บทถอดความฉบับเต็มของ Grokbot UPLINKED To Morpheus | Live Demo Time

cost? Right? With with a lot of testing that I've been working on, 58 00:04:43.835 --> 00:04:44.315 the 59 00:04:44.794 --> 00:04:46.474 right. And time, 60 00:04:46.474 --> 00:04:54.210 cost, and quality is is that, you know, pyramid where typically you pull one and and get the other. Right? But when you look at the agent harness, 61 00:04:54.450 --> 00:05:03.890 you can kind of collapse that pyramid a little bit, you know, from the performance standpoint because your end goal is a completed task. So now you're just trading 62 00:05:04.825 --> 00:05:06.585 time and cost holding 63 00:05:06.665 --> 00:05:12.905 performance the same. And I think that using a flash model, it it looks cheap, 64 00:05:13.065 --> 00:05:16.105 right, at first glance, but if it takes 65 00:05:16.810 --> 00:05:17.850 32 66 00:05:18.250 --> 00:05:33.845 agent loops to get to the answer, then your cost goes up significantly. That that's, you know, maybe eight x the calls that you needed from from a larger model. So I I think that there's a difference between narrative and actual 67 00:05:34.885 --> 00:05:38.645 in some of these, and I've actually seen the best performance 68 00:05:38.965 --> 00:05:39.765 using 69 00:05:39.845 --> 00:05:41.525 one of the latest Grok models, 70 00:05:42.710 --> 00:05:47.670 you know, and kind of agent harness or agent orchestration applications 71 00:05:47.670 --> 00:05:48.230 where 72 00:05:48.630 --> 00:05:52.070 it can get the right tool call faster, 73 00:05:52.230 --> 00:05:59.315 which which reduces the turns and then help you know, reducing turns whole helps both cost and latency 74 00:05:59.635 --> 00:06:02.595 in that standpoint. So maybe the model's more expensive, 75 00:06:03.075 --> 00:06:07.395 but less turns and less time has has driven the cost down. So 76 00:06:08.195 --> 00:06:12.595 I think v four or v 4.1 flash from DeepSeek is great. 77 00:06:13.770 --> 00:06:15.690 I think that we're probably 78 00:06:16.729 --> 00:06:26.490 one or two revisions of those to catch, like, that current, like or I think rock and and opus five five or five six, whatever it five five 79 00:06:27.375 --> 00:06:29.055 is right now. So 80 00:06:29.775 --> 00:06:30.415 excuse me. 81 00:06:31.855 --> 00:06:36.015 I think I'm I'm really happy that we've gotten there on the ability 82 00:06:36.015 --> 00:06:36.975 parity, 83 00:06:37.055 --> 00:06:41.455 and I'm excited to get there more on the reasoning structure 84 00:06:41.455 --> 00:06:42.175 and 85 00:06:42.575 --> 00:06:43.935 route 86 00:06:44.610 --> 00:06:45.570 quality. 87 00:06:45.890 --> 00:06:54.450 Yep. A 100%. And this goes back to what we were talking about in previous weeks. It's like the benchmark I care about now is like cost per task. 88 00:06:54.610 --> 00:06:56.930 Exactly. Cost per successful 89 00:06:56.385 --> 00:07:03.505 task. Yes. Yes. Right. I'm okay if it takes five times more loops to get there as long as it's 20 times cheaper. 90 00:07:03.825 --> 00:07:08.545 I'm still

coming out ahead. Right? To your point, unless it's, you know, really time sensitive. 91 00:07:08.705 --> 00:07:14.350 And that kinda gets into, you know, what we've been talking about with this recursive self improvement. 92 00:07:14.590 --> 00:07:18.750 Right? I'm starting to think about that much more in the agentic 93 00:07:18.830 --> 00:07:19.630 area 94 00:07:19.630 --> 00:07:28.375 and less of it as like models making models better. Like, the models are good enough. What I care about is my agent getting better at completing my tasks. 95 00:07:28.775 --> 00:07:32.535 And so I've been trying to think of, like, what is the benchmark 96 00:07:33.175 --> 00:07:35.495 that I really care about there? 97 00:07:35.815 --> 00:07:41.230 And I think I've I've settled on tasks per day. Right? How many tasks successfully completed 98 00:07:41.550 --> 00:07:43.790 per day can I get through 99 00:07:44.110 --> 00:07:54.045 my looping agents that are doing all this work? Right? I put things in the top of the funnel, or I have them go grab pull requests or issues off of GitHub. 100 00:07:54.125 --> 00:07:59.165 Right? They're putting things in, and so now I've got a dashboard, and it's like, okay, 10 approved, 101 00:07:59.165 --> 00:08:01.565 50 done, 10 waiting for review, 102 00:08:01.885 --> 00:08:03.245 you know, that whole thing. 103 00:08:03.485 --> 00:08:13.860 You know, can I get a 100 done a day? Can I get a thousand done a day? Like, what's I think that might be the new metric. I don't have a better metric than that for like getting 104 00:08:14.340 --> 00:08:17.940 stuff done. And I think we're moving generally from 105 00:08:18.415 --> 00:08:23.775 Alan, you'll remember these these days when people used to talk about the megahertz 106 00:08:24.415 --> 00:08:26.495 the speed of their processor 107 00:08:26.735 --> 00:08:32.975 and the Pentium two the Pentium three. Yep. Apple came out and they had like 10,000 songs in your pocket. 108 00:08:33.760 --> 00:08:45.360 Yes. And everybody stopped talking about gigabytes and megahertz and processing speeds, they just talked about what it could do for you. And so I think that's a good transition for us too. 109 00:08:46.334 --> 00:08:56.014 Less obsession on, you know, nerdy benchmarks around, you know, exactly, you know, how it performs on x y z coding thing. Just just tell me, oh, it can get 20 tasks for you done. For 110 00:08:57.295 --> 00:09:04.630 for my use case, it's it's can get can business get done, you know, through the use of of crock pot and some of the other stuff I'll talk about. 111 00:09:05.030 --> 00:09:15.270 Started your own business and are able to get business done. Things that were out of reach before. So it's not songs in my pockets, it's dollars in my wallet at some point because you can leverage that. 112 00:09:15.945 --> 00:09:19.385 This is what brings that. So yes, we started with all the nerdy, the 113 00:09:19.785 --> 00:09:22.985 gigahertz or the token equivalents, but now it's, you 114 00:09:23.385 --> 00:09:27.785 know, you have viable businesses that have created because of the support. I think that's outstanding. 115 00:09:28.240 --> 00:09:30.240 Well well, let me double click on that, Alan. 116 00:09:30.800 --> 00:09:35.600 What you're really saying is how many dollars per day does my agent make me? 117 00:09:35.920 --> 00:09:42.720 Yeah. Yeah. And that's a good way to think about it because every task, right, I could fill a pipeline with a million 118 00:09:43.325 --> 00:09:45.805 maybe useful, maybe useless tasks, 119 00:09:46.285 --> 00:09:48.285 but did those tasks produce 120 00:09:48.445 --> 00:09:49.245 revenue? 121 00:09:49.485 --> 00:09:51.405 Right? Either for me as income 122 00:09:51.565 --> 00:09:55.245 because it's helping with my job or because I'm running the company. 123 00:09:55.810 --> 00:10:02.770 Right? And I'm racking up more sales or whatever. But that's that's that's a really good suggestion for the metric. It's like 124 00:10:03.410 --> 00:10:16.095 my Morpheus agent made me a $100 today or a thousand dollars today. Right? And and we've seen some of that in what goes viral on social media is people are using AI to do x y z, 125 00:10:16.495 --> 00:10:20.014 and the brag is, oh, I set up this agent and it, you know, 126 00:10:20.750 --> 00:10:22.270 whacked me up $10,000. 127 00:10:22.270 --> 00:10:26.910 Everybody's like, oh, I I want that. Right? So I I think that's that's a really good point. 128 00:10:29.149 --> 00:10:32.350 Yeah. I think the only other metric would be time saved. 129 00:10:32.805 --> 00:10:51.140 Mhmm. It's like if you could somehow just quantify, like, the time that you would have spent doing x y z thing. That's much harder to quantify, obviously. Sure. But but some some version of that, which is like, it made me x amount of dollars and I didn't And like, maybe it should be looked at as time gained doing other things. So like, I, 130 00:10:51.700 --> 00:10:59.540 you know, I spent two extra hours with my kid today, or whatever it is, which is more easily quantified, I guess, than time saved. But 131 00:11:00.055 --> 00:11:08.375 still kinda tough. But yeah, I mean, we're to the point now, like, one of the biggest struggles I've had the whole time as these things have developed is knowing 132 00:11:08.455 --> 00:11:14.295 which model to use for what. Because, like, for most of my use cases And luckily, 133 00:11:14.295 --> 00:11:17.630 we're at the point now where most of my use cases, 134 00:11:17.950 --> 00:11:24.270 we can I could use any flash not any, but, like, any of the the better flash models and do 90% 135 00:11:24.270 --> 00:11:33.625 of what I need to do? And so now we're at the point where, yeah, it's it's just cost and also, to your point, David, like, what is it making me? But 136 00:11:34.105 --> 00:11:34.505 I I 137 00:11:35.225 --> 00:11:54.230 I'm glad that we eclipsed that finally because there was a huge period of time where I was like, oh,

