1 00:00:00,880 --> 00:00:05,040 S1: Unsupervised Learning is a podcast about trends and ideas in cybersecurity, 2 00:00:05,080 --> 00:00:09,960 S1: national security, AI, technology and society, and how best to 3 00:00:10,000 --> 00:00:17,640 S1: upgrade ourselves to be ready for what's coming. All right, 4 00:00:17,640 --> 00:00:19,840 S1: in this video, I'm going to talk about the AI 5 00:00:19,880 --> 00:00:22,840 S1: ecosystem that I think everyone is actually building and moving 6 00:00:22,880 --> 00:00:26,280 S1: towards without even realizing that they're doing so. And I'm 7 00:00:26,280 --> 00:00:28,600 S1: going to break it down into four pieces, which are 8 00:00:29,200 --> 00:00:34,960 S1: assistance APIs, agents, and augmented reality. And I think once 9 00:00:34,960 --> 00:00:38,199 S1: you see this model, you're going to realize that the 10 00:00:38,200 --> 00:00:41,120 S1: news coming in from OpenAI and anthropic and all these 11 00:00:41,120 --> 00:00:45,480 S1: different companies, it's all moving in this direction toward this model. 12 00:00:46,080 --> 00:00:48,639 S1: And I think it's going to be really helpful for 13 00:00:48,640 --> 00:00:51,400 S1: you to just have that mental model of it. So 14 00:00:51,400 --> 00:00:54,120 S1: let's jump into it. So I actually broke this down 15 00:00:54,120 --> 00:00:57,280 S1: into a much longer explanation that you could see above, uh, 16 00:00:57,280 --> 00:01:00,720 S1: back in December of 2023. And I wrote a kind 17 00:01:00,720 --> 00:01:02,720 S1: of a little bit of a crappy book about it 18 00:01:02,720 --> 00:01:05,600 S1: in 2016. I really just wanted to capture the ideas, 19 00:01:05,600 --> 00:01:08,039 S1: which it was decent at doing that. It was called 20 00:01:08,040 --> 00:01:10,200 S1: The Real Internet of Things. You could put a link 21 00:01:10,200 --> 00:01:12,400 S1: to that in there. Don't really need to read the book. 22 00:01:12,400 --> 00:01:15,800 S1: This is actually much better. But, um, this stuff is 23 00:01:15,800 --> 00:01:18,360 S1: actually starting to happen now. And what I want to 24 00:01:18,360 --> 00:01:21,400 S1: do today is take the concepts from the book and 25 00:01:21,400 --> 00:01:25,440 S1: from that video above, and show how they're actually falling 26 00:01:25,440 --> 00:01:28,679 S1: into this model and how it's actually starting to happen today, 27 00:01:29,160 --> 00:01:31,600 S1: which we could see throughout the news from all these 28 00:01:31,600 --> 00:01:36,040 S1: different companies. So the first a here is assistant or 29 00:01:36,040 --> 00:01:38,800 S1: what I call a digital assistant. So a million different 30 00:01:38,800 --> 00:01:42,560 S1: companies are actually building this piece in various ways. Some 31 00:01:42,560 --> 00:01:46,160 S1: companies are building like digital companions or like smart assistants 32 00:01:46,160 --> 00:01:50,240 S1: or personal agents. But to me like this is all 33 00:01:50,240 --> 00:01:53,000 S1: actually kind of the same thing. It is basically a 34 00:01:53,040 --> 00:01:55,800 S1: piece of tech that is the most intimate to you 35 00:01:56,280 --> 00:01:59,200 S1: because it knows everything about you your preferences, your calendar, 36 00:01:59,200 --> 00:02:03,250 S1: your contacts like health information. Your finances is going to 37 00:02:03,250 --> 00:02:07,890 S1: be our best advocate, our best tutor, right? It's a 38 00:02:07,890 --> 00:02:10,850 S1: filter against like incoming stuff that you don't want to see. 39 00:02:10,889 --> 00:02:14,130 S1: And hopefully that filter and those filters are actually assigned 40 00:02:14,130 --> 00:02:16,930 S1: by you and not by someone else. But, um, yeah, 41 00:02:16,970 --> 00:02:19,930 S1: filtering out messages and emails and stuff you don't want. 42 00:02:19,970 --> 00:02:24,330 S1: It's basically figuring out exactly what you want and figuring 43 00:02:24,330 --> 00:02:26,690 S1: out how to make that happen all the time. Now, 44 00:02:26,730 --> 00:02:30,890 S1: if you're insecurity like me, you're probably thinking, well, that's crazy. 45 00:02:31,010 --> 00:02:33,170 S1: All this stuff on the screen here, like, knows everything 46 00:02:33,169 --> 00:02:38,370 S1: about you. History, trauma preferences, remembers everything, has multiple agents. 47 00:02:38,570 --> 00:02:42,970 S1: Like nobody's going to actually put that information into their Da. 48 00:02:43,090 --> 00:02:45,090 S1: But I think we already know that's not true. We 49 00:02:45,090 --> 00:02:47,650 S1: already know that people are doing this. That's why these 50 00:02:47,650 --> 00:02:51,930 S1: digital companion companies are doing so well already. This is 51 00:02:51,930 --> 00:02:55,530 S1: functionality that is just so powerful. Like it's just going 52 00:02:55,530 --> 00:02:59,169 S1: to happen, right? And this is kind of the centerpiece 53 00:02:59,169 --> 00:03:01,130 S1: of this whole model that I'm going to break down. 54 00:03:01,290 --> 00:03:04,170 S1: This is the first one. This is the assistant. So 55 00:03:04,169 --> 00:03:07,329 S1: let's call RDA Kai hours is going to be called 56 00:03:07,370 --> 00:03:13,329 S1: Kai okay. So the second A is APIs. So your 57 00:03:13,330 --> 00:03:16,929 S1: Da isn't that useful if it can't do stuff for 58 00:03:16,930 --> 00:03:21,250 S1: you right. So the way it'll do stuff is through APIs. Uh, 59 00:03:21,250 --> 00:03:24,410 S1: I didn't say in 2016 which of like the Da 60 00:03:24,410 --> 00:03:26,330 S1: or the APIs was going to come first because I 61 00:03:26,330 --> 00:03:29,730 S1: didn't know. And I still really kind of don't know, 62 00:03:29,730 --> 00:03:31,570 S1: but it seems like they're kind of happening at the 63 00:03:31,570 --> 00:03:35,530 S1: same time. Basically, your Da Chi is over here being 64 00:03:35,530 --> 00:03:38,690 S1: an agent for you, right? It's being like your advocate, 65 00:03:38,690 --> 00:03:42,130 S1: like we already talked about. It's it is an agent. 