WEBVTT 00:00:00.000 --> 00:00:01.720 Do you ever get that feeling that technology 00:00:01.720 --> 00:00:04.639 is just moving almost too fast, like we're right 00:00:04.639 --> 00:00:07.000 on the edge of something? Oh, all the time. I 00:00:07.000 --> 00:00:09.140 mean, I certainly do. It feels like just yesterday, 00:00:09.439 --> 00:00:12.480 AI was, you know, this little novelty. It could 00:00:12.480 --> 00:00:14.880 write a decent email, maybe. A little stiff, 00:00:14.900 --> 00:00:18.480 but it worked. Exactly. And now today we're talking 00:00:18.480 --> 00:00:21.059 about systems that can execute this incredibly 00:00:21.059 --> 00:00:24.500 complex work that once took a whole team of human 00:00:24.500 --> 00:00:27.500 experts months. Yeah, the testing phase is definitely 00:00:27.500 --> 00:00:30.100 over. It's completely over. I think 2026 is going 00:00:30.100 --> 00:00:31.920 to be the year we really start living and working 00:00:31.920 --> 00:00:34.810 in a totally new way. Welcome to the deep dive. 00:00:35.030 --> 00:00:37.250 I'm so energized for this one because we are 00:00:37.250 --> 00:00:39.729 getting into some critical insights from, really, 00:00:40.250 --> 00:00:43.609 the cutting edge of tech reports. And for you, 00:00:43.810 --> 00:00:46.049 our listener, the learner, it makes sense to 00:00:46.049 --> 00:00:48.109 ask, you know, how do we make sense of all this 00:00:48.109 --> 00:00:50.689 acceleration? Our mission today is pretty specific. 00:00:51.170 --> 00:00:54.229 We're going to chart this huge shift from AI 00:00:54.229 --> 00:00:56.429 as just a simple tool, like that old chat bot 00:00:56.429 --> 00:00:59.750 model, to AI as a true collaborative partner. 00:00:59.920 --> 00:01:02.079 This is really the age of digital amplification. 00:01:02.460 --> 00:01:05.540 We have seven big trends to unpack here. We're 00:01:05.540 --> 00:01:07.620 going to start with that profound shift in how 00:01:07.620 --> 00:01:11.719 we work and then the intense new safety demands 00:01:11.719 --> 00:01:14.260 that come with it. Then we'll pivot a bit. We're 00:01:14.260 --> 00:01:16.519 going to look at how AI is dramatically improving 00:01:16.519 --> 00:01:19.700 things like global health and speeding up scientific 00:01:19.700 --> 00:01:22.719 discovery on a scale that honestly is hard to 00:01:22.719 --> 00:01:24.659 even imagine. And finally, we'll pull back the 00:01:24.659 --> 00:01:27.159 curtain on the massive global infrastructure, 00:01:27.379 --> 00:01:30.010 the sort of silent engine. that has to power 00:01:30.010 --> 00:01:32.349 this whole new world. OK, let's get into it. 00:01:32.409 --> 00:01:34.010 Let's do it. So we have to start with the big 00:01:34.010 --> 00:01:36.010 one, right? Yeah. The existential question that's 00:01:36.010 --> 00:01:39.569 on everyone's mind. Is AI going to take my job? 00:01:39.909 --> 00:01:42.310 It's a natural fear. It is. It is. But if you 00:01:42.310 --> 00:01:44.530 look at the consensus across the major reports, 00:01:45.090 --> 00:01:49.209 the goal for 2026 is amplification. It's not 00:01:49.209 --> 00:01:51.549 wholesale replacement. It's about collaboration 00:01:51.549 --> 00:01:54.599 at scale. I appreciate that, but. you know, these 00:01:54.599 --> 00:01:56.719 analogies can sometimes gloss over the complexity. 00:01:57.159 --> 00:01:59.540 I know we love the one about AI being like a 00:01:59.540 --> 00:02:02.019 set of super modern power tools. And the human 00:02:02.019 --> 00:02:03.859 is still making the fundamental design choice, 00:02:04.180 --> 00:02:06.480 but the tool gives you like 10 times the speed. 