Bản chép lời đầy đủ của Official AI pDOOM is at 10%? | Morpheus Crosses 10 Billion P2P Tokens

and again. They 66 00:04:19.150 --> 00:04:38.075 make an appearance. They're like, Alibaba's still here. Like, oh, yeah. That that that's right. And and Mini Max is doing their thing. H three on the video side is awesome. It's good that we finally got an open weight model that, like, get text accurate and all that for the videos. So but, yeah, that started to jump on you, but, yeah. No. 10,000,000,000. 67 00:04:38.075 --> 00:04:42.600 Here we are on the peer to peer. Yeah. No. No problem at all. It's it's it's amazing. And I think, 68 00:04:43.240 --> 00:04:53.800 you know, speaking about, like, the models jumping and then ultimately becoming a flash version, so then it's hitting, from an economics perspective, like, it's becoming super cheap as well. We're I I swear, we're only, like, 69 00:04:54.575 --> 00:05:07.215 one or maybe two iterations away for my use cases not being able to tell the difference. And so then just at least not, like, a real meaningful difference and then just having a workflow, you know, on a daily basis that is 70 00:05:07.970 --> 00:05:08.930 effectively 71 00:05:09.090 --> 00:05:19.889 free, quote, unquote, because of my more stick. I mean, it's obviously not really free, but, you know, perpetual is a better way of of looking at it. And in fact, I mean, I just set up a new 72 00:05:20.505 --> 00:05:21.945 multi agent workflow 73 00:05:22.185 --> 00:05:25.145 using Hermes, and maybe maybe I'll start off some time, 74 00:05:25.705 --> 00:05:27.545 all on Morpheus inference where 75 00:05:27.945 --> 00:05:32.185 I have an agent doing research every single morning by 07:30AM. 76 00:05:32.430 --> 00:05:44.350 It's then sending the research to a secondary agent who's organizing it into a newspaper specifically for me and then sending that to a third agent who's making it pretty. And so I have, like, my own personalized 77 00:05:44.350 --> 00:05:54.595 newspaper now every single morning that my agents are just making for me. Because I like reading in that format, but I wanted something, you know, I'm sick of hearing just doom and gloom mainstream, 78 00:05:54.595 --> 00:06:00.675 you know, everything all of the time, basically. I basically was like, tell me good news unless if it's really important about these topics. 79 00:06:01.980 --> 00:06:02.620 But, 80 00:06:03.100 --> 00:06:16.195 anyway, yeah, stuff like that. It's I mean, we're so close to just Well, if you wanna jump into the doom and gloom, I think the big thing this week was the Anthropic employee who came out. He's like, AI is dangerous. 81 00:06:16.195 --> 00:06:21.715 Like, I know we've said this before, but it really is. And I don't know. There's a lot of caveats. 82 00:06:21.875 --> 00:06:24.275 One the guy was there six weeks. 83 00:06:25.075 --> 00:06:41.840 So it's like it's not like a founding member of the team or, like, somebody who's been there for years. Dude went straight from OpenAI to Anthropic, and then was like, nope. And then just ducked out. So

it's hard to know because they've played this, like, fear mongering marketing thing so many times. 84 00:06:43.135 --> 00:06:47.215 Though a few other employees that are still there or, like, came out, oh, we have the same concerns. 85 00:06:47.695 --> 00:06:53.375 The thing that always drives me crazy, though, with these posts is he offers no solution. 86 00:06:53.695 --> 00:06:59.140 Yeah. Yeah. Yeah. And and it's like you've been doom and glooming for four years since '22, 87 00:06:59.540 --> 00:07:03.060 and you have no particular, like, we ought to do x 88 00:07:04.340 --> 00:07:07.300 other than, like, they should be in control, 89 00:07:07.540 --> 00:07:08.979 and they don't have a plan, 90 00:07:09.625 --> 00:07:15.705 but them being in control would be better than somebody else. And it's like, dude, that's not a real proposal. 91 00:07:16.025 --> 00:07:22.425 And and that's the thing that, you know, keeps me so focused and passionate about Morpheus is it feels like we have one of the few 92 00:07:22.970 --> 00:07:25.530 communities who have a clear vision, 93 00:07:26.090 --> 00:07:28.090 like an articulated goal, 94 00:07:28.330 --> 00:07:33.450 a set of principles that, like, let's get this power in the hands of as many people as possible, 95 00:07:33.850 --> 00:07:34.890 transparently, 96 00:07:35.050 --> 00:07:36.090 uncensored, 97 00:07:36.330 --> 00:07:39.675 and, like, coming out and, like, I have concerns, 98 00:07:39.755 --> 00:07:47.035 but no solution. It's like, okay. I don't know what to do with that. And it's you know, maybe it's boy that cried wolf, 99 00:07:47.275 --> 00:07:52.600 you know, and and there certainly are, you know, dangers, but I don't see any scenario 100 00:07:52.760 --> 00:07:54.600 in which we get a better outcome 101 00:07:55.000 --> 00:08:09.015 by, what, centralizing all the power into, like, one hyper scaled person that, you know, totally won't make mistakes or isn't aligned to everybody. So, like, I don't know. I saw all of those. Like, again, here we go again. 102 00:08:09.175 --> 00:08:11.335 So Yeah. I'm over it personally. 103 00:08:11.574 --> 00:08:14.615 You're muted, Alan. Alan, I'm 104 00:08:14.615 --> 00:08:20.530 over it personally. I don't believe them at this point. I think they're totally full of it. And it's one of those things where 105 00:08:21.650 --> 00:08:26.690 okay. I look at this similar to how I would look at, like, the presidential elections. 106 00:08:27.010 --> 00:08:48.480 They those people will stand on stage and call each other literal Hitler. And then the day after the election, they'll have coffee together in the White House lawn and be, like, yucking it up like they're just buddies. You know? Like like, if they actually believed in their core, the things that they were saying, their actions would look completely different than they do. And and that's how I feel about this. Like, if these people 107 00:08:48.800 --> 00:08:50.880 actually believed what they were saying,

108 00:08:51.360 --> 00:09:04.545 then they have a like like, if this person actually believes that about Anthropic, he has a moral obligation to, like, actually do something about it. But they but they won't because, like, they don't actually believe that, you know, and it's it's just It's 109 00:09:04.545 --> 00:09:19.339 more about the clicks. Yeah. Was you you read into it. You clicked into it, which which was the goal. The goal wasn't to solve the problem. It was to it was to click count. And so it's resist resist the need to click in and say, maybe he's got a solution. Melody. 110 00:09:19.660 --> 00:09:20.620 No, they don't. 111 00:09:21.980 --> 00:09:22.699 Although 112 00:09:23.100 --> 00:09:24.459 It's a cynical 113 00:09:24.459 --> 00:09:27.339 way to look at life, but it seems to be more and more the case is, 114 00:09:27.660 --> 00:09:27.980 you know, 115 00:09:31.035 --> 00:09:32.475 amazing headline, 116 00:09:32.795 --> 00:09:35.915 no substance. Click. That's all we needed. 117 00:09:36.395 --> 00:09:37.035 Anyway 118 00:09:37.275 --> 00:09:41.675 Yeah. And and most of it's not even written by a human anymore. Right? So, like, the whole, 119 00:09:42.250 --> 00:09:57.075 you know, what what's the theory referred to? The empty Internet or the dead Internet? Dead Internet. Yes. The Internet theory. It's like yeah. I think we're we're pretty much there. And so I use signal and other sources like my proof of human for like who you know, where's the real signal. 120 00:09:57.555 --> 00:10:04.675 Though on this subject is, like, what would advances look like? There was an interesting breakthrough 121 00:10:04.835 --> 00:10:06.355 with OpenAI, 122 00:10:07.075 --> 00:10:17.250 or at least it looked like that at first when they're like, we solved this math problem that is one of the millennial goals, and, like, it's this 100 year old thing. 123 00:10:18.450 --> 00:10:21.250 And and the two interesting things came out the next day. 124 00:10:22.035 --> 00:10:26.115 One, mathematicians were like, that's not how any of this works. 125 00:10:26.435 --> 00:10:32.355 And, like, the way you did this was basically just brute force with 10,000 agents 126 00:10:32.915 --> 00:10:37.990 exploring all possible solutions until they came to some proof, 127 00:10:38.230 --> 00:10:38.870 which 128 00:10:39.190 --> 00:10:48.390 I don't necessarily have a big objection to. I get the mathematicians don't like the brute force approach. But if you brute force the cure to cancer, I would be perfectly okay 129 00:10:48.935 --> 00:10:56.135 that you guys brute force the the cure to cancer. So so I'll give them the brute forcing is okay. But the interesting thing is the next day, 130 00:10:57.175 --> 00:10:58.135 apparently, 131 00:10:58.535 --> 00:11:00.774 a lot of the breakthroughs that they claimed 132 00:11:00.935 --> 00:11:03.415 were actually some of the mathematicians 133 00:11:03.590 --> 00:11:05.190 who used their system, 134 00:11:05.910 --> 00:11:08.870 who then they were training and using the data 135 00:11:09.030 --> 00:11:16.855 to fix the thing. And then the mathematicians was like, wait a minute. Is this is this our work? And they opened their eyes like, maybe, 136 00:11:17.334 --> 00:11:20.214 but Was your work, bro. Gotcha. 137 00:11:21.015 --> 00:11:24.615 Yeah. Like, you know, we're not gonna claim the prize. You guys can claim the prize. 138 00:11:25.015 --> 00:11:41.910 And then, like, wait a minute. This is really, really messed up. And and the people are like, you know, maybe we can ruin your career. And, like, the whole transcript of, like, them going back and forth about the announcement came out. And it's like, wow. Yeah. These aren't good people 139 00:11:42.545 --> 00:11:45.905 that are doing this. Right? They're they're very 140 00:11:46.945 --> 00:11:50.545 maniacal or, you know, sort of very 141 00:11:51.025 --> 00:11:51.905 mercantilistic, 142 00:11:51.905 --> 00:11:53.105 you know, transactional. 143 00:11:53.105 --> 00:12:06.570 They're like, oh, but we need an announcement before our IPO. And, like, what can we steal? It's like, jeez. Like, this this whole thing went from, like, oh, wow. That's a cool breakthrough to, oh, wow. They did it by stealing a bunch of other people. 144 00:12:07.209 --> 00:12:13.084 Which that's how it's all gonna happen, though. Yeah. I mean, none of it it it like, none of it's proprietary going. It's even 145 00:12:14.125 --> 00:12:16.445 I have a friend who works at a 146 00:12:16.764 --> 00:12:21.565 very well known AI company that I will not dox because I 147 00:12:21.805 --> 00:12:24.444 care about this person. But when 148 00:12:24.444 --> 00:12:34.280 when this person first found out, you know, about Morpheus and kinda that I was, like, you know, working on this with this community and, like, hey. What's this thing all about? Why are you so passionate about it? Whatever. 149 00:12:34.840 --> 00:12:35.720 This 150 00:12:35.720 --> 00:12:43.715 person was under the assumption that, well, we have these contractual agreements with these inference providers, so they they you know, it's fine. Like, they're not taking the information. 151 00:12:43.955 --> 00:12:47.395 And then does a little bit of digging into it and was like, oh, wait. Actually, 152 00:12:47.635 --> 00:12:53.555 there is two massive ways that they can they can use it. And the first is cloning the data because now it's a new dataset. 153 00:12:54.009 --> 00:12:59.930 And so it's not your dataset. It is our dataset that looks eerily similar like your dataset. 154 00:13:00.089 --> 00:13:01.450 And the second is, 155 00:13:01.690 --> 00:13:16.685 and in some cases, like, if there is potential for, like, security or any type of, like, public safety consideration or something like this where they can make the claim, oh, well, we had to retain this, you know, for, like, basically the greater good, like, for public safety. 156 00:13:16.845 --> 00:13:20.445 And and and