66 00:03:42,130 --> 00:03:44,690 S1: It has agents working for it. But ultimately it's like 67 00:03:44,690 --> 00:03:48,410 S1: one personality with a collection of agents behind it. It's 68 00:03:48,450 --> 00:03:51,770 S1: effectively kind of like one entity or one person, which 69 00:03:51,770 --> 00:03:54,050 S1: is why we give it a name. You know, we're 70 00:03:54,050 --> 00:03:56,650 S1: treating it like a person, like like a friend, right? 71 00:03:57,130 --> 00:04:02,420 S1: So ultimately it's encapsulated in one personality, in one sort 72 00:04:02,420 --> 00:04:06,660 S1: of entity. So Chi is constantly looking at your state 73 00:04:06,660 --> 00:04:09,180 S1: and trying to figure out how to make it better. 74 00:04:09,540 --> 00:04:13,340 S1: That's the core concept for the Da. That's what it's doing. 75 00:04:13,460 --> 00:04:16,419 S1: Are you hungry? Are you angry? Are you stressed out? 76 00:04:16,460 --> 00:04:19,300 S1: Do you have a meeting coming up? Right. That it 77 00:04:19,300 --> 00:04:21,700 S1: needs to help you prepare for. And all of this 78 00:04:21,700 --> 00:04:24,380 S1: is proactive. You haven't even asked it. Anything else yet. 79 00:04:24,420 --> 00:04:26,700 S1: So basically, these are going to do something I talked 80 00:04:26,700 --> 00:04:29,299 S1: about a couple of weeks ago, which is managing your 81 00:04:29,300 --> 00:04:35,300 S1: current state relative to your ideal state or your desired state. Right. 82 00:04:35,500 --> 00:04:39,659 S1: How could this current situation that I am in that 83 00:04:39,660 --> 00:04:42,700 S1: Chi is monitoring? How could that be better? So if 84 00:04:42,700 --> 00:04:45,860 S1: you're hungry, Chi will go find food. If if you're 85 00:04:45,860 --> 00:04:48,140 S1: worried they're going to find camera feeds to like see 86 00:04:48,140 --> 00:04:50,380 S1: around corners, like if you're worried about your security or 87 00:04:50,380 --> 00:04:53,780 S1: you're walking around or something. Or they'll listen to scanners 88 00:04:53,779 --> 00:04:56,420 S1: and see if there's, like police activity nearby. Or look 89 00:04:56,460 --> 00:05:00,060 S1: at crime stats for like the, the neighborhood that you're 90 00:05:00,060 --> 00:05:02,980 S1: in or whatever. Never. And that's actually what all these 91 00:05:02,980 --> 00:05:05,779 S1: APIs are over here that you see, right. These are 92 00:05:05,779 --> 00:05:10,140 S1: all the different things that Cai that your Da will 93 00:05:10,140 --> 00:05:17,539 S1: have access to. So APIs are essentially the representations of 94 00:05:17,860 --> 00:05:22,660 S1: people and companies and services. Basically everything becomes an API. 95 00:05:23,100 --> 00:05:25,700 S1: And we're already seeing this as we're going to talk 96 00:05:25,700 --> 00:05:29,820 S1: about we're already seeing this with MCC. This is actually 97 00:05:29,820 --> 00:05:32,500 S1: starting to happen. So so what I said in 2016 98 00:05:32,500 --> 00:05:38,659 S1: was basically everything gets an API. Every person objects, businesses 99 00:05:38,660 --> 00:05:42,380 S1: most importantly people and businesses, but also other objects. That's 100 00:05:42,380 --> 00:05:45,140 S1: why I called it the real Internet of Things. Basically, 101 00:05:45,140 --> 00:05:48,620 S1: everything gets an API and you have your Da navigating 102 00:05:48,620 --> 00:05:52,979 S1: those APIs for you on your behalf, right? So all 103 00:05:52,980 --> 00:05:55,860 S1: these APIs you see here, uh, except for there's going 104 00:05:55,860 --> 00:05:59,100 S1: to be millions of them, right? Eventually billions. But you 105 00:05:59,100 --> 00:06:02,260 S1: start off with thousands and then millions or whatever, but 106 00:06:02,260 --> 00:06:05,580 S1: every company will be an API. Every product will be 107 00:06:05,580 --> 00:06:10,059 S1: an API. People will be broadcasting APIs of ourselves, which 108 00:06:10,100 --> 00:06:12,620 S1: which I call demons is just a Greek word for 109 00:06:12,660 --> 00:06:16,660 S1: like soul, basically. And think of this like your own 110 00:06:16,660 --> 00:06:20,420 S1: personal like MCP server. And these are not designed to 111 00:06:20,420 --> 00:06:23,180 S1: be used by you or me. We can't read all 112 00:06:23,180 --> 00:06:27,300 S1: these APIs. You need help reading all these APIs. You 113 00:06:27,300 --> 00:06:29,420 S1: can't walk into a mall or walk into a city, 114 00:06:29,420 --> 00:06:32,220 S1: or walk down a road or whatever, and read all 115 00:06:32,220 --> 00:06:34,420 S1: the cars and the trees and all the people and 116 00:06:34,420 --> 00:06:37,020 S1: all the businesses. You can't do that. That's why you 117 00:06:37,020 --> 00:06:39,700 S1: need your Da to do that for you. Right. So 118 00:06:39,700 --> 00:06:43,700 S1: all these systems, all these APIs here and the agents 119 00:06:43,700 --> 00:06:46,339 S1: that sort of represent them, they're all designed to be 120 00:06:46,339 --> 00:06:51,380 S1: used by your Da, right? That's their purpose. So it's 121 00:06:51,380 --> 00:06:53,980 S1: like the interface