Agent Looping Towards Recursive Self Improvement | Token2049 Reflections 的完整文字稿

00:04:49.230 to the point where you have something that can, you know, be 46 00:04:49.550 --> 00:04:50.590 antifragile, 47 00:04:50.750 --> 00:04:51.390 basically. 48 00:04:52.350 --> 00:04:56.485 Yeah. 100%. If you go back to tech.more.org, 49 00:04:56.805 --> 00:04:57.685 it's 50 00:04:57.685 --> 00:04:59.045 interesting to see 51 00:04:59.685 --> 00:05:01.205 that if you scroll down, 52 00:05:01.685 --> 00:05:06.565 the most popular models are still really heavily the open weight models. 53 00:05:06.645 --> 00:05:10.330 Yeah. Even though lots of people are offering quad and other models, 54 00:05:10.490 --> 00:05:13.610 it's DeepSeek v four, GLM 5.2, 55 00:05:13.610 --> 00:05:14.970 GLM 5.3. 56 00:05:14.970 --> 00:05:19.770 Like, those are the dominant models. And if you look at a lot of the public data, 57 00:05:20.090 --> 00:05:21.530 the majority now 58 00:05:21.849 --> 00:05:23.050 of inference usage, 59 00:05:24.324 --> 00:05:28.565 it's those open weight models. Right? It's finally flipped 60 00:05:29.444 --> 00:05:31.365 from mostly proprietary 61 00:05:31.365 --> 00:05:38.450 to now majority are open weight. So, yeah, people are getting the picture. Like, I was spending $6,000 62 00:05:38.450 --> 00:05:40.130 a month on Claude 63 00:05:40.370 --> 00:05:45.570 in January. And when I switched to the open weight models, it cut my bill 90%. 64 00:05:46.210 --> 00:05:52.005 Right? And so any power user I talk to is like, how do I do that? Right? Okay. You need an agent. 65 00:05:52.405 --> 00:05:54.725 Give it the API key for Morpheus. 66 00:05:55.285 --> 00:06:01.125 Go to town. Right? And all of a sudden, you don't need a giant budget to do a ton of work. So it's not a of And 67 00:06:02.170 --> 00:06:06.730 we're just getting to the point too where, like, you don't need the jackhammer to 68 00:06:07.130 --> 00:06:16.825 nail a picture to your wall. You know? You can use the small hammer for that job. And we have agents that are smart enough to discern between when it does and doesn't need the super heavy lift. 69 00:06:17.305 --> 00:06:19.545 And and for builders, by the way, I mean, 70 00:06:20.025 --> 00:06:32.180 this is how you really should be thinking about the builds, right? It's like how much more token do I need staked to support the product or service that I'm building, Thinking through the fact that I can use really cheap inference for a lot of stuff. 71 00:06:32.740 --> 00:06:51.335 And I just met with a company yesterday who built a product that I think I'm gonna use. I'm just in in my other business that it's a great product that they built. I was trying to dig deep into, you know, what are you running on the back end for this thing to function. And they gave me, like, a little bit of a look under the hood. And, yeah, it's exactly that. It's how do I do the thing 72 00:06:51.655 --> 00:06:55.175 to a really high capability but also cut cost 73 00:06:55.335 --> 00:06:56.375 because I don't need, 74 00:06:57.330 --> 00:07:03.250 you know, the the frontier model for every task. And and in fact, you don't need it for most tasks. And 75 00:07:05.090 --> 00:07:07.890 and so it's been interesting. At the conference, 76 00:07:08.930 --> 00:07:10.765 you know, a huge 77 00:07:11.005 --> 00:07:14.525 undercurrent has been the rise of Zcash 78 00:07:14.685 --> 00:07:19.485 Mhmm. And near these other protocols. And I think it's because there's a large 79 00:07:19.965 --> 00:07:20.925 crossover 80 00:07:21.085 --> 00:07:26.750 between the people that believe in, like, private money and now who want private intelligence. 81 00:07:27.070 --> 00:07:33.870 Right? There's there's a pretty close match there. I love some of the memes. It's like Zcash not communism. 82 00:07:35.150 --> 00:07:38.510 And it's like, those are your choices. Right? If you don't have privacy, 83 00:07:39.155 --> 00:07:42.195 you're always going to be a victim of the state. 84 00:07:42.995 --> 00:07:44.595 Right? And that's true for, 85 00:07:45.074 --> 00:07:49.235 intelligence as it is for for money. Hey, Kyle. Good to see you, brother. 86 00:07:49.794 --> 00:07:56.540 Hey, guys. Sorry about that. Glad you could jump up. Oh, no worries. All good. Yeah. I was Knocking... 87 00:07:56.699 --> 00:07:58.300 Zcash pilled. Finally, 88 00:07:58.940 --> 00:08:05.660 David. I mean, I've been well aware of it for, you know, ever, basically. But I just never really took a dive into it and, like, 89 00:08:06.324 --> 00:08:08.324 really looked at it for utility. 90 00:08:08.405 --> 00:08:15.845 But, yeah, it it is very cool to see. There's a lot of crossover there between that community and the d a I d a I community. 91 00:08:16.884 --> 00:08:21.845 Well well... And there's there's a lot of parallels too. I was talking to some of the Zcash 92 00:08:22.780 --> 00:08:24.460 open source developers. 93 00:08:24.540 --> 00:08:27.420 And they're like, wait a minute. Morpheus doesn't have a company? 94 00:08:27.979 --> 00:08:31.580 Morpheus didn't have a pre mine? Like, it's all fair launch? 95 00:08:31.900 --> 00:08:45.875 I'm gonna look at this now. Right? Because there's such a presumption in the industry now that everything's a company. And the guy was complaining to me. He's like, everybody came up to our booth and, oh, what company do you work for? He's like, CCash is open source. There's no company. 96 00:08:46.275 --> 00:08:56.010 That's funny. But, like, really, what's your title? Are you the CEO? It's like, people have forgotten what a crypto protocol looks like. And, like, no. These are... This is actually a crypto protocol 97 00:08:56.250 --> 00:09:02.089 not run by a company. And, like, it's become a foreign concept as it's sort of gone mainstream, 98 00:09:02.490 --> 00:09:04.010 and people see other players 99 00:09:04.250 --> 00:09:06.250 like Coinbase and others. They just presume 100 00:09:06.985 --> 00:09:07.945 everything is, 101 00:09:08.505 --> 00:09:17.785 is, is a company. So it it was

refreshing to him that we're actually a protocol, and it's okay. It is neutral, and it is private, and it has all these aspects. 102 00:09:18.265 --> 00:09:20.265 That got his attention, which is really cool. 103 00:09:21.560 --> 00:09:22.840 That is really cool. 104 00:09:23.959 --> 00:09:25.080 Alright. 105 00:09:25.080 --> 00:09:31.400 So other updates quick. And Kyle, glad you jumped on. Thank you. I know we're doing some moving around with the timing here. 106 00:09:31.959 --> 00:09:35.720 A lot of travel to orchestrate, so we're trying to make it make it work. But 107 00:09:37.445 --> 00:09:46.085 Kyle, we know that from the node side of things, there was an update the other day. Is that something you're able to provide any context to by chance? 108 00:09:47.605 --> 00:09:49.125 I can give some 109 00:09:49.845 --> 00:09:52.565 context on it. I really would 110 00:09:53.120 --> 00:09:54.080 typically 111 00:09:55.360 --> 00:10:00.400 leave that specific one to Alan, but, a quick update is that, 112 00:10:01.279 --> 00:10:05.519 there were... And these are common. Right? We're we're making updates to the node, 113 00:10:05.920 --> 00:10:11.575 kind of all the time. Every week, you see a new release in there. There were some updates to 114 00:10:12.455 --> 00:10:14.295 some of the back end code, 115 00:10:14.535 --> 00:10:15.735 in terms of 116 00:10:16.055 --> 00:10:25.630 some different bugs that were identified through the the bug bounties or or through some other people doing audits on on the platform. It was just in relation to 117 00:10:26.110 --> 00:10:29.950 security hardening hardening on the provider node. 118 00:10:29.950 --> 00:10:30.589 So 119 00:10:31.470 --> 00:10:38.765 what it does is it... It's like belt and suspenders, right, where you have security, but there were some cases that, 120 00:10:39.565 --> 00:10:41.405 how do I explain this properly? 121 00:10:43.805 --> 00:10:45.245 Additional verification 122 00:10:45.245 --> 00:10:46.045 checks 123 00:10:46.045 --> 00:10:46.445 on, 124 00:10:49.030 --> 00:10:51.430 what is it exactly, essentially, 125 00:10:53.430 --> 00:10:54.470 who is... 126 00:10:56.550 --> 00:10:58.710 It's 127 00:10:58.710 --> 00:11:08.685 it's not on who's sending the request, but it it was essentially just on the more approval. So is the more approval exactly coming from the same consumer? And could someone 128 00:11:08.845 --> 00:11:10.285 potentially make 129 00:11:10.525 --> 00:11:12.765 an unauthorized 130 00:11:13.405 --> 00:11:19.610 request to you, not seeing someone else's data or anything, but essentially getting free use of Morpheus token. So it it wasn't, 131 00:11:19.850 --> 00:11:36.745 and everyone can look through the code. It's all public. All the PRs are public. But, you'll see that that bug fix went in. It's just more protection for providers. You know, we're gonna find little things here and there of of, you know, now that that Babel and Astra and everything, people find, like, workarounds that you'd never see before. So, 132 00:11:37.545 --> 00:11:47.070 we would recommend for everyone to update their provider notes just from normal safety practices, just like you get in iOS or in Microsoft update however often. 133 00:11:47.230 --> 00:11:51.790 Just great security hardening for the protocol, and and, you know, we're happy to 134 00:11:52.190 --> 00:11:58.685 find something else and and get it out there. But there's no, like, big red fire alarm or anything like that. 135 00:11:59.485 --> 00:12:05.165 Just kind of business per usual and stay updated to... On the latest code to get the latest security updates. 