and who's the arbiter of that? You know? Like, there is no 157 00:13:21.209 --> 00:13:32.010 actual ins and ins and outs of, like, what that framework looks like yet. And so, yeah, it's I I'm I'm bummed but not surprised that this is how that went. Yeah. Yeah. But but that's 158 00:13:32.170 --> 00:13:34.250 telling about base assumptions. 159 00:13:35.475 --> 00:13:38.595 People that are coming out of the AI community 160 00:13:38.675 --> 00:13:42.115 are coming from academia or they're coming from a big corporate world. 161 00:13:42.435 --> 00:13:45.715 And all they've ever had are contractual 162 00:13:46.115 --> 00:13:50.900 trust me, bro. We signed on the dotted line and swore we wouldn't do 163 00:13:51.220 --> 00:13:53.700 something that's, you know, not in that contract. 164 00:13:54.020 --> 00:13:57.140 But crypto people are used to an adversarial 165 00:13:57.140 --> 00:13:57.940 environment 166 00:13:58.260 --> 00:14:05.735 where bad actors will do anything and everything they can in the system, and the only thing that's real is a verifiable 167 00:14:05.735 --> 00:14:06.535 cryptographic 168 00:14:06.535 --> 00:14:07.415 proof. 169 00:14:07.415 --> 00:14:12.135 Right? The whole fifteen years of crypto can be summed up as 170 00:14:12.535 --> 00:14:18.920 verify, don't trust. Yeah. Right? And so all these people coming in now to these adversarial 171 00:14:18.920 --> 00:14:20.200 open systems, 172 00:14:20.520 --> 00:14:22.120 they're still on trust, 173 00:14:22.520 --> 00:14:27.720 and they haven't gotten to verify yet. But if they wanna survive this change, they gotta get to verify 174 00:14:28.040 --> 00:14:29.240 and not trust. 175 00:14:29.480 --> 00:14:35.975 Right? And and we can get there, but that's the cool thing we have to offer to the world is, hey. Here's all the cryptographic primitives 176 00:14:36.135 --> 00:14:44.695 we built for identity and validation and whatever. The last fifteen years, they're free. Come and use them. Right? And so 177 00:14:44.970 --> 00:14:46.810 but it's taken a while because 178 00:14:47.050 --> 00:14:52.490 crypto has become the word so associated with, like, meme coins and speculation 179 00:14:52.649 --> 00:14:55.290 Yeah. Opposed to cryptography, 180 00:14:55.769 --> 00:14:59.370 which is the math behind the proofs, which is the real foundation 181 00:14:59.915 --> 00:15:01.355 of Web three, crypto, 182 00:15:01.675 --> 00:15:09.275 and blockchain. So we'll we'll get them there, but, like, that's the education people in that ecosystem have to go through is is how we all learned 183 00:15:09.595 --> 00:15:24.190 the last fifteen years in crypto. We're so close to the breakout event of, like, the average normie realizing why it matters. And I I just had an example. I was just meeting with a group of people the other day, and they were talking about spam text messages 184 00:15:24.365 --> 00:15:36.045 that they're getting and, like, spam phone calls. And I was saying, well, just wait till, you know, your family members' voices are cloned, and now that spam phone call because one of them that somebody got was a text that was like, 185 00:15:36.285 --> 00:15:38.285 hey, this is, you know, their kid's name. 186 00:15:38.920 --> 00:15:46.360 My phone died, so I'm texting you here to get ahold of you, and it wasn't their kid, which is obviously scary. Well, what happens when, 187 00:15:46.760 --> 00:15:56.415 you know, your kid has a video of them talking on the Internet, and now all of a sudden their voice is cloned, and now you're getting a voicemail from someone who you'd swear is your kid because of the the similarities 188 00:15:57.135 --> 00:15:59.055 in the the voice and mannerisms, 189 00:15:59.055 --> 00:15:59.695 that's So a huge 190 00:16:00.335 --> 00:16:01.535 something like that 191 00:16:02.015 --> 00:16:08.415 and and then, of course, people, when I was meeting with them, were asking the question, well, like, how do you how do you stop that? I'm like, proof of identity. 192 00:16:09.190 --> 00:16:22.070 By the way, we have the technology and have had the technology for quite a while. It just hasn't been, like, largely adopted. But I'm sure, like, all of that all of those types of things will also be in some way, shape, or form on chain. 193 00:16:22.675 --> 00:16:26.595 Because it's the only way to verify that it's actually them. Right? I mean, like, 194 00:16:26.915 --> 00:16:29.875 it's it's coming, without a doubt. 195 00:16:30.435 --> 00:16:31.955 Yep. Speaking 196 00:16:31.955 --> 00:16:33.235 of all of this, 197 00:16:33.715 --> 00:16:40.720 I do wanna draw people's attention to and give a shout out here to you, David. This article that you wrote in 2016. 198 00:16:41.279 --> 00:16:41.839 So 199 00:16:43.279 --> 00:16:44.720 Soothsayer David 200 00:16:46.160 --> 00:16:54.405 wrote this article that I'm showing here, which you should check out on Medium, How Blockchain Check Tech Will Help Us Survive the coming of strong AI. 201 00:16:54.805 --> 00:16:57.285 This is a decade ago, you guys. 202 00:16:58.005 --> 00:17:01.045 And, I mean, I I definitely encourage I 203 00:17:01.605 --> 00:17:03.445 encourage everyone to check it out. 204 00:17:04.165 --> 00:17:06.885 But, David, can you just give, like, a really quick overview on, 205 00:17:07.610 --> 00:17:14.809 like, what it is that you wrote here and kinda why at that point in time this was something that you were thinking about? First of all, it's hard to believe it's been ten years. 206 00:17:15.690 --> 00:17:16.809 But I remember 207 00:17:17.050 --> 00:17:18.730 like, I'm not 208 00:17:20.265 --> 00:17:21.625 oblivious to risk. 209 00:17:21.865 --> 00:17:26.185 Right? And I'm not oblivious to being concerned about where these trends are headed. 210 00:17:26.505 --> 00:17:28.105 And so in 2016, 211 00:17:28.425 --> 00:17:33.145 I remember walking into a restaurant, and if if you show that that image on the article, 212 00:17:33.385 --> 00:17:34.665 like, saw