to the world, like changes. It's no 122 00:06:53,980 --> 00:06:56,940 S1: longer about what we see with our eyes, like Google 123 00:06:56,980 --> 00:07:01,099 S1: like old Google now. It's about what do agents see? 124 00:07:01,140 --> 00:07:04,270 S1: What do DAC write? That's the world that starts to 125 00:07:04,270 --> 00:07:06,469 S1: matter a lot more. And a big part of this 126 00:07:06,470 --> 00:07:08,950 S1: is going to be a bunch of APIs that are 127 00:07:08,950 --> 00:07:14,990 S1: actually just concatenations or lists or directories of other APIs, 128 00:07:15,310 --> 00:07:17,870 S1: because one of the things that Midia has to do 129 00:07:17,870 --> 00:07:20,470 S1: is it has to ask, hey, what's the best restaurant 130 00:07:20,470 --> 00:07:24,910 S1: or whatever? And I've got a few here, right? Best food. Um, yeah. 131 00:07:24,950 --> 00:07:27,790 S1: Lookups or whatever. These are all just be, you know, 132 00:07:27,830 --> 00:07:30,870 S1: third party services that do nothing but crawl all the 133 00:07:30,870 --> 00:07:34,430 S1: other ones and rate them so that when Kai reaches 134 00:07:34,430 --> 00:07:36,590 S1: out and says, hey, I need to find the best food, 135 00:07:36,790 --> 00:07:40,750 S1: you know, within like three minutes, uh, close to this location, 136 00:07:40,750 --> 00:07:43,670 S1: but it can't have chicken in it or whatever. All 137 00:07:43,670 --> 00:07:46,550 S1: those different criterias, it can find the right one, right? 138 00:07:46,590 --> 00:07:48,830 S1: So there'll be a whole bunch of like lookup service 139 00:07:48,870 --> 00:07:52,030 S1: type of things like that. Okay. So that takes us 140 00:07:52,030 --> 00:07:54,310 S1: right into the third one which is agent. So we've 141 00:07:54,310 --> 00:07:56,430 S1: got a few agents here. And the way I like 142 00:07:56,430 --> 00:07:59,190 S1: to define an agent, there's lots of different definitions. I 143 00:07:59,190 --> 00:08:03,150 S1: think the agent should be super obvious, like from the definition, 144 00:08:03,190 --> 00:08:05,430 S1: like what it actually means and why it has value. 145 00:08:05,910 --> 00:08:08,790 S1: So I say it's an AI system component that autonomously 146 00:08:08,790 --> 00:08:11,990 S1: pursues a goal by taking multiple steps that previously would 147 00:08:12,070 --> 00:08:14,910 S1: have required a human. I think that is a really 148 00:08:14,910 --> 00:08:20,190 S1: good definition. Um, it's autonomous and it's taking a goal, 149 00:08:20,750 --> 00:08:24,230 S1: and it's pursuing that with multiple steps in a way 150 00:08:24,230 --> 00:08:27,270 S1: that only humans could do before. The part that makes 151 00:08:27,270 --> 00:08:30,390 S1: an agent different than automation, this is really important. This 152 00:08:30,390 --> 00:08:32,230 S1: is why I have it in the definition. The part 153 00:08:32,230 --> 00:08:35,950 S1: that makes it different is the fact that when a 154 00:08:35,950 --> 00:08:38,709 S1: human is trying to get something done, like say you're 155 00:08:38,710 --> 00:08:40,630 S1: an assistant for your boss or whatever, and you're trying 156 00:08:40,630 --> 00:08:43,270 S1: to get something done, like you call the first place. 157 00:08:43,270 --> 00:08:45,870 S1: They don't answer the phone, you call the first place. 158 00:08:45,870 --> 00:08:49,230 S1: The phone number doesn't work. Life is just broken, right? Like, 159 00:08:49,230 --> 00:08:52,030 S1: all these different steps are broken. Now, if you have automation, 160 00:08:52,030 --> 00:08:56,229 S1: automation is static, right? It's a whole bunch of if thens. 161 00:08:56,350 --> 00:08:59,510 S1: Agents aren't. If then they are. I have all these 162 00:08:59,510 --> 00:09:02,880 S1: tools available. I'm going to keep going. I'm going to keep, 163 00:09:02,920 --> 00:09:07,720 S1: you know, exhausting my resources, trying different things to try 164 00:09:07,720 --> 00:09:11,120 S1: to get it done right. I will, you know, maybe, 165 00:09:11,120 --> 00:09:13,360 S1: maybe none of the things work. So I'm going to 166 00:09:13,360 --> 00:09:16,600 S1: do more research to find another, uh, API that I 167 00:09:16,600 --> 00:09:20,560 S1: could use or another service to find this person. Pizza. 168 00:09:20,600 --> 00:09:24,000 S1: Sarah wants pizza. I'm going to get Sarah pizza, and 169 00:09:24,000 --> 00:09:26,600 S1: she's going to do you know, the Da is going 170 00:09:26,600 --> 00:09:28,480 S1: to do multiple things to make sure that she gets 171 00:09:28,480 --> 00:09:31,600 S1: that pizza. That's the difference between automation and agent. So 172 00:09:31,600 --> 00:09:35,160 S1: that's our definition here. And we see that our Da 173 00:09:35,240 --> 00:09:39,080 S1: actually has the use of multiple agents. These agents might 174 00:09:39,080 --> 00:09:43,120 S1: be like researchers. They might be like security bots uh, 175 00:09:43,120 --> 00:09:46,079 S1: to lock down your infrastructure. That could be whatever. But 176 00:09:46,080 --> 00:09:50,040 S1: they all kind of work for Kai, right? Kai is, like, 177 00:09:50,040 --> 00:09:52,120 S1: the centerpiece here. And this is going to be a 178 00:09:52,120 --> 00:09:55,160 S1: theme we're going to see throughout. Agents all over the place, 179 00:09:55,160 --> 00:09:58,240 S1: including inside of companies like we have over here with like, 180 00:09:58,280 --> 00:10:02,000 S1: United or whatever. It's the concept of you're talking to 181 00:10:02,040 --> 00:10:04,360 S1: one agent, but behind it, it