136 00:12:05.644 --> 00:12:06.845 Cool. Thanks, Kyle. 137 00:12:07.485 --> 00:12:09.485 I do have a couple notes 138 00:12:09.805 --> 00:12:24.240 from Alan that he had sent me on it, which, by the way, I just suspect that anytime Alan sends me something, he's taking his own internal notes and then giving them to an agent and saying, okay. Now, like, you're seven. And then sending him over to me. 139 00:12:25.360 --> 00:12:27.600 But, yeah, basically, what what he had said is, 140 00:12:28.705 --> 00:12:46.900 in layman's terms, the node was too trusting about who owned a session. It could accept a key that did not match the wallet on that session, and a session opened with one provider could theoretically be used by another. After the session ended, the provider also stopped sending the closed receipt. So a normal close recorded on change 141 00:12:47.140 --> 00:12:56.900 and as a dispute, and the provider did not get credit. So basically, what you just said, Kyle, that it's just a a further verification that the correct provider is, 142 00:12:57.565 --> 00:13:00.925 and and the correct wallet are being connected with the correct more, 143 00:13:01.165 --> 00:13:04.685 basically. Yeah. I I guess, technically, there is another... 144 00:13:05.565 --> 00:13:06.125 Right? 145 00:13:06.605 --> 00:13:10.300 Yeah. Technically, someone could come in and and, I guess, dispute 146 00:13:10.300 --> 00:13:12.460 that the session was legitimate, 147 00:13:12.460 --> 00:13:20.940 and that would cause some lockup of more for a certain amount of time. But, again, corner case of a corner case Yeah. On that, we're, 148 00:13:21.420 --> 00:13:27.845 you know, close the corner cases, obviously, and update your node. But, I don't think this is, 149 00:13:28.885 --> 00:13:29.845 yep, good 150 00:13:30.325 --> 00:13:50.180 practice. Yeah. Yeah. I was gonna say, this is why I love, you know, our community and all you guys working on this type of thing because it is like the edge case of edge case of edge case, and yet it's still identified and fixed. And we're hearing about it after it was identified and fixed, you know, which is Yeah. The best best thing we can wish for. So awesome. Thanks, guys. I'll give one I'll give one more 151 00:13:50.345 --> 00:13:51.545 there that 152 00:13:51.945 --> 00:13:56.105 is important. And this one,

you know, anyone who is a... 153 00:13:58.185 --> 00:14:01.065 Very closely tracking or or get updates 154 00:14:01.065 --> 00:14:04.185 in the staging branch or the testing branch will see that 155 00:14:04.760 --> 00:14:07.080 more work has been done on... 156 00:14:09.320 --> 00:14:22.705 To to fix one concern that I think has been lagging a little bit. Right? And and we like to be super open with with where we think the protocol is amazing, where we think we need to spend a little time doing work. So one topic 157 00:14:22.705 --> 00:14:23.425 was 158 00:14:23.825 --> 00:14:26.465 we have a lot of decentralized 159 00:14:26.465 --> 00:14:48.155 providers in... You know, who are providing into the system. Right? And everyone's a little bit different. Some people are are hosting themselves in their computer in the closet, wink wink. Right? Some people are, you know, going to be running from massive data centers. Some people are going to be pointing through to OpenRouter or to Venice or to really name your name your provider. And 160 00:14:49.435 --> 00:14:53.115 each one could have a slightly different format 161 00:14:53.115 --> 00:14:57.195 that they accept or slightly different features that they support. 162 00:14:57.355 --> 00:14:58.955 Right? So when you call one model, 163 00:14:59.520 --> 00:15:04.640 if... You know, we need to make sure that if you call a model and say, 164 00:15:04.880 --> 00:15:08.000 I want this to run with low 165 00:15:08.000 --> 00:15:10.400 effort or low level of thinking. 166 00:15:10.560 --> 00:15:13.360 Right? That needs to work across the board 167 00:15:13.715 --> 00:15:16.115 no matter which provider you get. 168 00:15:16.595 --> 00:15:19.155 So what we're implementing is 169 00:15:20.355 --> 00:15:23.075 a type of routing, if you will, that says, 170 00:15:24.595 --> 00:15:25.555 I'm a provider. 171 00:15:26.139 --> 00:15:30.540 I'm running in this manner. This is how you talk to me. 172 00:15:31.100 --> 00:15:33.500 So that when a request comes in, 173 00:15:33.740 --> 00:15:37.580 it knows how to pass that information along 174 00:15:37.740 --> 00:15:40.220 to the provider in the language that they speak. 175 00:15:40.795 --> 00:15:43.835 So let's say Bo is hosting 176 00:15:43.915 --> 00:15:46.795 and he uses the the parameter, 177 00:15:48.315 --> 00:15:48.955 you know, 178 00:15:49.595 --> 00:15:51.355 thinking underscore level. 179 00:15:51.595 --> 00:15:54.715 And David is a provider and uses the... 180 00:15:55.035 --> 00:16:10.135 Uses it just as thinking as his parameter. And if I send thinking underscore level, he's like, this guy's talking French. I I don't understand French, so I'm I'm not gonna do anything here, and doesn't apply that proper parameter. 181 00:16:10.375 --> 00:16:14.135 So what we're trying to do is kind of create this 182 00:16:14.455 --> 00:16:20.935 this single unified language that's able to talk to Bo and talk to David to make sure that me as a user, 183 00:16:21.500 -->

00:16:25.260 I don't care if you're speaking French or or whatever. 184 00:16:25.980 --> 00:16:27.580 I just need you to do 185 00:16:27.899 --> 00:16:29.980 what I'm asking you. So 186 00:16:31.339 --> 00:16:37.180 that is something that we have in testing right now, and we're going through. As you can expect, it's fairly complex 187 00:16:37.795 --> 00:16:44.195 because there... There's just so much different variability with with what people are hosting and what people are running. But we recognize 188 00:16:44.275 --> 00:16:44.835 that 189 00:16:45.155 --> 00:16:46.115 stability 190 00:16:46.115 --> 00:16:47.555 and simplicity 191 00:16:47.715 --> 00:16:53.510 from a user interacting with the platform is is paramount to being able to use it in... 192 00:16:53.830 --> 00:16:59.910 At scale or in an enterprise environment or for an agent that... That's really doing something that's critical to you. So 193 00:17:00.870 --> 00:17:04.230 that's what we're in the process of doing, and and we're hoping to 194 00:17:04.534 --> 00:17:11.815 to get that out, quote, unquote, soon. Right? We're testing it. Maybe it's a day away. Maybe it's two weeks away. We we we just gotta work through it a little bit more. 195 00:17:15.495 --> 00:17:16.455 Awesome. Thank you, Kyle. 196 00:17:19.429 --> 00:17:20.549 On to 197 00:17:20.870 --> 00:17:24.470 the demo type things, which I know we don't have 198 00:17:24.870 --> 00:17:30.549 an exact... So for those who are maybe watching this week and didn't watch last week or just need a refresher, 199 00:17:31.304 --> 00:17:33.144 we have been talking about 200 00:17:33.625 --> 00:17:36.744 agents, and we wanna show off the things that agents can do, 201 00:17:37.065 --> 00:17:40.104 and also just things being built on Morpheus. 202 00:17:40.504 --> 00:17:41.065 And 203 00:17:41.625 --> 00:17:42.424 sometimes 204 00:17:42.664 --> 00:17:58.590 that's hard to do because we're talking about concepts more so. Like, it would be like... I think Kyle, you had this great quote last week where you were like, it would be like showing someone the code base of a website and being like, look, there it is. You know, like, here's here's the code base. Hope you can read code. 205 00:17:59.955 --> 00:18:01.634 However, like, talking about 206 00:18:02.115 --> 00:18:04.914 some of those concepts and things are really important 207 00:18:04.995 --> 00:18:12.995 as things get developed and we can show off, you know, tools and experiences and things like that. But the thread that we've been pulling on since last week was this idea of 208 00:18:13.940 --> 00:18:15.860 recursive self improvement, 209 00:18:15.860 --> 00:18:17.060 so RSI, 210 00:18:17.060 --> 00:18:19.539 specifically for in the agent context. 211 00:18:19.780 --> 00:18:20.980 So your agent 212 00:18:21.140 --> 00:18:24.900 paying attention and recording the things that you're doing or having it do rather, 213 00:18:25.875 --> 00:18:28.995 remembering it, learning from it, and

improving itself. 214 00:18:29.155 --> 00:18:31.715 And last week, David showed Bernardo, 215 00:18:31.715 --> 00:18:42.129 which is his open source agent that has a open source GitHub that you can pull things down from, a lot of awesome skills and things like that. And we talked about the agent's brain. 216 00:18:42.450 --> 00:18:45.329 This week, we were gonna talk a little bit about looping. 217 00:18:45.490 --> 00:18:52.690 And what what is looping? Like, how does that work in practice? Why does it matter? And, like, kinda where does it fit into RSI? 218 00:18:53.495 --> 00:18:56.695 David, can you, I guess, start on any or all of those things? 219 00:18:57.815 --> 00:18:59.335 I'd love to. So 220 00:18:59.575 --> 00:19:05.015 once you've got an agent, let's say you're using OpenCrawl or Hermes or something like that, 221 00:19:05.495 --> 00:19:06.135 and 222 00:19:06.300 --> 00:19:11.100 we talked a little bit about how you create a brain for it. Right? Load up context and 223 00:19:11.340 --> 00:19:14.780 map it, and then, you're able to use it. Okay. 224 00:19:15.180 --> 00:19:25.174 Your agent's pretty good at doing tasks if you tell it what to do, and it understands sort of the broader context of what it's done before and its best practices. 225 00:19:25.495 --> 00:19:27.735 It's got its its learnings from its brain. 226 00:19:28.375 --> 00:19:32.774 The next thing you can really improve on is this concept of looping. 227 00:19:33.175 --> 00:19:43.870 And so you can think of this... It's not really different agents, but think of it as, like, a series of cron jobs is one way you can implement it. And you can say, okay. 228 00:19:44.750 --> 00:19:47.389 The agent wakes up every hour, 229 00:19:47.950 --> 00:19:50.830 and it picks up the next task in the backlog. 