this 213 00:17:35.029 --> 00:17:37.669 sort of weird cartoon of, like, the chef, 214 00:17:37.830 --> 00:17:45.029 and the chicken is, like, self aware, and it's running away, and he catches it, and he kills it. And it's, like, pretty, you know, explicit 215 00:17:45.190 --> 00:17:47.350 in the sense that they had, you know, anthropomorphized 216 00:17:47.350 --> 00:17:57.484 this this chicken, and he slaughters it and and eats it. Right? And I and I was I had been thinking a lot about AI. I'd been building, you know, projects in AI all the way back in 2011, 2012. 217 00:17:57.725 --> 00:18:00.844 And I was like, we really gotta avoid being the chicken. 218 00:18:01.005 --> 00:18:05.325 Like, how how do you not be calories for the more intelligent 219 00:18:05.520 --> 00:18:08.640 entity? And so I thought about it very deeply, 220 00:18:08.960 --> 00:18:14.400 and what I came to is sort of the same thesis that I have today, and I think it's played out pretty well, 221 00:18:14.720 --> 00:18:16.240 which is TLDR. 222 00:18:17.760 --> 00:18:20.000 If good AI 223 00:18:20.395 --> 00:18:22.075 creates more outcomes, 224 00:18:22.155 --> 00:18:25.115 more money, more, you know, economic 225 00:18:25.115 --> 00:18:34.830 activity for people, people will reward it with power and compute and all these things. Right? So if there are bad AIs and good AIs, 226 00:18:34.830 --> 00:18:37.070 I think the good AI will win 227 00:18:37.790 --> 00:18:39.150 as long as 228 00:18:39.790 --> 00:18:42.030 it's creating more value in the world. 229 00:18:42.350 --> 00:18:44.750 You know, it it's it's very much if you've ever read Tolkien. 230 00:18:45.265 --> 00:18:49.424 Right? You know, evil cannot create. It can only corrupt. 231 00:18:49.665 --> 00:18:52.065 Right? That whole idea, by definition, 232 00:18:52.465 --> 00:18:58.144 all the theft in the world, all the credit card fraud and and crime is a tiny fraction 233 00:18:58.465 --> 00:18:59.745 of global GDP. 234 00:19:00.550 --> 00:19:07.190 Right? By definition, like, the the thieves can't steal something somebody didn't else create. Mhmm. Right? And 235 00:19:07.270 --> 00:19:07.909 usually, 236 00:19:08.310 --> 00:19:11.909 in a healthy society, the builders are better resourced 237 00:19:11.990 --> 00:19:12.790 than the thieves. 238 00:19:13.355 --> 00:19:30.989 Right? They have more money and and power and ability to protect their wealth than the thieves do have to steal it. Right. Any place that's not true is basically a completely broken dysfunctional society where, there are no property rights and it's just banditry or whatever. Right? But mostly, 239 00:19:31.470 --> 00:19:33.229 in a functional society, 240 00:19:34.190 --> 00:19:41.230 people that create value are gonna defend it, and they have more resources to defend it. Right? And so if you're concerned about bad AI, 241 00:19:41.925 --> 00:19:44.805 you really want good actors 242 00:19:44.805 --> 00:19:46.645 to have the best intelligence 243

00:19:46.885 --> 00:19:49.125 and the most access to these agents 244 00:19:49.205 --> 00:19:50.965 and able to protect 245 00:19:51.045 --> 00:19:56.485 what they already protect, their house, their family, their property rights, their wealth, you know, their food, 246 00:19:57.400 --> 00:20:04.039 their economic outcome. Right? And so I feel pretty good about that. Right? And that's why I was so concerned 247 00:20:04.680 --> 00:20:07.480 when the proposals were made to ban open 248 00:20:08.200 --> 00:20:10.360 weight AI back in 2023 249 00:20:10.360 --> 00:20:13.335 because that's one of the edge cases where it goes wrong. 250 00:20:13.655 --> 00:20:16.855 It's when, you know, all of the power ends up concentrated 251 00:20:17.015 --> 00:20:22.615 in a potential bad actor, whether that's a government or a large company, and everybody else is prevented 252 00:20:23.255 --> 00:20:26.375 from having access to the same intelligence to protect themselves. 253 00:20:26.809 --> 00:20:29.369 Right? If I were to use the historical analogy, 254 00:20:29.610 --> 00:20:32.009 you know, in in the industrial revolution 255 00:20:32.250 --> 00:20:35.369 and the age of guns, germs, and steel, 256 00:20:35.690 --> 00:20:39.929 you really wanted to be on the guns, germs, and steel side of history 257 00:20:40.774 --> 00:20:45.254 and not the one that didn't have access to that. And so it wasn't that guns weren't dangerous. 258 00:20:45.575 --> 00:20:47.815 It's that you wanted to decentralize 259 00:20:47.815 --> 00:20:49.254 their power to everybody 260 00:20:49.414 --> 00:20:54.294 so there wasn't as huge asymmetry where those with took advantage of those without. 261 00:20:54.710 --> 00:21:00.070 Right? And so I feel like we're in a pretty good place. Looking back on this article from ten years ago, 262 00:21:00.310 --> 00:21:01.990 we've got open weight models. 263 00:21:02.470 --> 00:21:07.830 We've got broad access to them. We're using, you know, through Morpheus crypto right now 264 00:21:08.195 --> 00:21:16.034 to offer access in a way that's is permissionless. So I I feel like we're in a pretty good place. But, yeah, I have I have been thinking about this for a while, 265 00:21:16.355 --> 00:21:19.235 but that's my answer is I think it's an economics answer. 266 00:21:19.635 --> 00:21:22.899 It's not a perfect technical solution. 267 00:21:23.059 --> 00:21:31.539 They're gonna be hacks. They're gonna be stuff like that. But what do you do with a 160 IQ bad actor? I want a 200 IQ defender. 268 00:21:31.620 --> 00:21:31.940 Yeah. 269 00:21:34.595 --> 00:21:38.754 I just still can't believe that this was stuff on your radar a decade ago 270 00:21:39.315 --> 00:21:57.200 and anyone's radar a decade ago. Like, how many nerds out there are there like you? What do you what do think? Just like a solid couple 100 maybe? It's it's probably a short list. But to be fair, AI's been sort of thought about and talked about and researched for forty or fifty years. Yeah. We just didn't know when it was gonna get to that tipping point. 271 00:21:57.520 --> 00:21:59.040 Right? And Transformers, 272 00:21:59.040 --> 00:22:05.924 right, the the seminal article by Ilia and the others at Google didn't come out till a year later in 2017. 273 00:22:06.405 --> 00:22:16.405 And that was sort of the moment like, oh, okay. You know, attention is all we need. If we put enough compute and data with this formula, we're gonna get better and better results. And for most people, it wasn't till '22 274 00:22:16.620 --> 00:22:24.460 when ChatGPT had that moment of breakout. They're like, oh, all the things people talked about AI doing now can actually do some of them. 275 00:22:24.860 --> 00:22:35.135 Yeah. That was for me, admittedly, the first time that this became, like, remotely real in my brain. Prior to that, I just looked at it as, you know, you might have software that has 276 00:22:36.095 --> 00:22:54.630 a lot of canned answers, so you can ask it a lot of things, but ultimately, like, it will have a limit and, like, it's not actually, like, creating anything new. And then, you know, GPT was the thing that was like, okay. Wow. This is like. But even by comparison, like, look at GPT, what was it, two in 2022 compared to everything now? It is, like, 277 00:22:54.950 --> 00:22:57.029 laughably different. I love that 278 00:22:57.670 --> 00:22:59.510 Will Smith eating spaghetti, 279 00:22:59.910 --> 00:23:02.150 you know, meme thing that they use as a test. 280 00:23:02.924 --> 00:23:05.484 This actually segues really nice into 281 00:23:06.684 --> 00:23:20.240 one of the craziest things. This was, like, something that just happened the other day and is now kinda happening in real time and was another one of those moments for me where it was like, oh, this just got real in, like, a very different way than it was previously. 282 00:23:20.240 --> 00:23:28.080 And it ties into what we were talking about right before we hit record where I told you, hold it because I wanna hear your guys' thoughts on this on air. But 283 00:23:28.960 --> 00:23:31.760 the the guy who mapped 284 00:23:32.115 --> 00:23:34.195 the fruit fly brain, 285 00:23:35.395 --> 00:23:39.635 like, all, looking at it right now, it's all a 166,700 286 00:23:39.635 --> 00:23:44.115 neurons inside Minecraft so that they could have simil simulated 287 00:23:44.115 --> 00:23:45.155 neuroactivity 288 00:23:45.155 --> 00:24:02.290 driving the brain the fly's movement. And now, I don't know if you guys have seen this, but people have been taking that because he open sourced it and, like, giving it a $100 and attaching it to a Bitcoin wallet. And, like, doing all these things to basically say, like, hey. How much brainpower do we need to do these, like, menial tasks? I saw 289 00:24:02.495 --> 00:24:13.215 a video of somebody attaching it to, like, a driving simulator. And I don't know how real any of this is, but it was the first thing that I was like, okay. Wow.