has a whole bunch 182 00:10:04,360 --> 00:10:07,199 S1: of other agents. So you give it the goal and 183 00:10:07,200 --> 00:10:09,880 S1: it breaks that down into sub goals and gives that 184 00:10:09,880 --> 00:10:12,320 S1: to the smaller agents, which are then doing the other 185 00:10:12,320 --> 00:10:16,319 S1: things like building a marketing campaign, uh, hacking a website, 186 00:10:16,320 --> 00:10:20,320 S1: doing whatever it is. Right. So that's really the concept 187 00:10:20,320 --> 00:10:24,200 S1: of agents. And Google actually just released a thing called 188 00:10:24,200 --> 00:10:26,400 S1: agent to agent, I think was the name of it. 189 00:10:26,400 --> 00:10:28,440 S1: And what it does is it makes it so that 190 00:10:28,440 --> 00:10:31,319 S1: all these different agents here, they could talk to each 191 00:10:31,320 --> 00:10:34,880 S1: other with a common protocol, which is very similar to MCP, 192 00:10:35,559 --> 00:10:39,000 S1: where it's a common protocol for creating APIs for an 193 00:10:39,000 --> 00:10:43,440 S1: application or, you know, a company or whatever. So we're 194 00:10:43,440 --> 00:10:47,040 S1: starting to see the glue, the protocol glue that's going 195 00:10:47,080 --> 00:10:49,600 S1: to make all this stuff possible with this agent agent 196 00:10:49,600 --> 00:10:53,760 S1: protocol in MCP and stuff like that. So the final 197 00:10:53,760 --> 00:10:57,120 S1: piece here, so we've got we've got the Da, we've 198 00:10:57,120 --> 00:11:00,480 S1: got the assistant, we've got the APIs, we've got the agents. 199 00:11:00,880 --> 00:11:04,920 S1: So the final piece or the final A here in 200 00:11:04,920 --> 00:11:09,400 S1: the four A's is AR or augmented reality. And this 201 00:11:09,400 --> 00:11:11,200 S1: is the one you might be thinking is fringe or 202 00:11:11,200 --> 00:11:13,640 S1: it's like ten years away, but it's actually much closer. 203 00:11:14,120 --> 00:11:17,160 S1: Meta and Apple are currently fighting over this now. Tim 204 00:11:17,160 --> 00:11:19,800 S1: Cook just recently said, look, I'm not going to let 205 00:11:19,800 --> 00:11:22,760 S1: anyone beat us here. They want to beat meta at 206 00:11:22,760 --> 00:11:27,240 S1: this game. Meta already has really good glasses. Um, they're 207 00:11:27,240 --> 00:11:29,760 S1: not actually displaying anything inside of it, but you can 208 00:11:29,760 --> 00:11:33,720 S1: see out of it that takes pictures like it's pretty decent. 209 00:11:33,880 --> 00:11:36,719 S1: And obviously it's not big and heavy and super expensive 210 00:11:36,720 --> 00:11:39,760 S1: like the Vision Pro. So that is a battle that 211 00:11:39,760 --> 00:11:44,560 S1: is happening right now. So we are all eventually I 212 00:11:44,559 --> 00:11:46,120 S1: don't know how long this is going to take. It's 213 00:11:46,120 --> 00:11:50,230 S1: hard to make like specific predictions. Right. So 2 to 214 00:11:50,230 --> 00:11:54,080 S1: 5 years, who knows. It's going to be something relatively soon. 215 00:11:54,280 --> 00:11:57,160 S1: Meta or Apple or maybe someone comes out of the 216 00:11:57,160 --> 00:12:00,680 S1: dark and just kind of crushes this. Who knows? But 217 00:12:00,679 --> 00:12:02,570 S1: the point is, we're all going to have these AR 218 00:12:02,650 --> 00:12:06,170 S1: glasses eventually, like contact lenses or something better than that. 219 00:12:06,170 --> 00:12:09,010 S1: But it's going to start off with glasses. And here's 220 00:12:09,010 --> 00:12:11,530 S1: the trick. This is how the whole ecosystem starts to 221 00:12:11,570 --> 00:12:14,370 S1: come together. Our Das are going to be showing us 222 00:12:14,370 --> 00:12:19,849 S1: stuff that is time based, and that is contextually relevant 223 00:12:19,850 --> 00:12:22,770 S1: to whatever we're doing at that moment. Remember, our Da 224 00:12:22,770 --> 00:12:25,850 S1: is trying to optimize everything according to our goals. It's 225 00:12:25,850 --> 00:12:28,730 S1: trying to get to our desired state from our current state. 226 00:12:28,890 --> 00:12:31,530 S1: So if we're walking down a street like this here, 227 00:12:31,770 --> 00:12:33,809 S1: we're walking down a street and like we think it's 228 00:12:33,809 --> 00:12:37,530 S1: kind of dangerous. Yeah, it's going to present this interface here, 229 00:12:37,530 --> 00:12:40,809 S1: which I've got over here coming from this, this demon 230 00:12:41,290 --> 00:12:44,530 S1: called bastion, which is really it's just a company. It's 231 00:12:44,530 --> 00:12:47,250 S1: a company. It's called bastion. And they have feeds called 232 00:12:47,250 --> 00:12:52,250 S1: get feed, poll cameras, poll microphones, query personal mics, get 233 00:12:52,250 --> 00:12:57,250 S1: local CCTV. Right. So maybe it could pull all the different, um, 234 00:12:57,570 --> 00:13:00,730 S1: people who are broadcasting their feed because people are going 235 00:13:00,730 --> 00:13:04,689 S1: to be wearing cameras as well. This is coming soon. Uh, basically. 