230 00:19:51.150 --> 00:19:51.470 Right? 231 00:19:52.515 --> 00:19:54.035 So number one, 232 00:19:54.355 --> 00:19:58.355 you have your agent start building a backlog. Hey. If I tell you something to do, 233 00:19:58.755 --> 00:20:03.315 add it to the backlog, give it a number, task zero zero one, zero zero two, whatever. 234 00:20:03.890 --> 00:20:09.410 And okay. Now I've got things that I've told my agent, hey. I want you to work on x y z. Great. 235 00:20:09.890 --> 00:20:14.290 If you already have a standard process for your agent to do development, 236 00:20:14.530 --> 00:20:18.290 it should usually be able to get tasks you put into the backlog done, 237 00:20:19.205 --> 00:20:22.965 but not always. Gets stuck, gets confused, 238 00:20:23.365 --> 00:20:25.044 runs out of context, 239 00:20:25.125 --> 00:20:28.245 all kinds of edge cases on why it could stop. So 240 00:20:28.565 --> 00:20:31.284 the concept of looping is very simple. 241 00:20:31.685 --> 00:20:33.605 You need a coordinator agent 242 00:20:34.190 --> 00:20:35.950 that checks in on the worker. 243 00:20:36.430 --> 00:20:39.950 So if the worker wakes up every hour and starts a task, 244 00:20:40.510 --> 00:20:43.549 I have the coordinator wake up every thirty minutes 245 00:20:43.870 --> 00:20:44.990 and see 246 00:20:45.070 --> 00:20:46.990 if the agent's being successful. 247 00:20:47.230 --> 00:20:51.934 Has the session stopped? Did it run out of context? Did, you know, it gets stuck or whatever? 248 00:20:52.255 --> 00:20:54.335 And if it's stuck, it restarts it. 249 00:20:54.735 --> 00:20:56.815 And so you sort of have an agent 250 00:20:57.054 --> 00:20:57.855 supervising 251 00:20:57.855 --> 00:20:58.815 an agent. 252 00:20:59.054 --> 00:21:05.700 Right? And that's the loop. Now it can go and it keeps going through that task until it gets to a finish line, 253 00:21:06.340 --> 00:21:12.179 and then it goes back to the beginning of the loop and starts in the next task in the backlog. 254 00:21:12.500 --> 00:21:15.140 So this is how you can really move 255 00:21:15.525 --> 00:21:16.245 from 256 00:21:16.325 --> 00:21:23.205 sitting there and talking to it and then waiting twelve minutes or fifteen minutes for it to get the next thing done to instead 257 00:21:23.525 --> 00:21:25.684 just focus on giving it tasks, 258 00:21:26.325 --> 00:21:35.900 adding to the backlog and knowing that it's gonna work through all of those items. So after setting up my looping, I think I've gotten it through a 173 259 00:21:35.900 --> 00:21:42.940 successful items, and I didn't have to babysit all those for hours at a time. Right? I have my agent 260 00:21:43.485 --> 00:21:45.325 who knows the best practices, 261 00:21:45.645 --> 00:21:48.524 check-in on the worker, make sure it's getting done. 262 00:21:48.925 --> 00:21:53.085 Right? And there's sort of things you can do to improve this type of loop. 263 00:21:53.725 --> 00:22:00.059 You might make the choice that, hey. That's great, but I still want it to come to me at the end 264 00:22:00.460 --> 00:22:01.580 of the process 265 00:22:01.740 --> 00:22:13.815 and give me all the the task that have been built. And before it goes public, I wanna I wanna see it, and I wanna push that button. That's a choice you can make. What I've ended up doing is 266 00:22:13.975 --> 00:22:15.575 I have an auditor 267 00:22:15.655 --> 00:22:16.774 in my loop 268 00:22:17.655 --> 00:22:20.295 that is looking at the final tasks. 269 00:22:20.535 --> 00:22:22.135 I call it my architecture 270 00:22:22.295 --> 00:22:22.695 agent, 271 00:22:23.290 --> 00:22:30.010 and it understands because I have the the brain already built, the broader context of what I'm trying to accomplish, 272 00:22:30.490 --> 00:22:32.330 and it can make a pretty good judgment. 273 00:22:32.570 --> 00:22:33.289 Is this 274 00:22:34.010 --> 00:22:41.855 on task and and something that I want for for the broader project? Is it sort of a scope creep? Like, is this task really, 275 00:22:42.255 --> 00:22:43.615 you know, properly 276 00:22:43.615 --> 00:22:44.335 built? 277 00:22:44.495 --> 00:22:46.095 And about half the time, 278 00:22:46.895 --> 00:22:48.655

it'll come back and say, 279 00:22:49.215 --> 00:22:58.580 this is trivial. It's on it's on task. Like, it it... It's something that I can... The architecture agent can approve without having to escalate to me. 280 00:22:58.980 --> 00:23:04.260 But a lot of times, it's like, oh, this is out of scope or oh, this is, you know, financial 281 00:23:04.340 --> 00:23:09.765 or it's got, you know, money or keys or something associated with it where it needs me to review. 282 00:23:10.085 --> 00:23:10.965 And so 283 00:23:11.285 --> 00:23:12.325 my goal 284 00:23:12.485 --> 00:23:15.445 for the loop is to continue improving 285 00:23:15.845 --> 00:23:17.445 my architecture agent 286 00:23:17.765 --> 00:23:27.490 and making it better and better so that six out of 10 tasks, seven out of 10 tasks, eventually, hopefully, nine out of 10 tasks, it can approve and push public. 287 00:23:27.730 --> 00:23:34.610 And in order to accomplish that, I also have a learning agent. And the learning agent is looking at my decisions 288 00:23:35.175 --> 00:23:36.935 and finding patterns 289 00:23:36.935 --> 00:23:39.495 and improving that architecture agent. 290 00:23:39.895 --> 00:23:41.255 Right? And so 291 00:23:41.575 --> 00:23:45.015 if you think about that whole process, you have the worker, 292 00:23:45.895 --> 00:23:47.495 you have the coordinator 293 00:23:47.655 --> 00:23:54.529 that, you know, keeps the worker on track, You have the architecture agent making sure that everything 294 00:23:54.770 --> 00:24:01.330 was, you know, as expected for the outcome, and then it's elevating it to me for final decisions. 295 00:24:01.570 --> 00:24:03.650 And it's taking those decisions and improving 296 00:24:04.155 --> 00:24:05.275 the architecture 297 00:24:05.435 --> 00:24:10.554 agent for for next time. One thing I didn't mention, at the very beginning of the process, 298 00:24:11.035 --> 00:24:15.195 before the workers even picked up the item in the backlog, 299 00:24:16.075 --> 00:24:17.515 I have an agent 300 00:24:17.899 --> 00:24:19.419 use my brain 301 00:24:19.580 --> 00:24:22.619 to enhance the prompt in the backlog. 302 00:24:23.500 --> 00:24:24.779 So this way, 303 00:24:25.340 --> 00:24:30.380 if all I told my agent was, hey, follow-up on item number 304 00:24:30.075 --> 00:24:32.075 for install OpenClaw, 305 00:24:32.075 --> 00:24:36.154 the improvement to do with, you know, updating the software for the leases. 306 00:24:36.554 --> 00:24:39.754 That might not be enough context for it to really know what the heck I'm talking about. 307 00:24:40.395 --> 00:24:48.070 But it's in my brain. It's in my OpenClaw brain. And so by taking the context from the open claw brain, 308 00:24:48.470 --> 00:25:10.165 looking in that backlog item, it will expand it to a couple of paragraphs. Like, here's the context for exactly what we did last time, and here are the tools, and the reference, and the memory you need, and, you know, the best practice to follow. And so that really improved my loop. That went from, like, it kinda guessing at my intent to really knows my intent, because I've had 6,000 309 00:25:10.165 --> 00:25:10.965 sessions 310 00:25:11.285 --> 00:25:12.325 with my agent, 311 00:25:12.880 --> 00:25:15.999 and it's learned from those, and now it's improving 312 00:25:16.000 --> 00:25:17.200 the quality 313 00:25:17.520 --> 00:25:19.919 of the write ups that start at the beginning 314 00:25:20.159 --> 00:25:21.359 of the backlog. 315 00:25:21.679 --> 00:25:22.399 So 316 00:25:22.480 --> 00:25:25.919 puts up in the backlog, gets worked on, keeps on track. 317 00:25:26.865 --> 00:25:28.465 At the end, it's audited, 318 00:25:28.625 --> 00:25:31.505 makes sure it makes sense, either gets promoted or 319 00:25:31.825 --> 00:25:37.825 elevated to me for a decision, and then it learns from those decisions. So if you can take that context... 320 00:25:37.825 --> 00:25:43.390 And I'm gonna put a copy of this up in in public. I've made a third party version of my looping 321 00:25:43.470 --> 00:25:46.030 that I'm putting up as a best practice on the GitHub. 322 00:25:46.430 --> 00:25:47.870 If you can implement that, 323 00:25:48.590 --> 00:25:52.190 it's a game changer because you go from building things one at a time 324 00:25:52.865 --> 00:25:59.505 to basically instead focusing on, like, what you want to build and let your agent really kinda churn through it. So... 325 00:25:59.905 --> 00:26:04.865 I'm really glad you said that because you you just answered my question, 326 00:26:05.025 --> 00:26:12.929 which was gonna be, okay. Now I'm dumb and don't know how to structure any of what you just said. Not dumb. I'm talking about me specifically. 327 00:26:14.130 --> 00:26:16.770 And don't know how to structure any of those things you just said, 328 00:26:17.090 --> 00:26:29.145 is there a tool that I can go to or something you can point me to that I can look at with my agent and say, hey. This is the thing I wanna implement. Help me organize it. Yes. Okay. Yeah. And and that's that's what I wanna get to is 329 00:26:29.385 --> 00:26:30.425 I publish... 330 00:26:30.505 --> 00:26:33.785 Here's the guide to how to implement looping. 331 00:26:33.865 --> 00:26:36.520 Here's an MD file. Hand this to your agent. 332 00:26:37.000 --> 00:26:39.559 And what it's gonna do is really actually pretty simple. 333 00:26:39.800 --> 00:26:47.320 It's gonna schedule a series of cron jobs that do each of the tasks I just talked about, and then you sort of have a good orchestration 334 00:26:48.315 --> 00:26:57.114 for how this works. And it's helped me moving from doing one task at a time to my agent getting 10 tasks done, a 100 tasks done. 335 00:26:57.434 --> 00:27:20.304 I see a roadmap pretty soon where my agent might be able to get a thousand tasks done for me. I'm feeding in pull requests now from GitHub and issues and the backlog of ideas and suggestions and bug reports and all this stuff that sort of builds up that