00:02:06.670 --> 00:02:09.189 But see, those power tools don't ask for a salary, 00:02:09.409 --> 00:02:11.990 and they don't introduce these massive security 00:02:11.990 --> 00:02:14.509 risks when you plug them into your company's 00:02:14.509 --> 00:02:16.889 financial system. Isn't that analogy kind of 00:02:16.889 --> 00:02:19.129 missing the complexity of adding a decision -making 00:02:19.129 --> 00:02:21.770 thing? That's a really fair challenge. I think 00:02:21.770 --> 00:02:23.770 the difference is in the evolution away from 00:02:23.770 --> 00:02:26.229 the old chat bot. That was the model where you 00:02:26.229 --> 00:02:28.310 had to micromanage everything, write this email, 00:02:28.389 --> 00:02:30.789 you wait, then... OK, now send this email. Right, 00:02:30.969 --> 00:02:32.889 step by step. We're moving toward what the industry 00:02:32.889 --> 00:02:35.610 is calling AI agents. And you should think of 00:02:35.610 --> 00:02:39.569 an agent less like a tool and more like a smart, 00:02:40.250 --> 00:02:42.930 goal -oriented intern who doesn't need constant 00:02:42.930 --> 00:02:45.530 check -ins. So you don't micromanage. You just 00:02:45.530 --> 00:02:47.490 give it a single high -level goal, something 00:02:47.490 --> 00:02:49.770 like, help me launch and sell these three new 00:02:49.770 --> 00:02:52.330 products. Exactly. And then that agent just... 00:02:52.270 --> 00:02:55.030 It goes, it handles the market research, it drafts 00:02:55.030 --> 00:02:57.389 the sales copy, it schedules the emails, and 00:02:57.389 --> 00:02:59.449 it even generates a report on conversion rates, 00:02:59.930 --> 00:03:02.449 all from that one command. Microsoft calls this 00:03:02.449 --> 00:03:04.750 amplification, right? Yeah, that's their term, 00:03:04.810 --> 00:03:06.770 because it fundamentally changes the scale of 00:03:06.770 --> 00:03:09.669 what a human can do. I mean, a small team of, 00:03:09.669 --> 00:03:12.530 say, three people can suddenly do the work of 00:03:12.530 --> 00:03:14.430 a medium -sized company. That's where the real 00:03:14.430 --> 00:03:17.629 value is. So, if this new agent is defined by 00:03:17.629 --> 00:03:21.250 its goal orientation... It's autonomy. What's 00:03:21.250 --> 00:03:23.550 the fundamental change in how we have to prompt 00:03:23.550 --> 00:03:25.870 it? You have to ask it to act like a manager 00:03:25.870 --> 00:03:28.210 or a partner. You're setting a high -level objective, 00:03:28.449 --> 00:03:31.210 not giving it a to -do list. That kind of power 00:03:31.210 --> 00:03:34.069 is incredible. But, you know, the second you 00:03:34.069 --> 00:03:36.830 give that digital intern the keys to the kingdom 00:03:36.830 --> 00:03:39.250 access to your client list, your company email, 00:03:39.550 --> 00:03:42.409 we immediately have to talk about risk. Oh, absolutely. 00:03:42.710 --> 00:03:45.289 The security risk just skyrockets. I mean, what 00:03:45.289 --> 00:03:48.189 if that autonomous agent gets tricked by a bad 00:03:48.189 --> 00:03:51.319 actor? Your helpful tool could be told to send 00:03:51.319 --> 00:03:53.520 out secret company data, and suddenly it's a 00:03:53.520 --> 00:03:55.560 spy working against you. They're calling this 00:03:55.560 --> 00:03:58.099 the double agent problem. And it's a genuinely 00:03:58.099 --> 00:04:00.300 terrifying thought that the AI you trust to help 00:04:00.300 --> 00:04:02.060 you is turned against you because you gave it 00:04:02.060 --> 00:04:04.699 too much access. Yeah. So the solution for 2026 00:04:04.699 --> 00:04:06.680 is all about building a really strong protection 