WEBVTT 00:00:00.000 --> 00:00:02.200 Could your brain waves, your heart rate, and 00:00:02.200 --> 00:00:05.759 even subtle muscle twitches predict 130 different 00:00:05.759 --> 00:00:08.720 diseases? I mean, including your long -term risk 00:00:08.720 --> 00:00:11.339 of death just by analyzing a single night of 00:00:11.339 --> 00:00:14.980 your sleep. That is the astonishing level of 00:00:14.980 --> 00:00:16.760 predictive tech we're going to get into today. 00:00:16.899 --> 00:00:19.039 It just raises these profound questions about 00:00:19.039 --> 00:00:21.760 health and data. But first, we really need to 00:00:21.760 --> 00:00:23.739 talk about some crucial new safety measures that 00:00:23.739 --> 00:00:26.140 are being put in place for these highly sensitive 00:00:26.140 --> 00:00:29.539 health conversations. Welcome to the Deep Dive. 00:00:29.600 --> 00:00:31.339 You've given us a whole stack of sources here, 00:00:31.440 --> 00:00:33.579 and our mission is to filter them down, to really 00:00:33.579 --> 00:00:35.299 grasp what's happening. We're going to move from 00:00:35.299 --> 00:00:39.100 super personalized healthcare right into massive 00:00:39.100 --> 00:00:42.060 enterprise adoption and this sort of quiet revolution 00:00:42.060 --> 00:00:44.539 in how we interact with our tech. Yeah, we've 00:00:44.539 --> 00:00:46.079 got a lot to get through, so let's just lay it 00:00:46.079 --> 00:00:48.020 all out. We'll unpack the launch of a really 00:00:48.020 --> 00:00:51.380 highly anticipated, secure... AI health space. 00:00:51.560 --> 00:00:54.000 Then we'll shift into practical stuff like how 00:00:54.000 --> 00:00:56.079 to automate complex client services, you know, 00:00:56.079 --> 00:00:58.719 24 -7. We'll check in on some major interface 00:00:58.719 --> 00:01:02.520 trends. And then we'll dedicate the time to decoding 00:01:02.520 --> 00:01:05.299 that Stanford AI breakthrough, the one predicting 00:01:05.299 --> 00:01:07.939 health from sleep signals. Okay. Let's get into 00:01:07.939 --> 00:01:11.680 it. So chat GPT health. This is a huge move. 00:01:11.939 --> 00:01:14.819 And the reason why... Seems pretty straightforward. 00:01:15.239 --> 00:01:19.019 The sources say 230 million users are already 00:01:19.019 --> 00:01:21.959 asking health questions every single week. They 00:01:21.959 --> 00:01:24.400 just need a system to manage that safely. Exactly. 00:01:24.540 --> 00:01:27.620 That demand is why they built this. It's basically 00:01:27.620 --> 00:01:30.280 a dedicated health tab that's sandboxed. And 00:01:30.280 --> 00:01:32.579 sandbox means what exactly? It means it has a 00:01:32.579 --> 00:01:34.640 totally separate memory structure. And this is 00:01:34.640 --> 00:01:37.780 so critical because it means your detailed conversations 00:01:37.780 --> 00:01:40.560 about wellness or symptoms won't accidentally 00:01:40.560 --> 00:01:43.700 bleed into your regular chats about, say, Planning 00:01:43.700 --> 00:01:45.379 a trip. Which is smart. And they've engineered 00:01:45.379 --> 00:01:47.319 a nudge mechanism to support that. Oh, yeah. 00:01:47.439 --> 00:01:49.739 So if you start talking about symptoms in a regular 00:01:49.739 --> 00:01:53.900 chat, the AI will actually proactively push you 00:01:53.900 --> 00:01:56.159 toward this protected health space. It's a safety 00:01:56.159 --> 00:01:58.239 feature that feels like convenience. And to make 00:01:58.239 --> 00:02:01.120 it really useful, connectivity is everything. 