236 00:13:04,850 --> 00:13:09,250 S1: Camera ahead of you. Camera behind. Behind you. And maybe 237 00:13:09,250 --> 00:13:14,090 S1: you sell your camera feed to Bastian. People will do this. 238 00:13:14,090 --> 00:13:16,809 S1: Trust me. It's going to happen. People are going to 239 00:13:16,809 --> 00:13:20,290 S1: sell their camera feeds to Bastian. Right. It's not going 240 00:13:20,290 --> 00:13:22,250 S1: to be for private stuff. Like it's going to get 241 00:13:22,250 --> 00:13:24,329 S1: turned off when you go home. Stuff like that. You 242 00:13:24,330 --> 00:13:26,970 S1: shouldn't trust that. You should also like cover the camera 243 00:13:26,970 --> 00:13:29,010 S1: or whatever. But the point is, if you're sitting in 244 00:13:29,010 --> 00:13:33,089 S1: Starbucks or whatever and say a fight altercation happens or 245 00:13:33,090 --> 00:13:36,610 S1: something like that, Bastian will be able to show that 246 00:13:36,610 --> 00:13:39,090 S1: to the police or show that to somebody else who's 247 00:13:39,090 --> 00:13:43,089 S1: worried about it. So my Da, while I'm walking down 248 00:13:43,090 --> 00:13:48,450 S1: the street, right? I'm walking down the street here, it's like, oh, 249 00:13:48,570 --> 00:13:52,450 S1: I this neighborhood feels unsafe. That's what I'm saying. I'm 250 00:13:52,450 --> 00:13:55,250 S1: saying this neighborhood feels unsafe or it hears me say 251 00:13:55,250 --> 00:13:57,850 S1: something in a conversation where I'm just like, I don't know, 252 00:13:57,850 --> 00:14:00,929 S1: it's kind of sketchy. I'm a little worried, right? I 253 00:14:00,929 --> 00:14:04,020 S1: say anything like that or even before I say it, 254 00:14:04,620 --> 00:14:08,540 S1: the da Chi goes out and looks at one of 255 00:14:08,540 --> 00:14:12,060 S1: these services to find the best security interface, the best 256 00:14:12,059 --> 00:14:16,540 S1: one for parsing feeds, uh, giving, you know, real time 257 00:14:16,580 --> 00:14:19,780 S1: HUD data and stuff like that. So it gets one back. 258 00:14:19,780 --> 00:14:23,540 S1: It's called bastion. So it starts pulling stuff like that, 259 00:14:23,940 --> 00:14:27,860 S1: it gets back the content. It then goes to another interface, 260 00:14:28,220 --> 00:14:32,100 S1: which is a whole separate company, which is the UI 261 00:14:32,460 --> 00:14:35,580 S1: for this content. Okay. You see these red. This is 262 00:14:35,580 --> 00:14:39,020 S1: a great example. But like let's say there's data here right. 263 00:14:39,060 --> 00:14:42,020 S1: Let's say there's like, oh, how many people are around. Um, 264 00:14:42,060 --> 00:14:46,340 S1: is anyone wearing a weapon. Let's do like gait analysis 265 00:14:46,340 --> 00:14:48,740 S1: to see if they're leaning because they're carrying a gun 266 00:14:48,740 --> 00:14:51,940 S1: or something like that. Right. All this stuff, all these 267 00:14:51,940 --> 00:14:57,620 S1: different individual pieces, different companies are better at, okay, somebody 268 00:14:57,620 --> 00:15:01,660 S1: is better at making this red, cool looking interface. Somebody 269 00:15:01,660 --> 00:15:05,300 S1: is better at doing voice analysis of microphones coming from 270 00:15:05,300 --> 00:15:10,220 S1: all around you. Somebody is better at doing camera analysis of, like, 271 00:15:10,220 --> 00:15:13,580 S1: all the different dangers on the street. All of those 272 00:15:13,820 --> 00:15:18,220 S1: are these right here. This is what every company becomes. 273 00:15:18,220 --> 00:15:21,780 S1: It becomes a specialized thing at doing a thing better 274 00:15:21,780 --> 00:15:28,220 S1: than everyone else, all judged by these indexing services, these 275 00:15:28,220 --> 00:15:32,900 S1: rating services which are marketing to your Da. It is 276 00:15:32,900 --> 00:15:36,700 S1: marketing to Chi. So when I'm walking down the street 277 00:15:36,700 --> 00:15:39,300 S1: and I say, hey, show me what's going on around 278 00:15:39,300 --> 00:15:41,820 S1: me or something like that, or I don't even have 279 00:15:41,820 --> 00:15:44,220 S1: to say it. It just knows I'm freaking out. Why? 280 00:15:44,380 --> 00:15:47,780 S1: Because Chi can see my heart rate. Chi can see 281 00:15:47,780 --> 00:15:50,620 S1: that we're in a place I've never been. Um, somebody 282 00:15:50,620 --> 00:15:52,820 S1: is laying on the street with, like, a needle sticking 283 00:15:52,820 --> 00:15:56,020 S1: out of their arm. Chi figures out this is kind 284 00:15:56,060 --> 00:15:58,300 S1: of seedy. It's a little bit dangerous. I don't like it. 285 00:15:58,620 --> 00:16:03,990 S1: And obviously, my principal Daniel, doesn't like it either. Therefore, 286 00:16:04,350 --> 00:16:09,550 S1: broom goes and searches, finds Bastian, finds a UI. The 287 00:16:09,550 --> 00:16:12,869 S1: best UI. Okay, the best UI is called UI Wizard. 288 00:16:13,310 --> 00:16:16,550 S1: Not too creative, but whatever it's called UI Wizard. UI 289 00:16:16,590 --> 00:16:19,990 S1: Wizard pops up. That's this red interface, and it starts 290 00:16:19,990 --> 00:16:23,510 S1: filling in data where the data come from. From Bastian, 291 00:16:23,510 --> 00:16:27,270 S1: it came from the Bastian service. Where does that go? 292 00:16:27,470 --> 00:16:32,150 S1: This interface is in these glasses, which is on my face. 293 00:16:32,150 --> 00:16:35,710 S1: So now watch this. We've got other scenarios here. Okay? 294 00:16:36,190 --> 00:16:39,670 S1: You start browsing for headphones. Your Da does this. It 295 00:16:39,670 --> 00:16:43,550 S1: uses these services and it gives you back a response. 296 00:16:43,550 --> 00:16:46,470 S1: So I'm looking for headphones. It goes and investigates all 297 00:16:46,470 --> 00:16:49,950 S1: these different things. You mentioned your friend that you're getting hungry. 298 00:16:50,110 --> 00:16:55,230 S1: It goes and looks researches all these different best food places, 299 00:16:55,510 --> 00:16:58,950 S1: parses all 713 different places in like a second and 300 00:16:58,950 --> 00:17:02,310 S1: a half. Gets back the results. Hey, you haven't had 301 00:17:02,430 --> 00:17:04,670 S1: Thai in a while. There's a great little place with 302 00:17:04,670 --> 00:17:07,629 S1: super high ratings if you take a right into blocks. 