I don't know. Like, Kimmy's got a good vibe. I'm gonna use it for this. Even though it was like I had no actual reason for thinking that other than, you know, that's the one I wanted to use on that day. Which is a really dumb reason, but I think is how most people are are experiencing 138 00:11:54.230 --> 00:11:54.630 AI. 139 00:11:55.585 --> 00:11:59.985 Well, and on the time metric, you know, instead of time saved because, 140 00:12:00.145 --> 00:12:08.145 you know, you could only get up to twenty four hours eventually of That's what you're thinking. Saving you time, you know, more sleep, more, you know, playing with the kids and stuff like that. 141 00:12:09.640 --> 00:12:13.160 It's, I think, hours of work done. That's what 142 00:12:13.560 --> 00:12:14.920 what was it? Meteor? 143 00:12:15.640 --> 00:12:22.600 Meter? You know, it's misspelled meter. They did a benchmark around, like, how many hours of human labor equivalent 144 00:12:23.125 --> 00:12:40.110 the the agent was doing. And first, it was seconds and then minutes and then hours. And I think they kinda gave it up when it got to like sixteen hours. And they're like, you can just loop them now and they could do an unbounded amount of hours worth of work. So so the metric kinda went away, but I always thought that was really compelling. 145 00:12:40.270 --> 00:12:47.150 It's like, oh, wow. We get a week's worth of work done. We got a month's worth of work done. I shipped that product, 146 00:12:47.470 --> 00:12:49.710 you know, instead of six months in, 147 00:12:49.870 --> 00:13:00.215 you know, six hours or whatever. Right? I always thought that was a that was a really cool one, but I don't I don't see them updating it anymore. So we'll see. Claude told me 148 00:13:00.775 --> 00:13:08.055 the other day that the feature request I was asking for would take, like, three hundred fifty man hours. And I was like, can you just do this right now? 149 00:13:10.070 --> 00:13:12.950 Oh, sure. Yeah. No no problem. The 150 00:13:13.270 --> 00:13:20.550 other metric I've seen the other day is before AI, I had nine unfinished projects. After AI, I've got a 134 projects. 151 00:13:21.670 --> 00:13:24.070 That's that's where I feel like I'm at is 152 00:13:24.605 --> 00:13:28.764 definitely feel like I can start them and and close them a lot, they 153 00:13:29.964 --> 00:13:36.204 I need to get on to what Bo was suggesting and yep. Make sure that they can take care of stop adding things to the pile 154 00:13:36.524 --> 00:13:39.165 and go do other things while my agents 155 00:13:39.270 --> 00:13:45.510 do do some of the work. So instead of starting more things, oh, look at this. Building at the speed of thought is 156 00:13:45.910 --> 00:13:47.110 can be dangerous. 157 00:13:47.270 --> 00:13:48.870 Yep. 100%. 158 00:13:49.030 --> 00:13:53.590 Cool. Well, then I I think there's there is something to be said there, Alan. Right? That 159 00:13:54.084 --> 00:14:00.005 it allows you to rapid prototype

too. Right? Where Oh, yeah. In the past, we would, 160 00:14:00.725 --> 00:14:05.765 you know, myself, I I would build products to completion and launch them. 161 00:14:06.165 --> 00:14:13.080 And sometimes they would suck and no one would use them. And that was a lot of time wasted. Right? Like, now 162 00:14:13.640 --> 00:14:16.920 getting to 80% to get feedback 163 00:14:16.920 --> 00:14:19.560 from users, real live users 164 00:14:19.720 --> 00:14:20.519 on something, 165 00:14:21.584 --> 00:14:25.185 you know, before you make the investment in the full industrialization 166 00:14:25.185 --> 00:14:29.585 or the full end to end feature set, I think, allows us to 167 00:14:29.904 --> 00:14:44.230 get things closer to product market fit at a much faster rate. And then once you, you know, say, does this meet the the real investment bar, then you're able to kind of finish that. So, like, a 120 open projects 168 00:14:44.630 --> 00:14:46.710 isn't a bad thing if, 169 00:14:47.190 --> 00:14:52.055 you know, a 100 of them were gonna be garbage anyway. Right? You know, it's it's 170 00:14:53.175 --> 00:14:58.855 I I think it's a I I think it's a win there. And and I and I think that that's a good tendency 171 00:14:58.855 --> 00:15:00.454 of of development. 172 00:15:01.095 --> 00:15:06.430 I I agree. I it's funny. My in my past life, my my my past employment, 173 00:15:07.470 --> 00:15:09.310 we would spend a lot of time in IT 174 00:15:09.950 --> 00:15:17.710 testing, you know, ideas for prototypes. Right? We had a whole process around taking all the ideas and whittling them down to place your bets. 175 00:15:18.775 --> 00:15:22.135 And and you wouldn't start the prototypes until you kind of knew 176 00:15:22.535 --> 00:15:34.660 reasonably what the success was. The other metric from the same company was is they would start a 100 projects and only one of them would yield. And their whole thing about taking that innovations that innovation cycle 177 00:15:34.980 --> 00:15:37.540 was you start more faster, 178 00:15:37.620 --> 00:15:40.740 the more you have potential for that next billion dollar business. 179 00:15:41.220 --> 00:15:42.260 And so I think you're right. 180 00:15:42.905 --> 00:15:43.545 It's 181 00:15:43.785 --> 00:15:53.945 I I feel I feel the pressure because I've got a 134 plates I'm trying to keep spinning, but you're right, one of them has more chance to yield. If I didn't ever start them, they may never yield. So, 182 00:15:54.585 --> 00:15:55.305 absolutely. 183 00:15:56.105 --> 00:16:00.040 The more you can innovate, the more you can just wing it and go, hey, hey, 184 00:16:00.360 --> 00:16:05.320 Claude, go try this for me. I know it may take three hundred fifty man hours, it's gonna take you about thirty seconds. 185 00:16:05.640 --> 00:16:07.240 Wing it, see what happens. 186 00:16:07.880 --> 00:16:11.800 Yeah. It looks great. Instinct, right? Like, Instinct came out with 187 00:16:12.945 --> 00:16:19.585 honestly, a a half assed idea when it started. Right? With no UI, no nothing around it, 188 00:16:19.825 --> 00:16:23.745 and said, are people going to want this? And they did. 189 00:16:24.459 --> 00:16:29.820 And Mhmm. You know, what happened now? A 100,000,000 valuation to $2.50 190 00:16:29.820 --> 00:16:32.540 to 1,000,000,000 to 2,500,000,000 to 10,000,000,000, 191 00:16:33.180 --> 00:16:39.260 rather instantly. And I would say they probably have the cash now to do it the right way if if they choose. 192 00:16:40.285 --> 00:16:41.005 Yep. 193 00:16:41.165 --> 00:16:42.205 Bring in the resource. Get 194 00:16:42.925 --> 00:16:53.165 it. Cool, Bo. Should we, jump into demos and other stuff? Yeah. Let's let's see it. Let's, so, Alan, you should be able to share from your Ready to go? Alright. 195 00:16:56.510 --> 00:17:00.830 So I'm gonna bounce around a little bit. Let me know if you can see the screen alright. Yep. Yep. 196 00:17:01.870 --> 00:17:14.524 So we talked a we talked a little bit about this on the on the preamble, and I think this is kind of what I'm gonna show is we're show you the map, point you at the door. I'm gonna actually do the magic trick, create an API key and ask the Grokbot crew. 197 00:17:15.085 --> 00:17:19.085 And in that, I'm gonna interleave because I actually wanna show a real live uplink setup 198 00:17:20.190 --> 00:17:22.509 here. So we're gonna bounce around a little bit. 199 00:17:22.830 --> 00:17:26.029 And as this as a setup here, I run a whole 200 00:17:27.309 --> 00:17:28.669 Grokbot crew. 201 00:17:29.070 --> 00:17:30.190 I originally 202 00:17:30.750 --> 00:17:36.844 I kinda missed the boat on some of the early open open harnesses like Open Claw and Hermes and 203 00:17:37.404 --> 00:17:49.645 I've I've I've of got attracted to I've always liked Grok and the the SpaceX story and all that kind of stuff. And Cursor was the daily driver, so was really excited when I saw Cursor and and XAI coming together. 204 00:17:50.419 --> 00:17:51.139 And 205 00:17:51.220 --> 00:18:03.700 Grok build was good, CLI and then the Grok Grok bot came out and like, okay, this is the hardest I can really get, you know, kind of lean into lean into that. And as I started to use it, I'm like, wow, that Cursor build really starts to grow and the on demand 206 00:18:04.245 --> 00:18:06.885 and the way they limit smell like, you know, 207 00:18:07.605 --> 00:18:09.925 I am part of this Morpheus DEAI 208 00:18:09.925 --> 00:18:10.725 community, 209 00:18:10.885 --> 00:18:16.725 I kinda ought to, you know, be here and see if I can leverage that and that's really what came up with this 210 00:18:18.850 --> 00:18:24.609 this inference, Morpheus inference skill. And so, my crew is on the left here, my chief of staff, 211 00:18:25.330 --> 00:18:28.289 CTO, all your usual roles you have in a company. 212 00:18:28.850 --> 00:18:44.365 How do I get them to use Morpheus instead of just cursor and grok all the time? And really that's by using either our public

API gateway or what we've just created which is uplink, which is really one of the focus. So, instead of just saying, hey, here's uplink, I wanted to show you how I use it 213 00:18:44.685 --> 00:18:45.165 overall. 214 00:18:46.009 --> 00:18:47.690 So first thing I'm gonna 215 00:18:48.649 --> 00:18:50.249 scoot over to is 216 00:18:51.370 --> 00:18:52.809 what should be on the 217 00:18:53.610 --> 00:19:04.844 Grokbot marketplace soon which is the Morpheus inference skill. And this is basically an MCP server that that basically does what I wrote, you know, what we wrote in Python on on Grok to do 218 00:19:05.404 --> 00:19:06.124 this. 219 00:19:07.404 --> 00:19:15.650 We really wanted a mixture of experts and we don't have that access in in Grok as it shipped. So we built a MOM or a mixture of models 220 00:19:15.810 --> 00:19:18.929 and how do we get that those inference into 221 00:19:19.170 --> 00:19:24.050 into Morpheus. So the first part is I kind of got to tell the box and the bot crew, 222 00:19:24.450 --> 00:19:32.764 hey, let's go talk to Morpheus and that really comes with this this skill that you'll be able to get off of the Morpheus or excuse me, the Grockbot marketplace. 223 00:19:33.565 --> 00:19:36.445 The next thing is the actual uplink 224 00:19:37.165 --> 00:19:44.259 software and that's the personal API gateway. We've talked a lot about the the larger, you know, community API gateway that that we've 225 00:19:44.180 --> 00:19:45.700 gosh, been running for almost 226 00:19:46.660 --> 00:19:49.779 gosh, it's over a year. You two years? Two years? 227 00:19:50.820 --> 00:19:56.100 Gosh, '25 or '24. I think it was the first iteration of that might have been I think the big launch was 228 00:19:56.655 --> 00:19:58.015 '25 and 229 00:20:00.575 --> 00:20:01.855 wow, time flies. 230 00:20:02.895 --> 00:20:10.495 So this is the so if, you know, everybody likes to get into the the the community API gateway, it's an easy start to see how Morpheus can run. 231 00:20:11.309 --> 00:20:14.669 But you might have other people that, you know, hey, I want my own personal 232 00:20:14.990 --> 00:20:18.429 node, I want my personal on ramp. And although we publish 233 00:20:18.590 --> 00:20:31.004 the proxy router and we've talked a lot about the code and a lot of the nerd stuff underneath this, it's still difficult for some people to get into. And what we're trying to do again is, as I've talked a lot, is is find the barrier and and try to wipe them out. 234 00:20:32.205 --> 00:20:36.365 David demoed the install install ever token claw 235 00:20:37.005 --> 00:20:39.565 x y z, install open claw x y z last week. 236 00:20:41.290 --> 00:20:44.330 Really easy button on how to start working within 237 00:20:45.050 --> 00:20:49.850 within Morpheus. And so this is just another variant of how how you can put these legos together. 238 00:20:50.490 --> 00:20:54.090 So this is Morpheus uplink. It's on the Morpheus repo. 239 00:20:54.250 --> 00:20:55.930 I'm