00:04:06.680 --> 00:04:10.379 layer right into the OS itself. Every single 00:04:10.379 --> 00:04:13.159 AI agent is going to need a digital ID card to 00:04:13.159 --> 00:04:16.680 do anything. And that ID is key because it strictly 00:04:16.680 --> 00:04:19.339 limits what the agent can see. It can only access 00:04:19.339 --> 00:04:22.079 certain folders. It's basically walled off from 00:04:22.079 --> 00:04:24.459 your most sensitive files. And critically, the 00:04:24.459 --> 00:04:26.759 agent cannot send any data outside the system 00:04:26.759 --> 00:04:29.399 without your explicit human consent. You have 00:04:29.399 --> 00:04:32.800 to physically click yes. It's mandatory human 00:04:32.800 --> 00:04:35.060 oversight. And this is where it gets challenging. 00:04:35.120 --> 00:04:37.240 I have to admit, even with all the safeguards, 00:04:37.439 --> 00:04:39.819 I still wrestle with prompt drift myself, just 00:04:39.819 --> 00:04:41.879 trying to maintain that vigilance when I'm delegating 00:04:41.879 --> 00:04:44.699 sensitive tasks. It's a constant battle to prevent 00:04:44.699 --> 00:04:47.019 accidental data leaks. That's a really powerful 00:04:47.019 --> 00:04:49.779 admission. So given that high stakes risk, what's 00:04:49.779 --> 00:04:52.000 the key piece of practical advice for our listeners 00:04:52.000 --> 00:04:55.649 today? Practice data safety. Never input bank 00:04:55.649 --> 00:04:58.610 passwords or client secrets into those free unregulated 00:04:58.610 --> 00:05:01.350 chat tools. That security issue leads us right 00:05:01.350 --> 00:05:03.870 into how AI is handling responsibility in the 00:05:03.870 --> 00:05:07.569 real world. Let's shift gears now out of the 00:05:07.569 --> 00:05:10.709 server room and into the clinic. Trend three, 00:05:11.230 --> 00:05:14.610 fixing the doctor shortage. This is a huge looming 00:05:14.610 --> 00:05:17.089 crisis. The World Health Organization is predicting 00:05:17.089 --> 00:05:20.509 a global shortage of 10 million health workers 00:05:20.509 --> 00:05:24.240 by 2030. I mean, that means huge numbers of people 00:05:24.240 --> 00:05:26.279 just won't see a doctor when they're sick. And 00:05:26.279 --> 00:05:28.939 AI is central to trying to solve this. The goal 00:05:28.939 --> 00:05:31.240 isn't replacement. It's about freeing the doctor's 00:05:31.240 --> 00:05:34.180 hands to deal with the most critical human cases. 00:05:34.240 --> 00:05:36.199 For sure. The applications are immediate. First, 00:05:36.300 --> 00:05:38.839 you've got triage. A nurse AI can look at initial 00:05:38.839 --> 00:05:41.319 pain level, temperature, and figure out the danger 00:05:41.319 --> 00:05:43.680 level to prioritize who needs to see a human 00:05:43.680 --> 00:05:46.259 right away. And second is specialized diagnosis 00:05:46.259 --> 00:05:48.600 support. Microsoft has been pushing a tool, I 00:05:48.600 --> 00:05:50.480 think it's called the Diagnostic Orchestrator. 00:05:50.699 --> 00:05:53.899 Yeah, and what's wild here is the data. In tests 00:05:53.899 --> 00:05:56.660 on really difficult, complex medical cases, that 00:05:56.660 --> 00:06:00.040 orchestrator hit an 85 .5 % accuracy rate. That's 00:06:00.040 --> 00:06:01.879 not a small improvement. That's literally saving 00:06:01.879 --> 00:06:05.439 lives right now. Wow, 85 .5 % on the hard cases. 