00:02:01.280 --> 00:02:03.540 You can link your data from places like Apple 00:02:03.540 --> 00:02:06.920 Health, your Peloton, MyFitnessPal, even Weight 00:02:06.920 --> 00:02:10.300 Watchers. Right. letting the AI analyze a much, 00:02:10.319 --> 00:02:13.659 much fuller picture. But the data protection 00:02:13.659 --> 00:02:17.400 part here is it's nuanced. They say they will 00:02:17.400 --> 00:02:20.439 not train their models on this health data. And 00:02:20.439 --> 00:02:23.139 they use enhanced encryption, which we should 00:02:23.139 --> 00:02:25.680 pause on because that is explicitly not end -to 00:02:25.680 --> 00:02:27.860 -end encryption. That's a really important distinction 00:02:27.860 --> 00:02:29.780 for you to know. It really is. I mean, enhanced 00:02:29.780 --> 00:02:32.479 encryption is a step up, for sure. But end -to 00:02:32.479 --> 00:02:34.900 -end means only the user and the recipient can 00:02:34.900 --> 00:02:37.280 read it. The fact that they are using E2E, it 00:02:37.280 --> 00:02:39.099 suggests there's still some layer of corporate 00:02:39.099 --> 00:02:41.539 oversight, which is pretty standard. But you 00:02:41.539 --> 00:02:43.080 need to understand that. To their credit, they 00:02:43.080 --> 00:02:44.969 did a lot of prep work. They worked with over 00:02:44.969 --> 00:02:48.110 260 physicians. They reviewed more than 600 ,000 00:02:48.110 --> 00:02:50.710 health responses just to try and get the guardrails 00:02:50.710 --> 00:02:53.009 right. But here's the absolute biggest point, 00:02:53.069 --> 00:02:55.509 the critical warning we have to shout. This is 00:02:55.509 --> 00:02:58.430 not for diagnosis or treatment. The fact is, 00:02:58.469 --> 00:03:01.250 AI hallucination is still possible. It's just 00:03:01.250 --> 00:03:03.550 an inherent risk with these models. Yeah, that 00:03:03.550 --> 00:03:06.189 immediately brings to mind those past AI health 00:03:06.189 --> 00:03:09.509 failures we've all seen. The, you know, the bizarrely 00:03:09.509 --> 00:03:13.289 bad diet advice or even worse, dangerous recommendations. 00:03:14.050 --> 00:03:16.289 This is just such high sensitivity territory 00:03:16.289 --> 00:03:19.349 that sandbox is a necessary barrier. So given 00:03:19.349 --> 00:03:21.310 all that, the dedicated space, the physician 00:03:21.310 --> 00:03:24.050 reviews, the pledge not to train, what do you 00:03:24.050 --> 00:03:26.020 think is... the biggest risk that still remains 00:03:26.020 --> 00:03:28.580 for the user. Hallucinations are still possible, 00:03:28.860 --> 00:03:31.120 meaning we must always double -check AI health 00:03:31.120 --> 00:03:34.199 advice. Okay, so let's shift from protection 00:03:34.199 --> 00:03:36.500 to application. We're seeing builders just move 00:03:36.500 --> 00:03:38.780 incredibly fast using these advanced AI agents 00:03:38.780 --> 00:03:41.500 to create real revenue -generating products. 00:03:41.599 --> 00:03:44.120 That's the real change. An AI agent system isn't 00:03:44.120 --> 00:03:46.620 just a chatbot. It's an AI designed to execute 00:03:46.620 --> 00:03:50.300 complex, multi -step tasks all on its own. Think 00:03:50.300 --> 00:03:52.319 of it like automating an entire workflow, not 00:03:52.319 --> 00:03:55.069 just getting one answer. And the sources show 00:03:55.069 --> 00:03:57.050 how accessible this is getting. There's a guide 00:03:57.050 --> 00:03:59.729 out there and how to build free professional 00:03:59.729 --> 00:04:03.229 full stack apps using Google's AI family. And 00:04:03.229 --> 00:04:06.090 it relies entirely on that agent system to sort 00:04:06.090 --> 00:04:08.569 of glue all the parts together. Right. And that 00:04:08.569 --> 00:04:11.770 ability translates directly to careers. The sources 00:04:11.770 --> 00:04:15.870 highlight five elite high income AI skills right 00:04:15.870 --> 00:04:18.290 now, skills that can create massive leverage, 00:04:18.410 --> 00:04:22.009 even if you're starting from zero with. profitability 00:04:22.009 --> 00:04:25.750 expected by 2026, all driven by this ability 00:04:25.750 --> 00:04:27.870 to manage agents. And this is where it gets really 00:04:27.870 --> 00:04:30.050 cool, I think, in the creative space. We've finally 00:04:30.050 --> 00:04:32.329 had a huge breakthrough in AI video. The big 00:04:32.329 --> 00:04:34.990 problem used to be continuity. You know, a character's 00:04:34.990 --> 00:04:36.870 face would change mid -scene or things in the 00:04:36.870 --> 00:04:39.709 background would just shift around. That inconsistency 00:04:39.709 --> 00:04:42.730 made AI video totally unusable for serious