you don't have time to possibly keep up with. And and to tie this all back into what's happening on the security side. 336 00:27:21.184 --> 00:27:23.424 Everybody's dealing with this. Every 337 00:27:23.825 --> 00:27:27.664 time a new model comes out now, everybody rushes 338 00:27:27.664 --> 00:27:31.729 to test and make sure they found all the bugs in their protocol because 339 00:27:31.730 --> 00:27:38.609 bad actors are also rushing to see if they can find honey pots and and issues. And how do you sort through 340 00:27:39.090 --> 00:27:40.609 all of those reports? 341 00:27:40.850 --> 00:27:42.850 Like, even if somebody is a white hat 342 00:27:43.255 --> 00:27:48.614 hacker and they're trying to help you and and they're putting in a report to pick up a bug bounty, 343 00:27:49.655 --> 00:27:51.335 you got 500 reports. 344 00:27:51.655 --> 00:27:55.174 How do you read all the reports and know which ones are legit? 345 00:27:55.669 --> 00:27:57.669 You need the agentic system. 346 00:27:57.990 --> 00:28:02.549 Right? And so the only way to keep up with agents is to implement agents. 347 00:28:02.630 --> 00:28:06.869 So that's where everybody's getting to. And so for me, it's been really powerful, 348 00:28:07.029 --> 00:28:13.465 and I could see a path where I can have a thousand tasks done by my agent, and there's no way I'm gonna be able to do that manually. 349 00:28:14.025 --> 00:28:19.144 So it's pretty wild, so I'll I'll stop there, but gonna open the source this. We'll include the link, 350 00:28:19.305 --> 00:28:23.785 all that kind of good stuff. But looping, it's very powerful. It's one of the big payoffs 351 00:28:24.140 --> 00:28:25.900 is you're able to multiply 352 00:28:25.900 --> 00:28:37.580 your output. So start an agent, build its brain, then you can do looping. And we'll talk more about RSI next week, but that's the next layer. Yeah. That is super helpful and exciting. And also 353 00:28:37.995 --> 00:28:38.715 just 354 00:28:39.355 --> 00:28:40.715 reiterates for me, 355 00:28:41.595 --> 00:28:43.274 we need a lot of inference 356 00:28:43.355 --> 00:28:55.999 to do. As you were talking, you were talking about a thousand tasks. I mean, I have, like, a handful of cron jobs basically that my agent does for me right now every day, and that chews through, like, a pretty decent amount of inference. Granted, 357 00:28:56.400 --> 00:28:59.679 I am certainly no specialist in making sure, 358 00:29:00.000 --> 00:29:07.975 the efficiency of my inference is being considered, so I'm sure there's ton... Like we were talking about earlier, there's tons of things that I am, 359 00:29:08.775 --> 00:29:13.654 using the jackhammer to nail a picture to the wall when I don't necessarily need to. 360 00:29:14.215 --> 00:29:17.894 But, I mean, I'm just picturing, you know, we have 361 00:29:18.775 --> 00:29:29.610 thousands, hundreds of thousands, millions of people like you who have an agent that's accomplishing thousands or more tasks for them all the time, and just like, 362 00:29:30.330 --> 00:29:38.714 everyone is underestimating how much compute we need. You know, it is it is so much more scarce and valuable than people realize and it's it's gonna become a 363 00:29:38.955 --> 00:29:45.115 question of literal do we have enough energy to do it to do all of the things. Yeah. 364 00:29:45.300 --> 00:29:47.460 I'm pretty sure the the 365 00:29:47.700 --> 00:29:50.580 demand for intelligence is pretty unbounded. 366 00:29:50.980 --> 00:30:05.914 Like Yeah. How how many agents do you want working for you? How many tasks do you want them to do? It's probably more than one. It it might be a 100. It might be a thousand because Yeah. We're sort of all this... The term people forget is latent demand. 367 00:30:06.395 --> 00:30:18.710 Mhmm. Because all these things I wanna do, I just don't have time to do. Right? I should be reading that article. I should be, you know, checking that email. I should be doing a lot of stuff. And today, it kinda just falls through the cracks. 368 00:30:18.870 --> 00:30:19.429 Right? 369 00:30:20.309 --> 00:30:23.990 All of a sudden, you can deploy your army of agents 370 00:30:23.990 --> 00:30:33.125 to catch up with this stuff. It's get... It gets pretty cool. Right? Because, you know, say I go to a conference like this. I meet a ton of cool people. 371 00:30:33.525 --> 00:30:36.565 Now I just give their contact 372 00:30:36.885 --> 00:30:38.005 to my agent, 373 00:30:38.485 --> 00:30:40.245 and I have a personal CRM. 374 00:30:40.740 --> 00:30:45.299 And it reminds me if something I'm doing is relevant to somebody I've met. 375 00:30:46.020 --> 00:30:48.660 Like, I could never keep 10,000 376 00:30:48.660 --> 00:30:58.345 people in my short term memory of like, oh, yeah. What I'm building is gonna be relevant to him, and I talked to him about that last year. All of a sudden, you can start discovering 377 00:30:58.345 --> 00:31:04.985 that value in the network you already have. So for me, that's huge. Yeah. I was gonna ask, have there been... 378 00:31:05.305 --> 00:31:13.440 Like like, from token $20.49 of last to this year, has there been a noticeable increase of, like, agent interactions between people? 379 00:31:14.240 --> 00:31:15.759 Yeah. 100%. 380 00:31:15.920 --> 00:31:18.879 A lot of people I'm talking to now have an agent, 381 00:31:19.360 --> 00:31:28.695 but they're still dealing with it like a single player type of deal. Like, they've got an agent and they message their agent and, oh, that's cool and I'll send them my agent that link. 382 00:31:29.335 --> 00:31:31.815 But they're still not thinking of it collaboratively. 383 00:31:32.055 --> 00:31:39.735 Mhmm. Right? So things like Buzz and other collaborative tools really haven't hit yet. Like, yet has realized 384 00:31:40.410 --> 00:31:42.730 that their agent's gonna work with your agent. 385 00:31:43.130 --> 00:31:45.130 And that maybe you shouldn't set up a call 386 00:31:45.450 -->

00:31:59.544 or have another meeting, but you just have your agents go ahead and execute on the context you've already got... Given them. Like, nobody's thinking about that yet. People are just getting to the point where, like, they have an agent, or they know they need to set up an agent, 387 00:31:59.785 --> 00:32:05.225 or they still think their agent is Claude, but now they're finally realizing, oh, that's a tool. 388 00:32:05.625 --> 00:32:06.985 It's not my agent. 389 00:32:07.305 --> 00:32:15.639 I need something that sits above my tools and can use all of them, right? So people are people are getting there though, but there's still definitely 390 00:32:15.960 --> 00:32:18.600 like getting their first agent. Twenty twenty 391 00:32:19.240 --> 00:32:23.914 six has been the year everybody got their first agent, at least like high agency, 392 00:32:23.915 --> 00:32:25.274 high capability folks. 393 00:32:25.435 --> 00:32:26.075 Yeah. 394 00:32:26.555 --> 00:32:28.315 Yeah. It's so interesting. I I 395 00:32:28.955 --> 00:32:32.715 can't wait to see just like how that develops and if by 396 00:32:33.435 --> 00:32:45.859 next year, for example, most people do have a personal agent and recognize the difference. Because, yeah, I think that there's still... Actually, just I know from interacting with people, there is still a huge swath of people that have not 397 00:32:46.340 --> 00:32:47.299 totally 398 00:32:47.299 --> 00:32:56.465 come to the realization of, like, my agent sits above the tools that I'm using, even if the tools that I'm using can do agentic things, 399 00:32:56.705 --> 00:32:59.985 like like a Claude, which is technically agentic, 400 00:33:00.145 --> 00:33:05.809 but not your agent. Like, it's not representing you. You don't own that, You know? And it's... 401 00:33:06.850 --> 00:33:09.970 Kyle, you sound like you're a wood chipper. Yeah. Sorry. 402 00:33:11.330 --> 00:33:15.730 So the... And I I can't switch the microphone midstream. Oh, it's okay. 403 00:33:17.090 --> 00:33:18.450 I think the 404 00:33:18.934 --> 00:33:21.254 something important to take into consideration 405 00:33:21.335 --> 00:33:21.975 is 406 00:33:23.335 --> 00:33:26.215 that not everyone is going to 407 00:33:26.855 --> 00:33:29.975 self host it either. Right? Like, there's... 408 00:33:30.615 --> 00:33:32.054 I think that adoption 409 00:33:32.870 --> 00:33:37.190 in AI agents, no matter where it comes from, is 410 00:33:37.430 --> 00:33:38.230 valuable 411 00:33:38.390 --> 00:33:49.994 for Morpheus. It's valuable for the progression of AI and and just the industry overall. And it's like a... You know, I'd said it last week, but a rising tide lifts all boats. Right? If people go in and 412 00:33:50.394 --> 00:33:50.794 learn 413 00:33:51.755 --> 00:33:54.794 how to use and what AI agents are from 414 00:33:54.875 --> 00:33:59.994 Dots or Rockbaud or Muse or Instinct or 415 00:34:00.075 --> 00:34:02.234 the other project from working on Liaison, 416 00:34:02.235 --> 00:34:05.999 then, like, that's a win. Right? That's a win for all of us because 417 00:34:06.480 --> 00:34:07.759 more, you know, 418 00:34:08.960 --> 00:34:10.240 more knowledge, 419 00:34:10.480 --> 00:34:11.200 more 420 00:34:11.840 --> 00:34:12.880 notoriety, 421 00:34:12.960 --> 00:34:14.240 more utilization, 422 00:34:14.240 --> 00:34:16.800 more development around this is 423 00:34:17.244 --> 00:34:18.205 just creating 424 00:34:18.685 --> 00:34:23.805 more demand and and just making it more part of our daily lives. So I think it's like... 425 00:34:24.285 --> 00:34:35.670 And, David, we said this so long ago that, like, we love the the competition. We love the the development in the AI. Even if someone's, like, in a similar realm as as what we're working on, it's, like, 426 00:34:36.230 --> 00:34:39.510 great. Do it. Like, we all have the same, 427 00:34:39.750 --> 00:34:44.390 you know, underlying goals of what we want for the world with this. And 428 00:34:45.444 --> 00:34:48.724 starting with closed source and a gradual 429 00:34:48.724 --> 00:34:51.445 transition to open source when people, like, 430 00:34:52.005 --> 00:34:56.325 you know, see those values and and want to go down that path, I think, is just... 431 00:34:57.880 --> 00:34:59.400 I'm not surprised, 432 00:34:59.400 --> 00:35:10.200 and and I'm happy about it as well on on how this is shaping out. And seeing the major companies invest in this technology and bring this to market is just a testament to 433 00:35:11.394 --> 00:35:16.595 the vision we had being the right vision. And, you know, we have a a leg up on it already. 