00:06:06.060 --> 00:06:08.360 It really shows you how fast this is moving when 00:06:08.360 --> 00:06:11.980 it's applied to a real human need. So if AI is 00:06:11.980 --> 00:06:14.439 being used for diagnosis, what's the critical 00:06:14.439 --> 00:06:16.660 boundary users have to respect with their own 00:06:16.660 --> 00:06:19.339 health data? AI should only be used for reference 00:06:19.339 --> 00:06:22.480 and simple explanation. A human doctor must always 00:06:22.480 --> 00:06:24.379 be the final authority on your health. Okay, 00:06:24.420 --> 00:06:26.459 let's talk scientific discovery. Trend four. 00:06:26.759 --> 00:06:29.160 We're moving beyond AI just, you know, writing 00:06:29.160 --> 00:06:31.360 summaries. Now it's becoming an active scientist 00:06:31.360 --> 00:06:33.319 itself. Right. Think about the old scientific 00:06:33.319 --> 00:06:36.199 method. You had researchers manually mixing substance 00:06:36.199 --> 00:06:38.879 A and substance B, hoping for something, failing, 00:06:38.980 --> 00:06:41.500 trying again. That could take years just to test 00:06:41.500 --> 00:06:44.399 one idea. The new AI way just shortcuts that 00:06:44.399 --> 00:06:46.759 entire process. It doesn't just read a few papers. 00:06:47.019 --> 00:06:49.560 It reads all global science papers, and it connects 00:06:49.560 --> 00:06:51.939 these non - obvious dots that no human team could 00:06:51.939 --> 00:06:55.240 ever see. And then it runs millions of virtual 00:06:55.240 --> 00:06:57.800 experiments in seconds. It simulates chemical 00:06:57.800 --> 00:07:00.240 reactions with perfect precision. And the result 00:07:00.240 --> 00:07:03.079 is these highly targeted suggestions that just 00:07:03.079 --> 00:07:06.519 slash R &D timelines. Things like, hey, why don't 00:07:06.519 --> 00:07:09.600 we mix compound X and metal Y? And that leads 00:07:09.600 --> 00:07:13.240 to faster cures new materials. Whoa. Just imagine 00:07:13.240 --> 00:07:16.379 that potential. When AI can suggest a new drug 00:07:16.379 --> 00:07:18.519 or a new battery material that would have taken 00:07:18.519 --> 00:07:21.240 thousands of man hours to test, and it does it 00:07:21.240 --> 00:07:23.800 in a few minutes, I mean, that changes the fundamental 00:07:23.800 --> 00:07:26.420 speed of civilization? That's a serious moment 00:07:26.420 --> 00:07:29.720 of wonder. It truly is. So what's a simple, real 00:07:29.720 --> 00:07:32.139 -world application of that pattern -finding mindset 00:07:32.139 --> 00:07:35.290 for someone who isn't a scientist? You can use 00:07:35.290 --> 00:07:38.410 AI to find non -obvious connections, like between 00:07:38.410 --> 00:07:41.569 fast fashion and local water challenges, to generate 00:07:41.569 --> 00:07:43.889 creative business solutions. Speaking of solutions, 00:07:43.889 --> 00:07:46.490 we have to talk infrastructure. Trend five, the 00:07:46.490 --> 00:07:49.410 energy problem. Every sophisticated AI query 00:07:49.410 --> 00:07:52.269 uses way more electricity than a simple Google 00:07:52.269 --> 00:07:54.250 search. Yeah. If everyone's using an agent all 00:07:54.250 --> 00:07:56.350 day, the demand for power is going to skyrocket. 00:07:56.490 --> 00:07:58.209 You can't just keep building more servers. It's 00:07:58.209 --> 00:08:00.829 terrible for the environment. So the 2026 solution 00:08:00.829 --> 00:08:03.199 is all about efficiency. And they're basically 00:08:03.199 --> 00:08:05.379 building a smart grid for intelligence using 00:08:05.379 --> 00:08:07.759 something called distributed computing. The easiest 00:08:07.759 --> 00:08:09.339 way to think about it is like a ride sharing 00:08:09.339 --> 00:08:12.660 app, like Uber, but for computer power. Exactly. 