.360 --> 00:06:11.079 Yeah, yeah, I think that's a good way to break 107 00:06:11.120 --> 00:06:14.680 it down, right, you sort of have like AI for crypto, 108 00:06:15.000 --> 00:06:17.480 so that means like how do you make models available 109 00:06:17.519 --> 00:06:19.879 on chain use them for trading and stuff like that, 110 00:06:20.120 --> 00:06:24.279 And then you have crypto for AI where you're using 111 00:06:24.319 --> 00:06:27.160 the tools of blockchain and for crypto to really advance 112 00:06:27.199 --> 00:06:30.040 AI progress. And that's where Vona sits, and so that's 113 00:06:30.040 --> 00:06:32.319 where I spend most of my time kind of thinking 114 00:06:32.360 --> 00:06:37.319 about the intersection. I think what's really powerful about crypto 115 00:06:37.560 --> 00:06:42.279 is its ability to like mobilize a massive amount of resources. Right, 116 00:06:42.319 --> 00:06:45.879 Like if you look at just the number of GPUs 117 00:06:46.120 --> 00:06:48.879 that were in the proof of workdays of ethereum set 118 00:06:48.959 --> 00:06:51.879 up to mine ethereum, it's like fifty times as much 119 00:06:51.879 --> 00:06:55.680 compute power as what was used to train CHATCHYBT. Right, 120 00:06:55.720 --> 00:06:58.439 So you have like a huge amount of resources, and 121 00:06:58.800 --> 00:07:02.000 AI models requires huge amount of resources to be trained, right. 122 00:07:02.000 --> 00:07:04.319 It can cost like ten million dollars of train a model. 123 00:07:04.879 --> 00:07:08.639 And that's just thinking about the compute like the GPU cost, 124 00:07:08.680 --> 00:07:11.120 which I has been talked about a lot, and not 125 00:07:11.240 --> 00:07:14.079 even yet thinking about the data costs. So I think 126 00:07:14.120 --> 00:07:19.279 that kind of using crypto incentives to allow for very 127 00:07:19.360 --> 00:07:22.399 large scale projects to be funded in more kind of 128 00:07:22.399 --> 00:07:26.839 community oriented, open source type ways is like what's most 129 00:07:26.839 --> 00:07:29.279 interesting to me. And what I would say too, is 130 00:07:29.319 --> 00:07:33.439 that's distinct from just sort of like the naive approach 131 00:07:33.480 --> 00:07:36.079 is basically Okay, let's put AI on the blockchain, right, 132 00:07:36.079 --> 00:07:37.920 but I think it's important to take a step back 133 00:07:37.959 --> 00:07:40.120 and be like, why do we even have a blockchain? Okay, 134 00:07:40.160 --> 00:07:43.879 it's because we value like censorship, resistance, like true self sovereignty. 135 00:07:44.120 --> 00:07:47.000 And then how do we create a system that uses 136 00:07:47.399 --> 00:07:51.160 some of that distributed consensus that's worked really well in blockchains, 137 00:07:51.800 --> 00:07:56.879 but applies it towards this AI native world that we're 138 00:07:56.920 --> 00:07:59.319 moving towards over the next five to ten years. 139 00:08:00.279 --> 00:08:03.680 So when did the idea come about to start VANA? 140 00:08:03.800 --> 00:08:04.879 And what's your mission? 141 00:08:05.680 --> 00:08:10.079 Yeah? Yeah, So VANA was first started actually back at 142 00:08:10.199 --> 00:08:14.519 MIT as like a research project super early on. This 143 00:08:14.600 --> 00:08:17.600 is like twenty eighteen of like, basically, what I had 144 00:08:17.639 --> 00:08:21.560 seen is the only thing that matters for AI models 145 00:08:21.519 --> 00:08:24.319 in the long run is data. Like data is really 146 00:08:24.360 --> 00:08:27.399 the new oil. It is just this immensely valuable thing. 