This is, like, now another level of 290 00:24:14.015 --> 00:24:14.815 intelligence, 291 00:24:14.815 --> 00:24:17.375 you know, that I didn't even consider really before. 292 00:24:18.019 --> 00:24:24.820 Yeah. Have you guys seen this? Oh, yeah. I saw it beating Beat Saber. Saw the Beat Saber. Yeah. Crazy. 293 00:24:25.059 --> 00:24:29.299 That was that was entertaining. It reminds me of if you ever read The Singularity is Near, 294 00:24:29.539 --> 00:24:30.259 which was 295 00:24:30.500 --> 00:24:32.259 Kurzweil's 2005 296 00:24:32.259 --> 00:24:40.335 book, This is basically sort of he puts everything on a nice curve, and it's like, okay. This is how much you compute you need for a fly brain and 297 00:24:40.575 --> 00:24:48.015 a a mammal brain and a monkey brain and a human brain, and here's the years according to Moore's Law that we get to those different milestones. 298 00:24:49.030 --> 00:24:51.110 And I think it was 299 00:24:51.350 --> 00:24:52.310 2025, 300 00:24:52.310 --> 00:24:53.669 end of twenty twenty five, 301 00:24:53.990 --> 00:25:00.390 we got to the first you know, with an enormous amount of compute for the first time, you had enough of it to get to, like, a human 302 00:25:00.715 --> 00:25:02.075 level of cognition 303 00:25:02.155 --> 00:25:02.955 across, 304 00:25:03.755 --> 00:25:05.435 let's say, a large language model 305 00:25:05.675 --> 00:25:08.475 given this many tokens equals this much 306 00:25:08.875 --> 00:25:09.675 human 307 00:25:10.155 --> 00:25:17.450 verbal output or something like that. So we've got sort of to that level, and now running a fly brain or whatever, 308 00:25:17.850 --> 00:25:31.984 it's gonna get weird, man. It's gonna get weird. Really weird because, like, where's the line where, like, okay. Now you're running enough neurons to be a human brain. Well, that's the part, okay, that's the part that I'm confused by is, like, what exactly it's measuring. So, for example, 309 00:25:32.945 --> 00:25:33.664 my 310 00:25:33.825 --> 00:25:48.879 understanding of this is they're using the same number of neurons that a fruit fly brain has, but the way that it's arranged is not the same way that it's arranged for a fruit fly. So in other words, it's saying if we took all of the computational, 311 00:25:49.120 --> 00:25:52.240 like, theoretical possibility of this many neurons, 312 00:25:52.695 --> 00:25:56.374 this is the outcome. Like, oh, it can play Beat Saber or it can do whatever. 313 00:25:57.815 --> 00:26:06.615 And then, yeah, to your point, like, we'll get we'll get to the point where it's like, okay. Well, now we've matched the number of neurons of a human brain, and so we can effectively, like, have 314 00:26:07.530 --> 00:26:09.290 an isolated incidence, 315 00:26:09.370 --> 00:26:11.610 or or instance of of a human, 316 00:26:11.850 --> 00:26:20.410 you know, and and there's a billion sci fis about this. Right? Like uploading your consciousness and measuring your neurons and doing that. I mean, it's feeling more and more 317 00:26:21.225 --> 00:26:22.745 just weirdly, 318 00:26:22.825 --> 00:26:28.664 like, maybe this can exist. I don't know. Maybe it's not just theoretical. I don't know. 319 00:26:29.065 --> 00:26:41.039 Yeah. It brings up a lot of questions. A lot of questions I don't think we really have answers for yet. So This was a jump for me, though, where I saw this and was like, okay. We're now doing something that is, like, truly different and goofy and 320 00:26:41.280 --> 00:26:54.195 is not is goofy today, but the goofiness of today is, like, very serious, you know, development of tomorrow a lot of the time. It's it's a lot of orders of magnitude, but it's gonna be interesting when we get to, like, the cat 321 00:26:54.435 --> 00:26:55.154 level. 322 00:26:55.555 --> 00:27:00.915 And they're like, oh, I'm gonna scan my my pet's brain and upload it, and now it plays in a metaverse. 323 00:27:01.300 --> 00:27:10.820 And, like, long after it passed away, like, you know I mean, there there there's a lot of difference between a fruit fly and a and a cat that's several orders of magnitude. But, yeah, like, 324 00:27:11.140 --> 00:27:25.575 it's gonna it's gonna get weird. It's gonna get weird and interesting. Well, and this is what what I had mentioned before when you guys were talking about, like, the human battery scenario. Right? Like, we're like, we're talking about the potential for p doom, and do we just effectively become, like, computational 325 00:27:25.575 --> 00:27:26.855 batteries for 326 00:27:27.220 --> 00:27:53.955 super intelligence in in the way that the original and, David, you were saying this, that the original matrix was that's what it was supposed to be. They they dumbed it down to batteries. Right? Yeah. The the studio had them dumb it down because they didn't think people would understand that brains had lots of neurons and computational power, and the machines were using that to power the matrix and all this stuff. And they're like, no. Use the battery analogy. We're using them for power. And but it was really confused. It's like one of the few confusing things about the movie. It's like 327 00:27:54.590 --> 00:27:56.750 they're the heat of the human 328 00:27:56.830 --> 00:28:01.390 body is being used for powering this? Like, okay. 329 00:28:01.710 --> 00:28:18.624 Like, they gotta wave their hands like, and a form of fission or fusion or whatever. And it's like but clearly, this studio kinda shoved that in there to make it easy for for people to to run the analogy. But, yeah, the brain one makes a lot more sense. Like, the human brain has an enormous amount of computational 330 00:28:19.184 --> 00:28:29.409 power, even compared to an entire data center full of h one hundreds. Right? The human brain, we we we've discovered over time is much more complex and much more powerful 331 00:28:29.570 --> 00:28:32.289 and works in ways we don't really understand. 332 00:28:32.850 --> 00:28:34.450 Maybe