00:08:12.879 --> 00:08:16.699 When server farms in, say, the US are quiet because 00:08:16.699 --> 00:08:18.959 it's the middle of the night, their power gets 00:08:18.959 --> 00:08:21.740 automatically routed to users in another region, 00:08:21.939 --> 00:08:24.500 like Vietnam, where it's daytime and demand is 00:08:24.500 --> 00:08:27.860 high. The goal is just maximum utilization. No 00:08:27.860 --> 00:08:31.079 server sits idle. As Mark Risenovich from Microsoft 00:08:31.079 --> 00:08:33.799 Azure said, we have to make every single watt 00:08:33.799 --> 00:08:36.139 of power produce intelligence. There's no more 00:08:36.139 --> 00:08:38.600 margin for waste. So beyond just saving power, 00:08:38.759 --> 00:08:41.320 what does distributed computing really enable 00:08:41.320 --> 00:08:44.500 for users around the world? It allows for continuous 00:08:44.500 --> 00:08:46.740 high -speed access to computing power globally 00:08:46.740 --> 00:08:49.299 by just maximizing the use of the hardware we 00:08:49.299 --> 00:08:51.179 already have. Okay, moving to trend six. Let's 00:08:51.179 --> 00:08:53.759 talk about coding, which is a huge field for 00:08:53.759 --> 00:08:55.600 so many of our learners. This evolution is a 00:08:55.600 --> 00:08:58.879 big one. For sure. The old AI coding cools, like 00:08:58.879 --> 00:09:01.159 the first version of GitHub Copilot, were basically 00:09:01.159 --> 00:09:03.960 just fancy autofill. They just guessed the next 00:09:03.960 --> 00:09:06.340 few words of code you were writing. But the new 00:09:06.340 --> 00:09:09.480 AI understands the entire deep context of your 00:09:09.480 --> 00:09:11.879 project, not just the one file you're in. They 00:09:11.879 --> 00:09:14.980 call this repository intelligence. The old AI 00:09:14.980 --> 00:09:17.860 tried to guess the story from one page. The new 00:09:17.860 --> 00:09:20.480 AI has read the whole book, the author's notes, 00:09:20.779 --> 00:09:23.320 and the publisher's style guide, the entire project 00:09:23.320 --> 00:09:26.210 folder. And the real value for a coder is system 00:09:26.210 --> 00:09:29.649 coherence. The AI knows that if you fix one little 00:09:29.649 --> 00:09:31.950 thing in page A, it's going to affect something 00:09:31.950 --> 00:09:35.889 15 files away in page B. It sees the whole architecture. 00:09:36.049 --> 00:09:38.509 And you can already see the impact. GitHub said 00:09:38.509 --> 00:09:41.889 their users' code updates jumped by 25 % to over 00:09:41.889 --> 00:09:44.909 a billion updates, just driven by this context 00:09:44.909 --> 00:09:47.620 -aware help. For a beginner, what's the most 00:09:47.620 --> 00:09:50.000 important takeaway about the value of that kind 00:09:50.000 --> 00:09:52.240 of intelligence? It's like the AI has read the 00:09:52.240 --> 00:09:53.840 architect's blueprint. So if you try to move 00:09:53.840 --> 00:09:55.759 a load -bearing wall, it instantly tells you 00:09:55.759 --> 00:09:57.860 the roof is going to collapse. Finally, trend 00:09:57.860 --> 00:10:00.919 seven, the frontier of speed itself, quantum 00:10:00.919 --> 00:10:03.039 computing. For years, this was, you know, science 00:10:03.039 --> 00:10:06.419 fiction. By 2026, it's becoming a real integrated 00:10:06.419 --> 00:10:08.700 factor. The simplest way to get the speed difference 00:10:08.700 --> 00:10:11.299 is this. The normal computer solves a maze by 00:10:11.299 --> 00:10:13.720 walking down one path at a time. It hits a wall, 00:10:13.720 --> 00:10:16.080 turns back, tries again. A quantum computer, 00:10:16.299 --> 00:10:18.620 because of how qubits work, can essentially flow 00:10:18.620 --> 00:10:21.679 into all paths at the same time like water. It 00:10:21.679 --> 00:10:24.240 finds the solution almost instantly. That