1 00:00:00,040 --> 00:00:03,279 Speaker 1: Hello, and welcome back to Thrilling Threads. It is February seventh, 2 00:00:03,399 --> 00:00:05,799 twenty twenty six, and I want you to just take 3 00:00:05,839 --> 00:00:09,720 a second right now wherever you are, and look at 4 00:00:09,720 --> 00:00:12,000 your phone, Look at the screen. Now, look at the 5 00:00:12,080 --> 00:00:14,160 room you're in. Maybe you're on a commute, maybe you're 6 00:00:14,199 --> 00:00:16,600 in a hun office. For the last I don't know, 7 00:00:16,600 --> 00:00:19,879 three or four years we've been talking about artificial intelligence. 8 00:00:19,960 --> 00:00:23,440 Is this thing that is coming like a weather forecast, 9 00:00:23,679 --> 00:00:26,440 you know, predicting a hurricane a week out. We were 10 00:00:26,440 --> 00:00:29,039 always saying when AI does this or if AI does that. 11 00:00:29,199 --> 00:00:32,039 Speaker 2: It was always speculative. It was always about the potential exactly. 12 00:00:32,359 --> 00:00:35,240 Speaker 1: But looking at the date today, I mean we're in 13 00:00:35,320 --> 00:00:37,880 early twenty twenty six. We aren't looking at a crystal 14 00:00:37,880 --> 00:00:41,159 ball anymore. This isn't some distant future. We are living 15 00:00:41,159 --> 00:00:45,640 in the shift. The storm made landfall. The landscape is flooded, 16 00:00:45,840 --> 00:00:47,520 and we're all just learning how to swim in it. Now. 17 00:00:47,719 --> 00:00:49,039 Speaker 2: That is such a good way to put it. That's 18 00:00:49,079 --> 00:00:51,079 the feeling, isn't it. You know? I was looking at 19 00:00:51,119 --> 00:00:53,799 the primary research report we're using for today's show, this 20 00:00:54,039 --> 00:00:57,759 massive analysis of the twenty twenty six landscape and The 21 00:00:57,840 --> 00:01:01,119 thing that really struck me wasn't the tech itself. It 22 00:01:01,159 --> 00:01:04,239 was the normalization, the fact that it's just everywhere. It's 23 00:01:04,560 --> 00:01:07,359 the fact that this year AI is projected to touch 24 00:01:07,799 --> 00:01:09,359 eighty percent of all work. 25 00:01:09,519 --> 00:01:12,319 Speaker 1: Eighty percent. Just pause on that for a second. That's 26 00:01:12,359 --> 00:01:13,760 not just you know, tech bros. 27 00:01:13,799 --> 00:01:15,640 Speaker 2: In Silicon Valley, No, not anymore. 28 00:01:15,719 --> 00:01:18,239 Speaker 1: It's not just coders. That's the barista, the barrister, the 29 00:01:18,280 --> 00:01:20,760 truck driver, the middle school teacher. 30 00:01:21,239 --> 00:01:25,359 Speaker 2: It's ubiquitous and the global market. I mean, it's pushing 31 00:01:25,400 --> 00:01:28,480 past four hundred billion dollars. That figure alone backs it 32 00:01:28,480 --> 00:01:28,879 all up. 33 00:01:28,920 --> 00:01:29,239 Speaker 1: Wow. 34 00:01:29,439 --> 00:01:31,400 Speaker 2: But we need to be really careful with that number. 35 00:01:31,439 --> 00:01:34,840 It's not just money being spent on like chatbot subscriptions. 36 00:01:35,319 --> 00:01:39,079 It represents a fundamental restructuring of the economy. We're seeing 37 00:01:39,200 --> 00:01:44,840 huge capital expenditure moving from hiring people to renting compute. 38 00:01:44,359 --> 00:01:45,359 Speaker 1: From people to compute. 39 00:01:45,519 --> 00:01:47,799 Speaker 2: That is the new normal everyone was talking about. But 40 00:01:47,840 --> 00:01:51,239 it looks, it looks very different than what everyone predicted 41 00:01:51,319 --> 00:01:53,200 back in say, twenty twenty three. 42 00:01:53,280 --> 00:01:55,799 Speaker 1: It really does. Yeah, and that is exactly why we're 43 00:01:55,799 --> 00:01:58,439 here today. We have this comprehensive report. It's going to 44 00:01:58,480 --> 00:02:01,079 be our roadmap for this whole compnversation, and we're going 45 00:02:01,159 --> 00:02:04,560 to pull apart the nine specific trends that are defining 46 00:02:04,599 --> 00:02:06,680 twenty twenty six. And I want to be clear to 47 00:02:06,680 --> 00:02:09,280 you listening, we aren't just going to list off some 48 00:02:09,319 --> 00:02:11,800 cool gadgets. Yeah, we're going to talk about how you 49 00:02:11,879 --> 00:02:14,560 talk to machines, how machines are actually you know, entering 50 00:02:14,560 --> 00:02:18,400 the physical world, and then the really heavy stuff, how 51 00:02:18,520 --> 00:02:22,800 nations are starting to fight over the brains of these machines. 52 00:02:22,840 --> 00:02:26,080 Speaker 2: The geopolitical angle is usually the one people ignore, but frankly, 53 00:02:26,800 --> 00:02:28,479 it's the one that's going to affect your life the 54 00:02:28,479 --> 00:02:29,360 most in the long run. 55 00:02:29,400 --> 00:02:31,719 Speaker 1: Oh for sure, we'll get there, but I have to 56 00:02:31,800 --> 00:02:34,800 tease the number one trend go on. Honestly, when I 57 00:02:34,800 --> 00:02:37,639 saw it on the list, I just laughed, because three 58 00:02:37,719 --> 00:02:41,439 years ago, everybody thought the future belonged to whoever had 59 00:02:41,479 --> 00:02:43,879 the biggest, the smartest AI model. 60 00:02:43,639 --> 00:02:46,800 Speaker 2: The god model, right, the race for the super brain, right. 61 00:02:47,240 --> 00:02:51,120 Speaker 1: And it turns out that might have been a total red. 62 00:02:50,919 --> 00:02:54,599 Speaker 2: Herring, a huge misdirection. But we have to earn that reveal. 63 00:02:54,719 --> 00:02:55,360 Speaker 1: We do we do. 64 00:02:55,439 --> 00:02:58,199 Speaker 2: We have to understand the interface first, how we even 65 00:02:58,240 --> 00:02:59,319 interact with this stuff. 66 00:02:59,400 --> 00:03:02,520 Speaker 1: So let's start there Segment one, the evolution of interaction. 67 00:03:02,599 --> 00:03:05,439 We're covering trends nine, seven, and four here. This is 68 00:03:05,479 --> 00:03:09,680 all about the user experience of well of being alive 69 00:03:09,680 --> 00:03:11,719 in twenty twenty six, and I want to start with 70 00:03:11,719 --> 00:03:15,400 trend number four, Context replaces prompt. 71 00:03:15,039 --> 00:03:17,000 Speaker 2: Engineering RIP prompt engineering. 