commercial 00:04:42.730 --> 00:04:45.870 work. But now people can create whole films, 00:04:45.889 --> 00:04:49.870 ads, long videos from just one prompt. They're 00:04:49.870 --> 00:04:52.029 using what they call a master grid formula. And 00:04:52.029 --> 00:04:54.170 that formula is specifically designed to maintain 00:04:54.170 --> 00:04:57.250 consistent characters, scenes, lighting, everything 00:04:57.250 --> 00:05:00.569 across every single frame you create. It basically 00:05:00.569 --> 00:05:02.850 solves that continuity problem. It turns a fun 00:05:02.850 --> 00:05:05.670 novelty into a real commercially viable tool. 00:05:05.930 --> 00:05:08.649 So if the goal here is to streamline professional 00:05:08.649 --> 00:05:11.610 production, why is that consistent character 00:05:11.610 --> 00:05:15.009 generation so fundamentally important? Consistency 00:05:15.009 --> 00:05:17.750 makes AI video usable for professional ads and 00:05:17.750 --> 00:05:20.269 long -form storytelling. Okay, let's drill down 00:05:20.269 --> 00:05:22.589 into a really specific automation use case because 00:05:22.589 --> 00:05:24.870 this highlights the architecture you need. We're 00:05:24.870 --> 00:05:28.430 talking about a 247 AI voice receptionist. I 00:05:28.430 --> 00:05:30.389 mean, this is gold for any service business. 00:05:30.550 --> 00:05:33.029 The whole thing runs on this sequential step 00:05:33.029 --> 00:05:35.329 -by -step logic. The goal is just automating 00:05:35.329 --> 00:05:38.250 reception, booking, client management. 247, no 00:05:38.250 --> 00:05:40.290 human needed. And the beauty is breaking the 00:05:40.290 --> 00:05:42.290 conversation down into these conditional blocks. 00:05:42.550 --> 00:05:45.439 So let's map it out. Block one, new caller logic. 00:05:45.680 --> 00:05:48.160 If the caller is new, the AI asks for a name 00:05:48.160 --> 00:05:51.060 and email, then it creates a new profile. Simple. 00:05:51.220 --> 00:05:53.720 Right. Then block two might be booking logic 00:05:53.720 --> 00:05:55.839 for someone who is already in the system. So 00:05:55.839 --> 00:05:58.300 the AI greets them by name, which means it has 00:05:58.300 --> 00:06:00.240 to look them up first, and then it checks the 00:06:00.240 --> 00:06:02.240 calendar before trying to book it. It sounds 00:06:02.240 --> 00:06:03.959 so simple when you lay it out like that, like 00:06:03.959 --> 00:06:06.600 stacking Lego blocks. Yeah. But designing these 00:06:06.600 --> 00:06:10.699 rigid multi -step flows. It requires immense 00:06:10.699 --> 00:06:13.379 focus to plan for every possibility. You know, 00:06:13.420 --> 00:06:15.939 I still wrestle with prompt drift myself. Where 00:06:15.939 --> 00:06:17.920 the AI's instructions just seem to wander off 00:06:17.920 --> 00:06:20.660 track during a long chat, it forgets its original 00:06:20.660 --> 00:06:23.259 mission. Designing this kind of rigid flow that 00:06:23.259 --> 00:06:26.139 stays anchored to those blocks, it requires intense 00:06:26.139 --> 00:06:29.079 focus to keep control. That's my vulnerable admission. 00:06:29.240 --> 00:06:32.040 It takes practice. That struggle is exactly why 00:06:32.040 --> 00:06:34.139 this structure is so valuable. The benefit of 00:06:34.139 --> 00:06:35.759 breaking it down into these small sequential 00:06:35.759 --> 00:06:40.050 blocks is twofold. It ensures reliability and 00:06:40.050 --> 00:06:42.029 lets you apply the exact same logic, the same 00:06:42.029 --> 00:06:44.230 blocks, to almost any other business workflow. 