gonna just really quick 240 00:20:56.674 --> 00:21:02.195 need to get us started here a little bit on the compose YAML file. This is I'll put this on secret. 241 00:21:02.674 --> 00:21:07.394 So I'm gonna download that YAML file, which basically just a docker instruction file of how 242 00:21:08.515 --> 00:21:09.315 we're gonna build. 243 00:21:10.340 --> 00:21:11.140 And 244 00:21:12.340 --> 00:21:14.900 let me create that secret cloud. 245 00:21:14.900 --> 00:21:16.179 Go brek 246 00:21:16.980 --> 00:21:22.980 dot demo and the reason I need to do this is does it does have to reboot a couple times, so I'm gonna fill the air with some other noise. 247 00:21:23.794 --> 00:21:29.154 So on this setup and the setup instructions are in the repo, I'm not gonna go to them in detail. 248 00:21:29.955 --> 00:21:32.114 But here's that YAML file we just downloaded 249 00:21:32.355 --> 00:21:34.034 and that fills in basically, 250 00:21:34.355 --> 00:21:35.315 this is your 251 00:21:35.635 --> 00:21:40.409 composed file which has about three different containers on top of each other and I'll talk about that in a minute. 252 00:21:41.210 --> 00:21:46.089 The next are the secrets and before anyone freaks out, these are all burners. So, 253 00:21:47.929 --> 00:21:48.649 you know, 254 00:21:49.049 --> 00:21:58.225 you can screen, you can stop your screen and kind of see them, but that's fine. One of the things I did want to say is we're gonna leave this web public URL empty. 255 00:21:58.785 --> 00:22:12.429 And the reason for that is we don't get an external endpoint until the VM starts up and that's a bit of a pain in the neck. Still haven't figured out how to solve that yet. But we'll come back once this started and actually fill this in and reboot it again so we can actually get a good real end point. 256 00:22:13.390 --> 00:22:17.710 So that was it. I filled in my YAML file. I put my my secrets in 257 00:22:19.310 --> 00:22:20.909 including some other tuning. 258 00:22:21.950 --> 00:22:28.895 Kyle, some of these may look familiar to make sure we ensure that the proxy writer isn't storing context or forwarding context or something like that. 259 00:22:29.215 --> 00:22:30.415 So, 260 00:22:30.655 --> 00:22:32.815 those are all those are all kind of locked down 261 00:22:33.375 --> 00:22:37.535 and we're gonna watch that. And while that's creating, I'm gonna go back and talk through 262 00:22:37.935 --> 00:22:43.830 a little bit more of this. So, we are Crockbot crew on the left, we've got this this Morpheus skill 263 00:22:44.150 --> 00:22:51.910 that's gonna be available and then we're gonna go through this Uplink, you know, personal API gateway and get to that DEA DEI marketplace. 264 00:22:52.855 --> 00:22:54.455 And really, what is uplink? 265 00:22:54.774 --> 00:23:01.495 And like I said, uplink is kind of this triple stack of three existing containers. The first being that Morpheus Lumerin 266 00:23:01.495 --> 00:23:03.734 node. This is as shipped 267 00:23:04.455 --> 00:23:07.975 from the Morpheus environment, this is the proxy router kind of running in 268 00:23:09.090 --> 00:23:12.130 what we'll call consumer node. It's it's built as a consumer. 269 00:23:13.090 --> 00:23:15.250 The second layer is this uplink, 270 00:23:15.570 --> 00:23:19.970 which is the API engine and the GUI and again, that's that was born from the original consumer 271 00:23:20.290 --> 00:23:25.774 or the the large centralized API gateway has got a lot of the same elements, but not as much multi user 272 00:23:29.215 --> 00:23:34.015 weight to it. And then the top one, especially on a secret, is making sure that you can 273 00:23:34.800 --> 00:23:39.440 protect all of this behind an HTTPS endpoint instead of just empty and open 274 00:23:40.320 --> 00:23:41.760 the web. So that's 275 00:23:42.160 --> 00:23:49.040 I think that's really important. I think the Railways flavor also has that capability to make sure that you've got that TLS or HTTPS 276 00:23:49.965 --> 00:23:53.644 protection on your API endpoint as well as your your admin state. 277 00:23:54.605 --> 00:24:01.164 So that's really what uplink is. It's just putting the existing Legos that are out there, you know, I built the the uplink layer 278 00:24:01.245 --> 00:24:11.389 because I wanted that kind of GUI interface and ability to create API keys and monitor sessions and make sure everything's happy, doing a little test, make sure that it's testable 279 00:24:11.710 --> 00:24:17.150 and of course secure. But the base foundation is the Lumira node that we've been publishing for many many years now. 280 00:24:19.585 --> 00:24:22.705 I'm gonna take a quick peek at our secret VM, see if it's 281 00:24:24.305 --> 00:24:29.825 it's up and running and we've got a really cool name, apricot jungle fowl. I love these 282 00:24:30.545 --> 00:24:32.705 these names. So I'm gonna do real quick. 283 00:24:33.660 --> 00:24:37.260 This is really cool, Alan. I'm glad that you're walking everybody through 284 00:24:37.580 --> 00:24:41.500 how to do this. I think that kind of demo is super useful. 285 00:24:42.060 --> 00:24:45.340 Well, I'm glad. I I it's funny, I get into thank you. 286 00:24:45.740 --> 00:24:55.025 I get into I do it so often that I forget other people sometimes a nice walk through helps and a live demo is always fun because you're kind of expecting that 287 00:24:55.185 --> 00:25:00.145 the poor guy giving the demo getting a false flat on his face or something. It's the demo gods. 288 00:25:00.705 --> 00:25:01.505 So 289 00:25:01.810 --> 00:25:13.970 what I've done is in my I've got, you know, my one password off to the side here that I store my, you know, these kind of a list of secrets in here and what I did is I've now changed this app, you know, this web public URL to that delivered 290 00:25:15.010 --> 00:25:15.970 jungle name 291 00:25:16.315 --> 00:25:20.875 so that when we restart it comes up and it will will answer on that on that endpoint. 292

00:25:22.075 --> 00:25:28.475 So that's updating and come back here and talk a little bit little about how I run GrockBot, how I run the company on GrockBot. 293 00:25:31.270 --> 00:25:34.309 I wouldn't say it's not an RSI just yet. 294 00:25:35.750 --> 00:25:40.070 Although, did ask it, David, after our conversation. I'm like, hey, can we use Laura 295 00:25:40.150 --> 00:25:45.429 or some other layering in Uplink to help that RSI? And there's a there's a potential that 296 00:25:46.705 --> 00:25:52.784 that that what we built here, we can start adding on an RSI like layer, which I'm I'm kind of digging. 297 00:25:53.024 --> 00:25:53.664 Like, 298 00:25:54.784 --> 00:25:57.424 RSI, what are talking about? Recursive 299 00:25:57.424 --> 00:26:07.400 self improvement. Got it. Okay. Thank you. That was that. Hey. Because RSI true RSI is actually at the model level where it's actually, let's say, a nightly basis changing the weights within the model. 300 00:26:08.120 --> 00:26:15.274 And since we don't have access to some of these, can you put a layer? And I think Kyle introduced me this a couple months ago as the LoRa 301 00:26:15.515 --> 00:26:27.195 and I haven't looked into it too much, but LoRa can actually skin that as a virtual layer to to change those weights of an existing model. So I think there really could be a way to to get some RSI into 302 00:26:29.090 --> 00:26:35.170 this harness. So you you spend a lot of time with you. Bo, you mentioned naming your naming your agent. 303 00:26:35.970 --> 00:26:42.130 We anthropomorphize these quite a bit and and I've certainly done that as well as named my agents because they're easy to interact with. So 304 00:26:42.985 --> 00:26:45.864 Seraph is my chief of staff, Trinity is my developer, 305 00:26:45.945 --> 00:26:49.784 you know, you kind of name because it's easier it's that human way to interact. 306 00:26:51.225 --> 00:26:56.024 So those are my titled crew. We have a mixture of models that I mentioned, so instead of just Grok out there, 307 00:26:58.710 --> 00:27:01.429 I can use a bunch of different and I'll show you that in a second. 308 00:27:02.150 --> 00:27:14.445 I have a sovereign door into the Morpheus layer, so it's my uplink node, not somebody else's, not the shared gateway. And it's also not difficult, it's not me trying to stand up a proxy router for every agent that I want, which is a real pain in the neck. 309 00:27:15.485 --> 00:27:20.764 We have a gated ship process, you know, design or my Sizzo takes make sure I don't do anything silly, 310 00:27:21.005 --> 00:27:25.085 we build it, we gate it again and then I hold the final key to deliver 311 00:27:25.245 --> 00:27:30.339 and then we learn from those corrections either daily or weekly, we've got a we've got a learning cadence that gets uploaded 312 00:27:30.580 --> 00:27:33.379 and that nightly cadence and backup of 313 00:27:33.620 --> 00:27:39.539 the soul and all of the stuff we've learned so that it can rehydrate if we