72 00:03:17,120 --> 00:03:19,479 Speaker 1: Seriously, yeah, I mean, do you remember twenty twenty three 73 00:03:20,280 --> 00:03:21,840 or even most of twenty twenty four? 74 00:03:21,960 --> 00:03:22,759 Speaker 2: Oh? Vividly. 75 00:03:23,039 --> 00:03:25,039 Speaker 1: It felt like you had to be a wizard casting 76 00:03:25,080 --> 00:03:27,960 a spell. Right, You'd open a textbox and type this long, 77 00:03:28,039 --> 00:03:31,280 convoluted thing. Oh yeah, act as a senior marketing executive 78 00:03:31,319 --> 00:03:34,199 with twenty years of experience in SAW sales and write 79 00:03:34,280 --> 00:03:37,039 me a polite but firm email to a client who 80 00:03:37,080 --> 00:03:40,400 is laid on payment using the structure of the Challenger 81 00:03:40,439 --> 00:03:41,000 Celle method. 82 00:03:41,199 --> 00:03:44,360 Speaker 2: And if you missed a comma or phrased it slightly wrong. 83 00:03:44,199 --> 00:03:47,479 Speaker 1: Garbage output total garbage, or it would just hallucinate something 84 00:03:47,520 --> 00:03:48,400 completely useless. 85 00:03:48,479 --> 00:03:50,800 Speaker 2: It was brittle. That's the perfect word for it. The 86 00:03:50,840 --> 00:03:54,439 system was just brittle. You were trying to coax intelligence 87 00:03:54,439 --> 00:03:57,520 out of a system that had almost no short term memory. 88 00:03:57,599 --> 00:03:59,840 Speaker 1: But people were selling courses on this. I remember seeing 89 00:03:59,840 --> 00:04:02,840 them everywhere by my PDF of the fifty magic prompts. 90 00:04:02,919 --> 00:04:05,759 Speaker 2: It was a whole cottage industry, but trend four. I 91 00:04:05,759 --> 00:04:08,879 mean it basically says that industry is dead, gone, and 92 00:04:08,919 --> 00:04:12,479 it died because of a brute force technical leap. The 93 00:04:12,520 --> 00:04:13,439 context window. 94 00:04:13,680 --> 00:04:16,360 Speaker 1: Okay, let's unpack context window, because I think a lot 95 00:04:16,360 --> 00:04:18,959 of people hear that term in their eyes just glaze over. Sure, 96 00:04:19,319 --> 00:04:20,639 what are we actually talking about. 97 00:04:20,839 --> 00:04:23,879 Speaker 2: Think of the context window as the working memory of 98 00:04:23,920 --> 00:04:27,519 the AI. Okay, not what it knows from its training data, 99 00:04:27,680 --> 00:04:30,839 like who George Washington was, but what it knows about 100 00:04:30,920 --> 00:04:34,000 you and this specific conversation right now. 101 00:04:33,800 --> 00:04:36,399 Speaker 1: The immediate short term memory exactly. 102 00:04:36,560 --> 00:04:38,560 Speaker 2: And in the old days, and again we're talking like 103 00:04:38,639 --> 00:04:39,600 eighteen months. 104 00:04:39,399 --> 00:04:41,439 Speaker 1: Ago, which feels like a decade it does. 105 00:04:41,519 --> 00:04:46,000 Speaker 2: Back then, that window was about eight thousand tokens. 106 00:04:45,800 --> 00:04:49,399 Speaker 1: Which translates to what in human terms a few pages 107 00:04:49,639 --> 00:04:52,600 roughly a long essay, maybe a short chapter of a book. 108 00:04:52,839 --> 00:04:55,079 Speaker 2: So if you fed it a fifty page contract, by 109 00:04:55,079 --> 00:04:57,720 the time it got to page fifty, it had technically 110 00:04:57,800 --> 00:05:00,639 forgotten what was on page one. Oh, dropped out of 111 00:05:00,639 --> 00:05:01,399 the memory. 112 00:05:01,120 --> 00:05:04,759 Speaker 1: Buffer, which is completely useless for any real complex work. 113 00:05:04,920 --> 00:05:07,560 Speaker 2: Totally useless, right, But look at the data and the report. 114 00:05:07,879 --> 00:05:11,759 In just eighteen months, that capacity exploded from eight thousand 115 00:05:11,839 --> 00:05:15,360 tokens to over two hundred thousand tokens. Wow, that is 116 00:05:15,399 --> 00:05:17,759 a twenty five hundred percent increase. 117 00:05:17,439 --> 00:05:19,560 Speaker 1: Two hundred thousand tokens. So we aren't talking about an 118 00:05:19,639 --> 00:05:20,319 essay anymore. 119 00:05:20,480 --> 00:05:23,439 Speaker 2: We are not. We are talking about the entire Harry 120 00:05:23,439 --> 00:05:26,360 Potter series, the whole thing, the whole thing. We're talking 121 00:05:26,399 --> 00:05:30,959 about your company's entire chaotic slack history from the last quarter, 122 00:05:31,720 --> 00:05:35,079 every single line of code in a massive software project. 123 00:05:35,680 --> 00:05:37,920 You can now dump all of that into the prompt. 124 00:05:37,959 --> 00:05:39,800 Speaker 1: But hold on a second. Yeah, I have to play 125 00:05:39,839 --> 00:05:42,279 Devil's Advocate here. Yeah, just because I can dump a 126 00:05:42,360 --> 00:05:44,920 library's worth of text into the chat box, Yeah, does 127 00:05:44,920 --> 00:05:47,920 that mean the AI actually understands it or finds the 128 00:05:47,920 --> 00:05:50,040 important parts? I remember reading about the lost in the 129 00:05:50,040 --> 00:05:53,399 middle phenomenon. Yeah, with the AI would remember the start 130 00:05:53,439 --> 00:05:55,079 and the end of a document, but it would kind 131 00:05:55,079 --> 00:05:57,600 of just hallucinate the stuff in the middle. Yeah, is 132 00:05:57,600 --> 00:05:58,519 that still a problem. 133 00:05:58,680 --> 00:06:01,639 Speaker 2: That's a really sharp question, And in twenty twenty four 134 00:06:01,680 --> 00:06:03,839 that was a huge, huge problem. You'd give it one 135 00:06:03,920 --> 00:06:07,079 hundred page legal brief and it would completely miss the 136 00:06:07,120 --> 00:06:11,199 critical loophole on page forty right, But the reasoning capabilities 137 00:06:11,199 --> 00:06:13,600 have really started to catch up with the memory size. 138 00:06:13,959 --> 00:06:17,120 The attention mechanisms that's the part of the AI that 139 00:06:17,279 --> 00:06:21,040 decides what's important, have gotten so much better at needle 140 00:06:21,040 --> 00:06:22,199 in a haystack. 141 00:06:21,800 --> 00:06:24,639 Speaker 1: Retrieval, so it can actually find the key detail in 142 00:06:24,680 --> 00:06:25,600 a see of text. 143 00:06:25,879 --> 00:06:28,639 Speaker 2: It can, so now you don't need to craft a 144 00:06:28,680 --> 00:06:31,959 perfect prompt, you just need to provide the perfect context. 