434 00:35:17.394 --> 00:35:21.395 Yeah. So far, the Morpheus community has been right about everything. I just wanna put that out there. 435 00:35:21.795 --> 00:35:28.490 And I don't see that slowing down anytime soon. And when I say that, I don't mean me. I mean the people the people that I'm listening to. 436 00:35:29.690 --> 00:35:30.330 But, 437 00:35:31.130 --> 00:35:58.050 yeah, I know. You're you're 100% spot on, Kyle. And, like, it's it's all just about staying ahead of the curve. I saw a really funny tweet or a tweet that made me laugh anyway this morning that was like, you guys, normies are not vibe coding their own software. Twitter is a bubble. Relax. I'm like, I know, but... And that's probably true. I'm sure the average person I'm bumping into in the grocery store is not sitting at home vibe coding their own apps, But it feels like everyone is, because we are in this in this bubble of like, 438 00:35:58.290 --> 00:36:04.130 just people shipping stuff left and right, and like, oh my gosh, like I need to move so fast to keep up. I 439 00:36:04.850 --> 00:36:07.735 think that it's what's cool about decentralized 440 00:36:07.735 --> 00:36:09.735 AI though relative to even, 441 00:36:09.895 --> 00:36:15.655 like, the earlier days of crypto, like the twenty seventeen ICO mania and stuff, is 442 00:36:16.135 --> 00:36:16.855

this 443 00:36:17.015 --> 00:36:18.695 has such a in your face 444 00:36:19.180 --> 00:36:20.540 practical application 445 00:36:20.540 --> 00:36:29.340 that you can interact with every day or just like mess around with. Like sometimes I'm bored and I'm just like, I'm gonna mess around with my agent. I'm just gonna book a flight. 446 00:36:30.380 --> 00:36:31.100 Yeah. Wish. 447 00:36:32.265 --> 00:36:38.744 No. Because that's like the only thing that actually works well on on DOS and Grokbat right now is Book a dinner 448 00:36:39.145 --> 00:36:47.464 reservations. Yeah. You're gonna find out that it's actually the airlines paying for it and developing it. It's just that's the only tool that works right now. That's so funny. Yeah. 449 00:36:48.240 --> 00:36:51.600 But but I agree, Bo. Like, the realness of 450 00:36:51.920 --> 00:36:55.760 I'm using Morpheus every day. Like, I'm so happy. Yeah. Every day. 451 00:36:56.160 --> 00:37:04.025 Three years of building it, like, I'm actually getting to use it all the time and not just, like, once in a while. Like, it's literally powering 452 00:37:04.345 --> 00:37:05.945 most of my workflows 453 00:37:06.105 --> 00:37:19.280 at this point. Like, running advertising campaigns, I've got my agent doing, right, building websites. I've got my agent doing it, like, doing coding on GitHub. I've got my agent doing it. Right? Auditing smart contracts, you name it. Right? So, 454 00:37:19.680 --> 00:37:22.400 yeah, it's it's pretty cool. I accept we're in the bubble, 455 00:37:22.720 --> 00:37:23.680 but, you know, 456 00:37:24.640 --> 00:37:29.680 we tend to be, like, a year, a couple years ahead, and, like, everybody's gonna get there. 457 00:37:30.325 --> 00:37:31.605 To Kyle's point, 458 00:37:32.485 --> 00:37:34.645 Morpheus can power Grok. 459 00:37:34.645 --> 00:37:35.285 And, 460 00:37:35.445 --> 00:37:40.725 you know, if you wanna use Hermes and Open Claw, go for it. Like, we're harness agnostic. 461 00:37:40.725 --> 00:37:45.045 Right? The whole point is to get people to the point where they own an AI 462 00:37:45.230 --> 00:37:48.109 and they have an agent and it's making their life better. 463 00:37:48.349 --> 00:37:50.670 Dumb. Right? So, yeah, it's, 464 00:37:51.150 --> 00:37:52.190 it's pretty cool. 465 00:37:53.950 --> 00:37:55.150 Couldn't agree more. 466 00:37:56.270 --> 00:38:00.829 I'm trying to think because that that was all, I guess, on the looping conversation things related to Morpheus. 467 00:38:01.265 --> 00:38:10.065 I didn't have any big news stories this week, specifically on AI, which is kinda crazy because it seems to be like... I'm... There's obviously stuff happening. Always Hey. 468 00:38:10.625 --> 00:38:13.025 Hey. Don't don't fade on Mistral. 469 00:38:13.265 --> 00:38:14.305 Don't Oh, okay. Don't... 470 00:38:14.810 --> 00:38:16.730 Tell me what was the latest from Mistral? 471 00:38:17.290 --> 00:38:19.050 Trillion parameter model. 472 00:38:19.609 --> 00:38:22.570 It's really hitting some some cool benchmarks. 473 00:38:23.930 --> 00:38:26.890 If your thing is like you're skeptical about 474 00:38:27.755 --> 00:38:32.235 Eastern open weight models, great. Here's a Western open weight model. 475 00:38:32.635 --> 00:38:35.035 It says in preview today, 476 00:38:35.355 --> 00:38:41.115 and they're gonna release the weights end of the month. And Mistral's a really good record of basically 477 00:38:41.210 --> 00:38:45.609 not only open sourcing, like, they'll literally put it on, you know, BitTorrent, 478 00:38:45.609 --> 00:38:50.010 and here's the, you know, magnet link that will never be takable downable. 479 00:38:50.089 --> 00:38:51.130 Right? So 480 00:38:51.529 --> 00:38:59.945 really cool to see sort of another Western open rate model coming in, because before they'd done a bunch of small models. And you look historically, 481 00:39:00.105 --> 00:39:04.425 at one point, were like 30% of inference. They had a huge 482 00:39:04.744 --> 00:39:08.940 market share, and it sort of got eaten up the last year or two 483 00:39:09.340 --> 00:39:16.060 by people releasing bigger and bigger models that are way more powerful. So this is their first trillion parameter model. 484 00:39:16.300 --> 00:39:17.900 They call it LaChunk. 485 00:39:18.140 --> 00:39:20.060 Yeah. I was gonna say, I'm looking at their Yeah. 486 00:39:21.545 --> 00:39:23.305 Is their is their mascot? 487 00:39:23.385 --> 00:39:28.905 So it's cool. Like, people were were speculating the last few months. Like, are they gonna release a big model? They finally did. 488 00:39:29.865 --> 00:39:39.520 Though people kinda gave them a hard time, like, one of the benchmarks it's best at is regulation and legal. And they're like, of course, the EU model is good at regulation 489 00:39:39.840 --> 00:39:43.200 and legal. They're like, oh, you guys did it. So... 490 00:39:43.600 --> 00:39:49.265 You know? But it's credit to those guys. Glad to see it. Another great open weight model. 491 00:39:49.505 --> 00:39:51.425 And it looks like it's benchmarking 492 00:39:51.905 --> 00:39:53.025 right between 493 00:39:53.425 --> 00:39:55.825 GLM five point two and five point three. 494 00:39:56.305 --> 00:40:03.410 So right around like DeepSeek v four pro ish, it looks pretty solid. Like, if those benchmarks hold, 495 00:40:03.650 --> 00:40:07.170 it'll be another sort of, third, fourth really good 496 00:40:07.410 --> 00:40:09.170 solid open weight model. 497 00:40:09.330 --> 00:40:13.650 And I don't think it is that many active parameters. It's, like, sixteen thirty two billion. 498 00:40:14.235 --> 00:40:28.795 And so I think it should be not too bad to run as far as, like, how much hardware you need. It definitely needs some powerful hardware, you know, like a terabyte of of of RAM, but still credit to those guys. Good to see another big overweight release. 499 00:40:29.300 --> 00:40:35.060 Yeah. Shout out to those guys. Sorry. I missed that announcement, mister O. But

shout out to you guys. LeChonk is 500 00:40:35.460 --> 00:40:41.700 fantastic name. Just 10 out of 10, whoever your BD dash marketing person is and or agent. Good job. 501 00:40:42.345 --> 00:40:49.865 Yeah. This ad's fantastic. I'll have to... I'm looking forward to trying it. Well, and to your point, David, yeah, you need some really extreme hardware to run this locally. 502 00:40:49.945 --> 00:40:53.065 Oh, just kidding. No, I don't, because I use Morpheus. 503 00:40:54.665 --> 00:40:56.585 I'm sure it'll be I'm sure it'll be up there, 504 00:40:57.380 --> 00:41:02.580 you know, very, very quickly, if not at launch. So that is that is exciting. 505 00:41:03.140 --> 00:41:04.500 Very cool. Good The 506 00:41:05.060 --> 00:41:06.420 other stuff in the news, 507 00:41:06.740 --> 00:41:07.940 I guess adjacent 508 00:41:07.940 --> 00:41:13.845 to d AI, but just AI in general, is a lot a lot of talk about us solving math just generally, 509 00:41:13.925 --> 00:41:30.770 and how that's gonna pose some massive issues for, encryption and, you know, security for crypto projects and wallets. And you have the likes of Justin Drake and Vitalik chiming in, which that's... You know that meme of the guy who's sitting there and then the people talk and he leans forward in the chair? 