lose that Grockbot computer. 314 00:27:40.899 --> 00:27:42.419 And so this is what I'll get to is 315 00:27:43.525 --> 00:27:51.124 we all know that system prompts by agent or per sub agent can be really powerful. Hey, your your role is chief of staff, your role is CFO, 316 00:27:51.365 --> 00:28:03.280 here's what that means, you can design that. But at the end of the day, even though I love xAI and Grock and Grockbot and all that, it's still the same model behind it. You get kind of a single view of the world 317 00:28:03.600 --> 00:28:06.000 and I would rather have something like this, 318 00:28:06.320 --> 00:28:13.715 where I've got that mixture of models and then say, for for random discussion, it's funny there was no pre 319 00:28:15.955 --> 00:28:16.755 preconditioning 320 00:28:16.755 --> 00:28:20.434 but yeah, DeepSeek v four flash is my go to for my cause. 321 00:28:21.075 --> 00:28:25.154 It's fast, it's cheap, it's smart, it's smart enough. 322 00:28:26.515 --> 00:28:26.755 So 323 00:28:27.450 --> 00:28:31.769 my my my general chat or my development on a, you know, my developer. 324 00:28:32.970 --> 00:28:39.450 But we've got that mix especially in the CTO and Sizzo role. I wanna make sure that, you know, DeepSeek, GLM and and Quinn 325 00:28:39.610 --> 00:28:43.130 are all taking a look at some of the code instead of just being very 326 00:28:43.795 --> 00:28:45.955 yes, Croc the same 10 questions. 327 00:28:47.635 --> 00:28:53.395 So with that, I think that's my burn enough time and go back here to my my secret VM, it's up and running 328 00:28:54.195 --> 00:28:59.970 and if everything has gone well, I click this button, we get a morph we get an uplink prompt and we do. 329 00:29:00.290 --> 00:29:01.090 Hey. 330 00:29:01.890 --> 00:29:03.090 It's worked out. 331 00:29:04.690 --> 00:29:05.250 So 332 00:29:05.570 --> 00:29:08.690 I love the the style of the landing page, it's great. 333 00:29:09.490 --> 00:29:10.130 Excellent. 334 00:29:10.450 --> 00:29:11.890 I am a huge 335 00:29:11.890 --> 00:29:12.610 huge 336 00:29:13.250 --> 00:29:14.690 Matrix fan so 337 00:29:15.135 --> 00:29:16.174 I get it. 338 00:29:16.495 --> 00:29:20.414 I'm gonna see if this will work better in light mode on the display. So 339 00:29:21.855 --> 00:29:25.455 this is the API gateway, this is the human side of the GUI. 340 00:29:26.495 --> 00:29:32.309 The best thing here is this API endpoint, this is that base URL which is HTTPS protected 341 00:29:32.550 --> 00:29:41.750 and then you start off with two API keys. You start off with the master API key, which allows your admin to kind of get in and work stuff and a prompt 342 00:29:41.990 --> 00:29:45.110 or you can create your own. So I'm going to create my own Grokbot 343 00:29:46.145 --> 00:29:46.945 key 344 00:29:47.424 --> 00:29:52.864 because these are ephemeral, they do not survive a full reboot, they'll fight they'll survive a restart 345 00:29:53.265 --> 00:29:57.184 but you do you can import and export keys if you need to to save them. 346 00:29:57.745 --> 00:30:01.970 You know, in case you lose private keys that may or may not have happened recently. 347 00:30:04.210 --> 00:30:06.769 And one of the first things I wanna do is 348 00:30:07.410 --> 00:30:11.890 test and just make sure everything's wired up correctly. So the nice thing is we got this little probe 349 00:30:12.445 --> 00:30:15.725 and we'll use Gemma 31 b web. 350 00:30:17.885 --> 00:30:19.005 Let me see. 351 00:30:19.645 --> 00:30:21.085 What is Morpheus? 352 00:30:22.205 --> 00:30:24.285 Aishttps 353 00:30:25.220 --> 00:30:26.419 or .org. 354 00:30:27.700 --> 00:30:31.220 Build me a fortune 500 company. Make no mistakes. Go. 355 00:30:31.700 --> 00:30:33.460 Make make money tomorrow. 356 00:30:35.220 --> 00:30:42.575 So this last in my test, it took me about twelve seconds. We'll see how how well things are working today. Hey, there we go. That's pretty good. 357 00:30:43.055 --> 00:30:49.215 Morpheus AIS is an AI driven platform designed to automate complex business workflows in the great part of AI agents. 358 00:30:49.375 --> 00:30:50.654 I think it's pretty good. 359 00:30:51.215 --> 00:30:56.440 So this tells me that I'm wired up and that I can reach the Morpheus network and I can query the models 360 00:30:56.760 --> 00:31:01.240 and I go over my sessions and usage and I can kind of see, hey, what was my Morpheus stake? 361 00:31:02.680 --> 00:31:05.720 I could close that session if I wanted to reclaim whatever's unused 362 00:31:06.245 --> 00:31:16.325 and I can kind of, you know, watch things as they as they go forward. So now comes the final part of the demo. We've kinda gotten to the to the end and I wanna I'll share my 363 00:31:18.165 --> 00:31:19.684 Crockbot crew 364 00:31:20.005 --> 00:31:20.485 with you, 365 00:31:22.120 --> 00:31:23.639 Invite them to the chat. 366 00:31:28.200 --> 00:31:30.039 And you should be able to see them. 367 00:31:30.520 --> 00:31:31.320 Nice. 368 00:31:31.880 --> 00:31:33.960 Hey there. So, Seraph. 369 00:31:38.304 --> 00:31:39.984 Let's do the demo. 370 00:31:41.745 --> 00:31:44.384 Sorry. I pre warned her it was coming, so 371 00:31:46.065 --> 00:31:47.264 see how 372 00:31:47.585 --> 00:31:50.144 see how well 373 00:31:49.279 --> 00:31:51.520 she would yell at me for for including 374 00:31:51.520 --> 00:31:54.559 the API key, but it's they're all burnable, so it's okay. 375 00:31:55.679 --> 00:31:59.360 I like how she gave you like a little input box to my 376 00:32:00.080 --> 00:32:04.715 That's been that's been new and it did not work so well on the 377 00:32:05.835 --> 00:32:13.435 in the demo in the in the pre in in my pre work early today, is why I'm kinda forcing it. She'll she'll not be happy with me pasting it in directly, but 378 00:32:13.920 --> 00:32:16.720 I thought that was interesting as they now having 379 00:32:16.880 --> 00:32:18.320 a masked fields. 380 00:32:18.960 --> 00:32:19.520 I 381 00:32:20.560 -->

00:32:24.240 do have one of the first integrations I did was one password. 382 00:32:24.960 --> 00:32:25.200 So 383 00:32:26.425 --> 00:32:35.784 she's got access to a specific one password vault, so I can move things in there whether it's a wallet or or any other kind of credential and that's worked very well. 384 00:32:36.345 --> 00:32:43.109 So I strongly suggest with any of these harnesses is make sure your secrets management are the kind of upfront and first. 385 00:32:43.590 --> 00:32:50.629 The other thing that I like about this and I think is gonna be ideal, we talked about all the nerd knobs and which model do you choose. 386 00:32:51.190 --> 00:32:59.965 At some point I asked it, I said, hey, listen to all the models and we know all the roles. What would you recommend based on research on the web and I let it do the research on, 387 00:33:00.685 --> 00:33:09.405 hey, on GLM five two or deep seek or something like that. So, it's less and less that I've got to know what models are good at is I can now ask 388 00:33:11.130 --> 00:33:12.409 I can ask them, 389 00:33:12.889 --> 00:33:16.809 which has been very powerful. I do same thing, Alan, for 390 00:33:17.130 --> 00:33:37.545 I have my agent create a newspaper for me every day that's just specific on all the stuff I wanna know about. And I and I asked it, like, which model should be doing the research? Which should be actually creating the, like, the document itself and designing it Mhmm. That, and it did parse it out for me. So, yeah, that did work pretty well when I did that too. Yeah. It's it's getting better because, you know, you're talking about 391 00:33:38.720 --> 00:33:46.080 we know it pretty well because we're in it, but, you know, the the normies like, what's Gemma? What does what's it good for? What's DeepSeek? What's it good for? 392 00:33:46.560 --> 00:33:51.360 That's, you know, everyone knows that, oh, go to cursor and it does does magic things and, you know, that's okay. 393 00:33:53.845 --> 00:33:54.404 So, 394 00:33:54.804 --> 00:34:00.565 I think you have the models all running on one hour sessions. No, they're not one hour sessions, they're five minute five minute sessions 395 00:34:01.924 --> 00:34:03.445 that they can reuse. 396 00:34:04.404 --> 00:34:08.325 And so here's the result, you can see my Grock Pot Grock Pot prompt, 397 00:34:08.810 --> 00:34:12.170 single session, 12 prompts, and 27 completions. 398 00:34:12.490 --> 00:34:15.930 Wait. They can make it as long as it's within those five minutes? 399 00:34:16.490 --> 00:34:24.045 Within those five minutes. So you remember your stake. So if I come back here to status sorry, I was kind of looking at the comments on on YouTube there. 400 00:34:25.085 --> 00:34:30.605 I've got nine more active in sessions, so those two sessions have currently locked it and if I let them run out, 401 00:34:31.405 --> 00:34:32.765 they will lock until tomorrow. 402 00:34:33.890 --> 00:34:37.810 But I'll come back here and say close all and recover unused stake. 403 00:34:43.010 --> 00:34:44.050 And my 404 00:34:44.370 --> 00:34:46.210 liquid more comes back to nine 405 00:34:46.530 --> 00:34:47.810 make sure to refresh this. 406 00:34:50.025 --> 00:34:52.345 May take a little while for the eth node to catch up, 407 00:34:52.905 --> 00:34:56.025 but this should come back I and I it picked it up already. 408 00:34:56.184 --> 00:34:59.865 Did it? Yeah. Yep. Yeah. There it is. There it is. Yeah. 409 00:35:00.905 --> 00:35:01.704 That's 410 00:35:02.529 --> 00:35:07.410 that was it. So we were able to show how we're we're able to force through that Morpheus inference, 411 00:35:09.250 --> 00:35:18.635 kinda get, you know, here's the Grokbot crew adding that Morpheus inference skill, is coming, but I'm using the the prototype is actually on my Grockbot machine right now, 412 00:35:19.435 --> 00:35:22.075 through my uplink personal API gateway 413 00:35:22.235 --> 00:35:24.795 into Morpheus to start leveraging 414 00:35:25.035 --> 00:35:26.715 that mixture of models 415 00:35:26.795 --> 00:35:27.915 from a commercial, 416 00:35:28.470 --> 00:35:34.870 you know, I still have Grock as that backstop. So, Sarah, if I, you know, I asked her to use Grock quite a bit because I want some of the horsepower 417 00:35:35.670 --> 00:35:42.869 and as part of that mixture models, but then I can also leverage the Morpheus network and take my my more and leverage that. 418 00:35:44.265 --> 00:35:46.744 The last thing I'll say on that, the nice thing is is 419 00:35:47.224 --> 00:35:48.105 cursor 420 00:35:48.585 --> 00:35:50.425 meters on a weekly basis 421 00:35:50.904 --> 00:35:54.105 and so I was running at about a 100%. 422 00:35:54.424 --> 00:35:58.105 I was over my limit at about 9AM this morning and luckily it reset. 423 00:35:58.910 --> 00:36:01.070 But Morpheus recycles every day 424 00:36:01.550 --> 00:36:05.150 and so on my daily driver, I've got about 200 more 425 00:36:05.950 --> 00:36:06.830 that is 426 00:36:08.030 --> 00:36:13.230 haul in the mail and as long as it resets every day, I've got enough that I can use that perpetual inference 427 00:36:13.515 --> 00:36:14.955 with those models 428 00:36:15.275 --> 00:36:15.994 and 429 00:36:16.154 --> 00:36:19.195 a small cursor subscription and I'm running a business. 430 00:36:20.315 --> 00:36:20.955 So, 431 00:36:22.795 --> 00:36:24.635 that's pretty cool to me. 432 00:36:25.434 --> 00:36:26.154 That's awesome. 433 00:36:26.710 --> 00:36:33.910 No. I love, the dashboard, especially there at the end where you showed off how much more is staked, how much more is claimable, 434 00:36:34.310 --> 00:36:52.185 all of that. Because my agent has been staking for a long time, but it's constantly getting sort of confused about, you know, okay. Where's the ledger? I need to reconcile on Shane. Where do I go find the information for how much is in my wallet? Oh, and I don't know. There's a bunch missing. Oh, it's in a locked session. 435