145 00:06:32,079 --> 00:06:35,279 Speaker 1: Ah okay, so the skill shifts. It's not about knowing 146 00:06:35,360 --> 00:06:37,920 how to talk to the robot anymore. It's about knowing 147 00:06:38,040 --> 00:06:40,160 what information to give the robot precisely. 148 00:06:40,199 --> 00:06:43,639 Speaker 2: It's become context dumping. You don't say act as a marketer. 149 00:06:43,879 --> 00:06:47,120 You upload your last fifty marketing campaigns and you just say, 150 00:06:47,600 --> 00:06:48,639 write one like these. 151 00:06:49,040 --> 00:06:52,839 Speaker 1: So the AI derives the style from the data, not 152 00:06:53,040 --> 00:06:54,360 from your description. 153 00:06:54,000 --> 00:06:56,360 Speaker 2: Of the style exactly. It's so much more effective. 154 00:06:56,480 --> 00:06:59,199 Speaker 1: That's a huge relief. I always hated trying to describe 155 00:06:59,240 --> 00:07:01,279 my own writing style. It's like trying to describe your 156 00:07:01,279 --> 00:07:01,759 own face. 157 00:07:01,920 --> 00:07:05,680 Speaker 2: It's impossible, right, And the data totally supports this shift. 158 00:07:06,160 --> 00:07:10,360 The report says context aware, so searches based on uploaded 159 00:07:10,360 --> 00:07:13,680 documents increased by eight hundred and ninety percent in twenty 160 00:07:13,720 --> 00:07:18,160 twenty five alone. Wow, people just stopped asking generic questions 161 00:07:18,199 --> 00:07:21,240 and started asking really specific questions based on their own 162 00:07:21,360 --> 00:07:22,360 uploaded data. 163 00:07:22,399 --> 00:07:25,839 Speaker 1: But okay, this leads to a massive friction point, right, Yeah. 164 00:07:25,879 --> 00:07:29,720 If I have to upload my life context, all my emails, 165 00:07:30,040 --> 00:07:34,680 my project history, my personal preferences, every single time I 166 00:07:34,720 --> 00:07:38,759 open a new tab, that is incredibly annoying. It is 167 00:07:38,839 --> 00:07:41,000 I don't want to reintroduce myself to my own assistant 168 00:07:41,040 --> 00:07:41,879 every single morning. 169 00:07:41,959 --> 00:07:44,160 Speaker 2: And that is the perfect segue into trend number nine, 170 00:07:44,560 --> 00:07:48,720 which is AI becomes persistent and personal. This is the 171 00:07:48,800 --> 00:07:50,759 end of what I call the groundhog Day effect. 172 00:07:50,879 --> 00:07:54,160 Speaker 1: Groundhog Day, great movie, absolutely terrible user experience. 173 00:07:54,279 --> 00:07:57,279 Speaker 2: It's the worst, and it was the default for so long. 174 00:07:57,319 --> 00:08:00,399 You close the tab, the AI basically gets a lobotomy. Yep, 175 00:08:00,519 --> 00:08:02,839 it forgets who you are, what you were talking about, everything. 176 00:08:03,160 --> 00:08:05,360 But the report shows that twenty twenty six is the 177 00:08:05,439 --> 00:08:08,839 year this finally stops. It found that eighty two percent 178 00:08:08,920 --> 00:08:13,079 of users either implicitly or explicitly want an AI that 179 00:08:13,120 --> 00:08:14,560 remembers them across sessions. 180 00:08:14,600 --> 00:08:17,959 Speaker 1: Eighty two percent. That number actually seems high to me. Yeah, 181 00:08:18,079 --> 00:08:21,439 I mean, aren't people terrified of the privacy implications? If 182 00:08:21,439 --> 00:08:24,199 the AI remembers that I was stressed about my mortgage 183 00:08:24,560 --> 00:08:27,759 last Tuesday, and then today I ask it about planning 184 00:08:27,759 --> 00:08:31,199 a vacation, isn't that getting a little too close to 185 00:08:31,439 --> 00:08:32,600 big brother territory? 186 00:08:32,720 --> 00:08:35,120 Speaker 2: It absolutely is, And I don't want to gloss over 187 00:08:35,200 --> 00:08:38,159 the privacy nightmare this could become. It's a huge concern, 188 00:08:38,879 --> 00:08:44,039 but human behavior almost always follows the path of least resistance. 189 00:08:44,399 --> 00:08:48,200 Congeniens almost always beats privacy concerns in the long run. 190 00:08:48,240 --> 00:08:50,159 Speaker 1: That's a depressing thought, but you're probably right. 191 00:08:50,240 --> 00:08:52,519 Speaker 2: Well, think about the utility of it. If the AI 192 00:08:52,639 --> 00:08:55,200 remembers your mortgage stress from last week and then you 193 00:08:55,240 --> 00:08:58,639 ask it for a vacation plan, it might proactively suggest 194 00:08:58,679 --> 00:09:01,799 a more budget friendly itinerate without you even having to say, hey, 195 00:09:01,799 --> 00:09:02,759 by the way, I'm broke. 196 00:09:02,960 --> 00:09:04,080 Speaker 1: It connects the dots for you. 197 00:09:04,200 --> 00:09:07,559 Speaker 2: It connects the dots, and the report notes that personalized 198 00:09:07,559 --> 00:09:11,080 assistance the ones with this long term memory show three 199 00:09:11,240 --> 00:09:15,480 point five times higher engagement than the generic forgetful. 200 00:09:14,960 --> 00:09:17,799 Speaker 1: Ones three and a half times. That's a huge difference. 201 00:09:17,879 --> 00:09:18,440 Speaker 2: It's mats. 202 00:09:18,519 --> 00:09:20,960 Speaker 1: I guess it's because it stops feeling like a tool, 203 00:09:21,200 --> 00:09:24,159 like a calculator, and starts feeling more like a colleague. 204 00:09:24,200 --> 00:09:27,000 Speaker 2: That's the perfect word. A colleague remembers that you hate 205 00:09:27,000 --> 00:09:29,600 early morning meetings. A calculator doesn't care. 206 00:09:29,840 --> 00:09:30,039 Speaker 1: Right. 207 00:09:30,279 --> 00:09:34,799 Speaker 2: Persistent AI means the software builds a model of you 208 00:09:34,919 --> 00:09:38,879 over time. It's kind of it's creating a theory of 209 00:09:39,039 --> 00:09:40,840 mind about its user. 210 00:09:40,679 --> 00:09:44,360 Speaker 1: Which is both incredibly cool and incredibly creepy at the 211 00:09:44,399 --> 00:09:47,360 same time. It is. But okay, let's assume I'm on 212 00:09:47,480 --> 00:09:49,720 board with the AI knowing my soul. How am I 213 00:09:49,720 --> 00:09:52,000 interacting with it? Yeah, because for the last few years, 214 00:09:52,000 --> 00:09:54,519 my thumbs have been doing all the work. I'm just 215 00:09:55,279 --> 00:09:57,559 typing constantly. 