303 00:17:07,990 --> 00:17:11,949 S1: I can call you in an order if you want. Right. 304 00:17:11,990 --> 00:17:16,510 S1: This is the model. These four A's. This is the model. 305 00:17:16,510 --> 00:17:20,350 S1: This is where this is all heading. This is the direction, right? 306 00:17:20,550 --> 00:17:25,389 S1: So I'm telling you. I'm telling you this. This is 307 00:17:25,390 --> 00:17:31,030 S1: what's happening. It is absolutely exciting to see this starting 308 00:17:31,070 --> 00:17:34,430 S1: to unfold. Right. There's a million different companies working on 309 00:17:34,430 --> 00:17:39,070 S1: this part. There's multiple companies working on the AR glasses part. 310 00:17:39,109 --> 00:17:42,470 S1: Everything is turning into an API already. This is MCP 311 00:17:42,510 --> 00:17:46,109 S1: over here. This is the unification of it. And then 312 00:17:46,109 --> 00:17:48,590 S1: of course, over here we have what's happening on the 313 00:17:48,590 --> 00:17:52,350 S1: corporate side where agents are basically going to be doing 314 00:17:52,350 --> 00:17:55,310 S1: a whole bunch of work. You'll have a humans kind 315 00:17:55,310 --> 00:17:57,909 S1: of in charge of things. The leaders and the in 316 00:17:57,950 --> 00:18:01,470 S1: the extreme, SMEs will be human for quite some time, 317 00:18:01,470 --> 00:18:03,750 S1: I think. I mean, it's going to be pretty hard 318 00:18:03,750 --> 00:18:06,030 S1: to automate everyone away, but a lot of the work 319 00:18:06,030 --> 00:18:08,710 S1: that was getting done is going to be getting done 320 00:18:08,710 --> 00:18:12,470 S1: by agents and teams of agents. So that's agents inside 321 00:18:12,470 --> 00:18:16,630 S1: the corporate place. But as far as the consumer side, 322 00:18:16,630 --> 00:18:20,150 S1: as far as the stuff that you're seeing, like in the, 323 00:18:20,950 --> 00:18:24,709 S1: you know, the OpenAI and anthropic and most of the 324 00:18:24,710 --> 00:18:27,670 S1: stuff that they're talking about is mostly about the consumer 325 00:18:27,670 --> 00:18:33,590 S1: and stuff like that. This is it. This is the structure. Okay. 326 00:18:33,630 --> 00:18:36,590 S1: So another example of this is like let's say you're 327 00:18:36,590 --> 00:18:39,670 S1: in like a live conversation and you're having a conversation 328 00:18:39,670 --> 00:18:43,550 S1: with somebody and it's like somebody you've never met before 329 00:18:43,670 --> 00:18:45,830 S1: and you're considering whether to go into business with them 330 00:18:45,830 --> 00:18:48,110 S1: or whatever, and they're making a whole bunch of claims. 331 00:18:48,109 --> 00:18:50,510 S1: They're like, oh, yeah, I used to work with so-and-so 332 00:18:50,510 --> 00:18:52,390 S1: and blah blah, blah. And actually I helped him start 333 00:18:52,390 --> 00:18:56,510 S1: his business and, uh, yeah. Do you know Sarah? Yeah. Sarah. 334 00:18:56,670 --> 00:18:58,750 S1: You know, I went to college with her and blah, blah, blah. 335 00:18:59,590 --> 00:19:03,440 S1: So again, you're wearing glasses. Everyone's wearing glasses. The person 336 00:19:03,440 --> 00:19:07,359 S1: you're talking talking to is actually wearing these glasses as well, 337 00:19:07,840 --> 00:19:11,120 S1: and you're having this conversation. But in the whole time 338 00:19:11,119 --> 00:19:15,040 S1: you're wondering like, is this actually true? Is all the 339 00:19:15,040 --> 00:19:18,440 S1: stuff that this person claimed happened or the people that 340 00:19:18,440 --> 00:19:21,200 S1: they claim they know or whatever? Is this all true? Right. 341 00:19:21,320 --> 00:19:25,560 S1: So what will happen is you'll have like something going 342 00:19:25,560 --> 00:19:28,880 S1: off to voice analysis. This depends how many things you're 343 00:19:28,880 --> 00:19:31,320 S1: subscribed to. It depends how far along we are in 344 00:19:31,320 --> 00:19:34,960 S1: this cycle. You know what all your Da can actually do. 345 00:19:34,960 --> 00:19:37,800 S1: But this is this is all sort of being built 346 00:19:37,800 --> 00:19:42,200 S1: right now. So like, if there's tension in their voice, like, uh, 347 00:19:42,200 --> 00:19:44,840 S1: analyzing the claims that they're making, doing research on it, 348 00:19:44,880 --> 00:19:47,720 S1: did they actually go to college? Did did Sarah and 349 00:19:47,720 --> 00:19:50,000 S1: this person actually were they in college at the same time? 350 00:19:50,000 --> 00:19:52,119 S1: That should be on LinkedIn. Let's go find that out. 351 00:19:52,720 --> 00:19:54,840 S1: So if you're waiting on a delivery and this goes 352 00:19:54,840 --> 00:19:56,720 S1: back to the R side, if you're waiting on the 353 00:19:56,720 --> 00:20:00,480 S1: delivery you will see a timer timing down right. Just 354 00:20:00,480 --> 00:20:03,760 S1: like you have on on your phone. Now that will 355 00:20:03,760 --> 00:20:05,760 S1: be in your interface so you don't have to pick 356 00:20:05,760 --> 00:20:08,480 S1: up a phone. The whole point is with AR glasses 357 00:20:08,680 --> 00:20:11,840 S1: is to have to do much less with an actual 358 00:20:11,840 --> 00:20:14,480 S1: physical device that you pick up and have to interact with. 359 00:20:14,640 --> 00:20:17,560 S1: It'll be a lot more. It's visual here, and you're 360 00:20:17,560 --> 00:20:21,000 S1: just talking to your Da, and your Da is doing 361 00:20:21,000 --> 00:20:24,440 S1: most of the work yourself using all these different services, right. 