147 00:08:28.040 --> 00:08:30.959 And so the question I was asking along with my 148 00:08:31.040 --> 00:08:36.159 co founder, who I met he was doing his godduate 149 00:08:36.240 --> 00:08:38.600 degree at Harvard when I was doing my undergrad at MIT, 150 00:08:38.799 --> 00:08:41.360 and so we had taken a class together at the 151 00:08:41.440 --> 00:08:45.279 MIT Media Lab. This question was like, how can you 152 00:08:45.919 --> 00:08:48.840 leverage the power of like many people rather than just 153 00:08:48.919 --> 00:08:53.559 having like centralized institutions training a single model to all 154 00:08:53.759 --> 00:08:58.039 collectively contribute. At the time, we were doing like data labeling, 155 00:08:58.080 --> 00:09:00.639 where we would have people label data from their phone 156 00:09:00.720 --> 00:09:05.279 super early on and earn based on that. And then 157 00:09:05.320 --> 00:09:08.240 I started working on Vona full time a little over 158 00:09:08.360 --> 00:09:10.799 like three and a half years ago. That was still 159 00:09:10.919 --> 00:09:15.600 quite early in the crypto AI space. In twenty twenty two, 160 00:09:15.679 --> 00:09:18.360 we got our patent for kind of non custodial data, 161 00:09:18.440 --> 00:09:21.320 which was kind of a way for you to be 162 00:09:21.360 --> 00:09:23.919 able to manage your data in a similar way to 163 00:09:23.960 --> 00:09:26.120 how you can manage your funds with a crypto wallet, 164 00:09:26.200 --> 00:09:28.159 so you can use your meta mask to manage your data. 165 00:09:28.159 --> 00:09:33.840 That's on Vana, and yeah, our mission is to kind 166 00:09:33.879 --> 00:09:38.080 of like shift the power of data and AI towards 167 00:09:38.480 --> 00:09:43.639 many people rather than just one centralized institution. The economic 168 00:09:43.720 --> 00:09:46.600 framing I have it is to ensure that users own 169 00:09:46.639 --> 00:09:48.639 their data and the value that it creates. But that's 170 00:09:48.639 --> 00:09:51.679 a little bit abstract, so it's really just about how 171 00:09:51.720 --> 00:09:54.240 do you make it so that things that get created 172 00:09:54.440 --> 00:09:57.480 from your personal data that you've created you kind of 173 00:09:57.559 --> 00:10:00.120 are attributed to you and you earn from an and 174 00:10:00.159 --> 00:10:04.200 also allowing AI progress to be pushed forward through that m. 175 00:10:05.440 --> 00:10:07.879 You know, for years, I've been thinking a lot about 176 00:10:07.919 --> 00:10:10.759 this and data going on the blockchain and where I 177 00:10:10.799 --> 00:10:13.639 can lend it out to advertisers or I could do 178 00:10:13.679 --> 00:10:16.159 whatever I want and monetize it. And it sounds like 179 00:10:16.200 --> 00:10:20.799 you guys are building that solution. And I think in 180 00:10:20.879 --> 00:10:24.399 the world we're headed towards where with digital identity and 181 00:10:25.000 --> 00:10:29.679 updates to different advertising laws, GDPR and privacy and all 182 00:10:29.720 --> 00:10:33.720 these things. Do you see that eventually all of us 183 00:10:33.799 --> 00:10:37.200 will have wallets, it'll have our digital identity, our data 184 00:10:37.240 --> 00:10:40.799 and so forth on a smart contract, locked and whatever 185 00:10:40.840 --> 00:10:43.399 it may be, and then we choose how who we 186 00:10:43.440 --> 00:10:45.039 want to give it to or lend it out to 187 00:10:45.120 --> 00:10:45.720 and so forth. 