quantum's involved. 333 00:28:34.530 --> 00:28:38.954 Who knows? Oh. But like, Yeah. The plastic the plasticity alone 334 00:28:39.674 --> 00:28:46.154 to be able to move and rewire compute computers are finite, right? They have this pathway or that pathway. 335 00:28:46.475 --> 00:28:50.830 The the brain is amazingly plastic to go and rewire 336 00:28:50.990 --> 00:28:53.309 and repath and reinforce. 337 00:28:53.309 --> 00:28:58.830 It's there's yeah, you're right. There's some things in there that were not all transparent just yet. Yep. 338 00:28:58.990 --> 00:29:02.190 Exactly. But coming back to like the takeaway, 339 00:29:03.615 --> 00:29:07.775 you know, I think the the goal ought to be, okay, how fast can we get 340 00:29:08.095 --> 00:29:11.455 an agent into the hands of as many humans as possible 341 00:29:11.535 --> 00:29:12.815 and preserve 342 00:29:13.135 --> 00:29:17.830 agency and privacy and transparency with with these systems. 343 00:29:18.230 --> 00:29:25.350 So, Bo, I don't know if this is a a good segment to to cover it, but I know we wanted to take a look at the current state 344 00:29:25.590 --> 00:29:27.429 of install OpenCall. 345 00:29:27.510 --> 00:29:33.750 And Oh, sure. And I'm a bit tempting the demo gods here because I literally just pushed him out to to 346 00:29:34.175 --> 00:29:35.535 the code base. And 347 00:29:36.095 --> 00:29:52.619 it's like, you know, yeah. Okay. Let's let's test in in live. But, know, if it doesn't work, you know, we'll we'll we'll we'll do it again and and record a little session. But I was working earlier today. Let's let's see if it works now. I but we need somebody Bo, I don't know if you have an account, an email account that you haven't set up 348 00:29:53.419 --> 00:29:55.419 and install OpenCloud with. 349 00:29:55.500 --> 00:29:59.340 So if you have a Gmail account that you haven't set up yet, it should work. 350 00:30:01.145 --> 00:30:04.264 And you'll just go to installopenclaw.xyz. 351 00:30:04.665 --> 00:30:05.304 And 352 00:30:06.345 --> 00:30:16.149 my hope here is and I've got a couple 100 beta testers now, and thank you everybody who's joined and is using it. It's awesome to see more people flowing in every day. 353 00:30:16.710 --> 00:30:17.509 You go, 354 00:30:17.750 --> 00:30:20.869 you give your agent a name, you push a launch button. 355 00:30:21.270 --> 00:30:22.149 That's it. 356 00:30:23.030 --> 00:30:29.684 Click on Google, log in with whatever you want, But I'm trying to get this, like, under sixty seconds 357 00:30:30.085 --> 00:30:31.124 so that 358 00:30:31.285 --> 00:30:32.085 anybody 359 00:30:32.165 --> 00:30:36.885 who wants an agent doesn't have to download anything, doesn't have to touch terminal, all that stuff. So 360 00:30:37.540 --> 00:30:48.100 let's see if it works. There's there's some new new instances in the buffer. So, you know, let's let's give it a name and Cool. Let's see. We want this to be 361 00:30:49.620 --> 00:31:00.674 I hope you like the title. Get Open Claw. This time on easy mode. Like, after after helping a lot of friends, like, spend hours trying to do it, it's like, no, no, no. Gotta have an easy mode. So doctor 362 00:31:01.155 --> 00:31:02.195 Samajay 363 00:31:02.195 --> 00:31:02.995 junior. Okay. 364 00:31:03.700 --> 00:31:06.980 I don't know if that just it just came to mind. Sure. 365 00:31:07.140 --> 00:31:11.140 That's it's all good. Whatever whatever you wanna call your agent. So 366 00:31:11.460 --> 00:31:13.700 hit the launch button. Let's let's see what happens. 367 00:31:13.940 --> 00:31:15.540 Give me just one second. 368 00:31:17.855 --> 00:31:25.135 This is this is where we get boat accidentally docks all his credentials. I know. I also just realized that I'm gonna have to actually look at this name. 369 00:31:25.535 --> 00:31:28.575 So we're gonna change it to Sama Sucks. 370 00:31:28.815 --> 00:31:29.535 Okay. 371 00:31:30.655 --> 00:31:33.759 Much better. Why you want your agent called that? I don't know. 372 00:31:35.600 --> 00:31:36.719 Alright. Launch. 373 00:31:37.360 --> 00:31:44.239 Alright. So now it should let you log in with Google or whatever account you want. You could use email. You could use Google 374 00:31:45.440 --> 00:31:58.835 off. It has other social options. You can connect to Wallet if you want, but I tried to set this up so it works with web two social people are sort of used to. And because you picked email, it'll ask you for a code. So you just check 375 00:31:59.480 --> 00:32:01.559 your email and put in 376 00:32:02.520 --> 00:32:04.520 that code, and it should 377 00:32:05.160 --> 00:32:11.720 connect your agent to your credentials. So this was really important for me. I wanted it to be an agent that's really owned by you. 378 00:32:12.365 --> 00:32:13.485 And because 379 00:32:14.045 --> 00:32:17.085 this is your account, only you can access this agent. 380 00:32:17.885 --> 00:32:23.325 And there you go. All set. So there's a big green button. This is launch agent 381 00:32:23.325 --> 00:32:24.044 now. 382 00:32:24.365 --> 00:32:26.445 Let's see if that does a thing. 383 00:32:27.179 --> 00:32:28.619 And if it does a thing, 384 00:32:28.940 --> 00:32:32.619 it will connect you. Remember, this is hosted on 385 00:32:32.940 --> 00:32:34.139 manifest network 386 00:32:34.299 --> 00:32:37.019 into a new tab, and it'll show 387 00:32:37.340 --> 00:32:38.539 your agent, 388 00:32:39.100 --> 00:32:42.940 and then you'll be able to talk to it. And it'll look like an install 389 00:32:43.315 --> 00:32:44.595 of OpenClaw, 390 00:32:45.235 --> 00:32:46.035 presuming 391 00:32:46.355 --> 00:32:51.155 the, you know, refresh Oh, can't see it. Share this. Sorry. Oh, yeah. Alright. 392 00:32:52.675 --> 00:33:03.980 No. You got it. Yeah. Sorry. I thought it was gonna switch over. Yeah. Boom. And we're live already. New agent. Yay. Sam Assass. So so yeah. So, you know, that's it. Like, now you've got an open claw. 393 00:33:04.220 -->

00:33:07.340 It's connected. You can see it's defaulting, 394 00:33:08.059 --> 00:33:09.580 I think, to DeepSeek 395 00:33:09.660 --> 00:33:13.164 as the model. Won't tempt fate too much more. 396 00:33:13.804 --> 00:33:17.404 Like, that's that's it. Name it, launch, 397 00:33:17.405 --> 00:33:18.845 log in, boom. 398 00:33:19.085 --> 00:33:20.364 Now you've got an agent, 399 00:33:20.605 --> 00:33:22.365 and and it should just do a thing. 400 00:33:23.179 --> 00:33:28.700 So yeah. For some reason, I'm not connected. I don't know what happened, but that's okay. But but, yeah, I mean, 401 00:33:29.340 --> 00:33:35.420 there's obviously, there's something happening. But, yeah, that that's crazy. How, David so let's say I was using this, 402 00:33:36.385 --> 00:33:46.625 you know, to set up I mean, you could set up a job like what I was talking about earlier, like, make yourself a newspaper, make yourself some daily research, like, have it do a cron job for you. Let's 403 00:33:46.625 --> 00:33:56.999 say that I did that. Right? And now I have this agent, but I wanted to, at some point, export it and, like, you know, quote, unquote, take custody. Like, what does that process look like? 404 00:33:57.880 --> 00:34:03.559 So the nice thing is and if you go back to the other page, you'll see there's a dashboard button, 405 00:34:04.775 --> 00:34:07.255 And this is where you can kinda control 406 00:34:07.255 --> 00:34:16.375 your agent. The thing I'm adding in right now, it doesn't work on this current version. This is what I'm working on today, is the download agent button. Oh, got it. And literally, 407 00:34:16.790 --> 00:34:22.710 it works in the code like you would have to type in, you know, export my agent and it would do it. But I'm trying to get it to, like, a one click. 408 00:34:22.950 --> 00:34:24.550 I wanna put it on my machine. 409 00:34:25.349 --> 00:34:33.365 And so that's what I'm working on right now. This is also a way to get back to your agent. If you, like, close that tab at the top, it's you know, open your agent, and it'll reopen 410 00:34:33.845 --> 00:34:37.285 your agent. You can see your usage on here. You can see 411 00:34:37.605 --> 00:34:41.285 the models. You know, you see it loads the gateway, and then it populates 412 00:34:41.605 --> 00:34:43.525 the agent and and opens it up again. 413 00:34:44.570 --> 00:34:46.650 To go back to the dashboard 414 00:34:46.890 --> 00:34:49.290 yeah. Open call. Oh, there it is. Alright. 415 00:34:49.530 --> 00:34:57.945 So you go down there. You can upgrade plans like, oh, I ran out of inference. Cool. Upgrade to pro, upgrade to power, you know, whatever scale you need. 416 00:34:58.345 --> 00:35:11.610 You don't wanna subscribe to anything? That's fine. Here's credits. Just buy whatever amount of access or inference you need. Honestly, DeepSeek v four is so cheap. It lasts for a long time. You can connect wallet like you see there, 417 00:35:12.330 --> 00:35:14.890 the whole thing. Right? So this sort of just gives you 418 00:35:15.609 --> 00:35:19.290 instead of having to poke around your agent in, like, the really 419 00:35:19.690 --> 00:35:25.345 complex and confusing settings inside of OpenCloud, this just gives you a way to control all these things 420 00:35:25.825 --> 00:35:42.770 with one click. So, yeah, I'm I'm pretty excited to where it's gotten to. I'm seeing five, ten. I had 18 people sign up randomly the other day, which is a good record considering we haven't done much announcements or anything. But I wanna get this down to, like, okay, here's a quick video. 421 00:35:43.329 --> 00:35:50.130 Click, click, click. Now you've got your agent, and people are off and running. But it's it's a free tier, anybody can set it up, use it how they want. 422 00:35:50.705 --> 00:35:54.305 Yeah. I see it's defaulting to DeepCP for Flash. 423 00:35:55.105 --> 00:36:08.110 And so, you know, I I think this is the key. Like, let's get Morpheus. And again, this is defaulting to Morpheus. Let's get all the major harnesses. Hey, success. Yeah. There you go. It was on my side. It wasn't your side. Yeah. 424 00:36:09.069 --> 00:36:13.549 I mean, this is, like, so incredible, David. This is the exact type of thing 425 00:36:15.465 --> 00:36:24.105 community so that you can literally show your friends and family how to set up an agent and set it up for them with an email, a two FA code, 426 00:36:24.345 --> 00:36:34.609 and a button click, and they're live. And and they can use it with the free inference. Obviously, like, there's only so much, like, free inference to kinda get them going, show them this, that it works. 427 00:36:34.930 --> 00:36:35.730 Upload 428 00:36:35.730 --> 00:36:47.965 credits, stake more, you know, connect to wallet, whatever makes the most sense for you, and boom, you you have an agent. And, like, the alternative to this, especially for those nontechnical, by the way, is spending, like, five hours 429 00:36:48.285 --> 00:36:51.565 learning how to set this up on your own. Ask me how I know. 430 00:36:52.045 --> 00:37:10.740 And it is did I learn a lot? Yes. I, you know, I I now understand command lines in a way that I didn't used to understand. So that was cool, but the reality is most people aren't gonna wanna do that. You know? Like, it's it's not I didn't even wanna do it. No. I'm just kidding. But, like, yeah, most people are not gonna wanna do that, and this is just like an instant onboard. 431 00:37:11.060 --> 00:37:21.175 Yeah. Awesome. And and you'll notice too when you go to the subscription or the credits on the dashboard, it uses Stripe. So if you wanna pay with a credit card, you don't have crypto, 432 00:37:21.495 --> 00:37:23.735 it's an option, but it's not a requirement. 433 00:37:23.815 --> 00:37:27.175 Right? So hopefully, this has removed all the barriers. 434 00:37:27.975 --> 00:37:31.735 Right? You you can pay with a card, you can log in with Google, 435 00:37:32.430 --> 00:37:36.110 It