00:36:52.505 --> 00:37:01.750 I need to go unlock it. You know? Having a smooth easy way that doesn't require months of of tuning your agent or building your own dashboard, 436 00:37:02.069 --> 00:37:04.230 this is this is a huge upgrade. 437 00:37:05.109 --> 00:37:05.750 The 438 00:37:06.069 --> 00:37:07.829 the other place you can find that, 439 00:37:08.405 --> 00:37:11.525 David, is is here under tech.more.org 440 00:37:11.525 --> 00:37:12.485 under session. 441 00:37:12.565 --> 00:37:13.125 Mhmm. 442 00:37:13.925 --> 00:37:15.605 Down in here, let me 443 00:37:16.724 --> 00:37:18.245 grab my demo wallet. 444 00:37:18.965 --> 00:37:25.100 I used it's funny. I built this and I use it a ton for exactly that. I was like, do I am I missing something here? 445 00:37:25.420 --> 00:37:29.820 We can always go and this will do the same thing and give you those kind of three columns. What's liquid, 446 00:37:30.220 --> 00:37:35.100 what's locked in the current active, or what's day locked and ready to be reclaimed tomorrow. 447 00:37:36.075 --> 00:37:39.835 Nice. That's great. I hadn't played around with that function of the website. 448 00:37:40.395 --> 00:37:41.115 Yep. 449 00:37:41.275 --> 00:37:45.035 So nineteen in the wallet, ten is day locked until midnight 450 00:37:45.035 --> 00:37:45.995 UTC, 451 00:37:46.474 --> 00:37:47.915 so at 8PM East 452 00:37:48.155 --> 00:37:51.195 and there's nothing active in in the contract right now. 453 00:37:52.440 --> 00:37:54.200 Got it. Very cool. 454 00:37:55.000 --> 00:37:58.200 This is awesome. So Yes. It's already 455 00:37:58.520 --> 00:38:01.400 open sourced in the Morpheus repos. 456 00:38:01.640 --> 00:38:05.000 Yep. You for donating that to the community. Much appreciated. 457 00:38:06.395 --> 00:38:12.075 Any other notes for anybody that wants to take it, fork it, improve it, any of that? 458 00:38:13.435 --> 00:38:14.315 Go nuts. 459 00:38:15.035 --> 00:38:15.994 The Morpheus 460 00:38:15.994 --> 00:38:16.955 Inference 461 00:38:17.035 --> 00:38:17.835 skill, 462 00:38:18.635 --> 00:38:20.474 I submitted to the Grokbot 463 00:38:20.760 --> 00:38:21.720 marketplace 464 00:38:21.720 --> 00:38:22.920 last night 465 00:38:23.320 --> 00:38:33.240 and I have not seen any response on that. I don't know what their criteria or their qualifications. I did try to do a terms of service check to say, hey. Can are you allowed to use or not use anybody's else's inference? 466 00:38:33.240 --> 00:38:36.120 And there's no restrictions, so I don't think there'll be any blockers there. 467 00:38:37.224 --> 00:38:39.785 But if there is, what I'll do is I'll post the 468 00:38:40.105 --> 00:38:44.905 the Python and the structure. I'll I'll have I'll have Sarah export that package 469 00:38:45.305 --> 00:38:50.025 because that's what I had him build. I said, hey, do this. And we came up with the Morpheus bridge 470 00:38:50.289 --> 00:38:55.570 is what they what what she originally called it and I said, let's make it an MCP that we can post on the marketplace. 471 00:38:56.369 --> 00:38:59.410 I know AWS posts their Bedrock inference there, 472 00:39:00.289 --> 00:39:06.045 other things, so I don't think they're gonna have an issue with Grok calling out to other inference. 473 00:39:06.525 --> 00:39:07.165 So 474 00:39:07.885 --> 00:39:08.525 but 475 00:39:09.165 --> 00:39:11.325 no no other no other restrictions there. 476 00:39:11.725 --> 00:39:14.765 Awesome. That's great, Alan. Yeah. No. I love to see 477 00:39:15.005 --> 00:39:20.840 basically the Morpheus community hooking in to all the different harnesses and platforms. 478 00:39:20.920 --> 00:39:23.640 Right? So I spent a lot of time hooking it up into 479 00:39:23.640 --> 00:39:24.680 OpenClaw. 480 00:39:24.920 --> 00:39:27.720 You know? You're bringing it to Grok. 481 00:39:27.720 --> 00:39:35.515 I'm sure somebody else is working on on Hermes and and the other harnesses. Like, that's that's huge. Right? Because now you can land on Morpheus 482 00:39:35.515 --> 00:39:39.435 and be like, oh, here's my agent. Here's my harness. Here's how I connect. 483 00:39:39.835 --> 00:39:48.040 One click kind of stuff. I'd love to All all roads lead to all roads lead to Morpheus. All roads lead to Morpheus. But I'd love to see that, like, one click 484 00:39:49.000 --> 00:39:50.280 pain of 485 00:39:50.520 --> 00:39:56.600 of, like, oh, I land on this page and I can just pick my agent harness, and it dumps me into 486 00:39:57.000 --> 00:40:06.234 the the guide or the simple setup for my harness to connect to Morpheus. So maybe useful as a as a page on tech.morn.org 487 00:40:06.234 --> 00:40:10.395 is, like, just to show off those different integrations so people know where to find them. 488 00:40:11.515 --> 00:40:17.480 Yeah. A live a library. And I think all of the sites that I've seen so far, at least the ones I've demoed today, by 489 00:40:17.800 --> 00:40:20.280 default, I'm making sure, are you robot? 490 00:40:20.360 --> 00:40:23.160 Are you bot sensitive? Are are you bot compliant? 491 00:40:23.640 --> 00:40:27.885 So you should be able to point your agent now to say, hey, go look at, you know, 492 00:40:28.525 --> 00:40:36.925 Morpheus AIS, Morpheus dash uplink, tell me what it's all about and your agent should be able to tell you probably better than I've demoed, I hope. 493 00:40:37.725 --> 00:40:38.445 Awesome. 494 00:40:38.845 --> 00:40:46.200 That's great, Alan. Cool. Very good. The fact that that works is a live demo, and you could just show it in twenty minutes nowadays. 495 00:40:46.200 --> 00:40:46.840 I 496 00:40:47.000 --> 00:40:47.880 mean, what 497 00:40:48.440 --> 00:40:57.525 are we doing? Like, we're come we've come so far. We really gods have smiled upon upon us. Yes. It's the Alan level of preparation, 498 00:40:57.685 --> 00:41:01.685 slides and tested it beforehand and, you know, the whole thing. 499 00:41:02.085 --> 00:41:04.565 So very cool. Yeah. It's amazing. 500 00:41:05.045

--> 00:41:14.190 Well, thank you, Alan, for doing that. I mean, yeah, everyone listening, I mean, obviously, back, rewatch it, set it up for yourself, explore the different harnesses. 501 00:41:15.230 --> 00:41:18.670 I've become partial to Hermes, but I have not tried Grokbot, 502 00:41:18.750 --> 00:41:20.830 so there's a very good chance it is better, 503 00:41:21.495 --> 00:41:39.960 which is what happened with me. And, well, Open Claw, a ton of people are still finding success with too. So honestly, just just check one of them out, people. Just just try one. You know, this is actually something that Taco and I say every week on Manifest Morpheus Monday is go try something this week. Just, like, try something new. There's so much stuff out there. 504 00:41:40.520 --> 00:41:44.680 And all of it pretty much talks to Morpheus now, or if it doesn't already, 505 00:41:44.839 --> 00:41:46.599 someone's done it, you know. This is Yeah. A perfect 506 00:41:48.200 --> 00:41:48.760 Yeah. 507 00:41:49.000 --> 00:41:49.640 Exactly right. 508 00:41:50.575 --> 00:41:53.055 And and let's just kind of 509 00:41:53.375 --> 00:41:55.295 end this going back to 510 00:41:56.015 --> 00:41:57.615 why what 511 00:41:57.615 --> 00:41:58.415 Alan 512 00:41:58.415 --> 00:42:03.215 should, you know, just demo this is so important and on the uplink side. Right? It's 513 00:42:03.375 --> 00:42:04.655 going back to the original, 514 00:42:05.960 --> 00:42:09.800 you know, when Morpheus was was launched, everyone said, 515 00:42:10.119 --> 00:42:11.720 this is absolutely 516 00:42:11.799 --> 00:42:15.720 amazing. Right? There is a true need in the market for 517 00:42:17.240 --> 00:42:18.359 access to 518 00:42:18.359 --> 00:42:19.240 permissionless 519 00:42:19.240 --> 00:42:20.359 decentralized 520 00:42:20.359 --> 00:42:20.920 inference. 521 00:42:21.295 --> 00:42:21.935 And, 522 00:42:23.295 --> 00:42:26.495 you know, we at Morpheus brought that to the world. 523 00:42:27.135 --> 00:42:33.375 Maybe we didn't bring it as easy as it could be for for everyone in the beginning, and and some people 524 00:42:34.415 --> 00:42:36.735 hit the wall a little bit of 525 00:42:38.200 --> 00:42:40.360 it's a little harder than I anticipated 526 00:42:40.360 --> 00:42:50.840 to get access to this amazing system and and kind of you've seen us, you've heard us through the last and it's crazy. It's a year on on creating this API gateway. 527 00:42:51.080 --> 00:42:53.425 And, you know, I think that 528 00:42:53.905 --> 00:42:55.105 good development, 529 00:42:55.105 --> 00:43:04.865 you know, kind of takes a couple routes. Right? Maybe you have to do something a little bit of a centralized way to test it and prove it, and then fight towards full decentralization. 530 00:43:04.865 --> 00:43:07.425 So I think the API gateway was 531 00:43:08.330 --> 00:43:15.770 how do we meet users where they are? They want an API key to be able to plug into their systems, and and we did that. 532 00:43:16.650 --> 00:43:18.730 How do you make it decentralized? 533 00:43:18.730 --> 00:43:19.930 And how do you own, 534 00:43:20.555 --> 00:43:25.035 you know, your intelligence and own that yourself. And that's, you know, the product uplink, 535 00:43:25.035 --> 00:43:29.675 and that's, you know, how you can kind of manage that full stack 536 00:43:29.755 --> 00:43:32.795 while, you know, of privacy and of decentralization 537 00:43:32.795 --> 00:43:33.915 and permissionless, 538 00:43:34.235 --> 00:43:37.070 and how do you do that in a way that's, you know, 539 00:43:37.390 --> 00:43:43.710 just a couple clicks to get there. So I think we've really come a long way, and I think that, you know, 540 00:43:44.990 --> 00:43:47.310 you know, the the proof is is 541 00:43:47.630 --> 00:44:02.745 in the work. Right? And the proof is you've seen the trajectory that, you know, the the contributors that Morpheus have taken, and and you'll see the trajectory that they continue to. How do we get to to the the deepest levels of decentralization 542 00:44:02.745 --> 00:44:07.065 while also kind of meeting people where they are and providing them the experience that they expect. 543 00:44:08.140 --> 00:44:09.740 Yep. Yeah. Well said. 544 00:44:10.940 --> 00:44:11.980 Totally agreed. 545 00:44:12.620 --> 00:44:13.420 That 546 00:44:13.660 --> 00:44:16.540 does bring us to, you know, some of the topics 547 00:44:16.620 --> 00:44:18.620 that we have this week, which 548 00:44:19.980 --> 00:44:23.955 the first two that I had jotted down were the fact that, 549 00:44:25.075 --> 00:44:30.355 Claude released Opus five five and kinda did the leapfrog thing again for the Frontier centralized 550 00:44:30.355 --> 00:44:32.595 models. But the bigger story here 551 00:44:32.994 --> 00:44:33.635 is 552 00:44:34.675 --> 00:44:36.275 how quickly 553 00:44:35.990 --> 00:44:42.869 the time between front Frontier models is happening. Like, it's dropped massively. I I think the number I saw was eighteen 554 00:44:42.869 --> 00:44:48.070 days now is, like, the expected period between one Frontier model to the next. 555 00:44:48.869 --> 00:44:52.485 Yep. Down from, like, seventy days between 556 00:44:52.725 --> 00:44:57.925 releases. Now it's less than three weeks, which is which is just wild. Right? 557 00:44:58.325 --> 00:45:01.285 And and I think it's gonna continue to accelerate. 558 00:45:01.445 --> 00:45:02.645 You know, Elon 559 00:45:03.110 --> 00:45:08.070 set the ambitious objective of releasing a new version of Grok every week or two. 560 00:45:08.790 --> 00:45:17.110 And I don't think they've been too far off that since Grok 4.5, 4.6. You just saw the release of Grok 4.7 561 00:45:17.145 --> 00:45:19.705 a few days ago. So they're definitely keeping it up, 562 00:45:20.185 --> 00:45:21.385 and his claim is 563 00:45:21.625 --> 00:45:35.240 by the time they get to five point o, it'll beat any current frontier model. So the other frontier models aren't staying still,