216 00:09:57,039 --> 00:09:59,480 Speaker 2: And typing is a bottleneck. It's actually a very low 217 00:09:59,519 --> 00:10:02,720 bandwidth way to communicate. You can speak so much faster 218 00:10:02,799 --> 00:10:04,440 than you can type, and you can show something with 219 00:10:04,480 --> 00:10:06,639 the camera faster than you can describe it in words. 220 00:10:07,080 --> 00:10:09,799 This is trend number seven Site and Sound. 221 00:10:10,159 --> 00:10:12,840 Speaker 1: The stats here are pretty interesting. Fifty five percent of 222 00:10:12,879 --> 00:10:15,919 smartphone users now use voice AI on a weekly. 223 00:10:15,639 --> 00:10:18,320 Speaker 2: Basis, and that's up from what maybe ten percent who 224 00:10:18,399 --> 00:10:20,519 are just using Siri for the rather a few years ago. 225 00:10:20,840 --> 00:10:23,559 Speaker 1: Totally, but this isn't just sirius set a timer for 226 00:10:23,600 --> 00:10:27,039 ten minutes. This is full, long, back and forth conversation. 227 00:10:27,320 --> 00:10:30,559 Speaker 2: It's latency. That was the big unlock in twenty twenty six, 228 00:10:30,720 --> 00:10:35,320 the lag between you speaking and the AI answering. It's gone. 229 00:10:35,480 --> 00:10:37,720 It's instantaneous. It feels like a real phone call. 230 00:10:37,879 --> 00:10:38,200 Speaker 1: Okay. 231 00:10:38,480 --> 00:10:40,960 Speaker 2: But the bigger shift, the one I think is actually 232 00:10:41,039 --> 00:10:42,399 more transformative, is. 233 00:10:42,440 --> 00:10:45,679 Speaker 1: Vision video understanding. The report calls it that. 234 00:10:45,960 --> 00:10:49,639 Speaker 2: Yes, queries for video understanding surged by six hundred and 235 00:10:49,679 --> 00:10:51,879 seventy percent in twenty twenty five. 236 00:10:52,120 --> 00:10:55,360 Speaker 1: Give me a real world example of that, because you know, 237 00:10:55,480 --> 00:10:57,360 I can see using it to identify a bird in 238 00:10:57,399 --> 00:11:00,799 my backyard. But six hundred and seventy percent growth that 239 00:11:00,919 --> 00:11:03,080 implies people using this for actual work. 240 00:11:03,200 --> 00:11:05,960 Speaker 2: Okay. Sure, Imagine you're a junior field technician for a 241 00:11:06,000 --> 00:11:09,080 big telecom company. You're standing in front of this massive, 242 00:11:09,279 --> 00:11:11,240 messy server rack you've never seen. 243 00:11:11,039 --> 00:11:13,279 Speaker 1: Before, okay, a mess of wires. 244 00:11:13,000 --> 00:11:14,679 Speaker 2: A total mess. In the old days, you'd have to 245 00:11:14,679 --> 00:11:16,960 find the right manual, try to read the tiny schematic, 246 00:11:17,000 --> 00:11:18,759 maybe call a senior tech for help, and. 247 00:11:18,720 --> 00:11:21,039 Speaker 1: Hope the senior tech actually picks up the phone. Right. 248 00:11:21,120 --> 00:11:23,879 Speaker 2: Good luck with that, ye, But now you just point 249 00:11:23,919 --> 00:11:26,759 your phone's camera at the rack. The AI sees all 250 00:11:26,759 --> 00:11:28,519 the wires in real time. You say which one of 251 00:11:28,559 --> 00:11:31,679 these is the secondary uplink, and the AI just highlights 252 00:11:31,679 --> 00:11:34,279 it on your screen in ar and says the blue cable, 253 00:11:34,440 --> 00:11:35,399 third from the left. 254 00:11:35,480 --> 00:11:38,399 Speaker 1: So it's not just seeing an image, it's analyzing context 255 00:11:38,399 --> 00:11:39,440 and providing instruction. 256 00:11:39,720 --> 00:11:43,480 Speaker 2: It's understanding the physical world. Humans don't experience the world 257 00:11:43,559 --> 00:11:47,039 in text. We experience it in photons and sound waves. 258 00:11:47,320 --> 00:11:51,399 Trends seven is all about AI finally matching human sensory input. 259 00:11:52,039 --> 00:11:57,679 It makes the AI feel present, present in your physical. 260 00:11:57,200 --> 00:11:59,720 Speaker 1: Reality, which you know, honestly that sets off some alarm 261 00:11:59,759 --> 00:12:01,840 bills for me. Oh so well, if the AI is 262 00:12:01,840 --> 00:12:04,639 present in my physical reality through a camera, the logical 263 00:12:04,720 --> 00:12:07,759 next step is for to actually touch the physical reality 264 00:12:07,960 --> 00:12:08,600 interact with it. 265 00:12:08,679 --> 00:12:10,639 Speaker 2: You're anticipating our next segment perfectly. 266 00:12:10,799 --> 00:12:13,360 Speaker 1: I am. Let's move to segment two, the physical world 267 00:12:13,399 --> 00:12:16,120 and the economy, because we aren't just talking about software anymore, 268 00:12:16,159 --> 00:12:17,799 are we We're talking about hardware. 269 00:12:18,200 --> 00:12:21,320 Speaker 2: Trend number six is humanoid robots, and this is the 270 00:12:21,320 --> 00:12:23,279 one where everyone usually rolls their eyes. 271 00:12:23,639 --> 00:12:25,480 Speaker 1: I am rolling my eyes right now. I've seen the 272 00:12:25,519 --> 00:12:27,879 demos I saw the robots walking like they had a 273 00:12:27,919 --> 00:12:32,000 full diaper. I saw them folding laundry at like one 274 00:12:32,120 --> 00:12:35,399 tenth the speed of a human. Is this actually a 275 00:12:35,440 --> 00:12:37,799 trend for twenty twenty six or is this just hype 276 00:12:37,840 --> 00:12:38,919 for stock prices. 277 00:12:39,399 --> 00:12:41,200 Speaker 2: Look, if you asked me that in twenty twenty four, 278 00:12:41,200 --> 00:12:43,639 I'd have said it was ninety percent hype. Yeah, But 279 00:12:43,720 --> 00:12:46,159 looking at the twenty twenty six data, it's very real. 280 00:12:46,559 --> 00:12:49,679 The market is projected to hit thirty eight billion dollars 281 00:12:49,720 --> 00:12:52,600 by twenty thirty. But that's not the telling stat What 282 00:12:52,759 --> 00:12:55,360 is the competition? In twenty twenty three, there were maybe 283 00:12:55,559 --> 00:12:58,399 what seven major companies in the space. Now there are 284 00:12:58,440 --> 00:12:59,120 twenty three. 285 00:12:59,039 --> 00:13:01,879 Speaker 1: Twenty three, including all the big players, right. 