362 00:20:25,000 --> 00:20:29,359 S1: So where is the data coming from? Right. How is 363 00:20:29,359 --> 00:20:32,520 S1: it being displayed in the glasses. That's exactly what we 364 00:20:32,520 --> 00:20:36,160 S1: talked about earlier. The data is coming from these services 365 00:20:36,840 --> 00:20:40,360 S1: moving through a UI being displayed in the glasses. Right. 366 00:20:41,000 --> 00:20:43,199 S1: So in the case of like trying to determine if 367 00:20:43,200 --> 00:20:46,560 S1: someone is lying, right. Let's say there's like a lie 368 00:20:46,560 --> 00:20:50,040 S1: o meter interface inside of this little UI here inside 369 00:20:50,040 --> 00:20:54,439 S1: your AR. Well, somebody is providing that lie o meter interface, right? 370 00:20:54,480 --> 00:20:57,800 S1: There is some one of these companies is actually the 371 00:20:57,800 --> 00:21:02,240 S1: voice analysis, uh, providing back data. And that's just like 372 00:21:02,330 --> 00:21:07,050 S1: Jason coming back. That is like, um, saying, like the 373 00:21:07,090 --> 00:21:10,170 S1: chances of them being lying about this particular thing, like, 374 00:21:10,210 --> 00:21:14,290 S1: according to voice analysis or like, according to the research 375 00:21:14,290 --> 00:21:17,850 S1: that was done. Right. All these things can be combined together. 376 00:21:17,890 --> 00:21:21,330 S1: Like that stuff could actually just be returned. Roar back 377 00:21:21,330 --> 00:21:24,770 S1: to Chi, and Chi could look at all that and 378 00:21:24,770 --> 00:21:29,370 S1: send another feed into the UI to update that Leo meter, right? 379 00:21:29,490 --> 00:21:32,730 S1: There's so many options here because because these are APIs, 380 00:21:32,770 --> 00:21:35,170 S1: this is just JSON or whatever. The protocol is going 381 00:21:35,210 --> 00:21:38,370 S1: to be flowing back and forth, right? That's the power 382 00:21:38,369 --> 00:21:41,850 S1: of this entire thing. So let's zoom out again and 383 00:21:41,850 --> 00:21:44,730 S1: let's just take a look at this entire thing. Okay. 384 00:21:44,970 --> 00:21:48,090 S1: Again just as a review here. Look at the four A's. 385 00:21:48,250 --> 00:21:52,010 S1: Chi knows what you want at all times. Constantly calling APIs, 386 00:21:52,010 --> 00:21:55,689 S1: making requests, summarizing things, creating reports for you, researching the 387 00:21:55,690 --> 00:21:58,810 S1: best options, anticipating your needs throughout the day and week 388 00:21:58,810 --> 00:22:02,530 S1: and month or year or whatever. Adjusting your ah, interface. 389 00:22:02,609 --> 00:22:06,810 S1: Constantly switching. Look, you're in the house. You maybe you 390 00:22:06,810 --> 00:22:09,570 S1: see your books. Maybe you see how hungry everyone is. 391 00:22:09,570 --> 00:22:13,010 S1: Like this. This interface is constantly changing. Chi is changing 392 00:22:13,010 --> 00:22:15,810 S1: it for you using these different APIs. Right. And the 393 00:22:15,810 --> 00:22:18,689 S1: interface that Chi is using will be coming from multiple 394 00:22:18,690 --> 00:22:23,210 S1: companies as well. Maybe, maybe Chi's generic interface actually gets 395 00:22:23,250 --> 00:22:27,169 S1: good enough. Maybe one of these companies is a generic 396 00:22:27,170 --> 00:22:31,929 S1: interface creator, so you don't actually have to use individual ones. Right. 397 00:22:31,970 --> 00:22:34,770 S1: And Kyle just switched using that one. So when you're 398 00:22:34,770 --> 00:22:37,369 S1: looking at books you're inside of a library. It's a 399 00:22:37,369 --> 00:22:39,490 S1: different HUD. You know, walking down the street it's a 400 00:22:39,490 --> 00:22:42,650 S1: different HUD talking to someone different HUD. Right. So if 401 00:22:42,690 --> 00:22:45,010 S1: you take a step back and you look at this 402 00:22:45,010 --> 00:22:48,169 S1: interface here, right, think of all the news that you've 403 00:22:48,170 --> 00:22:51,570 S1: heard from the last few years in. I think about 404 00:22:51,570 --> 00:22:54,170 S1: the latest news from OpenAI. They just added like long 405 00:22:54,250 --> 00:22:59,690 S1: term memory, where it's going to remember all your previous conversations. Right. Um, 406 00:22:59,730 --> 00:23:03,690 S1: think about digital companion companies, uh, digital helper apps that 407 00:23:03,690 --> 00:23:06,130 S1: will like go and do tasks for you. Siri and 408 00:23:06,130 --> 00:23:09,889 S1: Gemini on the mobile device. And of course, you all 409 00:23:09,890 --> 00:23:12,330 S1: heard the stuff about Siri. Right? Where they're trying to 410 00:23:12,369 --> 00:23:14,890 S1: make this thing a better assistant. They're trying to give 411 00:23:14,890 --> 00:23:18,010 S1: it access to more and more data about you. Right. 412 00:23:18,330 --> 00:23:21,010 S1: And they're trying to do it in a secure way, obviously. 