188 00:10:46.360 --> 00:10:50.039 Exactly. Yeah, you're just pitching VANA right back. So there 189 00:10:50.039 --> 00:10:50.320 you go. 190 00:10:52.919 --> 00:10:55.960 But let's talk a bit about the economics of that. 191 00:10:56.120 --> 00:11:00.399 So would it be like let's say, Apple compute, right, 192 00:11:01.279 --> 00:11:04.200 I don't know, and I'm spitballing here because I don't 193 00:11:04.279 --> 00:11:07.879 fully understand all the concepts of this yet. Would it 194 00:11:07.919 --> 00:11:10.080 be I give them my data and they pay me 195 00:11:10.399 --> 00:11:13.799 a monthly fee or a one time fee, Like, how 196 00:11:13.840 --> 00:11:14.799 do you think that might work? 197 00:11:15.440 --> 00:11:18.519 Yeah, that's a good question. And I think that it's 198 00:11:18.559 --> 00:11:21.679 sort of been this shift as people have become aware 199 00:11:21.759 --> 00:11:24.320 of the immense value that AI creates. Right, Like, if 200 00:11:24.360 --> 00:11:27.559 you think sort of five years back, the value proposition 201 00:11:27.639 --> 00:11:30.000 of hey, we're going to give you twenty dollars for 202 00:11:30.039 --> 00:11:32.519 your data, it's honestly not that interesting. Like to some 203 00:11:32.600 --> 00:11:34.960 people it might make sense, but like to the average person, 204 00:11:35.039 --> 00:11:38.519 it's just like not that interesting, of like why would 205 00:11:38.519 --> 00:11:41.559 I want that? Whereas if the value prop is like, hey, 206 00:11:42.360 --> 00:11:45.759 contribute your data in exchange for ownership in an AI model, 207 00:11:45.879 --> 00:11:48.759 right like own the part of the next CHATCHYBT, and 208 00:11:48.799 --> 00:11:51.360 it acts much more like equity, like a stake in 209 00:11:51.399 --> 00:11:54.960 an AI model, rather than a one time payment. I 210 00:11:55.000 --> 00:11:58.639 think that that is much more compelling for users from 211 00:11:58.639 --> 00:12:01.720 what we've seen at Vana, and I think that it 212 00:12:01.840 --> 00:12:06.639 also points towards like this economic shift that's happening where 213 00:12:06.679 --> 00:12:09.600 a huge amount of economic value is created through AI, 214 00:12:09.919 --> 00:12:13.080 right like chat GBT or open ai. They hit three 215 00:12:13.120 --> 00:12:16.200 point four billion dollars in revenue this summer, Right, that's 216 00:12:16.200 --> 00:12:19.360 just an immense amount of money that's primarily coming from 217 00:12:20.480 --> 00:12:24.440 intelligence that is trained on everyone's knowledge, right, everyone who's 218 00:12:24.480 --> 00:12:27.559 created data has created that knowledge which has helped teach 219 00:12:27.600 --> 00:12:30.440 that AI. And our view is that if you have 220 00:12:30.480 --> 00:12:33.679 the right incentive system, you can build much better AI 221 00:12:34.399 --> 00:12:38.600 and also align incentive. So yeah, kind of moving towards like, Okay, 222 00:12:38.600 --> 00:12:41.960 what does it look like tactically today, Like the Reddit 223 00:12:42.039 --> 00:12:44.240 data daw which is one of the projects that's built 224 00:12:44.279 --> 00:12:48.159 on VONAM. They aggregated like one hundred and forty thousand 225 00:12:48.320 --> 00:12:51.879 users data they trained an AI model from that. It's 226 00:12:51.919 --> 00:12:54.600 really good at sounding like a snarky redditor. I've heard 227 00:12:54.600 --> 00:12:56.759 some people joke it's sort of like the leading shit 228 00:12:56.840 --> 00:12:59.600 posting model, right, So like in terms of it's not 229 00:12:59.720 --> 00:13:02.559 yet feeding CHATGBT, but like at some point there will 230 00:13:02.600 --> 00:13:05.639 be enough data across all these different data sources that 231 00:13:05.720 --> 00:13:08.840 it will and every time that model is used, all 232 00:13:08.879 --> 00:13:11.519 