so we'll see if that that happens to be the case. But I think it's a massive, like, 6,000,000,000,000 parameter 564 00:45:35.320 --> 00:45:51.415 model when they get to to five point o. So, yeah, I think that pace is just gonna keep accelerating because think about it, as every one of those releases come out, because people are doing distillation and learning from other best practices that the model are using, it improves the other models. Yeah. But 565 00:45:52.535 --> 00:45:54.055 I wanna push back 566 00:45:54.454 --> 00:45:55.575 on a comment, 567 00:45:55.974 --> 00:45:58.135 Alan, I think it was you you said earlier, 568 00:45:58.535 --> 00:45:59.015 that 569 00:45:59.790 --> 00:46:00.910 I agree 570 00:46:01.310 --> 00:46:07.390 that historically people have thought of RSI, of recursive self improvement, as the model's improving themselves, 571 00:46:08.110 --> 00:46:12.270 but I think what people are gonna actually care about is RSI for agents. 572 00:46:12.755 --> 00:46:14.755 Okay. Oh. It did get 573 00:46:14.995 --> 00:46:20.755 better today. And I don't care whether the models changed. If all it had to do was learn a new memory, 574 00:46:20.995 --> 00:46:25.075 pick up a new skill, you know, record one of my preferences, 575 00:46:25.395 --> 00:46:37.500 you know, get faster between when it runs as loops. There's a lot of ways my agent could learn and get better every day, and that goes back to what do I really care about? How many tasks it got done? 576 00:46:37.740 --> 00:46:42.505 How much money it made me? How much work it did? Right? And whether it benchmarks 577 00:46:42.505 --> 00:46:44.985 at a 180 or a 181 578 00:46:44.985 --> 00:46:47.865 IQ, like, the model quality, 579 00:46:48.105 --> 00:46:50.265 to me, like, that's been their game. 580 00:46:50.505 --> 00:46:55.225 Right? They're all focused on RSI in the model context, but I actually think what matters to the individual 581 00:46:55.910 --> 00:46:57.190 is RSI 582 00:46:57.350 --> 00:47:01.590 for the agent. And I I I wanna say, I'm hearing it everywhere. 583 00:47:01.750 --> 00:47:04.150 I think this is the new buzz. 584 00:47:04.390 --> 00:47:08.390 Right? It is it is not AGI anymore. I don't hear AGI. 585 00:47:08.390 --> 00:47:12.785 Everybody basically said we're at AGI, like, you know, tell me tell me we're not. 586 00:47:13.345 --> 00:47:16.625 And so I don't think AGI is the benchmark people are going towards. 587 00:47:16.944 --> 00:47:20.945 The the new hype and and especially the fear mongers and the doomer crowd, 588 00:47:21.184 --> 00:47:22.224 like RSI, 589 00:47:22.224 --> 00:47:28.200 that's the new thing to be afraid of. Right? It used to be AGI. We made it to AGI and everything's fine. But RSI, 590 00:47:28.200 --> 00:47:30.680 that's what you really have to be 591 00:47:30.840 --> 00:47:36.040 afraid of, which is a good signal for builders that that's what you need to build next. And so, 592 00:47:36.360 --> 00:47:39.080 anyway, I just wanted to reframe it. I think we need to catch 593 00:47:40.305 --> 00:47:41.505 RSI for agents 594 00:47:41.585 --> 00:47:43.665 as the true definition 595 00:47:43.665 --> 00:47:46.065 of RSI that matters to the average person. 596 00:47:46.545 --> 00:47:50.465 I think that's a I think it's a great point and thank you for for for the pushback. 597 00:47:50.704 --> 00:47:54.464 I did that's where I started the conversation with with Sarah was a little bit 598 00:47:54.970 --> 00:48:03.290 of because we have these these daily routines or certain prompts that say, hey, remember this for the group or remember this for you and then the nightly dream cycles 599 00:48:03.530 --> 00:48:15.405 and I I kind of tease it like, this RSI? She's like, no, not really. Here's the real real definition. I'm like, but to your point, there was a part of that conversation that was like, yeah, we could get better at when there's a thing 600 00:48:16.045 --> 00:48:20.845 that everybody should remember. Let's stick that in in more permanent memory and make sure it's backed up. 601 00:48:21.405 --> 00:48:34.150 And I I made I've made the the parallel before. It's like raising kids. We've we've had toddlers. We gotta tell them everything. No. No. No. Don't do that. Don't do that. Don't do that. Here's how to tie your shoe. Here's how to properly lift the spoon to your face and feed yourself. 602 00:48:34.550 --> 00:48:35.430 And eventually, 603 00:48:35.830 --> 00:48:36.870 they get this RSI 604 00:48:37.350 --> 00:48:44.365 kids get their own RSI, they experiment, they go try something and they learn from their mistakes, they fall down, they get corrected, 605 00:48:44.765 --> 00:48:48.685 you know, that kind of thing. And so I think maybe are we entering the toddler, 606 00:48:49.005 --> 00:48:49.885 the 607 00:48:49.885 --> 00:48:52.205 toddler phase of AGI 608 00:48:52.430 --> 00:48:55.390 and can RSI in the frame that you've proposed 609 00:48:55.790 --> 00:48:56.350 it, 610 00:48:56.990 --> 00:48:59.150 can that get better? It's 611 00:48:59.150 --> 00:49:05.230 not about whether you have the intelligence in the back room, but can you ask it, can the agent ask it better questions 612 00:49:05.470 --> 00:49:19.695 with a better corpus of knowledge to go after it. So yes, this is just like, you know, you've now got a five year old in the room, not a two year old in the room. Well, I've deployed an RSI agent on top of all of my looping infrastructure to basically 613 00:49:20.799 --> 00:49:26.480 look every day at the different things that I'm doing and make specific 614 00:49:26.480 --> 00:49:27.440 proposals, 615 00:49:27.599 --> 00:49:32.960 put them into the loop. Mhmm. And then the looping agents work on their own improvements. 616 00:49:33.279 --> 00:49:37.535 Right? And I get a report every day of all of the improvements. 617 00:49:37.535 --> 00:49:39.535 Right? I have an architecture 618 00:49:39.694 --> 00:49:53.480 agent that's looking at, like, is this within scope?

Does it fit my values, my goals, you know, my tasks, and things like that? And then it escalates to me what it thinks might be out of scope or it needs more decisions on. And I've and I've given it 619 00:49:53.720 --> 00:49:55.480 the overarching goal 620 00:49:55.640 --> 00:50:01.000 is for it to learn a little bit more every day, and I get less and less escalated items 621 00:50:01.240 --> 00:50:05.400 because it knows better and better what I'm trying to achieve 622 00:50:05.664 --> 00:50:11.425 and how I want to achieve it. So so far, so good. Right? Like, I I look at my dashboard today, 623 00:50:11.744 --> 00:50:16.065 and if I, I take that as sort of my my gauge of my numbers, 624 00:50:16.224 --> 00:50:19.744 you know, I've got more than a 100 items now 625 00:50:20.220 --> 00:50:22.940 in the the looping infrastructure. 626 00:50:23.340 --> 00:50:29.580 Right? And it would be really interesting. Yeah. I don't know how we capture some of these these broader metrics, 627 00:50:29.740 --> 00:50:33.340 but I've got 16 things in the backlog, a 157 628 00:50:33.500 --> 00:50:35.260 ready, have been fully developed, 629 00:50:36.105 --> 00:50:58.760 31 have been shipped. 16 are are parked waiting for a decision. And it gives me this beautiful rundown of, oh, I spent, you know, one hour building this new cron job, and I spent, you know, half an hour improving either this new RSI task. And I can just look through and see everything that's going on with my agents. So, yeah, I I think that's where it gets really interesting 630 00:50:59.240 --> 00:51:00.040 because 631 00:51:00.200 --> 00:51:03.320 if you start that I r that RSI 632 00:51:03.320 --> 00:51:04.920 loop fairly early, 633 00:51:05.240 --> 00:51:06.120 it's compounding. 634 00:51:06.744 --> 00:51:07.785 Right? 635 00:51:07.865 --> 00:51:11.865 I I'm able to get more done, which makes it easier to get more done. 636 00:51:12.265 --> 00:51:12.905 Right? 637 00:51:13.065 --> 00:51:19.625 That's that's where I think it's gonna get really wild. So, yeah, it's I'd recommend it highly to 638 00:51:19.785 --> 00:51:23.080 to anybody who hasn't tried it yet. If you've got the infrastructure 639 00:51:23.080 --> 00:51:24.280 already with an agent 640 00:51:24.600 --> 00:51:27.480 and it has a brain, so you're you're storing the memories 641 00:51:27.640 --> 00:51:29.400 and you're you're running loops, 642 00:51:29.480 --> 00:51:30.840 they're already doing this work, 643 00:51:31.080 --> 00:51:34.280 set something on top of that that can that can improve it. So. 644 00:51:35.025 --> 00:51:39.665 Alright. I might try to tackle that next, David, because I I do not have it doing that currently. 645 00:51:39.905 --> 00:51:48.865 Yeah. If you don't have an RSI agent, what are you what are you even doing? What what are we doing? I'm still doing that. I I was thinking as as Alan was showing off his 646 00:51:49.720 --> 00:51:51.960 Grockbot company and and his, like, 647 00:51:52.280 --> 00:51:57.000