speed 00:10:24.240 --> 00:10:27.639 is mind -blowing. But there's a huge catch. Quantum 00:10:27.639 --> 00:10:30.480 computers are still super unstable, really sensitive 00:10:30.480 --> 00:10:33.190 to temperature, and hard to keep coherent. So 00:10:33.190 --> 00:10:36.769 the reality for 2026 isn't a total switch to 00:10:36.769 --> 00:10:39.669 quantum. It's the hybrid system. We're not using 00:10:39.669 --> 00:10:41.870 quantum for everything. We're integrating it 00:10:41.870 --> 00:10:44.289 carefully. Yeah, the hybrid system is this powerful 00:10:44.289 --> 00:10:47.230 three -part team. AI handles pattern recognition. 00:10:47.649 --> 00:10:49.950 Classical supercomputers do the massive stable 00:10:49.950 --> 00:10:52.090 math. And then the quantum computers are safe 00:10:52.090 --> 00:10:53.970 for the hardest problems, like in chemistry. 00:10:54.250 --> 00:10:56.490 That's why we still need those big supercomputers 00:10:56.490 --> 00:10:58.669 for all the stable math. Quantum is a specialist, 00:10:59.169 --> 00:11:00.990 perfect for modeling molecular bonds, things 00:11:00.990 --> 00:11:03.909 like that. Right. And Microsoft's Majorana One 00:11:03.909 --> 00:11:06.750 chip is a huge step here. It's basically trying 00:11:06.750 --> 00:11:09.490 to build a reliable interface to make these unstable 00:11:09.490 --> 00:11:12.190 quantum systems predictable enough for actual 00:11:12.190 --> 00:11:15.370 work. Given that instability, why is it so important 00:11:15.370 --> 00:11:18.049 to keep pushing this hybrid approach? It lets 00:11:18.049 --> 00:11:21.009 us leverage quantum speed for those very specific 00:11:21.009 --> 00:11:25.269 hard problems while we rely on stability for, 00:11:25.269 --> 00:11:27.730 you know, 99 % of everything else. Okay, so let's 00:11:27.730 --> 00:11:30.350 recap the big idea here. The transition we've 00:11:30.350 --> 00:11:33.909 been talking about I think is complete. AI is 00:11:33.909 --> 00:11:36.789 no longer a calculator. It's not a toy. It's 00:11:36.789 --> 00:11:39.049 a partner. And it's designed for amplification 00:11:39.049 --> 00:11:42.190 and goal execution in every major field, from 00:11:42.190 --> 00:11:45.009 your job to finding global health cures. The 00:11:45.009 --> 00:11:47.769 world is changing incredibly fast. But the key 00:11:47.769 --> 00:11:50.289 is not to run from it. The goal is to drive the 00:11:50.289 --> 00:11:52.529 technology, to learn how to be a better collaborator 00:11:52.529 --> 00:11:55.529 with it. We pulled three immediate actions from 00:11:55.529 --> 00:11:57.750 the sources you can start today. First, change 00:11:57.750 --> 00:12:00.350 your mindset. Stop treating AI like an answer 00:12:00.350 --> 00:12:03.049 machine. Ask it to plan, to critique, to act 00:12:03.049 --> 00:12:05.990 like a partner. Second, protect yourself. Be 00:12:05.990 --> 00:12:08.230 extremely careful with personal data, especially 00:12:08.230 --> 00:12:10.509 in those free tools. That double agent problem 00:12:10.509 --> 00:12:13.879 is very real. And third, start small. Just pick 00:12:13.879 --> 00:12:16.039 one of these new agent -style tools, use it for 00:12:16.039 --> 00:12:18.519 one small task this week, and just get comfortable 00:12:18.519 --> 00:12:20.539 with delegation. Learn the limits of your new 00:12:20.539 --> 00:12:22.879 partner. The future really belongs to the people 00:12:22.879 --> 00:12:26.299 who actively drive this technology, not the ones 00:12:26.299 --> 00:12:28.259 who stand on the sidelines and hope it passes 00:12:28.259 --> 00:12:30.240 them by. So are you ready to be a driver?