of the users actually get paid out, right, And so 233 00:13:11.960 --> 00:13:15.039 you have kind of this nice flow where, yeah, in 234 00:13:15.080 --> 00:13:18.360 the future, it's like every time someone's interacting with an 235 00:13:18.360 --> 00:13:22.279 AI model like chat GBT or claude, the people who 236 00:13:22.279 --> 00:13:26.080 have contributed towards teaching it the knowledge that it has 237 00:13:26.480 --> 00:13:29.200 are the ones that are rewarded, and so you have 238 00:13:29.279 --> 00:13:31.759 this really strong kind of data flywheel in place. 239 00:13:32.279 --> 00:13:35.480 That's so fascinating because you know, you mentioned about the 240 00:13:35.519 --> 00:13:38.240 shit posting and so forth. So it made me think 241 00:13:38.240 --> 00:13:42.200 of you have these niche AI models, so to speak, 242 00:13:42.960 --> 00:13:46.799 backed by communities. They have their own personality, their own 243 00:13:47.279 --> 00:13:51.039 take on things. That's so fascinating to me. So is 244 00:13:51.039 --> 00:13:54.039 that the world we're we're headed to where it's different AI? 245 00:13:54.600 --> 00:13:57.919 Am I artickling it well? Is in models or bots? 246 00:13:57.960 --> 00:13:58.759 I don't know what to call it. 247 00:13:58.840 --> 00:14:02.320 Yeah, yeah, no models or boss however you want to 248 00:14:02.360 --> 00:14:05.879 say it. That's both of those are perfectly correct. I 249 00:14:05.879 --> 00:14:10.720 think that, yeah, the world we're headed towards. Yeah. I 250 00:14:10.799 --> 00:14:12.840 think there's sort of this debate right now of like 251 00:14:13.279 --> 00:14:15.600 what is truth right because we're trying to all build 252 00:14:15.639 --> 00:14:17.919 a single AI model and be like the AI model 253 00:14:17.960 --> 00:14:20.440 knows this and this is true. But that's a really 254 00:14:20.440 --> 00:14:23.360 hard question, right, Like, actually we have not solved that 255 00:14:23.399 --> 00:14:25.879 as human beings aside from an AI context. Like that 256 00:14:26.039 --> 00:14:29.080 is a massive debate that happened in the social media 257 00:14:29.200 --> 00:14:31.159 days and now we're kind of having the same thing 258 00:14:31.200 --> 00:14:34.120 and in the AI days and our view is just 259 00:14:34.159 --> 00:14:37.120 having this healthy ecosystem. I think it aligns actually a 260 00:14:37.159 --> 00:14:39.879 lot with the open source work that Meta is doing, 261 00:14:40.240 --> 00:14:43.559 just putting AI models out there and allowing many people 262 00:14:44.360 --> 00:14:48.200 to sort of modify them, customize them, align them towards 263 00:14:48.240 --> 00:14:52.799 a particular set of preferences or views depending on what 264 00:14:52.840 --> 00:14:56.240 they'd like. Right, so you have this balanced, healthy ecosystem 265 00:14:56.480 --> 00:15:01.960 rather than one single massive AI model that is controlled 266 00:15:01.960 --> 00:15:04.080 by just one or two companies. 267 00:15:05.440 --> 00:15:08.440 So and give me some rope here. Let's say with 268 00:15:08.519 --> 00:15:11.120 the Thinking Crypto podcast, I have a community of people. 269 00:15:11.200 --> 00:15:13.840 Let's say it's a total is like over two hundred 270 00:15:13.840 --> 00:15:17.440 thousand people that listened to me? Right? Can I build 271 00:15:17.480 --> 00:15:21.759 an AI model to teach newbies about crypto and my 272 00:15:22.039 --> 00:15:26.679 community sources that info to that that model, and then 273 00:15:26.919 --> 00:15:29.679 people can pay to have that model teach him crypto. 274 00:15:29.759 --> 00:15:32.919 Would that be the economics or business models? 