00:06:44.370 --> 00:06:47.350 That just dramatically increases its value. So 00:06:47.350 --> 00:06:49.910 what's the core benefit of breaking down automation 00:06:49.910 --> 00:06:52.870 into small sequential blocks like this? It ensures 00:06:52.870 --> 00:06:55.430 reliability and allows the same logic to be applied 00:06:55.430 --> 00:06:58.300 to other workflows. Okay. Moving into some rapid 00:06:58.300 --> 00:07:00.420 -fire industry trends. If you're thinking about 00:07:00.420 --> 00:07:03.139 a career shift, LinkedIn's list of fast -growing 00:07:03.139 --> 00:07:07.339 AI roles for 2026 is out. And it confirms that 00:07:07.339 --> 00:07:09.839 not all of them need deep coding skills. A lot 00:07:09.839 --> 00:07:12.579 of them are in prompt engineering, data synthesis, 00:07:12.939 --> 00:07:15.899 that kind of thing. And that hiring is directly 00:07:15.899 --> 00:07:19.199 fueling how big companies are scaling up. Sources 00:07:19.199 --> 00:07:22.920 are detailing seven key trends shaping how enterprises 00:07:22.920 --> 00:07:24.980 need to scale their infrastructure this year. 00:07:25.300 --> 00:07:28.129 Whoa! I mean, just imagine the engineering needed 00:07:28.129 --> 00:07:29.910 to scale enterprise infrastructure to handle 00:07:29.910 --> 00:07:33.029 a billion AI queries a day. That's the level 00:07:33.029 --> 00:07:34.610 of demand we're talking about right now. That 00:07:34.610 --> 00:07:38.050 scale is just... It's staggering. It's massive. 00:07:38.250 --> 00:07:39.990 And we're also seeing this interface revolution 00:07:39.990 --> 00:07:42.250 where our whole relationship with screens might 00:07:42.250 --> 00:07:44.790 just fundamentally change. Meta demoed their 00:07:44.790 --> 00:07:47.189 EMG wristband. That's a device that reads the 00:07:47.189 --> 00:07:48.810 electrical signals from your muscle movements. 00:07:49.089 --> 00:07:51.370 Yeah, they showed this at CES. It allows for 00:07:51.370 --> 00:07:54.050 real hands -free finger -based screen control. 00:07:54.230 --> 00:07:56.329 You don't have to tap or swipe or even use your 00:07:56.329 --> 00:07:58.629 voice. It just reads the intent from your muscle 00:07:58.629 --> 00:08:00.709 signals. You can manipulate a virtual interface 00:08:00.709 --> 00:08:02.870 just by thinking about it. And then on the gaming 00:08:02.870 --> 00:08:05.829 side, Razer showed off a totally different kind 00:08:05.829 --> 00:08:08.889 of AI, a Grok -based AI hologram that lives in 00:08:08.889 --> 00:08:11.370 this little glowing capsule. It watches you play, 00:08:11.610 --> 00:08:13.670 it learns your style, and then gives you real 00:08:13.670 --> 00:08:16.149 -time personalized tips. A little digital coach. 00:08:16.410 --> 00:08:19.430 And following that, VC predictions are suggesting 00:08:19.430 --> 00:08:23.050 AI concierges will win big, but that human services 00:08:23.050 --> 00:08:27.110 will still keep their moats. areas AI can't easily 00:08:27.110 --> 00:08:29.329 cross, like relationship building. They even 00:08:29.329 --> 00:08:31.310 floated the radical idea that physical screens 00:08:31.310 --> 00:08:33.769 might just become extinct, replaced by AR and 00:08:33.769 --> 00:08:35.649 VR. And finally, you have to look at the money. 00:08:35.830 --> 00:08:38.309 Anthropic is reportedly raising $10 billion. 00:08:38.710 --> 00:08:40.789 That just underscores the fierce competition 00:08:40.789 --> 00:08:44.429 and the almost unimaginable sums flowing into 00:08:44.429 --> 00:08:46.769 AI leadership right now. So when we look at these 00:08:46.769 --> 00:08:48.950 new interfaces, the wristband or the hologram, 00:08:49.110 --> 00:08:51.909 which one feels closer to a really major shift? 00:08:52.429 --> 00:08:55.350 The hands -free control via the EMG wristband 00:08:55.350 --> 00:08:57.750 could radically change how we interact with all 00:08:57.750 --> 00:09:00.210 technology. So let's end with that jaw -dropping 00:09:00.210 --> 00:09:02.509 health breakthrough from Stanford. They published 00:09:02.509 --> 00:09:04.889 something called Sleep FM. That's a foundation 00:09:04.889 --> 00:09:08.330 model. A massive AI trained on tons of data to 00:09:08.330 --> 00:09:10.990 be super adaptable. And it can predict over 130 00:09:10.990 --> 00:09:13.690 diseases using nothing but overnight recordings. 00:09:14.049 --> 00:09:16.169 And this is where the scale is just so important. 