286 00:13:01,960 --> 00:13:04,480 Speaker 2: Tesla is the loudest one. Of course, They plan to 287 00:13:04,519 --> 00:13:07,320 deploy thousands of their Optimist robots in their own factories 288 00:13:07,360 --> 00:13:08,600 by the end of twenty twenty six. 289 00:13:08,679 --> 00:13:12,200 Speaker 1: Thousands. Okay, So let's talk about so wet here. If 290 00:13:12,200 --> 00:13:14,159 I'm a factory worker, should I be terrified? 291 00:13:14,360 --> 00:13:16,399 Speaker 2: I'd say you should be paying attention, for sure. But 292 00:13:16,559 --> 00:13:19,360 the use case right now isn't skilled labor. Okay, it's 293 00:13:19,399 --> 00:13:21,360 not replacing the guy who knows how to well to 294 00:13:21,360 --> 00:13:25,200 a complex joint for memory. It's boring practicality. It's moving 295 00:13:25,200 --> 00:13:27,440 a box from palad A to palette B for eight 296 00:13:27,480 --> 00:13:28,600 hours straight without a break. 297 00:13:28,799 --> 00:13:32,480 Speaker 1: So the backbreaking stuff, the repetitive motion, the. 298 00:13:32,480 --> 00:13:35,480 Speaker 2: Stuff that causes injuries, the stuff that has incredibly high 299 00:13:35,519 --> 00:13:38,200 turnover because nobody wants to do it. The trend is 300 00:13:38,279 --> 00:13:40,919 that AI is finally getting hands, but there is a 301 00:13:40,960 --> 00:13:45,039 massive economic barrier here. Robots are expensive. The compute needed 302 00:13:45,039 --> 00:13:47,759 to run them is expensive, which brings us to the 303 00:13:49,519 --> 00:13:52,360 let's say, the ugly side of the economy. How do 304 00:13:52,399 --> 00:13:52,960 we pay for. 305 00:13:52,919 --> 00:13:55,879 Speaker 1: All of this? Friend number five advertising. 306 00:13:55,399 --> 00:13:57,039 Speaker 2: The monetization reality check. 307 00:13:57,200 --> 00:13:59,279 Speaker 1: This one hurts a little. I think for a while. 308 00:13:59,320 --> 00:14:01,600 We all live in this nice little bubble where AI 309 00:14:01,840 --> 00:14:03,759 was this pure helpful tool. 310 00:14:03,919 --> 00:14:04,799 Speaker 2: We have a fun bubble. 311 00:14:04,919 --> 00:14:07,639 Speaker 1: It was, But the source material says the AI advertising 312 00:14:07,679 --> 00:14:10,200 sector is projected to hit one hundred and seven billion 313 00:14:10,240 --> 00:14:11,639 dollars by twenty twenty eight. 314 00:14:11,759 --> 00:14:14,279 Speaker 2: There's no such thing as a free lunch. We are 315 00:14:14,320 --> 00:14:17,320 burning through just massive amounts of energy and GPU time 316 00:14:17,399 --> 00:14:19,919 to run these huge context windows we love so much. 317 00:14:20,039 --> 00:14:24,080 The big companies open AI, Google anthropic. They need revenue. 318 00:14:24,080 --> 00:14:24,960 They have to pay the bills. 319 00:14:25,080 --> 00:14:27,039 Speaker 1: So what does an AIAD even look like is like 320 00:14:27,080 --> 00:14:28,879 a banner ad at the top of my chat. 321 00:14:28,960 --> 00:14:33,000 Speaker 2: That's the benign version, the safe version, But the dangerous version, 322 00:14:33,080 --> 00:14:35,399 the one the whole industry is grappling with right now, 323 00:14:36,000 --> 00:14:37,840 is recommendation as ad. 324 00:14:38,080 --> 00:14:39,000 Speaker 1: Okay, explain that. 325 00:14:39,200 --> 00:14:42,399 Speaker 2: You ask your persistent AI, the one that knows you, Hey, 326 00:14:42,600 --> 00:14:44,480 I need to book a flight to London and a 327 00:14:44,519 --> 00:14:48,120 hotel for three nights. I want something modern and cool, right. 328 00:14:48,480 --> 00:14:51,120 The AI comes back and says, okay, great, I've booked 329 00:14:51,120 --> 00:14:53,600 you on Virgin Atlantic and put you in the Marriott. 330 00:14:53,919 --> 00:14:57,960 Speaker 1: And my question is, did it choose the Marriott because 331 00:14:57,960 --> 00:15:01,039 it's really the best modern hotel for me, okay? Or 332 00:15:01,039 --> 00:15:04,000 did it choose the marriout because Marriott paid for that 333 00:15:04,360 --> 00:15:06,039 top recommendation slot. 334 00:15:05,879 --> 00:15:08,919 Speaker 2: Exactly, And unlike a Google search where it's as sponsored 335 00:15:08,960 --> 00:15:10,200 in a tiny, tiny. 336 00:15:09,960 --> 00:15:11,799 Speaker 1: Text at the top, which no one reads. 337 00:15:11,679 --> 00:15:15,600 Speaker 2: Right, conversational AI completely blurs that line. It feels like 338 00:15:15,639 --> 00:15:18,080 advice from a trusted friend. And if a friend recommends 339 00:15:18,120 --> 00:15:20,399 a restaurant, you trust them, Yeah, But if you find 340 00:15:20,399 --> 00:15:22,799 out your friend recommended that restaurant because the restaurant paid 341 00:15:22,799 --> 00:15:26,000 them twenty dollars to do it. They're not your friend anymore. 342 00:15:26,039 --> 00:15:26,639 They're a shill. 343 00:15:27,000 --> 00:15:30,480 Speaker 1: And that completely destroys the trust. The whole persistent personal 344 00:15:30,519 --> 00:15:34,240 system thing we just talked about. That whole relationship relies 345 00:15:34,279 --> 00:15:36,879 on me trusting that the AI is working for me. Yes, 346 00:15:37,200 --> 00:15:40,240 If it's actually working for the highest bidder, the entire 347 00:15:40,320 --> 00:15:42,879 relationship breaks. The value is gone. 348 00:15:43,000 --> 00:15:46,600 Speaker 2: It's an incredibly delicate balance. The report says fifty eight 349 00:15:46,600 --> 00:15:50,240 percent of users would accept ads to keep their tools free. Sure, 350 00:15:50,519 --> 00:15:52,720 we've basically been trained by the Internet for the last 351 00:15:52,720 --> 00:15:56,600 twenty five years to trade our attention for free software, right, 352 00:15:56,639 --> 00:15:58,600 But the intimacy of AI, as you said, it makes 353 00:15:58,639 --> 00:16:01,120 that betrayal sting a whole lot more. 354 00:16:01,240 --> 00:16:03,519 Speaker 1: The intimacy of AI. That is a phrase I am 355 00:16:03,559 --> 00:16:06,120 definitely going to be thinking about later tonight. But I 356 00:16:06,159 --> 00:16:08,120 want to shift gears. We've talked about the user, we've 357 00:16:08,120 --> 00:16:10,279 talked about the robots and the ads. Now I want 358 00:16:10,320 --> 00:16:11,440 to look under the hood. 359 00:16:11,399 --> 00:16:12,080 Speaker 2: The engine room. 360 00:16:12,159 --> 00:16:16,080 Speaker 1: The engine room segment three. This covers trends three, two, 361 00:16:16,200 --> 00:16:19,399 and one, and this is where the entire narrative really 362 00:16:19,440 --> 00:16:22,799 flips for me. Let's start with trend three. The technical 363 00:16:22,840 --> 00:16:24,840 divide disappears. 