275 00:15:33.200 --> 00:15:33.600 Totally? 276 00:15:34.759 --> 00:15:37.879 Hmm. That is fascinating. I have to talk to you 277 00:15:37.960 --> 00:15:42.440 after this, and I. 278 00:15:42.440 --> 00:15:44.600 Think the way to think about that is like, so, 279 00:15:45.720 --> 00:15:48.720 think about what AI models can't do today, and then 280 00:15:48.720 --> 00:15:51.960 think about the knowledge that you uniquely have or your 281 00:15:52.000 --> 00:15:56.080 community uniquely has beyond what's included in AI models today, 282 00:15:56.159 --> 00:15:58.960 and that's really where the sort of like alpha is, 283 00:15:59.080 --> 00:16:02.279 right of like where what information is contained in the 284 00:16:02.399 --> 00:16:06.320 data that most models don't have access to. So generally 285 00:16:06.399 --> 00:16:09.120 the Internet you can publicly scrape. That means just like 286 00:16:09.159 --> 00:16:11.720 all the sites you can visit without logging into them. 287 00:16:12.080 --> 00:16:13.879 So that's sort of already in a lot of these 288 00:16:13.879 --> 00:16:16.799 AI models. But there's a huge amount of information that, 289 00:16:16.879 --> 00:16:20.360 like you know and your community knows that not very 290 00:16:20.440 --> 00:16:23.159 many other people know, right, So having the ability for 291 00:16:23.279 --> 00:16:28.120 you to train an all powerful AI model based on 292 00:16:28.159 --> 00:16:31.600 that unique knowledge and have that truly belong to your community. 293 00:16:33.519 --> 00:16:36.519 And I guess, like you said, aside from what's scraped 294 00:16:36.559 --> 00:16:38.840 on the web and what's already out there, the nuance 295 00:16:39.000 --> 00:16:42.000 of hey, I'm from this part of the United States 296 00:16:42.080 --> 00:16:44.799 or this part of the world, I'm thirty something years old, 297 00:16:44.799 --> 00:16:48.759 I'm twenty something years old. Here's what my investment story 298 00:16:48.799 --> 00:16:51.600 and journey was here, here's how it impacted my life. 299 00:16:51.720 --> 00:16:54.919 That's something the AI can pull from it anywhere except 300 00:16:55.120 --> 00:16:57.159 someone sharing that story, right totally. 301 00:16:57.279 --> 00:17:00.840 Yeah, that's fascinating I think too. So often the data 302 00:17:00.879 --> 00:17:05.359 that's really valuable is this like contextual data on a person, right, 303 00:17:05.400 --> 00:17:07.640 So it's like, hey, like here's maybe how I got 304 00:17:07.640 --> 00:17:10.519 into crypto, and then also here's kind of my here 305 00:17:10.519 --> 00:17:13.440 are my demographics, here's my perspective on the world. Kind 306 00:17:13.480 --> 00:17:16.000 of being able to share like both your tweets, but 307 00:17:16.079 --> 00:17:19.400 also your financial history, which could include like your trades, 308 00:17:19.480 --> 00:17:22.359 your credit card history, and maybe even your journal entries. 309 00:17:22.759 --> 00:17:25.519 And there's really no single tech company that has access 310 00:17:25.599 --> 00:17:27.440 to all of that, right, they might have access to 311 00:17:27.440 --> 00:17:29.599 a small slice of it, but if you want to 312 00:17:29.640 --> 00:17:33.319 get all of those different data sources across a given user, 313 00:17:33.799 --> 00:17:36.359 you really need to work with the user directly. And 314 00:17:36.400 --> 00:17:40.319 so that's kind of the power of using decentralization. We 315 00:17:40.400 --> 00:17:44.519 can actually leverage data regulation like GDPR which you had 316 00:17:44.519 --> 00:17:48.720 mentioned CCPA in California to ensure that a user can 317 00:17:48.759 --> 00:17:51.640 get that data out of where whichever platform they've kind 318 00:17:51.680 --> 00:17:54.839 of posted it, and then actually contribute it to these 319 00:17:54.920 --> 00:17:56.160 powerful at models. 