works on on mobile. And think about it, it's now cross device. 436 00:37:36.750 --> 00:37:50.105 If you logged in with Gmail, you can do that on your phone or your computer or your laptop or wherever you want, and you can take your agent with you, and you can download and back up the context onto whatever additional machine you want. So 437 00:37:50.825 --> 00:38:01.450 I think we're we're this is this is huge. Right? Now we have an easy install for OpenClaw. I wanna do this for Hermes. I wanna do this for PY. I wanna do this for the other popular harnesses. 438 00:38:01.930 --> 00:38:04.250 And just give people an easy way because 439 00:38:04.410 --> 00:38:07.050 even using the new version of OpenClaw, man, 440 00:38:07.690 --> 00:38:25.015 the place you always get stuck is like, okay. Now I need a key. So I have to subscribe to something else, and I have to go sign up to Claude and figure out how to, you know, get their model connected in here. And that's usually where people get lost. Yeah. So my take on what happened with OpenClaw 441 00:38:25.140 --> 00:38:28.420 is you had all these people excited about a personal AI. 442 00:38:28.980 --> 00:38:30.260 They rushed in. 443 00:38:30.660 --> 00:38:36.260 Some people got through the door, but it was really hard to use, and you sort of exhausted the number of people 444 00:38:36.740 --> 00:38:45.255 that could get through that process. And it hasn't gotten a lot easier. A little bit, but not a lot. Right? Yeah. In Herbie's, it was much easier. 445 00:38:45.655 --> 00:38:47.335 That sort of took over. 446 00:38:47.575 --> 00:38:55.160 Right? Grock Grock Bot's even easier than that, but you're sort of getting away from from really owning it. Right? It's more of a third party. 447 00:38:55.880 --> 00:38:58.840 And so the key is let's let's make it easy. 448 00:38:59.240 --> 00:39:11.775 And that's my whole mission right now is I think how we get a lot of people onto Morpheus. And and huge credit to the Manifest guys for supporting me in the whole development process and offering a real alternative to something like AWS. 449 00:39:11.935 --> 00:39:15.615 Yeah. Because this would be pretty lame if it was on Morpheus, 450 00:39:15.615 --> 00:39:30.250 but it was also just entirely centrally hosted. Yeah. Right? As opposed to on a web three alternative. So it's cool. The timing's great. I don't think you could have built this a year ago. You probably couldn't have built it six months ago. You needed the open weight models to be good enough 451 00:39:30.730 --> 00:39:32.010 and cheap enough, 452 00:39:32.650 --> 00:39:38.665 and for Manifest to to host all the, you know, containers in a decentralized way, and Morpheus 453 00:39:38.825 --> 00:39:42.905 to be ready with all the pieces. So, yeah, this is a really cool milestone. 454 00:39:43.305 --> 00:39:44.665 Yeah. Absolutely. 455 00:39:44.665 --> 00:40:09.265 I I think, like, the the magic part about it too from a user's perspective is I remember, like, when I was when OpenCost first came out and I was first learning how to set it up, I just could not get wait to get to the point where it's like, can just talk to it in natural language and it can tell me what it can do. This just bypasses everything before that. Like, someone can get into it immediately and then say, hey. What can you do? 456 00:40:09.825 --> 00:40:17.665 How like, teach me how the like, teach me how to teach you, basically. Teach me what you can do and how I can use you. And this just gets you from zero to one instantly, 457 00:40:18.069 --> 00:40:22.869 which is it's awesome. I love it. This is fantastic. Looking forward to the Hermes version. 458 00:40:23.270 --> 00:40:28.950 I did switch over my personal agents all over to Hermes a little while back, and it's been a good experience. 459 00:40:29.910 --> 00:40:33.115 So, yeah, looking looking forward to that. It's it's awesome. 460 00:40:33.515 --> 00:40:36.155 Let me just look here what else we had on 461 00:40:36.635 --> 00:40:38.075 the list this week. 462 00:40:38.715 --> 00:40:42.395 Actually, Kyle, we haven't heard from you yet. What's going on over there? 463 00:40:45.610 --> 00:40:57.315 Maybe he's not in the spot. He can talk. That's okay. I'm not there. Just jump in if you are at any point. Yeah. We can we can talk about the the node release this week. I've got some provider messages if you want to Please. Yeah. Talk some of that. Yeah. 464 00:40:58.355 --> 00:41:01.875 So for all of the of you fans out there for the 465 00:41:02.675 --> 00:41:04.195 Morpheus Lumiren node, 466 00:41:04.355 --> 00:41:07.155 we did update to 7.11.0 467 00:41:07.155 --> 00:41:08.115 a couple days ago. 468 00:41:09.010 --> 00:41:12.370 This was a pretty cool update for a couple reasons. 469 00:41:12.610 --> 00:41:16.290 The biggest being that it had three big community submissions in it. 470 00:41:17.330 --> 00:41:21.970 Both two from our our Asia folks, Renee and 471 00:41:23.785 --> 00:41:25.545 Zac Zac 90 five 472 00:41:26.505 --> 00:41:30.105 and then from the from closer to Shore Bob, 473 00:41:31.145 --> 00:41:35.145 one of the contributors and discuss people, he's been on the 474 00:41:35.859 --> 00:41:36.900 spaces before 475 00:41:37.380 --> 00:41:42.500 but really good support on the on the proxy router, found some issues, some improvements. 476 00:41:43.060 --> 00:41:44.900 And one of the key things that 477 00:41:45.300 --> 00:41:46.660 came through was 478 00:41:46.820 --> 00:41:49.700 I talked last week about our active.more.org, 479 00:41:49.700 --> 00:41:50.580 our status site 480 00:41:51.125 --> 00:41:56.725 and one of the things that's happening, there was some there were some issues and I do shout out to Bob for 481 00:41:58.005 --> 00:42:00.805 pulling that out and we found some some pretty 482 00:42:02.005 --> 00:42:06.900 pretty gnarly errors and what they were is that we've started having the providers self report 483 00:42:07.860 --> 00:42:10.020 their

models and their bids 484 00:42:10.100 --> 00:42:23.335 and some of the behaviors to make sure that they're healthy or can you really provide them because no one wants to go open a session and realize that that bid really isn't there. And one of the things that was happening is that if if the model as 485 00:42:23.335 --> 00:42:25.335 it was registered didn't have any tags, 486 00:42:25.975 --> 00:42:29.095 the provider didn't know how to probe it and so just let it alone. 487 00:42:29.335 --> 00:42:34.750 And that left a big gap, lot of a lot of bids out there that were that were unknown when they were really real. 488 00:42:35.230 --> 00:42:35.790 So 489 00:42:36.030 --> 00:42:38.910 seven point one one fixes that, is that it does attempt 490 00:42:39.550 --> 00:42:45.069 some of those probes to make sure that if it's a healthy bid, it shows as a healthy bid on the status page. 491 00:42:45.684 --> 00:42:51.365 So, encourage for all providers who see this, watch this, please communicate to your friends or your non friends 492 00:42:51.605 --> 00:42:56.085 to please upgrade the, if you're a provider upgrade to 7.11.0 493 00:42:56.085 --> 00:42:56.565 and 494 00:42:57.204 --> 00:43:00.164 that will help to make sure that if you're bidding on models like 495 00:43:00.520 --> 00:43:03.080 Cloud Opus five was untagged, 496 00:43:04.040 --> 00:43:08.600 it some of the bids didn't come through because that said that self report wasn't there. 497 00:43:08.920 --> 00:43:09.560 Also 498 00:43:10.040 --> 00:43:14.435 because of that activity, we updated on techmore.org 499 00:43:14.755 --> 00:43:18.595 a new explainer page called active.html. 500 00:43:18.595 --> 00:43:21.075 So tech.more.org/active 501 00:43:21.395 --> 00:43:29.330 and what this is, is talks a little bit more about that status page and the filters we use. You have the raw models and bid data that's out there on chain. 502 00:43:29.809 --> 00:43:33.730 Well, what we did for active and for the for the API gateway was 503 00:43:34.050 --> 00:43:40.385 very similar to what the consumer node will do when you go to open a session, it knocks on the door says, are you healthy? Are you okay? 504 00:43:40.705 --> 00:43:47.105 Well, instead of having, you know, multiple consumers do that for all the providers and all 486 505 00:43:47.105 --> 00:43:48.225 or so bids 506 00:43:48.625 --> 00:43:49.345 is 507 00:43:49.665 --> 00:43:57.440 we've got our own cron job that runs every five minutes that pulls those providers. Are you good? Are you okay? What's your latest health check? Are you fine? 508 00:43:57.920 --> 00:44:02.160 And then consolidates that data into that S3 and into that status page. 509 00:44:02.640 --> 00:44:05.040 And what was happening is we did not have clear 510 00:44:05.785 --> 00:44:06.825 views 511 00:44:07.305 --> 00:44:09.785 of different layers of the network. 512 00:44:09.945 --> 00:44:12.025 Raw you can get on chain all the time. 513 00:44:12.745 --> 00:44:15.305 All means