364 00:16:24,360 --> 00:16:26,799 Speaker 2: The democratization of code. This one is huge. 365 00:16:26,960 --> 00:16:29,720 Speaker 1: For thirty years, if you wanted to build software, you 366 00:16:29,759 --> 00:16:32,480 had to speak the language of the machine. You needed 367 00:16:32,480 --> 00:16:37,000 to know C plus Python, JavaScript. If you didn't, you 368 00:16:37,080 --> 00:16:38,240 were just a consumer, or. 369 00:16:38,200 --> 00:16:41,440 Speaker 2: You were the idea guy. And nobody likes the idea guy. 370 00:16:41,840 --> 00:16:44,960 Speaker 1: H the worst. I have an amazing ad idea, I 371 00:16:45,000 --> 00:16:47,919 just need a developer to build it for me for equity. 372 00:16:47,679 --> 00:16:50,960 Speaker 2: The classic trap. Yeah, but just look at twenty twenty five. 373 00:16:51,080 --> 00:16:54,000 The report says forty three percent of all new applications 374 00:16:54,000 --> 00:16:57,360 were built using AI assisted, no code or low code 375 00:16:57,440 --> 00:17:00,480 tools forty three percent. And here's the real kicker, fifty 376 00:17:00,519 --> 00:17:04,039 two percent of so called citizen developers. These are people 377 00:17:04,079 --> 00:17:08,359 with zero formal code and training successfully shipped functional apps. 378 00:17:08,559 --> 00:17:10,839 Speaker 1: Shipped is the key word there. They didn't just build 379 00:17:10,839 --> 00:17:11,720 a prototype, they. 380 00:17:11,680 --> 00:17:15,400 Speaker 2: Launched, They launched. This is the single biggest leveler in 381 00:17:15,440 --> 00:17:19,160 the history of Silicon Valley. Innovation is no longer gated 382 00:17:19,200 --> 00:17:22,039 by your ability to write syntax. It's gated by your 383 00:17:22,079 --> 00:17:24,359 imagination and your understanding of a workflow. 384 00:17:24,519 --> 00:17:26,680 Speaker 1: But I have to be the cynic again. If my 385 00:17:26,680 --> 00:17:30,039 accountant is building apps and my hr rep is building apps, 386 00:17:30,640 --> 00:17:32,920 isn't that a massive security nightmare? We USTI to call 387 00:17:33,000 --> 00:17:33,599 that shadow it. 388 00:17:33,960 --> 00:17:34,319 Speaker 2: Oh it is. 389 00:17:34,640 --> 00:17:36,319 Speaker 1: Just because you can build an app doesn't mean you 390 00:17:36,400 --> 00:17:38,400 know how to properly encrypt a database. 391 00:17:38,480 --> 00:17:41,920 Speaker 2: Oh, it's a gigantic headache for CIOs everywhere. Suddenly you 392 00:17:41,960 --> 00:17:47,000 have five hundred unvetted insecure apps running on your corporate network. 393 00:17:47,400 --> 00:17:50,599 But from a trend perspective, it's unstoppable. The barrier to 394 00:17:50,759 --> 00:17:53,839 entry has completely collapsed. You don't need to be an 395 00:17:53,920 --> 00:17:55,920 architect to build a shed anymore. 396 00:17:55,640 --> 00:17:57,640 Speaker 1: But you still need an architect to build a skyscraper. 397 00:17:57,759 --> 00:18:01,319 Speaker 2: Exactly. Engineers are not obsolete. Role just moves up the stack. 398 00:18:01,559 --> 00:18:06,319 They handled the complex, scalable secure infrastructure. The citizen developers 399 00:18:06,400 --> 00:18:08,640 handled the local specific problems. 400 00:18:08,880 --> 00:18:11,960 Speaker 1: Now, speaking of how these things work, last year twenty 401 00:18:12,000 --> 00:18:15,519 twenty five, there was this massive hype cycle around agents, 402 00:18:15,680 --> 00:18:17,000 autonomous agents. 403 00:18:16,680 --> 00:18:18,559 Speaker 2: To set it and forget it dreams the dream. 404 00:18:18,839 --> 00:18:23,720 Speaker 1: But trend number two is agents versus workflows, and looking 405 00:18:23,720 --> 00:18:27,400 at this report, it seems like the agents they kind 406 00:18:27,440 --> 00:18:27,960 of failed. 407 00:18:28,039 --> 00:18:31,119 Speaker 2: They face planted hard. The report found that sixty seven 408 00:18:31,200 --> 00:18:34,039 percent of companies that deployed fully economous agents scaled them 409 00:18:34,079 --> 00:18:35,160 back within six months. 410 00:18:35,480 --> 00:18:38,960 Speaker 1: Why. I remember seeing the demos. You'd say, go plant 411 00:18:38,960 --> 00:18:42,240 a vacation to Italy, and the agent would browse the web, 412 00:18:42,480 --> 00:18:44,759 use your credit card and book the whole thing. What 413 00:18:44,880 --> 00:18:45,359 went wrong? 414 00:18:45,599 --> 00:18:49,559 Speaker 2: Loops, infinite loops and just wild hallucinations. You'd tell an 415 00:18:49,559 --> 00:18:53,680 agent research this competitor company. It would get ninety percent done. 416 00:18:53,839 --> 00:18:56,400 It would hit a broken link on their website, and 417 00:18:56,440 --> 00:18:59,359 then it would spend five hundred dollars of API credits 418 00:18:59,680 --> 00:19:01,960 just trying to click that one broken link over and 419 00:19:02,000 --> 00:19:04,440 over and over again. It got stuck, or it would 420 00:19:04,480 --> 00:19:08,319 make a decision that was like legally disastrous for the company. 421 00:19:08,680 --> 00:19:11,119 It lacked judgment, it lacked a brake pedal. That's a 422 00:19:11,119 --> 00:19:13,039 great way to put it. It lacked a brake pedal. 423 00:19:13,440 --> 00:19:15,880 So the big pivot, and this is trend too, is 424 00:19:15,960 --> 00:19:19,920 toward human in the loop workflows. The report says eighty 425 00:19:20,000 --> 00:19:23,279 nine percent of successful AI implementations use this model. 426 00:19:23,359 --> 00:19:25,880 Speaker 1: So instead of telling it go do this entire job 427 00:19:25,920 --> 00:19:28,480 by yourself, it's more like, do step one, then show 428 00:19:28,519 --> 00:19:30,559 me the result. Okay, that looks good. Now do step two, 429 00:19:30,599 --> 00:19:31,319 then show me again. 