320 00:17:57.599 --> 00:18:00.359 Now let's go back to the redded data DOWT question. 321 00:18:00.680 --> 00:18:04.480 I guess first part is why DOOS did decentralize a 322 00:18:04.519 --> 00:18:09.039 toon of autonomous organizations? And is that the only way 323 00:18:09.079 --> 00:18:11.519 to do this where you can have the crowd sourcing 324 00:18:11.640 --> 00:18:13.240 and to have it in a decentralized way. 325 00:18:13.839 --> 00:18:16.160 Yeah, that's a good question, and I think like doos 326 00:18:16.279 --> 00:18:19.759 have a pretty mixed reputation in crypto, right, Like, if 327 00:18:19.799 --> 00:18:22.640 we look over the past maybe five years of dows, 328 00:18:22.680 --> 00:18:26.759 I think there are a few breakout success stories, but 329 00:18:27.119 --> 00:18:30.319 as a whole, like, it's actually very hard to get 330 00:18:30.319 --> 00:18:33.839 a decentralized group of people to govern something, and so 331 00:18:33.920 --> 00:18:36.559 it can be a challenge. What we've seen with data 332 00:18:36.599 --> 00:18:39.519 dows is that often you have a large number of 333 00:18:39.519 --> 00:18:42.480 people come in initially and connect their data, and then 334 00:18:42.519 --> 00:18:45.640 a smaller percent of them stay and help manage that 335 00:18:45.799 --> 00:18:49.160 data and sort of like decide, hey, this company wants 336 00:18:49.200 --> 00:18:51.599 to buy access to it, or this machine learning engineer 337 00:18:51.640 --> 00:18:54.079 wants to train an AI model on it. Should we 338 00:18:54.119 --> 00:18:58.200 grant them access? What are fair economics, et cetera. And 339 00:18:58.279 --> 00:19:01.640 so by default all that is decided by the TAO, 340 00:19:01.880 --> 00:19:04.319 because it's very hard to price data, right like, how 341 00:19:04.319 --> 00:19:08.920 should you decide whether? So we basically we see the 342 00:19:09.000 --> 00:19:12.799 doos as making that decision ultimately. But in the future 343 00:19:12.799 --> 00:19:14.839 there are ways too where you don't need to create 344 00:19:14.880 --> 00:19:16.920 a DOO and there can just be kind of stable 345 00:19:16.960 --> 00:19:21.440 coin based payments for a particular data pool. And this 346 00:19:21.519 --> 00:19:23.240 is a little bit technical, but it's kind of the 347 00:19:23.279 --> 00:19:26.880 distinction between what we call a data liquidity pool, so 348 00:19:26.960 --> 00:19:30.000 a DLP on vana. Most of them are data DAOs, 349 00:19:30.039 --> 00:19:32.559 but they actually don't have to be a DOO if 350 00:19:32.559 --> 00:19:36.839 they don't want a token affiliated with their particular data set. 351 00:19:37.039 --> 00:19:40.440 So yeah, we have this slightly more general framing, which 352 00:19:40.480 --> 00:19:43.319 is a data liquidity pool, which is just about kind 353 00:19:43.359 --> 00:19:46.839 of building up that data liquidity, getting many people's data 354 00:19:46.880 --> 00:19:50.359 all pooled together, because your data is only valuable once 355 00:19:50.400 --> 00:19:52.400 you combine it with others. Right, you can't train an 356 00:19:52.440 --> 00:19:55.240 AI model on just one person's data. You really want 357 00:19:55.240 --> 00:19:58.640 a lot of data. And so that's that's a good question, 358 00:19:58.759 --> 00:20:00.400 and there is some nuance there. 