510 00:41:31.330 --> 00:41:40.155 That's that's what I do when those two guys post. When either of those two guys post something long for him, I'm like, oh, wait. Gotta lock in. What are they talking about here? 511 00:41:41.275 --> 00:41:55.220 And it was, it was pretty interesting. I mean, all just feedback on how cryptography is obviously AI vulnerable and what the plans are for that. And, I don't know if you guys have any, like, specific takes or thoughts on this, but just, you know, it's becoming more and more, 512 00:41:55.700 --> 00:41:58.660 in the forefront of, you know, what people are paying attention to. 513 00:41:59.700 --> 00:42:05.700 So this was a big subject. I literally just came back from the quantum and privacy 514 00:42:06.260 --> 00:42:07.140 day 515 00:42:08.474 --> 00:42:09.595 put on by 516 00:42:09.914 --> 00:42:10.795 Qantas 517 00:42:11.115 --> 00:42:17.194 and the guys at Zcash and Near. Like, they had some of the world's best cryptographers. 518 00:42:17.194 --> 00:42:17.755 Right? 519 00:42:18.795 --> 00:42:24.500 Zucco was there. A lot of the people that built some of the cryptographic algorithms and zero knowledge proofs were there. 520 00:42:24.820 --> 00:42:26.820 And it was an interesting discussion. 521 00:42:27.140 --> 00:42:29.620 I would say, generally, the math 522 00:42:29.859 --> 00:42:30.740 solutions 523 00:42:30.740 --> 00:42:32.660 that OpenAI put out 524 00:42:33.380 --> 00:42:34.180 weren't 525 00:42:34.420 --> 00:42:34.980 mostly 526 00:42:35.484 --> 00:42:36.605 crypto related. 527 00:42:36.765 --> 00:42:43.005 It's clear that they may have gotten a request from a certain set of people that they don't release 528 00:42:43.244 --> 00:42:50.520 something that would would break cryptographic proofs, but it shows that, like, we're getting close to that point. Right? 529 00:42:50.920 --> 00:42:54.920 I I I put out a couple of tweets on this. If you see my post from today, 530 00:42:55.160 --> 00:42:57.480 I took a number of screenshots, 531 00:42:57.720 --> 00:43:00.840 and certain aspects are more vulnerable than others. 532 00:43:01.080 --> 00:43:01.640 Like, 533 00:43:02.040 --> 00:43:03.320 hashes, lattices. 534 00:43:03.320 --> 00:43:07.855 There's like, these are pretty secure. There's not a lot of attack vectors. But, oh, over here, 535 00:43:08.335 --> 00:43:15.535 these set of things are are really getting opened up as as attack vectors. But some of the cryptographers put it in in perspective. 536 00:43:15.855 --> 00:43:17.900 It was like, okay. Before, 537 00:43:18.299 --> 00:43:20.300 it would have taken the entire 538 00:43:20.460 --> 00:43:26.940 timeline of the universe to crack this thing, and the big breakthrough is it only take half the time of the universe 539 00:43:27.259 --> 00:43:31.180 to crack this thing. But the fact that it's moving at all 540 00:43:31.615 --> 00:43:37.135 might mean that they can keep compressing that, and it might get to some reasonable amount of time 541 00:43:37.295 --> 00:43:38.975 to crack it. And so 542 00:43:39.295 --> 00:43:45.295 a lot of people at the event were... Took it pretty seriously, and they're like, most of the countries are trying to get post quantum. 543 00:43:45.740 --> 00:43:48.619 So most of these vulnerabilities go away 544 00:43:48.859 --> 00:43:50.859 by next year. China 545 00:43:51.500 --> 00:43:55.260 is trying to do it by '26. The US has a timeline for '27. 546 00:43:55.260 --> 00:43:58.140 Most of the countries expect to have it done by '28, 547 00:43:58.625 --> 00:44:02.305 and Google is set for all their commercial stuff, it'll be by '29. 548 00:44:02.385 --> 00:44:06.625 Because people used to talk about quantum as in like a 2035, 549 00:44:06.625 --> 00:44:07.825 2040 thing. 550 00:44:08.065 --> 00:44:11.585 Now everybody's talking about it more as a the next couple years. 551 00:44:11.665 --> 00:44:13.505 And if there's enough of these breakthroughs, 552 00:44:13.585 --> 00:44:15.490 it might be the next couple of months. 553 00:44:15.890 --> 00:44:16.610 So 554 00:44:16.850 --> 00:44:18.770 hopefully people are taking it seriously, 555 00:44:19.250 --> 00:44:20.050 and a 556 00:44:20.370 --> 00:44:21.970 lot of people work very hard. 557 00:44:22.530 --> 00:44:29.170 My hope is it'll be like Y two k. Bo, you're a little too young to remember. I was alive, but you know. 558 00:44:30.685 --> 00:44:40.445 I'm just giving you a hard time. But Y2K is like everybody was hyped about like how crazy it would be if the computers lost track of the the date and like what was gonna happen. 559 00:44:41.405 --> 00:44:58.495 Nothing happened. Like Yeah. No no data centers crashed, and no planes fell out of the

sky, and things. My computer at my house, the clock went haywire, and I hit reset on the clock, and it was fine. Right? So... But that's because a lot of people did a bunch of hard work to fix the problem. 560 00:44:58.735 --> 00:45:04.415 And so, hopefully, if people are taking it seriously, we'll have a y two k kinda moment. Q day will arrive and 561 00:45:04.415 --> 00:45:06.415 not a lot of critical things will break. 562 00:45:06.655 --> 00:45:12.940 So Yeah. That's that's the optimistic scenario, but there's some things in crypto that aren't easy to switch. 563 00:45:13.980 --> 00:45:17.420 Right? What do you do with Satoshi's coins? What do you do with 564 00:45:18.299 --> 00:45:20.060 wallets that are hard to upgrade? 565 00:45:20.299 --> 00:45:25.260 There's there's some the sticky questions, but I'm I'm pretty optimistic that most of the protocols 566 00:45:25.835 --> 00:45:30.075 are moving pretty quickly to address it. Well, that's easy. You call Craig Wright. 567 00:45:33.115 --> 00:45:35.835 Happens with Satoshi's coins on 568 00:45:36.154 --> 00:45:37.674 the hard forks of the past? 569 00:45:39.740 --> 00:45:41.099 That's 570 00:45:41.099 --> 00:45:43.580 part of why there have been so few hard forks. 571 00:45:44.220 --> 00:45:50.780 They don't like hard forks. They try to make everything a soft fork. And and for those in the audience that aren't familiar with the difference, 572 00:45:51.020 --> 00:45:52.780 hard forks are breaking changes. 573 00:45:53.585 --> 00:45:56.385 So... And I'm not aware of any breaking changes 574 00:45:56.464 --> 00:45:57.984 that have fundamentally 575 00:45:57.984 --> 00:45:58.945 changed 576 00:45:59.505 --> 00:46:00.704 the encryption 577 00:46:00.704 --> 00:46:01.665 around 578 00:46:01.744 --> 00:46:04.625 historical wallets. So that's what you would need to do. 579 00:46:05.505 --> 00:46:10.140 But if the user's not around to do a transaction to upgrade to the new thing, 580 00:46:10.860 --> 00:46:14.220 what do you do? Right? And there's 500,000, 581 00:46:14.220 --> 00:46:15.180 million coins. 582 00:46:15.340 --> 00:46:19.740 So they're they're... The Satoshi coins are kinda sitting there as a honeypot 583 00:46:19.900 --> 00:46:24.994 as a canary in the coal mine. Is anybody does manage to crack, 584 00:46:25.234 --> 00:46:31.155 the wallet encryption and and the original coins were on a a weaker version than what's used today. 585 00:46:32.115 --> 00:46:34.035 You know, we're gonna find out pretty quick. 586 00:46:34.595 --> 00:46:34.915 So... 587 00:46:35.670 --> 00:46:40.470 And that may be part of the hedging that's going on with people leaning on leaning 588 00:46:40.470 --> 00:46:41.990 in on Zcash. 589 00:46:42.309 --> 00:46:48.309 They're like, I wanna go with the thing that's recent encryption and post quantum and has all this baked in versus 590 00:46:48.615 --> 00:47:08.570 I don't know what the upgrade path looks like in Bitcoin. But like I said, I'm still optimistic they'll get there in general. The community's made us a bunch of proposals like, oh, maybe we should freeze those coins or no. They should be, you know, saved but, you know, recoverable in some complex way. We'll see how it plays out. But I don't think it's gonna be in five years. 591 00:47:08.970 --> 00:47:10.730 I'm starting to think it might be in 592 00:47:11.050 --> 00:47:12.970 a year or two, maybe sooner. 593 00:47:13.290 --> 00:47:15.210 Can I ask a Zika question? 594 00:47:15.290 --> 00:47:16.250 Oh, sorry. Go ahead, Kyle. 595 00:47:18.665 --> 00:47:20.905 Yeah. Let me just give this one on the topic. 596 00:47:21.785 --> 00:47:26.585 It's funny. We we talk about the situation claims. We talk about crypto security. 597 00:47:28.585 --> 00:47:31.945 We got way bigger problems if encryption 598 00:47:31.945 --> 00:47:33.305 is broken than 599 00:47:34.280 --> 00:47:38.840 crypto. Right? What's the market cap of crypto versus the market cap of 600 00:47:39.320 --> 00:47:41.320 the entire financial world? 601 00:47:42.599 --> 00:47:44.600 So, like, I feel like the... 602 00:47:45.400 --> 00:47:49.535 Things are going to move fast, and and there will be proof of concepts, 603 00:47:50.095 --> 00:47:52.015 you know, before 604 00:47:52.255 --> 00:47:58.415 before we get before we get to this. This will... I think this will be proven. People with more at stake 605 00:47:59.215 --> 00:48:01.215 will will have an answer 606 00:48:01.935 --> 00:48:03.615 before 607 00:48:04.760 --> 00:48:06.360 before this happens. 608 00:48:06.920 --> 00:48:13.320 Well... And and to their credit, a lot of people have already made the jump. So signal is already post quantum. 609 00:48:13.640 --> 00:48:17.575 They upgraded their algo a few years ago. And the banks, 610 00:48:18.295 --> 00:48:20.935 you know, call it a feature, call it a bug, 611 00:48:21.655 --> 00:48:22.615 aren't immutable. 612 00:48:23.015 --> 00:48:24.535 And so if somebody 613 00:48:24.615 --> 00:48:26.055 screws up their ledger, 614 00:48:26.455 --> 00:48:28.935 they can kinda just go to a backup because 615 00:48:28.935 --> 00:48:29.975 it's not 616 00:48:30.420 --> 00:48:31.220 permissionless 617 00:48:31.220 --> 00:48:36.820 or or immutable and things like this. Right? And so banks, to a certain extent, are are are 618 00:48:37.059 --> 00:48:39.700 insulated a bit by their slowness 619 00:48:39.700 --> 00:48:40.820 and friction 620 00:48:40.980 --> 00:48:44.415 and, you know, the fact that they can just use the database. 621 00:48:44.735 --> 00:48:46.655 They could just change the database 622 00:48:47.215 --> 00:48:49.855 and say the money's back there because there's nothing 623 00:48:50.095 --> 00:48:53.215 actually backing the numbers on their database. Right? So 624 00:48:54.095 --> 00:49:02.329 we'll see how it plays out, but I but I take your point. There's a lot of systems out there that are probably more vulnerable and have a lot more assets behind them. 625 00:49:02.650 --> 00:49:07.290