430 00:19:31,440 --> 00:19:34,400 Speaker 2: Exactly, It's a workflow, not an agent. The AI is 431 00:19:34,400 --> 00:19:37,359 the powerful engine but the human is the conductor. It 432 00:19:37,400 --> 00:19:41,799 turns out businesses don't actually want full autonomy. They want accountability. Right. 433 00:19:42,000 --> 00:19:45,160 Speaker 1: If the AI screw something up, you can't fire it. 434 00:19:45,440 --> 00:19:47,519 You need a human who signed off on the work. 435 00:19:47,759 --> 00:19:49,240 Speaker 2: That makes me feel a little bit better about my 436 00:19:49,279 --> 00:19:50,119 own job security. 437 00:19:50,119 --> 00:19:52,400 Speaker 1: Honestly it should. The human in the loop is the 438 00:19:52,440 --> 00:19:54,079 critical safety valve, which brings us. 439 00:19:54,000 --> 00:19:56,559 Speaker 2: To the number one trend, the one we teased at 440 00:19:56,559 --> 00:20:01,880 the very beginning trend. Number one models stuff being the differentiator. 441 00:20:02,240 --> 00:20:05,559 Speaker 1: This is the one that changes the entire business landscape completely. 442 00:20:05,640 --> 00:20:08,960 Speaker 2: Okay, explain this because for years the entire game was 443 00:20:09,400 --> 00:20:11,839 my AI is bigger and smarter than your AI. 444 00:20:12,079 --> 00:20:14,839 Speaker 1: It was a pure arms race. Yeah I have GPT four, 445 00:20:15,000 --> 00:20:18,759 you only have GPT three, so I win. Yeah, but 446 00:20:18,880 --> 00:20:22,160 look at the API pricing. The report notes that since 447 00:20:22,200 --> 00:20:25,559 twenty twenty three, the raw cost of top tier intelligence 448 00:20:25,559 --> 00:20:27,079 has dropped by eighty percent. 449 00:20:27,200 --> 00:20:29,319 Speaker 2: It's just crashing. The price is falling through the floor. 450 00:20:29,400 --> 00:20:32,680 Speaker 1: It's commoditizing. Intelligence is becoming like electricity. You don't care 451 00:20:32,720 --> 00:20:35,319 which power plant your electricity comes from. You just care 452 00:20:35,359 --> 00:20:36,920 that when you flip the switch, the light turns on. 453 00:20:37,039 --> 00:20:38,799 Speaker 2: So if I'm a startup in twenty twenty six. I 454 00:20:38,839 --> 00:20:41,400 can't go to VCS and I have my pitch be 455 00:20:41,599 --> 00:20:45,200 we have a really really smart model. No, they'll laugh 456 00:20:45,240 --> 00:20:48,160 you out of the room because your competitor has access 457 00:20:48,200 --> 00:20:51,279 to the exact same model for pennies. The moat is gone. 458 00:20:51,559 --> 00:20:54,200 If everyone has access to a super genius brain for 459 00:20:54,240 --> 00:20:56,880 next to nothing, then having a super genius brain isn't 460 00:20:56,920 --> 00:20:57,759 special anymore. 461 00:20:58,039 --> 00:21:00,359 Speaker 1: So what is special? If the brain itself is a 462 00:21:00,400 --> 00:21:03,480 cheap commodity, what's the luxury good? What's the differentiator? 463 00:21:03,880 --> 00:21:09,480 Speaker 2: Two things, primarily proprietary data and user experience. If you 464 00:21:09,519 --> 00:21:11,920 have a data set that nobody else on earth has, 465 00:21:11,960 --> 00:21:16,079 like fifty years of specific medical records for a rare disease, 466 00:21:16,400 --> 00:21:18,680 that's a mote, okay. Or if you have a workflow 467 00:21:18,680 --> 00:21:21,599 a user interface that is just delightful and intuitive to use, 468 00:21:21,599 --> 00:21:25,119 that's a mote. But the tech itself, the AI part, 469 00:21:25,200 --> 00:21:26,480 it's just table stakes. Now. 470 00:21:26,519 --> 00:21:28,440 Speaker 1: This is a massive shift. It means the tech war 471 00:21:28,519 --> 00:21:31,160 isn't about the algorithm anymore. It's about the product that's 472 00:21:31,160 --> 00:21:32,240 wrapped around the algorithm. 473 00:21:32,240 --> 00:21:35,240 Speaker 2: It shifts power from the research scientists to the product managers. 474 00:21:35,480 --> 00:21:38,319 Speaker 1: Okay, so we've covered the tech, the users, the business models, 475 00:21:38,400 --> 00:21:40,799 the economy. But we have to zoom out. Now, we 476 00:21:40,839 --> 00:21:44,119 have to look at the map, the literal world map. 477 00:21:44,400 --> 00:21:49,160 H Segment four is the geopolitical landscape trend number eight. 478 00:21:50,039 --> 00:21:51,640 AI as statecraft. 479 00:21:52,160 --> 00:21:53,880 Speaker 2: This is the heavy one. This is the one that 480 00:21:54,000 --> 00:21:56,039 keeps the lights on at the Pentagon and in national 481 00:21:56,039 --> 00:21:57,799 security agencies all over the world. 482 00:21:57,960 --> 00:22:00,480 Speaker 1: We aren't talking about Google versus Microsoft anym. We are 483 00:22:00,480 --> 00:22:04,559 talking about the US versus China, versus Europe versus everyone else. 484 00:22:04,720 --> 00:22:08,200 Speaker 2: AI has officially graduated from a corporate asset to a 485 00:22:08,319 --> 00:22:12,799 national imperative. The number of countries with formal, government funded 486 00:22:12,880 --> 00:22:16,160 national AI strategies has tripled in just a few years. 487 00:22:16,599 --> 00:22:19,079 Speaker 1: Why I mean, is it just national pride? We have 488 00:22:19,119 --> 00:22:19,799 an AI too. 489 00:22:20,000 --> 00:22:22,920 Speaker 2: It started that way, but now it's about economic sovereignty. 490 00:22:23,319 --> 00:22:25,759 Think about it. If the AI models that run your 491 00:22:25,759 --> 00:22:29,880 country's hospitals, your banks, and your school systems are all 492 00:22:29,920 --> 00:22:33,720 hosted on servers in California, do you really control your 493 00:22:33,720 --> 00:22:34,359 own country? 494 00:22:34,400 --> 00:22:36,359 Speaker 1: If the US government decides to turn off the. 495 00:22:36,319 --> 00:22:39,720 Speaker 2: Tab exactly, your economy grinds to a halt. Or what 496 00:22:39,759 --> 00:22:41,960 if the US decides to just peak at all the 497 00:22:42,039 --> 00:22:45,000 data flowing through those servers. So we are seeing this 498 00:22:45,319 --> 00:22:48,240 massive push for sovereign. 