359 00:20:01.359 --> 00:20:06.839 So the Reddit data dow, how are they monetizing that data, 360 00:20:07.039 --> 00:20:09.319 what are they selling it for? And what's the currency 361 00:20:09.319 --> 00:20:09.839 being used? 362 00:20:10.440 --> 00:20:14.559 Yeah, yeah, so they created their own currency which is 363 00:20:14.680 --> 00:20:18.480 used as a governance. It's basically the DOW kind of 364 00:20:18.559 --> 00:20:22.480 token that's used to govern the Reddit data set, and 365 00:20:22.559 --> 00:20:27.440 its primary purpose is governance, basically voting on hey, should 366 00:20:27.480 --> 00:20:30.640 we let new people come in and contribute their data 367 00:20:30.720 --> 00:20:33.880 and reward them, and also should we let someone access 368 00:20:34.039 --> 00:20:38.000 our data? So one of the folks they granted access 369 00:20:38.039 --> 00:20:42.240 to the data is actually like ex Google Brain machine 370 00:20:42.279 --> 00:20:46.079 learning engineer who trained an AI model that like ship 371 00:20:46.079 --> 00:20:48.599 posting model that I had mentioned, so that it's in 372 00:20:48.640 --> 00:20:51.440 the phase of like research preview right now and then 373 00:20:51.440 --> 00:20:54.039 they're rolling it out and to interact with that, you'll 374 00:20:54.079 --> 00:20:57.759 have to actually sort of like hold a minimum amount 375 00:20:57.960 --> 00:21:03.720 of that Reddit data that governance token. They're also sort 376 00:21:03.720 --> 00:21:07.240 of finalizing a data sale in the works right now, 377 00:21:08.319 --> 00:21:11.400 and so the way I'm actually not sure exactly how 378 00:21:11.440 --> 00:21:14.880 they structured those economics. It could be that they're having 379 00:21:14.920 --> 00:21:18.200 the data buyer burn the governance token, or it could 380 00:21:18.240 --> 00:21:23.200 be that they're distributing revenue to the governance tokenholders. So 381 00:21:23.359 --> 00:21:25.680 it could be either of those two, and it depends 382 00:21:25.720 --> 00:21:27.359 on what the two decided. 383 00:21:29.400 --> 00:21:32.640 A couple questions of follow questions on that. Do you 384 00:21:32.799 --> 00:21:34.519 have to create your own token or can you use 385 00:21:34.519 --> 00:21:38.079 an existing token or a stable coin as you mentioned before, Yeah, 386 00:21:38.119 --> 00:21:40.240 because if you get into the creation of a token, 387 00:21:40.720 --> 00:21:44.319 you know, could you have securities issues, security laws issues, 388 00:21:44.319 --> 00:21:45.119 I should say. 389 00:21:45.400 --> 00:21:47.680 Yeah, yeah, you can use a stable coin, you can 390 00:21:47.759 --> 00:21:50.319 use a pre existing token, and so you can kind 391 00:21:50.359 --> 00:21:53.319 of collectivize however you'd like. And I'll mention too the 392 00:21:53.359 --> 00:21:56.799 Reddit data now actually only intended for their token to 393 00:21:56.799 --> 00:21:59.519 be used for kind of like governance and voting, and 394 00:21:59.559 --> 00:22:03.680 so that was their intention from the beginning. I think 395 00:22:03.680 --> 00:22:06.119 there was some trading activity that came up on it, 396 00:22:06.400 --> 00:22:10.000 but it's primarily used for voting, so I think in 397 00:22:10.039 --> 00:22:13.039 the long run we'll see sort of some activities like that. 398 00:22:14.400 --> 00:22:17.119 And yeah, depending on kind of your risk profile and 399 00:22:17.160 --> 00:22:19.359 how you structure it, right, each daya now has their 400 00:22:19.400 --> 00:22:23.119 own preferences in terms of what they're looking for from 401 00:22:23.119 --> 00:22:27.279