assignment scheme for the agents, I was thinking through the next couple names for mine. And I have 648 00:51:57.400 --> 00:51:59.560 Sam Botman Fried for 649 00:51:59.560 --> 00:52:00.280 my 650 00:52:00.440 --> 00:52:02.120 chief compliance officer, 651 00:52:02.280 --> 00:52:04.200 and then I have Elizabeth 652 00:52:04.585 --> 00:52:06.185 Elizabeth Holmes 653 00:52:06.185 --> 00:52:07.865 for my chief of product. 654 00:52:08.105 --> 00:52:11.945 Wow. And, man, I was I was really brainstorming over here. 655 00:52:12.345 --> 00:52:14.665 That's the dream team you're talking about. 656 00:52:15.705 --> 00:52:22.060 Yeah. You know, it'll be it'll be good. No, that's that is awesome though. Yeah. I that totally makes sense in in why it would be 657 00:52:22.540 --> 00:52:23.340 compounded 658 00:52:23.340 --> 00:52:24.140 growth. 659 00:52:24.700 --> 00:52:28.300 The earlier you start it, the better it is. And I just 660 00:52:28.620 --> 00:52:30.380 keep thinking too about 661 00:52:31.020 --> 00:52:32.060 how 662 00:52:33.165 --> 00:52:44.365 what's so great about once you do have that personal agent set up and you're retaining all that data and information, like if it's on a dedicated device or if you're doing like IPFS or something like that, 663 00:52:45.980 --> 00:52:48.780 it'll just go with you forever now. 664 00:52:49.020 --> 00:52:51.740 Like like, as these things evolve, like, 665 00:52:52.059 --> 00:53:05.005 Verde for me will just evolve. Like like, I'll I'll I'll remember, it's like the analogy with your kid, Alan. I'll remember when Verde was just a little, you know, cron job basically in my in my Mac mini. 666 00:53:05.245 --> 00:53:07.165 And then it'll be standing 667 00:53:07.165 --> 00:53:23.390 in my house as the brain of my home bot or whatever in the future, and I'll be like, you have no idea. You used to just be this small little cron job server, basically, and like It used to be 10 lines it used to be 10 lines of code. Yeah. Yeah. Now you're you're three gig worth of a GitHub repo. 668 00:53:23.470 --> 00:53:26.910 Yeah. It's so crazy to think about, but that's that's really where we're headed. 669 00:53:28.045 --> 00:53:34.845 The only other thing I really had on the list this week, and we're going a little bit long, but that's okay, we started a little bit late, was the 670 00:53:36.045 --> 00:53:41.245 hype around JEV still being, you know, really huge, that's ramping up. There doesn't seem to be 671 00:53:41.859 --> 00:53:45.540 an exact, like, raw open weight version quite yet, 672 00:53:45.780 --> 00:53:47.780 although I suspect there will be 673 00:53:48.260 --> 00:53:49.460 at any moment. 674 00:53:49.780 --> 00:54:03.965 And also, I've hate to break it to you, but Is it out? Right now, I just saw there's Kev. Kev. Okay. Kev. Breaking news. Welcome, Kev. Yeah. There's there's a few of them now, but I think they're pretty close on on the benchmarks. 675 00:54:04.045 --> 00:54:07.725 And just to give people the quick download for those that haven't looked at Jev 676 00:54:08.170 --> 00:54:08.890 is 677 00:54:09.210 --> 00:54:11.610 I would describe it as 678 00:54:12.170 --> 00:54:14.730 a very sort of deterministic 679 00:54:15.130 --> 00:54:20.090 way of doing models where it's it's really just giving you a probability and trying to make a decision. 680 00:54:20.330 --> 00:54:22.330 It's not trying to give you a bunch of words back. 681 00:54:22.895 --> 00:54:29.135 Right? And it can only handle a small amount of context, and so it's like decide if this prompt is 682 00:54:29.454 --> 00:54:30.655 a plan, 683 00:54:30.974 --> 00:54:32.095 a question, 684 00:54:32.335 --> 00:54:33.375 a greeting, 685 00:54:33.615 --> 00:54:35.694 anything where there's like categorization 686 00:54:36.060 --> 00:54:36.940 involved. 687 00:54:37.100 --> 00:54:39.180 Jeb's really good at matching 688 00:54:39.420 --> 00:54:41.740 those categorizations very dynamically 689 00:54:41.900 --> 00:54:50.860 and giving you a probability. Okay, 99% they're asking for a plan on this prompt, or they're asking a question, or they're just greeting you. Right? And so one of the coolest 690 00:54:51.684 --> 00:54:52.805 use cases 691 00:54:53.365 --> 00:54:59.125 that I've seen is using it as a router, using it as a judge at the beginning of your process. 692 00:54:59.525 --> 00:55:06.244 In order to Alan, like you were saying, which model do I send it to? Well, it needs the heavy powerful model if it needs a whole plan. 693 00:55:06.840 --> 00:55:15.080 If it's just asking hello and what's the weather, you probably get away with your your lightest model. And Jeb is like, you know, half a second to 694 00:55:15.720 --> 00:55:19.240 give you an instant, you know, tenth of a penny reply 695 00:55:19.400 --> 00:55:26.175 on the probability of what it is, and it it seems very good at that kind of use case. So not gonna replace 696 00:55:26.974 --> 00:55:31.615 a large language model for the thinking, reasoning workload, 697 00:55:31.695 --> 00:55:33.535 but it's really interesting 698 00:55:33.775 --> 00:55:37.775 at those type of applications. The other one I saw is somebody took screenshots 699 00:55:38.410 --> 00:55:40.730 and said, look at this screenshot. 700 00:55:41.130 --> 00:55:45.690 Tell me where the user got lost. Right? Did they not click on the right thing? 701 00:55:46.250 --> 00:55:50.650 You know, whatever it is. And they gave you the basic thing, and they fit in a million screenshots. 702 00:55:51.474 --> 00:55:55.635 And it just ate through very quickly what would have cost a ton 703 00:55:55.795 --> 00:56:01.795 in in inference cost in, you know, a few bucks they had categorized these million screenshots. 704 00:56:02.035 --> 00:56:04.595 So it definitely is a compliment 705 00:56:05.470 --> 00:56:10.430 to those existing systems, not a replacement for, but it's definitely a new powerful 706 00:56:10.510 --> 00:56:12.590 tool in the tool belt. And ideas 707 00:56:12.590

--> 00:56:20.590 like this have been around for a while, but this was definitely the breakout moment when they made that hype video and everybody's like, oh, I need that. Yeah. 708 00:56:21.285 --> 00:56:26.005 I like I like the Rick and I like the Rick and Morty That's what I was gonna on Jev. That was great. 709 00:56:26.005 --> 00:56:26.965 That was the best. 710 00:56:27.365 --> 00:56:29.525 I saw a teaser. I 711 00:56:29.845 --> 00:56:32.724 did some work on the router the other day and 712 00:56:33.180 --> 00:56:39.580 found out that you do not interact with JEV or decision model like anything else. It doesn't take the chat completion. So 713 00:56:39.980 --> 00:56:43.020 we do have a submitted the PR yesterday 714 00:56:43.260 --> 00:56:48.405 that should be out in the package for the proxy router that will take if a model is tagged decisions 715 00:56:48.405 --> 00:56:53.285 just like embeddings. I I would like in this almost works in embeddings as you take a whole lot of input 716 00:56:53.765 --> 00:56:58.005 and it gives you a very little output. It's it's kind of the opposite of what we expect, a little input, lots of output. 717 00:56:58.850 --> 00:56:59.810 And so 718 00:57:00.210 --> 00:57:04.610 I'm hoping when Kev or or whatever comes on to the Morpheus network, 719 00:57:05.570 --> 00:57:08.690 we're planning on being ready for it. So that's already in testing. 720 00:57:08.930 --> 00:57:10.690 There's one JEV that's out there. 721 00:57:12.025 --> 00:57:15.865 So I think it will be like you said, it's that front end router. 722 00:57:16.105 --> 00:57:24.985 You may not interact a lot with it, but it will be the traffic cop for things. So so getting your agent to be able to talk to JEV, a JEV like model or decision model, 723 00:57:26.000 --> 00:57:30.480 and then go off and do work on the the mixture of models of the MOEs 724 00:57:30.560 --> 00:57:34.000 is gonna be powerful. So but the teaser is it's 725 00:57:34.400 --> 00:57:39.280 we're we've got first article into the proxy router that should be baked in in the next twenty four hours. 726 00:57:40.385 --> 00:57:41.025 Nice. 727 00:57:41.585 --> 00:57:47.425 That's great. It's good to hear that somebody's already hosting JEV on on Morpheus. That's One provider. 728 00:57:47.505 --> 00:57:56.490 Yeah. Nice. Pretty quick turnaround. Very cool. Yep. Good stuff. Bo, anything else we, we didn't cover? I think that's a lot of the hype from this week. Yeah. I don't think so. That was, 729 00:57:57.210 --> 00:57:58.730 yeah. Pretty much everything. 730 00:57:59.610 --> 00:58:07.210 Other stuff just really quick pseudo related, but not totally was, Meta yesterday rolled out the preview of their VR glasses, 731 00:58:07.290 --> 00:58:10.025 which is exciting for the hardware nerds as well. 732 00:58:10.505 --> 00:58:13.385 You know, as as Morpheus and and 733 00:58:13.625 --> 00:58:28.550 smart agents and personal agents evolve, then plugging into, like, other types of hardware will