499 00:22:47,759 --> 00:22:49,559 Speaker 1: Clouds, sovereign cloud. 500 00:22:49,400 --> 00:22:53,119 Speaker 2: Countries building their own massive data centers on their own soil. 501 00:22:53,599 --> 00:22:56,519 The Middle East is spending tens of billions. Europe is 502 00:22:56,559 --> 00:22:59,400 frantic to build their own compute capacity to catch up. 503 00:22:59,480 --> 00:23:01,359 Speaker 1: And this is a physical thing, right, We're talking about 504 00:23:01,359 --> 00:23:02,640 a race for computer chips. 505 00:23:02,759 --> 00:23:06,440 Speaker 2: It's chips and it's energy. That's the bottleneck. Nobody discusses enough. 506 00:23:06,920 --> 00:23:10,480 These data centers consume terrifying amounts of electricity. We are 507 00:23:10,519 --> 00:23:15,279 literally seeing geopolitical struggles over access to power grids. Who 508 00:23:15,359 --> 00:23:18,039 has the energy to run the brain of the future. 509 00:23:18,319 --> 00:23:20,680 Speaker 1: It feels like the oil wars of the twentieth century, 510 00:23:21,200 --> 00:23:22,480 but for silicon it. 511 00:23:22,519 --> 00:23:25,240 Speaker 2: Is exactly that. We are moving from a tech war 512 00:23:25,799 --> 00:23:29,359 to an economic solientty war that will be defined by 513 00:23:29,359 --> 00:23:33,680 compute power. Where the GPU clusters live, That's where the 514 00:23:33,680 --> 00:23:35,599 future economy lives. 515 00:23:34,960 --> 00:23:38,920 Speaker 1: That is really sobering. It certainly puts my annoyance about 516 00:23:38,920 --> 00:23:43,079 context windows into perspective. We're worrying about our little chatbots 517 00:23:43,119 --> 00:23:46,440 while nations are worrying about the fundamental infrastructure of reality. 518 00:23:46,519 --> 00:23:48,799 Speaker 2: It's all connected, though, isn't it. The little chatbot on 519 00:23:48,839 --> 00:23:52,160 your phone only works because of that massive, energy hungry, 520 00:23:52,359 --> 00:23:54,519 geopolitically sensitive infrastructure. 521 00:23:54,880 --> 00:23:59,079 Speaker 1: Right, it all flows together. So let's try to synthesize this. 522 00:23:59,519 --> 00:24:02,359 We have a lot of ground here. We have. We 523 00:24:02,400 --> 00:24:04,920 went from the death of prompt engineering, all the way 524 00:24:04,960 --> 00:24:09,920 to robots moving boxes, to the commoditization of intelligence, and 525 00:24:10,000 --> 00:24:12,640 finally to global chip wars. 526 00:24:12,920 --> 00:24:14,960 Speaker 2: It's a complex tapestry for sure. 527 00:24:15,119 --> 00:24:19,519 Speaker 1: If you had to pick on one single theme for 528 00:24:19,599 --> 00:24:21,799 twenty twenty six, what would it be. 529 00:24:21,839 --> 00:24:22,400 Speaker 2: Integration? 530 00:24:22,680 --> 00:24:23,200 Speaker 1: Integration. 531 00:24:23,359 --> 00:24:26,279 Speaker 2: Okay, Up until now, AI was a novelty. It was 532 00:24:26,319 --> 00:24:29,200 a destination you went to chat GPT to do a thing. 533 00:24:29,319 --> 00:24:33,000 Right now it's integrated. It's in your phones operating system, 534 00:24:33,240 --> 00:24:35,079 it's in the robot, in the factory, it's in the 535 00:24:35,160 --> 00:24:38,480 national defense strategy. It's no longer a separate layer on 536 00:24:38,559 --> 00:24:42,079 top of everything. It is the infrastructure. It's the concrete 537 00:24:42,160 --> 00:24:43,119 of the digital world. 538 00:24:43,279 --> 00:24:45,440 Speaker 1: And for the listener, you know, trying to navigate this 539 00:24:45,519 --> 00:24:47,680 new world, what's the big personal takeaway? 540 00:24:47,799 --> 00:24:50,759 Speaker 2: I'd say, look at trend three and trend two together. Okay, 541 00:24:50,880 --> 00:24:53,319 the technical barriers are down. You can build things you 542 00:24:53,400 --> 00:24:55,799 never could have before. If you have an idea you 543 00:24:55,799 --> 00:24:59,119 can probably make it a reality, but don't trust the 544 00:24:59,160 --> 00:25:02,000 machine blindly. Keep your hand on the wheel. The human 545 00:25:02,039 --> 00:25:04,119 in the loop is you. You're the one that matters. 546 00:25:04,359 --> 00:25:06,079 Speaker 1: When I have a question, I want to leave everyone 547 00:25:06,119 --> 00:25:09,319 with a final provocation, go for it. If intelligence is 548 00:25:09,359 --> 00:25:11,519 now cheap and abundant, like we said in trend one, 549 00:25:12,240 --> 00:25:14,759 and the technical skills to build things are no longer 550 00:25:14,799 --> 00:25:18,119 a barrier like in trend three, what becomes the most 551 00:25:18,240 --> 00:25:19,599 valuable human skill? 552 00:25:19,799 --> 00:25:22,039 Speaker 2: That is the billion dollar question, isn't it. 553 00:25:21,920 --> 00:25:24,079 Speaker 1: What do we bring to the table? Is it taste? 554 00:25:24,599 --> 00:25:27,880 Is it judgment? Is it empathy? When the machine can 555 00:25:27,920 --> 00:25:30,240 do all the math and write all the code, maybe 556 00:25:30,279 --> 00:25:33,160 the most valuable thing you can do is just be human. 557 00:25:33,279 --> 00:25:36,279 Speaker 2: I think the future belongs to the curators, the people 558 00:25:36,279 --> 00:25:38,039 who have the judgment and the taste to know what 559 00:25:38,079 --> 00:25:40,039 to ask the AI to do, not just how to 560 00:25:40,079 --> 00:25:40,559 ask it. 561 00:25:41,519 --> 00:25:43,880 Speaker 1: I love that the curators, Well, we want to hear 562 00:25:43,920 --> 00:25:46,839 from you. Seriously. Where do you stand on this whole 563 00:25:47,000 --> 00:25:50,039 human in the loop idea? Are you relieved that the 564 00:25:50,079 --> 00:25:53,680 autonomous agents struggled and you're still needed or are you 565 00:25:53,720 --> 00:25:55,839 annoyed because you just wanted the robot to do your 566 00:25:55,920 --> 00:25:57,680 job so you could finally go to the beach. 567 00:25:57,880 --> 00:25:59,640 Speaker 2: I definitely know my answer to that one. 568 00:26:00,079 --> 00:26:02,480 Speaker 1: On the comments. Thanks for joining us on this thread, 569 00:26:03,119 --> 00:26:04,759 keep pulling at it, See you next time.