all active bids, 514 00:44:15.945 --> 00:44:18.505 all active things that are not expired, bids and models. 515 00:44:19.880 --> 00:44:22.120 Then there's the active which says, look, 516 00:44:22.440 --> 00:44:24.520 we remove things that are broken, 517 00:44:24.680 --> 00:44:25.560 unhealthy, 518 00:44:25.560 --> 00:44:26.520 unreachable, 519 00:44:26.680 --> 00:44:28.600 a T failed attestation, 520 00:44:28.840 --> 00:44:30.840 you know, those kind of things we remove 521 00:44:31.080 --> 00:44:39.435 because you don't want to, again, you don't want to knock on the door and have a failed session That cost gas, that cost time and at the bottom was the gateway 522 00:44:39.435 --> 00:44:40.315 layer. 523 00:44:40.795 --> 00:44:49.115 That says, we even remove the ones that we skipped. It says, look, the gateway is only gonna use things that we know are healthy. We're not gonna guess, we're not gonna punt 524 00:44:49.480 --> 00:44:54.360 and so really we've got three sets of files. Each one has like all 525 00:44:54.680 --> 00:44:58.040 models dot JSON or all bids JSON 526 00:44:58.760 --> 00:45:00.920 or gateway at the at the very bottom 527 00:45:01.494 --> 00:45:03.015 that make sure that 528 00:45:04.055 --> 00:45:17.710 if you build an application, so the API gateway is an application and you want to say, hey, what's the list of models and bids that I have access to? You can choose as an application developer, an actual application builder to consume either any of those three layers 529 00:45:18.030 --> 00:45:21.790 to do what you want. And you could, you know, so you could say look, want the all 530 00:45:22.750 --> 00:45:25.869 and then all filter in my application, I'll filter stuff out. 531 00:45:26.589 --> 00:45:36.125 So that was pretty good. So again, I thank Bob for bringing that to our attention and I think it's a much better and as providers update to 7.11we'll 532 00:45:36.125 --> 00:45:37.485 see that fix. 533 00:45:38.045 --> 00:45:42.204 And then what's coming up next is as we continue to enhance that provider probe 534 00:45:42.980 --> 00:45:46.420 is we'll pull out capabilities. What we don't see behind it is, 535 00:45:46.819 --> 00:45:49.140 are you a thinking model? Are you non thinking bid 536 00:45:49.700 --> 00:45:55.299 on the same model? And so now we're trying to get that subtleties of bids and providers and capabilities 537 00:45:55.785 --> 00:46:00.185 that we can expose on those health checkpoints. So you can better steer your inference 538 00:46:00.585 --> 00:46:02.185 for context size, 539 00:46:02.265 --> 00:46:03.385 for web, 540 00:46:03.385 --> 00:46:06.665 can you search web, are you thinking, non thinking, what's the temperature? 541 00:46:07.065 --> 00:46:07.625 So 542 00:46:08.105 --> 00:46:11.545 there's more coming with those provider probes. It's really helped show 543 00:46:11.890 --> 00:46:16.130 that network status. So that's all I wanted to just upgrade to 7.11 544 00:46:16.130 --> 00:46:17.810

and things will be better. 545 00:46:18.290 --> 00:46:21.330 And thank you to the community members who contributed. That was a good set, 546 00:46:21.650 --> 00:46:23.730 good release this week. I'm happy that 547 00:46:24.369 --> 00:46:25.570 we had some good submissions. 548 00:46:26.625 --> 00:46:31.185 That's awesome. Yeah. And big credit to Renee and Mazin and Bob and everybody that 549 00:46:31.425 --> 00:46:35.745 contributed. I know there's a lot of hard work with that went into that. So you guys are awesome. Yeah. 550 00:46:36.945 --> 00:46:43.680 And that's what it's all about. Right? It's open source. Anybody can jump in there. And I think these are really important upgrades, Alan, because 551 00:46:44.160 --> 00:46:46.640 as people have better tools to, 552 00:46:46.880 --> 00:46:50.960 you know, like, curate the list, especially as we have more and more providers 553 00:46:51.360 --> 00:46:54.480 Mhmm. That's gonna be really helpful. It's, like, getting good results. 554 00:46:55.145 --> 00:46:56.185 Right. Yep. 555 00:46:56.825 --> 00:46:57.545 So 556 00:46:58.025 --> 00:47:05.785 It also helps. I mean, Bo, if you're if you're on there, if it also helps the providers. Like, hey. Why is my bid not getting getting taken? 557 00:47:05.944 --> 00:47:09.065 Scroll in the search. Go to Cloud Opus five. 558 00:47:10.080 --> 00:47:11.600 If you would type that in. 559 00:47:13.280 --> 00:47:14.400 Right here. 560 00:47:14.960 --> 00:47:20.080 Yeah. And then open him up. So click on click on him and that'll open up. Oh, yep. There you go. 561 00:47:20.880 --> 00:47:25.935 So you can see the here are the providers that are actually providing that and you can see at the top level, 562 00:47:26.255 --> 00:47:35.455 I'm pointing at the screen, I hope you can see I'm pointing at the screen. At the top, next to the model IDs LLM and it's untagged, that means on our model registry 563 00:47:35.694 --> 00:47:36.895 that's untagged 564 00:47:37.020 --> 00:47:41.180 and that's why the two bottom providers in that list are skipped because 565 00:47:41.660 --> 00:47:46.380 they don't know how to guess. Now, found a couple in the wild, people have patched their own 566 00:47:46.860 --> 00:47:48.380 proxy router provider 567 00:47:49.020 --> 00:47:52.780 to to come overcome this. So, we've rolled that into our core product, 568 00:47:53.715 --> 00:47:57.795 but people could say oh, that's why it's skipped, that's an actual provider signal 569 00:47:58.035 --> 00:47:58.995 that says, 570 00:47:59.635 --> 00:48:04.675 hey, we did not approve that, that should go away if those two providers upgrade to 7.1.1.0. 571 00:48:05.150 --> 00:48:07.310 So they would have active 572 00:48:07.310 --> 00:48:09.710 good bids for Cloud Opus five and 573 00:48:11.230 --> 00:48:14.350 other things can be admitted on that same line like, 574 00:48:14.750 --> 00:48:17.390 t attestation. I think if you look at 575 00:48:20.734 --> 00:48:22.255 well, search for TEE. 576 00:48:22.335 --> 00:48:24.255 I think this should work. 577 00:48:24.734 --> 00:48:25.775 Yeah. The 578 00:48:26.895 --> 00:48:28.174 scroll down a little bit, please. 579 00:48:31.550 --> 00:48:34.750 Yeah. It's not I don't think it's gonna come up with the the free text search. 580 00:48:36.590 --> 00:48:40.510 Can I filter for it? Yep. Yep. Yeah. It should be. There you go. 581 00:48:40.750 --> 00:48:42.030 And just remove yeah. 582 00:48:42.270 --> 00:48:43.070 Open him up. 583 00:48:43.704 --> 00:48:44.905 He's actually good. 584 00:48:46.825 --> 00:48:48.265 Yep. So he's healthy. 585 00:48:49.305 --> 00:48:50.345 Remove the 586 00:48:50.905 --> 00:48:54.665 the name the name field if you would. If you go back to your filter, 587 00:48:56.025 --> 00:48:58.105 make sure there's no names. Yeah. Use t. 588 00:48:59.980 --> 00:49:02.780 For some reason, it's showing just Oh, yeah. You're clear. 589 00:49:03.260 --> 00:49:09.420 Yeah. There's well, because the other two, I think, are defunct, and they should fail. I wanna show failed t attestation. 590 00:49:09.420 --> 00:49:11.340 So all the way back up to the top, hit t, 591 00:49:12.225 --> 00:49:14.145 goes up farther and 592 00:49:15.665 --> 00:49:19.825 see the other two pill filters, blue, green, yellow, skipped. 593 00:49:20.385 --> 00:49:24.385 Click on unknown and offline to the right. They're off by default. 594 00:49:25.150 --> 00:49:28.350 Oh, there we go. Yeah. There we go. Those click on Gemma. 595 00:49:30.350 --> 00:49:35.310 Yeah. There we go. That's what I wanted to show. So that Silverren VM secret labs 596 00:49:35.470 --> 00:49:36.990 is a t provider, 597 00:49:37.230 --> 00:49:38.350 but behind it, 598 00:49:38.915 --> 00:49:45.555 it can't verify the model. So, says look, I can't verify tea so I'm gonna peace out. So, this helps providers 599 00:49:45.635 --> 00:49:47.635 go, oh, why isn't my bid getting? 600 00:49:47.875 --> 00:49:50.994 You can actually do some some tests. So, if you click on Silver Ren 601 00:49:51.720 --> 00:49:54.840 real quick, that should there you go. So one 602 00:49:55.400 --> 00:49:56.760 is verifiable 603 00:49:56.760 --> 00:50:03.000 and there's something wrong behind the provider node that says t is unverified and most likely that's because, 604 00:50:03.880 --> 00:50:08.345 there's an endpoint that's wrong or endpoint that's missing. And So Alan, just to make sure, like, 605 00:50:09.705 --> 00:50:13.545 just layman's terming it down a little bit. So so basically, 606 00:50:13.865 --> 00:50:14.585 this is 607 00:50:15.065 --> 00:50:16.265 confirmation 608 00:50:16.265 --> 00:50:21.865 that they are hosting what they're claiming they're hosting. So they can't say, hey, we're hosting 609 00:50:22.160 --> 00:50:26.080 Gemma, but actually it's v four flash or something else, 610 00:50:26.320 --> 00:50:29.600 but they're registering it wrong on on the network. 611 00:50:30.000 --> 00:50:38.635 Correct? Okay. Because well, that's mostly in the t