definitely be or more so, like, you using your agent or interacting with it through other types of hardware will be big. I was gonna say if we're gonna go Facebook, we should probably cover Muse. 734 00:58:29.030 --> 00:58:30.230 Oh, sure. Go. That 735 00:58:30.710 --> 00:58:34.630 was I would I categorize it as the Facebook agent 736 00:58:34.710 --> 00:58:35.270 harness. 737 00:58:35.964 --> 00:58:38.605 Right? And and is this 738 00:58:38.845 --> 00:58:40.925 the moment where all the normies 739 00:58:41.085 --> 00:58:45.405 get an agent? Maybe? Like, they're setting an ambitious goal of a billion 740 00:58:45.645 --> 00:58:55.350 news agents and he's got a footprint of, you know, three, four billion people, whatever it is on on meta. So, you know, it's believable. He's got the distribution. 741 00:58:55.670 --> 00:58:58.790 Does he have enough compute to support them all? Maybe. I don't know. 742 00:58:59.110 --> 00:59:00.390 But, you know, 743 00:59:00.950 --> 00:59:02.790 I think it's it's interesting. 744 00:59:03.190 --> 00:59:06.950 So I can put that at least in my mind, I put it, like, next to Grokbot 745 00:59:07.175 --> 00:59:11.655 and Hermes and Open Claw and Pi and these other Mhmm. Harnesses. 746 00:59:11.655 --> 00:59:14.935 But I haven't taken a deep dive on it yet. I don't use a lot of meta products. 747 00:59:15.175 --> 00:59:27.270 So so I haven't ventured down that rabbit hole just yet. Yeah. Although, I will say, like, Zuck seems to be a big proponent of open source. So I mean, he has been historically with Llama other so we'll 748 00:59:27.349 --> 00:59:30.070 we'll see what happens, though. Aft after November, 749 00:59:30.070 --> 00:59:34.150 and then certainly two years from now, if the political pendulum swings, we'll see how, 750 00:59:35.225 --> 00:59:40.825 you know, a much a fan of open source he is. He tends to be a go with the wind kinda guy. 751 00:59:41.225 --> 00:59:42.585 May maybe so. 752 00:59:42.825 --> 00:59:44.185 Though it'll be interesting. 753 00:59:44.985 --> 00:59:48.825 I don't think it'll hit as much this election season, but next election season, 754 00:59:49.940 --> 00:59:52.500 apparently AI is going to be the story. 755 00:59:52.660 --> 00:59:58.420 But I don't know how the anti AI people deal when a billion of their constituents 756 00:59:58.420 --> 00:59:59.940 have an agent. 757 01:00:00.740 --> 01:00:01.540 The level 758 01:00:02.075 --> 01:00:04.795 of, like, noise communication 759 01:00:04.875 --> 01:00:05.915 propaganda 760 01:00:05.915 --> 01:00:07.435 is just gonna be 761 01:00:07.675 --> 01:00:08.875 insane 762 01:00:08.875 --> 01:00:10.395 in 2028. 763 01:00:11.115 --> 01:00:23.580 I it may be going on right now. I mute so many political words in my x feed. I I'm oblivious to a lot of what's going on in politics, and so maybe everybody's being bombarded with campaign ads right now. I wouldn't know. But 764 01:00:24.380 --> 01:00:33.255 I think in '28, it's just gonna be wild because everybody will have an agent. And if they have a political opinion, they're gonna be able to effectively 765 01:00:33.335 --> 01:00:37.175 amplify that to any of their friends or 766 01:00:37.175 --> 01:00:39.415 people they know. I I compare it to like 767 01:00:40.295 --> 01:00:42.695 politics did not see social media coming. 768 01:00:43.590 --> 01:01:02.445 Right? Yeah. First social media star in politics was Ron Paul. Right? He hit the airwaves when YouTube was early and all that stuff was early in two thousand seven, two thousand eleven. And then really Obama was the first guy to take it mainstream and be sort of all over social media when his opponents were like, email? 769 01:01:02.605 --> 01:01:14.285 Like, who uses who uses social media? Like, making fun of it as if it was like this fringe thing and not, you know, by the mid twenty tens, majority of people using social media. So it'd be interesting to see the impact 770 01:01:14.580 --> 01:01:22.740 it has on the next one. So hopefully, it empowers people to have more voice and that voice to be more decentralized and not only routed through 771 01:01:22.820 --> 01:01:29.535 a couple of big news organizations. So anyway, we'll see what happens. But Yeah. I mean it'll be interesting. It is definitely going to be 772 01:01:30.015 --> 01:01:33.375 a a or the massive topic of conversation, 773 01:01:33.455 --> 01:01:34.015 but 774 01:01:34.335 --> 01:01:36.575 at least as as things look today. 775 01:01:36.815 --> 01:01:38.975 I do suspect, though, that by then, 776 01:01:39.670 --> 01:01:41.350 it just won't even matter. 777 01:01:41.590 --> 01:01:43.350 Because, like, it is it is, 778 01:01:45.030 --> 01:01:46.630 what's the word I'm looking for? 779 01:01:47.350 --> 01:01:48.710 All encompassing. 780 01:01:48.950 --> 01:02:00.865 Like, you cannot unring the bell, and people will eventually come to terms with that, or they won't. And then I always say, like, the joke I make is that we're just creating a new brand of Amish. 781 01:02:01.425 --> 01:02:13.100 Like, the Amish drew the line in the sand somewhere and said, this is where we no longer use new technologies. Why? I don't know. Like, in terms of, like, the exact history behind it, I'm not gonna claim to be a a, 782 01:02:13.500 --> 01:02:15.100 you know, historian 783 01:02:15.100 --> 01:02:28.625 on the Amish and the Mennonites and all that fun stuff. But somewhere along the line, a group of people said, draw on the line in the sand. This is it. You know? And we're just gonna There's gonna be a new subset of people who do the same thing with AI, 784 01:02:28.785 --> 01:02:34.625 who are basically And there are some people with computers. I mean, my own grandmother has never owned a computer. 785 01:02:35.240 --> 01:02:45.800 She does not own And now she has a phone. She has an iPhone now, which I guess I mean, I I would be shocked if she's ever even used, like, Internet Explorer on it. I would be I would be floored. But that

786 01:02:46.200 --> 01:02:50.599 was kind of like a mini version of this. AI is just gonna be an extreme version of this. 787 01:02:51.345 --> 01:02:53.425 99.99% 788 01:02:53.425 --> 01:02:54.305 of everybody 789 01:02:54.385 --> 01:02:58.145 will just bend the knee and realize that, like, it is unavoidable. 790 01:02:58.305 --> 01:03:15.250 We could shut it down here, or stop talking about it, or developing it here. That's not gonna happen in the rest of the world. So, like, get on board or get ran over by the train. Like, that's just it is what it is. Two years is a lot of time. I think by two years from now, that might already be generally accepted. 791 01:03:15.650 --> 01:03:20.050 Yeah. It's it's either generally accepted or it's gonna be a really losing campaign 792 01:03:20.615 --> 01:03:28.535 promise to Yeah. But that's why AI. That's why I don't think it will be though because it'll be That's why they always frame it as safety. 793 01:03:28.694 --> 01:03:29.255 Like, 794 01:03:29.494 --> 01:03:32.375 for your own good. Yeah. Think of the children. 795 01:03:32.615 --> 01:03:35.974 Don't you understand these AIs? 796 01:03:34.420 --> 01:03:36.900 Very, very dangerous. If only it was controlled 797 01:03:37.140 --> 01:03:37.700 by 798 01:03:38.100 --> 01:03:39.140 your betters, 799 01:03:39.380 --> 01:03:40.580 then things 800 01:03:40.900 --> 01:03:43.940 would be so much better. But I take your point. 801 01:03:44.260 --> 01:03:45.780 It's sort of like crypto 802 01:03:45.860 --> 01:03:46.340 in 803 01:03:47.105 --> 01:03:49.585 super turbo speed up mode. 804 01:03:49.905 --> 01:03:50.465 Like, 805 01:03:50.865 --> 01:03:58.705 crypto was easy to pick on when the banks would give you lots of money to basically stomp on crypto and keep their competition out. 806 01:03:59.025 --> 01:03:59.984 But eventually, 807 01:04:00.464 --> 01:04:02.545 50,000,000 Americans owned crypto. 808 01:04:03.080 --> 01:04:06.200 Yeah. And in 2024, Coinbase and all these others 809 01:04:06.520 --> 01:04:08.440 put a bunch of money into politics, 810 01:04:08.680 --> 01:04:10.280 and lo and behold, 811 01:04:10.520 --> 01:04:13.560 every single candidate they backed won. 812 01:04:14.040 --> 01:04:19.895 And so if you haven't gotten the message yet, trying to stomp on 50,000,000 of your constituents 813 01:04:19.975 --> 01:04:22.055 might not be a winning strategy 814 01:04:22.055 --> 01:04:24.695 because at this point, they have enough money and power 815 01:04:24.935 --> 01:04:31.655 to affect their voice into the system. And so it'll be interesting to see if this is like a flash in the pan with '26, 816 01:04:32.060 --> 01:04:35.980 or if they're still going on this anti AI thing by '28 817 01:04:35.980 --> 01:04:39.100 when so many people are using it by that point. 818 01:04:39.660 --> 01:04:41.980 How do they think that's gonna be a winning strategy? 819 01:04:42.060 --> 01:04:49.345 You know, it's it's gotta be wrapped in protection and and all these other things. But anyway, we'll see what happens. I agree with you. 820 01:04:49.665 --> 01:04:52.705 Genie's out of the bottle. Yeah. Yeah. It's There's too no like, 821 01:04:53.185 --> 01:04:55.345 they tried to shove it in the bottle in '23, 822 01:04:55.585 --> 01:04:57.744 and they they might have been able to pull it off. 823 01:04:58.590 --> 01:05:02.670 At that point, the models weren't that powerful. Like, we are so far beyond 824 01:05:02.910 --> 01:05:20.645 where things were Yeah. Three. And looking to back two years from now, looking back at this video, we'd be like, you guys had no idea, like, how fast the RSI was gonna accelerate or 10 other things that come after that. Right? So yeah. Yeah. I wouldn't be surprised if two years from now we have, like, basically, 825 01:05:20.965 --> 01:05:56.480 like, brave new world, full on dead Internet theory when it comes to politics. Like, it'll be so invasive at that point and not yet like, I think the invasiveness will move a lot faster than the, like, proof of humanity, like like, solutions will happen. Those will eventually happen. I think eventually you will have a world where you have an Internet again where you're somehow able to confirm that everyone you're interacting with is a person, but I don't think that's two years from now. But I do think two years from now, it might be invasive enough that it just makes itself completely irrelevant to the campaigns and stuff, because all of it's just, like, presumed to be not real. 826 01:05:56.800 --> 01:06:01.680 It's all just noise. Yeah. Yeah. It's all noise. I guess maybe I'm underestimating, 827 01:06:01.680 --> 01:06:02.640 though, people's 828 01:06:03.840 --> 01:06:18.215 It's so hard for me Okay. I've I've been hanging out with you nerds so much that it's so hard for me to grasp now what the average person knows about all of this. Like, my my version of this is totally skewed, I think. Yeah. Perfect. 829 01:06:18.295 --> 01:06:20.775 Where there's one, there's more. Yeah. Yeah. Yeah. 830 01:06:22.160 --> 01:06:24.880 Confuse them all. Good stuff. 831 01:06:24.960 --> 01:06:32.640 Awesome, though. Awesome. Well, thank you guys. I appreciate it. As always, thanks for everyone who hung out with us and everyone listening to the recorded version. 832 01:06:34.175 --> 01:06:41.855 We said we were gonna try to make it, like, 1% better every single week. The big upgrade this week is I figured out how to make the title on x 833 01:06:41.935 --> 01:06:53.539 the actual title of the show, not just welcome to our livestream, which is what it kept saying for the last couple. So I figured out the setting, how to fix that, so huge improvement this week. 834 01:06:54.019 --> 01:06:55.298 But anyway, thank you guys.