00:09:16.429 --> 00:09:20.419 They trained it on 600 ,000 hours of sleep. For 00:09:20.419 --> 00:09:23.100 more than 65 ,000 different patients, that is 00:09:23.100 --> 00:09:26.120 just a gigantic data pool. And the AI is analyzing 00:09:26.120 --> 00:09:28.220 signals like brain waves, breathing patterns, 00:09:28.399 --> 00:09:30.240 heart rate, these tiny muscle twitches that happen 00:09:30.240 --> 00:09:32.559 when you sleep. But the true genius here wasn't 00:09:32.559 --> 00:09:35.159 just collecting all that sleep data. The crucial 00:09:35.159 --> 00:09:38.100 and I'm sure monumentally expensive step -step 00:09:38.100 --> 00:09:41.480 was matching that specific data against 25 years 00:09:41.480 --> 00:09:44.039 of actual documented patient health records. 00:09:44.259 --> 00:09:46.820 That's the gold standard. That longitudinal data 00:09:46.820 --> 00:09:48.460 link is everything because it lets the model 00:09:48.460 --> 00:09:51.080 look forward, not just at the present. The model 00:09:51.080 --> 00:09:53.440 flags these moments when signals are out of sync, 00:09:53.500 --> 00:09:55.759 like say your brain is in deep sleep, but your 00:09:55.759 --> 00:09:58.120 heart is suddenly racing. That mismatch acts 00:09:58.120 --> 00:10:00.970 as a red flag for future disease. And the accuracy 00:10:00.970 --> 00:10:03.590 stats are just. They're genuinely remarkable. 00:10:03.990 --> 00:10:06.350 We're talking highly accurate predictions for 00:10:06.350 --> 00:10:09.990 really serious conditions. 89 % for Parkinson's, 00:10:09.990 --> 00:10:13.470 85 % for dementia, 81 % for heart attacks. And 00:10:13.470 --> 00:10:16.169 predicting long -term risk of death, 84 % accuracy. 00:10:16.490 --> 00:10:19.049 Right. And traditional detection needs blood 00:10:19.049 --> 00:10:21.909 work or scans or these uncomfortable visits. 00:10:22.730 --> 00:10:24.590 This just needs to plug into your sleep signals. 00:10:24.990 --> 00:10:27.350 That decades -long health history, that's the 00:10:27.350 --> 00:10:29.570 killer advantage. It turns the AI from a simple 00:10:29.570 --> 00:10:31.929 detector into a true predictor. So what makes 00:10:31.929 --> 00:10:34.350 that decades -long data set so essential for 00:10:34.350 --> 00:10:36.629 a model like this one? Longitudinal data allows 00:10:36.629 --> 00:10:39.389 the AI to predict future outcomes, not just detect 00:10:39.389 --> 00:10:41.389 current conditions. So if we tie this all together 00:10:41.389 --> 00:10:43.610 for you, what we've seen is AI moving to protect 00:10:43.610 --> 00:10:45.929 extremely sensitive data with the health sandbox, 00:10:46.210 --> 00:10:49.129 while also paradoxically unlocking this profound, 00:10:49.529 --> 00:10:52.279 almost unbelievable predictive power with sleep 00:10:52.279 --> 00:10:55.000 FM. And right in the middle, we saw how builders 00:10:55.000 --> 00:10:57.179 are streamlining these high leverage workflows 00:10:57.179 --> 00:11:00.120 like that 2047 receptionist. And they're pushing 00:11:00.120 --> 00:11:02.100 the limits of how we interact with computers 00:11:02.100 --> 00:11:04.940 from the subtle signals of that wristband to 00:11:04.940 --> 00:11:08.399 the idea of a hologram coach. The future isn't 00:11:08.399 --> 00:11:11.860 just fast, it's hyper specific and deeply integrated. 00:11:12.399 --> 00:11:14.620 So we encourage you to think about those sequential 00:11:14.620 --> 00:11:17.100 rigid logic flows we talked about for that AI 00:11:17.100 --> 00:11:19.799 receptionist. How can you apply that intense 00:11:19.799 --> 00:11:22.799 focus on step -by -step automation to a complex 00:11:22.799 --> 00:11:24.899 process in your own professional life this week? 00:11:25.120 --> 00:11:27.259 And here's a final thought to mull over, especially 00:11:27.259 --> 00:11:29.899 with those accuracy numbers. If your sleep can 00:11:29.899 --> 00:11:32.419 reveal your risk of death with 84 % accuracy, 00:11:32.779 --> 00:11:35.000 what is the fundamental value of a traditional 00:11:35.000 --> 00:11:38.139 one -time physical exam? When the secret to your 00:11:38.139 --> 00:11:40.100 long -term health is already streaming silently 00:11:40.100 --> 00:11:41.440 every single night.