413 00:23:21,250 --> 00:23:23,930 S1: And Gemini is competing in that space as well. Samsung 414 00:23:23,930 --> 00:23:27,730 S1: has their own version. Right. This is all like heading 415 00:23:27,730 --> 00:23:30,410 S1: in that direction of like the Da. All of them 416 00:23:30,410 --> 00:23:34,290 S1: are heading in this direction of the unified right. That's 417 00:23:34,290 --> 00:23:37,090 S1: the easiest way to see this. So now think about 418 00:23:37,090 --> 00:23:41,050 S1: all the news around MCP and APIs and how they'll 419 00:23:41,050 --> 00:23:43,090 S1: be able to talk to each other with the agent 420 00:23:43,130 --> 00:23:46,730 S1: agent protocol and all of those different things. Basically, that's 421 00:23:46,730 --> 00:23:49,610 S1: what we're talking about with everything that's an API. And 422 00:23:49,609 --> 00:23:51,530 S1: you already hear all the talk about agents. That's all 423 00:23:51,530 --> 00:23:54,210 S1: everyone talks about now. And then you think about the 424 00:23:54,210 --> 00:23:58,650 S1: news about Meta and Apple fighting about AR glasses. And 425 00:23:58,650 --> 00:24:03,300 S1: that's this piece over here. Right. So everyone is moving 426 00:24:03,540 --> 00:24:07,260 S1: towards these four A's. This is it. This is the 427 00:24:07,260 --> 00:24:11,540 S1: ecosystem that we're creating. It's absolutely insane. And it's starting 428 00:24:11,540 --> 00:24:14,820 S1: to fill in. And like again, now that you've seen it, 429 00:24:14,820 --> 00:24:17,740 S1: I think you're going to realize that all the new 430 00:24:17,740 --> 00:24:20,740 S1: news is just filling in pieces of this and eventually 431 00:24:20,740 --> 00:24:23,939 S1: getting us to this. And it's very sort of cyberpunk 432 00:24:23,980 --> 00:24:27,620 S1: y and very future oriented. But so many of these 433 00:24:27,619 --> 00:24:32,140 S1: pieces are already exist, like this is just HTTP going 434 00:24:32,140 --> 00:24:35,180 S1: back and forth like these protocols are not too difficult. 435 00:24:35,180 --> 00:24:37,860 S1: The only difficult part right now is just like the 436 00:24:37,859 --> 00:24:41,460 S1: hardware challenge of like the AR stuff is really difficult, 437 00:24:41,780 --> 00:24:44,340 S1: and that's probably the thing that's going to take the longest. 438 00:24:44,540 --> 00:24:48,060 S1: But agents are getting better. Like the AI itself is 439 00:24:48,060 --> 00:24:52,420 S1: getting smarter and smarter. Like context size is a huge 440 00:24:52,420 --> 00:24:56,140 S1: one and memory is a huge one that is dramatically 441 00:24:56,140 --> 00:24:59,700 S1: getting better. Like for one now has a million tokens. Um, 442 00:25:00,460 --> 00:25:04,900 S1: a couple of the Google ones now have 4 million tokens, 443 00:25:04,900 --> 00:25:07,500 S1: I think, for context. So all of this is starting 444 00:25:07,500 --> 00:25:09,820 S1: to come together. All right. So anyway, that's what I 445 00:25:09,820 --> 00:25:13,300 S1: wanted to share. And I think this is a really 446 00:25:13,300 --> 00:25:16,140 S1: powerful way of just interpreting what's coming in in terms 447 00:25:16,140 --> 00:25:19,619 S1: of the news and putting that into a context and 448 00:25:19,660 --> 00:25:21,820 S1: showing how it fits into a model. I think it's 449 00:25:21,820 --> 00:25:24,379 S1: just useful to be able to parse things in that 450 00:25:24,380 --> 00:25:27,020 S1: way and make sense of it. Keep in mind, we 451 00:25:27,020 --> 00:25:30,220 S1: don't actually know the details of any of these pieces, right? 452 00:25:30,500 --> 00:25:33,340 S1: Is MCP going to win? Who knows? That new agent 453 00:25:33,340 --> 00:25:35,460 S1: to agent protocol that Google came out with? Is that 454 00:25:35,460 --> 00:25:38,419 S1: actually going to be useful? Who knows. Like it might 455 00:25:38,420 --> 00:25:41,820 S1: not be useful at all. Like nobody could adopt it. 456 00:25:41,820 --> 00:25:45,540 S1: Like who's going to win with the glasses, right. Is 457 00:25:45,540 --> 00:25:46,980 S1: it is it going to be Apple. Is it going 458 00:25:46,980 --> 00:25:49,740 S1: to be meta? Is it going to be someone completely separate? 459 00:25:49,780 --> 00:25:53,740 S1: I have no idea. Right. These things are not super predictable. 460 00:25:54,260 --> 00:25:59,180 S1: I believe that my 2016 like outlook and predictions from 461 00:25:59,180 --> 00:26:02,260 S1: back then, it's human based. It's based on the fact 462 00:26:02,260 --> 00:26:04,949 S1: that I know what I want my agent to be 463 00:26:04,950 --> 00:26:08,350 S1: able to do. And you know, I've been in tech 464 00:26:08,350 --> 00:26:11,270 S1: for so long that I understand all these protocols and stuff, 465 00:26:11,270 --> 00:26:13,750 S1: so I just assumed this is the way it was going. 466 00:26:13,830 --> 00:26:15,990 S1: And it turns out to be happening. But you can't 467 00:26:15,990 --> 00:26:19,230 S1: predict the companies. You can't predict the timelines, you can't 468 00:26:19,230 --> 00:26:22,030 S1: predict any of this. So that's what makes it so exciting. 469 00:26:22,310 --> 00:26:24,510 S1: I just hope that this model helps you sort of 470 00:26:24,550 --> 00:26:27,510 S1: make sense of the news as it comes in, and 471 00:26:27,510 --> 00:26:30,950 S1: hopefully moves it in this direction towards this model, and 472 00:26:30,950 --> 00:26:33,470 S1: just makes it easier for you to parse the news 473 00:26:33,630 --> 00:26:36,710 S1: and make sense of it. So, uh, do me a 474 00:26:36,710 --> 00:26:39,429 S1: favor and subscribe and I'll see you in the next one. 475 00:26:40,230 --> 00:26:44,190 S1: Unsupervised learning is produced on Hindenburg Pro using an Sm7 476 00:26:44,230 --> 00:26:47,870 S1: B microphone. A video version of the podcast is available 477 00:26:47,869 --> 00:26:51,510 S1: on the Unsupervised Learning YouTube channel, and the text version 478 00:26:51,510 --> 00:26:56,630 S1: with full links and notes is available at Daniel Miessler newsletter. 479 00:26:57,230 --> 00:26:58,229 S1: We'll see you next time.