But I think most people are gonna get there. Signal's upgraded. Most of the major 626 00:49:07.769 --> 00:49:23.655 comm systems are getting upgraded this year or already did last year. So... Yep. No. No. I I I agree. It's just, like, sometimes I like to level set. It's like, alright. The first thing that goes down is not going to be crypto. Like, we're going to be having this conversation in many, many places. 627 00:49:24.295 --> 00:49:25.415 And crypto 628 00:49:25.900 --> 00:49:27.820 being the tech enthusiast, 629 00:49:27.820 --> 00:49:30.620 will, I think, be ahead of the curve 630 00:49:31.340 --> 00:49:33.820 on this rather than behind it. 631 00:49:34.380 --> 00:49:38.940 Just agree, Kyle. Our doge wallets are the most important thing to 632 00:49:39.180 --> 00:49:40.700 the solvency of this world. 633 00:49:42.085 --> 00:49:46.245 Yes. It's it's more likely that someone takes 634 00:49:46.485 --> 00:49:50.805 the private key that I have on the post it in my mailbox 635 00:49:50.805 --> 00:49:51.605 than... 636 00:49:53.845 --> 00:49:55.365 Yeah. True. We... 637 00:49:56.330 --> 00:49:59.690 No. It's funny. The thing that makes, I think, Q Day so 638 00:49:59.850 --> 00:50:15.265 different... Well, obviously, like, what the actual, you know, capabilities are. But but Y two k had, like, a very specific moment in time where it was like, it's going to happen at this moment. Q Day, we... I don't even know, like, are we gonna even know when it happens? Like, who... 639 00:50:15.985 --> 00:50:19.585 Has it already happened? Like, could be months 640 00:50:19.585 --> 00:50:22.705 after it actually occurs or is possible that 641 00:50:23.105 --> 00:50:29.740 we even hear about it. Or maybe not. I don't know. I mean, that's that's the part that I think is so odd about this. 642 00:50:30.620 --> 00:50:35.580 I I think we'll know within an hour. Oh, you think so when someone figures it out? 643 00:50:36.220 --> 00:50:39.900 Yeah. Because a lot of random stuff just starts breaking. 644 00:50:40.375 --> 00:50:41.015 Like, 645 00:50:41.735 --> 00:50:48.215 that's weird. And we've we've had events like that where, like, you have some major piece of Internet infrastructure, 646 00:50:48.615 --> 00:50:52.055 cloud flare flare has a problem or, like, some 647 00:50:52.375 --> 00:51:04.390 system involved in the Internet, you know, has has some fundamental issue. And, like, it has this cascading effect, and everybody breaks for an hour. You're like, why did that happen? And then it comes out like, oh, x y z got compromised. 648 00:51:05.109 --> 00:51:07.190 I think it's gonna be more like that. Because 649 00:51:08.655 --> 00:51:21.215 once somebody can do it and somebody else is gonna notice they they can do it, they'll reverse engineer pretty quickly how they did it. Mhmm. And then it just, like, the incentives are to move really fast. 650 00:51:21.375 --> 00:51:30.619 Yeah. Right? And, you know, but also the defense works the same way. Right? People lock down their systems. They, you know, freeze whatever they can freeze as far as like, 651 00:51:31.019 --> 00:51:35.795 you know, access points and endpoints and there's there's a lot of ways. So, you know, 652 00:51:36.995 --> 00:51:38.835 these systems are antifragile 653 00:51:38.835 --> 00:51:41.155 because people are antifragile. 654 00:51:41.155 --> 00:51:51.310 Right? We're not we're not statically sitting around waiting for it to happen. Like, there's a lot of things people are doing to prepare. So, you know, again, I hope it's a y two k moment, 655 00:51:51.790 --> 00:51:55.070 but some stuff could break. But anything that breaks, 656 00:51:55.230 --> 00:52:00.350 like, we know how to fix it. It's just, you know, how much data and, 657 00:52:00.830 --> 00:52:07.805 you know, money get lost along the way and, you know, who's gonna back it up. So we'll see. Well, it's interesting and exciting. 658 00:52:08.045 --> 00:52:13.645 I mean, I was talking to somebody about this the other day. How lucky are we to get to be around 659 00:52:13.885 --> 00:52:18.605 for the earliest days of, you know, computing, the Internet, and 660 00:52:19.250 --> 00:52:26.770 artificial intelligence and or super intelligence depending on whether or not you live in California or the other 48 states... Or 49 states, rather. 661 00:52:27.730 --> 00:52:28.609 Yeah. It is, 662 00:52:29.170 --> 00:52:36.575 how lucky are we to get to, you know, be here to see all of this? I I think that even every time I see, like, a really stupid slop video, 663 00:52:36.895 --> 00:52:53.760 I'm like, slop videos are gonna be a flash in time. Of like, remember when AI videos were like this weird, like, slop crap, and now it's, you know, everything's perfect all the time? You're trying to replicate the slop, but you can't? Like, it won't be that long from now that we'll look back on it and be like, remember that? That was crazy. 664 00:52:55.040 --> 00:52:57.120 And we got to live for that and that's cool. 665 00:52:57.840 --> 00:53:00.880 You know, it's funny, Bo. I saw a survey recently. 666 00:53:02.154 --> 00:53:03.994 Every single generation 667 00:53:04.075 --> 00:53:04.795 is 668 00:53:05.434 --> 00:53:06.795 increasingly concerned 669 00:53:06.795 --> 00:53:08.635 about AI except for one. 670 00:53:09.755 --> 00:53:11.835 There's one group of people 671 00:53:11.994 --> 00:53:16.070 that know how technology worked because they came of age 672 00:53:16.310 --> 00:53:20.710 when computers and the Internet, it's the millennials. Millennials. Yeah. 673 00:53:21.190 --> 00:53:25.750 Yeah. Millennials are not overly worried. The Gen z don't know how it works 674 00:53:26.244 --> 00:53:29.125 because, you know, especially, like, Gen Alpha, 675 00:53:29.205 --> 00:53:31.525 like, they've only experienced these 676 00:53:31.845 --> 00:53:34.005 abstraction layers, and they've got a phone. 677 00:53:34.325 --> 00:53:45.530 They might not even have a laptop, and they've... They they

don't remember the early days. And the boomers, like, were totally out of, you know, phase for when this stuff came. But millennials, like, are the tech support 678 00:53:45.770 --> 00:53:57.435 for most of the generations up and down. And they're like, no. Like, we're gonna be okay. Like, I've been through this a few times. We'll we'll figure it out. But it was kinda funny too. Yeah. The 679 00:53:57.435 --> 00:54:05.275 only generation... Yeah. We're not worried. Like That is crazy. It is kinda like the defining feature of millennials. I mean, it really is because you're old enough. 680 00:54:06.250 --> 00:54:09.050 Well, so I guess my birth year is '95, 681 00:54:09.050 --> 00:54:16.010 so it's either the first gen z year or the last millennial year, depending on who you ask. And then the millennials are... What... Was it 85 682 00:54:16.250 --> 00:54:21.745 is the... On the other end of the spectrum, or 80 maybe? The other end of the spectrum. 84 to 2004. 683 00:54:21.745 --> 00:54:35.010 I go by the twenty year generation. Okay. Some people fifteen or ten years for some reason. Yeah. But, yeah. But, yeah. It's it's just so interesting. And, like, that's the thing I was saying to this other person is, you know, thousands of years of history happened 684 00:54:35.090 --> 00:54:36.370 for basically 685 00:54:36.450 --> 00:54:43.650 a twenty year period where information just changed, like, at a scale that was not even fathomable prior, 686 00:54:43.810 --> 00:54:47.995 and we got to be there to see it. You know, it's like, it's really, really cool. 687 00:54:49.035 --> 00:55:02.155 But, yeah. Anyway, I mean, that was that was pretty much all I had this week. The only other thing to put a pulse on is, I mean, this just dropped like an hour ago or so, was there's a trailer that Netflix is doing a series 688 00:55:03.090 --> 00:55:07.010 on FTX. I don't know if you saw that, the Altarist. So 689 00:55:07.570 --> 00:55:13.890 we did it, Crypto. We made we made it to the big time. That's that's great. Really, really happy about that. 690 00:55:14.370 --> 00:55:14.770 Yes. 691 00:55:15.795 --> 00:55:23.315 Oh, man. They're always gonna pick, like, whatever worst thing ever happened in the industry. The weird thing for me is I I 692 00:55:23.635 --> 00:55:25.635 don't think they were part of the industry. 693 00:55:26.994 --> 00:55:29.155 In a lot of ways, FTX 694 00:55:29.155 --> 00:55:30.675 was... It was TradFi. 695 00:55:31.779 --> 00:55:36.740 Right? It was a centralized hedge fund that turned into a centralized 696 00:55:36.740 --> 00:55:37.619 exchange 697 00:55:38.099 --> 00:55:41.540 that tried to get The US to outlaw DeFi 698 00:55:41.859 --> 00:55:43.619 and then kinda blew up 699 00:55:44.945 --> 00:55:49.105 and and went bankrupt because everybody realized they didn't have the assets 700 00:55:49.505 --> 00:55:50.625 on chain, 701 00:55:51.025 --> 00:55:52.865 and then it all kind of unwound. 702 00:55:52.865 --> 00:55:55.345 But that's the weird thing. Like, 703 00:55:55.825 --> 00:55:56.865 I never met the guy. 704 00:55:58.290 --> 00:56:01.570 I've spoken at a 120 crypto conferences 705 00:56:01.890 --> 00:56:03.650 over fourteen years. 706 00:56:03.810 --> 00:56:06.450 Yeah. He wasn't there in the early days. 707 00:56:06.690 --> 00:56:09.170 Like, he showed up so late. 708 00:56:09.330 --> 00:56:10.690 I remember his interview 709 00:56:11.405 --> 00:56:13.485 when he was debating Eric Voorhees. 