add the the t. If you're running a t provider and a t model, it must attest all the way through the chain. 612 00:50:39.035 --> 00:50:47.275 And so this is what this tells me is that that provider node which is Silverren and Secret Labs is up and reachable. So the nodes good 613 00:50:47.435 --> 00:50:52.010 and one model behind it is fully valid. There's something else wrong 614 00:50:52.490 --> 00:51:00.570 between the provider node and the actual LLM behind it for Gemma and Quinn that it's not there and so 615 00:51:01.690 --> 00:51:07.095 and basically for the gateway, like if we don't attest, we shouldn't provide it for the gateway. Yeah. 616 00:51:07.415 --> 00:51:16.775 Because then you then we're serving things through the API gateway that is not that are not truthful and that's why, again, why we had those different views like, look, you can show them all, you can give it a shot 617 00:51:17.680 --> 00:51:19.120 But for the gateway, 618 00:51:19.360 --> 00:51:20.240 it's 619 00:51:20.240 --> 00:51:31.760 just a waste of resources to try to bang on the door that we know something's not gonna work. Which just comes right back to the idea of verify, don't trust. Like, build build as many trustless systems as possible 620 00:51:33.494 --> 00:51:35.415 effectively. Physics. 621 00:51:35.415 --> 00:51:40.615 Well, I'll go back to David's earlier comment. I had a I had a laugh. You come from big enterprises and big corporations. 622 00:51:40.775 --> 00:51:47.730 The reason you trust is because you've got an army of lawyers. So my new catchphrase is physics, not not lawyers. Physics, not lawyers. I like it. Yeah. 623 00:51:48.370 --> 00:51:50.050 It's true. No. A 100%. 624 00:51:50.450 --> 00:51:55.010 It's it's good to see people getting that though. I tweeted out a compliment 625 00:51:55.250 --> 00:52:00.684 to Peter, the founder of Open Claw. And I was like, honestly, OpenClaw 626 00:52:00.684 --> 00:52:02.605 has done more for alignment 627 00:52:02.845 --> 00:52:07.405 than every Frontier Labs ethics group ever has. 628 00:52:07.724 --> 00:52:27.960 And it was cool to see him like that that that tweet. Right? Because he's like, yeah. Like, these are the practical things you have to do to make sure it's the model you expect and the outcome you expect and the harness you expect. And, like, those are the real things you can do. Not this, like, wave your hands about, you know, somebody do something. Like, no. We have we have a solution. Let's do it. Right? So 629 00:52:29.744 --> 00:52:30.785 100%. 630 00:52:32.065 --> 00:52:32.785 Yeah. 631 00:52:33.105 --> 00:52:40.065 Sorry. I was getting distracted because as you were talking, Josh Wolfe, who's the founder of Lux Capital, just tweeted that 632 00:52:40.625 --> 00:52:48.190 after further reviewing the work being done at the Frontier Labs, he has come to the conclusion with a 100% certainty that we're all going to die, 633 00:52:48.829 --> 00:52:52.750 and just unclear if it's five years or

fifty years. So, 634 00:52:53.150 --> 00:52:55.070 you know Well, I hate to say, you know, 635 00:52:55.630 --> 00:53:00.675 Ben Franklin in death and taxes. He's not wrong. It's just are you speculating on the cause? 636 00:53:01.795 --> 00:53:06.195 Guys, I'm gonna make a bold claim. We are all gonna die sometime between the next 637 00:53:06.994 --> 00:53:08.035 tomorrow and 638 00:53:09.075 --> 00:53:16.690 three hundred years from now. So Come come on, Bo. Between now and the heat death of the universe. And the heat death of the universe. The slow, 639 00:53:17.010 --> 00:53:21.410 slow heat death of the universe. There's actually this amazing YouTube video. 640 00:53:22.450 --> 00:53:28.215 What's it called? There's this YouTube video that's like obnoxiously long for what it is. I think it's like a 30 641 00:53:29.015 --> 00:53:34.135 Though to give an optimistic note, you got a few trillion years until the heat death of the universe. 642 00:53:34.455 --> 00:53:35.655 But, you know, 643 00:53:36.530 --> 00:53:37.570 technically 644 00:53:37.570 --> 00:53:40.450 speaking, black holes are the most computationally 645 00:53:40.450 --> 00:53:41.410 dense 646 00:53:41.490 --> 00:53:55.495 entities in the universe. If we could just figure out how to use those for compute then, you know Yeah. Have we simply considered harnessing black holes? Yeah. Then then maybe we can go past that. So it does heat death of the year doesn't have to be the end. Was just, you know, maybe. 647 00:53:55.734 --> 00:53:56.375 Maybe. 648 00:53:56.695 --> 00:54:06.860 It's true. There there's this YouTube video called time lapse of the future, a journey to the end of time. It's from seven years ago. I highly recommend everyone just, like, grab a beer sometime and throw it on. It's twenty nine minutes long. 649 00:54:07.100 --> 00:54:32.450 And it just goes through, like, the entire heat death of the universe. And for example, like, Earth dying happens in the first, like, fourteen seconds of this video, and there's another twenty nine minutes to go, basically, and it gets all the way down to, like and then things will just continue to collide into each other throughout space time over the course of, like, hundreds of billions of years until it gets broken down so microscopic that it's undetectable, and, like, it's it's great. Everyone should watch it. 650 00:54:33.490 --> 00:54:38.450 But, yeah, we in that time frame, we'll probably it'll probably be curtains for everybody. 651 00:54:40.609 --> 00:54:42.130 I guess I guess we'll see, though. 652 00:54:44.255 --> 00:54:52.015 Yeah. I think you guys like, in terms of news and stuff, that's all I had. I did wanna make one note. I saw in the chat earlier someone had mentioned that 653 00:54:52.895 --> 00:55:05.290 we had mentioned the Will Smith, like, spaghetti eating test for how far AI video is coming, and they had said that they haven't seen anyone do a good skateboard video. Like, it can't get the physics down for skateboarding. 654 00:55:05.450 --> 00:55:08.329 And so I think that's a good supplemental test. Like, 655 00:55:08.809 --> 00:55:10.250 I'd love to see if, 656 00:55:11.130 --> 00:55:19.695 if anyone knows who Chris Jocelyn and elf Tre Flipping El Toro, what that means. If someone can make that the new gold standard for 657 00:55:19.855 --> 00:55:22.015 AI video, like capturing 658 00:55:22.015 --> 00:55:25.855 an AI video of that event happening, that would be that would be amazing. 659 00:55:28.829 --> 00:55:30.270 Oh, David, you're muted. 660 00:55:31.230 --> 00:55:33.470 Reminds me of all the, like, gymnastics, 661 00:55:34.109 --> 00:55:48.215 you know, comma you know, comedy wherein, like, AI tries to end and the bodies all just twist up and you're like, oh, okay. Not not quite. But, you know, a little more physics, a little more phys a little more training data, it'll it'll get there. Yeah. Yeah. For sure. 662 00:55:48.775 --> 00:55:51.015 But, yeah, guys, that that was all I had this week. 663 00:55:51.895 --> 00:55:57.510 Appreciate you guys hopping on. I I guess we'll throw a little teaser for next week. We're we're in talks with, 664 00:55:58.390 --> 00:56:01.830 a pretty big, well known, successful project, 665 00:56:02.470 --> 00:56:07.109 to potentially come on and join us on on air next week and talk about 666 00:56:07.775 --> 00:56:09.295 what a future 667 00:56:11.055 --> 00:56:11.935 collaboration 668 00:56:11.935 --> 00:56:14.255 with the MOR token could look like with them. 669 00:56:14.575 --> 00:56:19.535 So I'll kinda kinda leave it at that, but it it'd be a pretty big deal if if it, you know, does happen and there's adoption.