710 00:56:14.445 --> 00:56:15.725 And so Good. 711 00:56:16.445 --> 00:56:18.845 Didn't know about Mt. Gox 712 00:56:19.244 --> 00:56:21.485 and what happened to Mt. Gox. 713 00:56:21.565 --> 00:56:22.445 And it's like, 714 00:56:22.845 --> 00:56:25.645 how do you be in this industry and not know, 715 00:56:26.559 --> 00:56:38.080 like, about the collapse of Mt. Gox and, like, what led to it? And I was like, oh, that doesn't that doesn't bode well for your, you know, risk assessment criteria for, like, you don't know what happened to Mt. Gox. 716 00:56:38.960 --> 00:56:43.325 That, oh, you showed up a couple years ago. You weren't there 717 00:56:43.725 --> 00:56:53.325 in '13 or even '17 showed up way late in the industry. Yeah. It was like '19 or so that they showed up. Yeah. Yeah. And they showed up out of nowhere and scaled infinitely. 718 00:56:53.610 --> 00:56:57.210 Oh, we have infinite dollars. We're buying football stadiums, 719 00:56:57.210 --> 00:57:03.370 and we're really smart. You guys are really dumb, and this should all be regulated, and all we should we should get the license. So everybody's like, 720 00:57:04.490 --> 00:57:05.770 what? Like, 721 00:57:06.010 --> 00:57:10.385 get out of here. Yeah. And then it all fell apart. So don't know. 722 00:57:10.705 --> 00:57:12.865 There's so many things in 723 00:57:13.025 --> 00:57:13.984 2021 724 00:57:13.984 --> 00:57:15.665 that were, like, LARPing 725 00:57:16.145 --> 00:57:17.585 as blockchain 726 00:57:18.224 --> 00:57:22.305 that were actually just companies using the words, but they weren't using 727 00:57:22.809 --> 00:57:25.690 smart contracts and none of it was decentralized. 728 00:57:25.770 --> 00:57:28.490 Like all the was open source. Yeah. 729 00:57:28.970 --> 00:57:36.089 Yeah. All the real DeFi stuff did fine. Like Uniswap's doing great. Aave's doing great. Compound's doing great. Like the actual DeFi stuff 730 00:57:36.775 --> 00:57:40.215 kept on working after all these things that were LARPing 731 00:57:40.375 --> 00:57:44.295 as DeFi went away. So anyway, that's that's kind of my thing. It's like, 732 00:57:46.215 --> 00:57:50.135 Bernie Madoff was the stock market. Like, he was one of the chairmans 733 00:57:50.135 --> 00:57:53.110 of, like, the markets. Like, he was, 734 00:57:53.510 --> 00:57:56.950 you know, part of that machine, but it never really felt like... 735 00:57:58.230 --> 00:58:05.990 I don't know. It's just... Yeah. That's a really good take actually, David. It's interesting. I I I...

Like, I knew that, but I never really thought of it through that lens. And and also, 736 00:58:06.815 --> 00:58:07.935 there's been like... 737 00:58:08.975 --> 00:58:14.895 It's funny. Crypto is like one of those industries where it's been around for, you know, what, fifteen years now, give or take. 738 00:58:15.535 --> 00:58:22.160 And and then DeFi maybe a little less than that, but, you know, more than a decade. And it's still the... It's like the perpetual, 739 00:58:22.160 --> 00:58:23.200 we're early 740 00:58:23.520 --> 00:58:24.080 thing. 741 00:58:24.640 --> 00:58:26.480 And and there's a lot of people who 742 00:58:26.800 --> 00:58:36.065 think like, oh, well, how could that possibly still be true, you know, a decade or more into an industry existing? But it's because a lot of the things that they're benchmarking weren't actually, 743 00:58:36.705 --> 00:58:41.025 like, built on what you and I are talking about when we say crypto, 744 00:58:41.505 --> 00:58:45.105 or when we say, like, DeFi, or when we say decentralized AI. 745 00:58:46.040 --> 00:58:46.760 And if 746 00:58:47.160 --> 00:59:02.745 you only drill down to those, like, really core principle things, they have not been around that long. Or maybe the idea has been around for a while, but you couldn't actually execute on it until pretty recently for a lot of this stuff. Mean, think about even the ICO craze. 747 00:59:03.065 --> 00:59:07.385 How many of those projects could do the thing that they said they were trying to do? 748 00:59:07.785 --> 00:59:11.465 Sure. Well, it's... We're about to hit the ten year anniversary. 749 00:59:11.785 --> 00:59:15.065 So so my friend Vlad, I've been on his podcast, 750 00:59:15.560 --> 00:59:16.600 he had 751 00:59:17.000 --> 00:59:19.400 one of the key ceremony computers, 752 00:59:19.400 --> 00:59:25.000 or what was left of one of the key ceremony computers because they destroyed them all from the creation of Zcash. 753 00:59:25.160 --> 00:59:31.000 And he had all these pieces in a box, and it's like got the original shipping label, and it's October 2016. 754 00:59:31.575 --> 00:59:34.215 We're about to hit the ten year anniversary 755 00:59:34.455 --> 00:59:35.495 of Zcash 756 00:59:35.575 --> 00:59:47.590 and Zuko was on stage today and he's like, honestly, the breakthrough's been the last year when all these wallets got so good and it became easy to use. Yeah. Right? And so I checked out, you know, the different 757 00:59:47.670 --> 00:59:48.470 Zcash 758 00:59:48.470 --> 00:59:53.510 people and I I was impressed. Like, it was one click kind of onboarding stuff now 759 00:59:53.830 --> 00:59:56.790 for stuff that was really technical 760 00:59:57.174 --> 01:00:03.575 even three to five years ago. So, yeah, I mean, that's a really good point. And that should be our challenge 761 01:00:03.974 --> 01:00:05.734 in decentralized AI. 762 01:00:06.135 --> 01:00:06.775 Like, 763 01:00:07.095 --> 01:00:13.890 we can't just leave it as a protocol. We gotta keep building these layers that make it easier and easier. So 764 01:00:13.890 --> 01:00:15.250 we're definitely getting there, 765 01:00:15.970 --> 01:00:21.330 and we just gotta keep building up that ease of use so that people can onboard 766 01:00:21.410 --> 01:00:32.095 with one click and, like, we're gonna get there. It's only a matter of time, but it really encourages you to see usage running up. I I do have... Actually, that reminded me, I do have a Zcash question if we got an extra minute. 767 01:00:32.575 --> 01:00:42.335 If if you guys know. I I remember earlier this year, there was a bunch of FUD around some type of thing that was discovered, and then it became, like, a question as to whether the actual 768 01:00:43.230 --> 01:00:56.430 total supply was the actual total supply. Do you do you know, like, what ended up coming of that? Because obviously, the... It's performed really well since then, which tells me people much smarter than me have realized that, oh, this was FUD and it's probably fine. 769 01:00:57.515 --> 01:00:59.915 So, you know, no protocol 770 01:01:00.155 --> 01:01:19.270 doesn't have issues. Like Yeah. Bitcoin had an inflation bug. Bitcoin had had an accidental hard fork, and people fixed it. Right? And things keep moving. Right? So all software is imperfect. But to the credit of the Zcash guys, I think the incidents you're referring to is a powerful new model came out. 771 01:01:20.150 --> 01:01:25.585 The guy that always runs the audits, like the model came out on a Thursday. 772 01:01:26.145 --> 01:01:28.545 He ran a new audit on a Friday, 773 01:01:29.185 --> 01:01:34.865 found and disclosed the bug to the open source developers, and they shipped a fix by the Tuesday. 774 01:01:35.185 --> 01:01:35.825 So 775 01:01:36.545 --> 01:01:37.265 it's like 776 01:01:37.680 --> 01:01:44.240 in theory, somebody could have abused it, but, like, the people that really cared were on top of the security 777 01:01:44.480 --> 01:01:46.560 Yeah. And implemented the fix very quickly. 778 01:01:46.960 --> 01:01:50.240 And Zcash has sort of, like, these epochs 779 01:01:50.454 --> 01:01:51.735 with their upgrades, 780 01:01:52.055 --> 01:01:57.734 and you can know after an epoch whether like everything passed through the turnstile 781 01:01:57.734 --> 01:02:03.015 is now like, yes, the numbers all add up Mhmm. And everything's added up. So 782 01:02:04.240 --> 01:02:09.280 it kinda seems like a nothing burger, like people that wanna attack CCash were like, look, you had an error. 783 01:02:09.760 --> 01:02:12.720 But it was caught by the community, fixed by the community, 784 01:02:12.800 --> 01:02:15.200 and there's no evidence of abuse. So, 785 01:02:16.160 --> 01:02:27.835 like you're saying, I think the consensus is like they did a really good job of, like, catching something and fixing it very quickly. So, like, beyond reasonable doubt, it's fair to say it's probably was not abused. 786 01:02:27.994 --> 01:02:36.990 I mean, we can't know for sure, obviously, but, like, based on the checks and balances as they exist, which are kind of always the checks and balances that you're measuring against, 787 01:02:37.390 --> 01:02:53.745 there's no evidence that it was a problem. Okay. That makes sense. Yeah. I knew it was something like that. I just didn't know exactly kinda what had happened. And But then it's one of those things where, like, the price recovers and it continues, and then I'm like, alright, well, I know people personally who I trust their technical ability far more than mine. 788 01:02:54.065 --> 01:03:04.940 They're interacting with this, so it's social proof, right? Which is not the best way to do things all the time, but, you know, it is the reality of less technical people, I mean, most of the time. 789 01:03:05.420 --> 01:03:06.060 Yep. 790 01:03:06.460 --> 01:03:15.555 Cool. Well, you guys, I mean, thank you as always, David. I know it's extremely late. It is tomorrow where you are, so we can get we can get out of here, unless if you guys have anything else. 791 01:03:16.835 --> 01:03:22.915 That's it for me. Keep being awesome, guys. Thanks for hosting, Bo. Yeah. You got it. I will talk to you guys soon. Take care, everybody.