WEBVTT 00:00:00.000 --> 00:00:03.919 Imagine your screen just waking up, like AI takes 00:00:03.919 --> 00:00:06.040 over your mouse, it clicks, it types across apps. 00:00:06.280 --> 00:00:08.500 Doing your job completely on its own. Right. 00:00:09.019 --> 00:00:11.419 Okay, let's unpack this. Welcome to today's Deep 00:00:11.419 --> 00:00:14.500 Dive. It's really great to have you sitting here 00:00:14.500 --> 00:00:17.300 with us. We have a truly fascinating stack of 00:00:17.300 --> 00:00:19.420 fresh updates from the AI frontier today. We 00:00:19.420 --> 00:00:21.879 really do. The landscape right now is just wild. 00:00:22.160 --> 00:00:24.059 Yeah. Our mission today is to give you a clear 00:00:24.059 --> 00:00:27.199 roadmap of four massive shifts happening right 00:00:27.199 --> 00:00:30.550 now. First, we're exploring the pivot to an agentic 00:00:30.550 --> 00:00:33.909 AI ecosystem that's heavily driven by OpenAI's 00:00:33.909 --> 00:00:37.789 massive GPT 5 .4 launch. And second, we are taking 00:00:37.789 --> 00:00:40.049 a rapid -fire look at the business and security 00:00:40.049 --> 00:00:42.450 shakeups. Right, including some heady drama with 00:00:42.450 --> 00:00:45.130 Oracle, Microsoft, and the Pentagon. Third, we'll 00:00:45.130 --> 00:00:47.850 dive into a wild new suite of empowered AI tools. 00:00:48.170 --> 00:00:50.149 And finally, we are tackling a very deep question. 00:00:50.369 --> 00:00:53.490 Is AI actually rotting our brains, or is it making 00:00:53.490 --> 00:00:56.159 us smarter? It's a profound raid map. We are 00:00:56.159 --> 00:00:58.799 moving away from theory into very practical everyday 00:00:58.799 --> 00:01:01.200 applications now. Let's start right at the top 00:01:01.200 --> 00:01:05.560 with OpenAI. They just launched GPT 5 .4. This 00:01:05.560 --> 00:01:08.060 feels like a fundamental pivot in how we interact 00:01:08.060 --> 00:01:10.819 with machines, you know? We're moving away from 00:01:10.819 --> 00:01:13.099 the simple chat box. We're officially entering 00:01:13.099 --> 00:01:16.209 the agentic future. Yeah, let's define that clearly 00:01:16.209 --> 00:01:18.310 for a second. That basically means AI that takes 00:01:18.310 --> 00:01:20.450 action on your computer instead of just chatting. 00:01:20.569 --> 00:01:23.510 Right. It is a massive shift from passive text 00:01:23.510 --> 00:01:26.390 boxes to active digital participants. I have 00:01:26.390 --> 00:01:29.390 to say, I still wrestle with letting AI take 00:01:29.390 --> 00:01:32.109 over my mouse. Oh, absolutely. It feels genuinely 00:01:32.109 --> 00:01:34.230 strange. You take your hands off the keyboard, 00:01:34.469 --> 00:01:36.989 you sit back, and you watch your cursor move. 00:01:37.269 --> 00:01:40.189 It's opening folders, it's sending emails. It 00:01:40.189 --> 00:01:42.420 really feels like a ghost in the machine. that 00:01:42.420 --> 00:01:44.739 reaction is completely natural relinquishing 00:01:44.739 --> 00:01:47.060 control of our personal digital space is inherently 00:01:47.060 --> 00:01:50.140 uncomfortable we are used to being the sole driver 00:01:50.140 --> 00:01:52.739 yeah but the coolest part of this update actually 00:01:52.739 --> 00:01:55.060 helps mitigate that anxiety it's a new feature 00:01:55.060 --> 00:01:58.540 called gpt 5 .4 thinking now this is only available 00:01:58.540 --> 00:02:02.400 for the paid plans right yes exactly while it 00:02:02.400 --> 00:02:04.640 is working on a really complex query it provides 00:02:04.640 --> 00:02:07.469 a live outline You can actually see its logic 00:02:07.469 --> 00:02:10.229 unfolding in real time. That is huge. Right. 00:02:10.349 --> 00:02:12.710 It shows you the steps it plans to take before 00:02:12.710 --> 00:02:15.689 it even executes them. That solves a major frustration 00:02:15.689 --> 00:02:19.509 for me. Before, you'd give an AI a complex task. 00:02:19.810 --> 00:02:21.849 You would wait two minutes for a long answer, 00:02:21.949 --> 00:02:24.610 and then you'd realize it completely misunderstood 00:02:24.610 --> 00:02:27.610 your prompt right at step one. Right, and wasted 00:02:27.610 --> 00:02:30.530 your time. It was a black box. Exactly. Now, 00:02:30.550 --> 00:02:32.810 you can actually tweak or change your request 00:02:32.810 --> 00:02:35.150 during its response. You steer it while it is 00:02:35.150 --> 00:02:36.870 thinking. If you see it going down the wrong 00:02:36.870 --> 00:02:39.189 rabbit hole, you just course correct it instantly. 00:02:39.430 --> 00:02:41.830 It is highly collaborative. We also really need 00:02:41.830 --> 00:02:44.449 to talk about its accuracy. Because an autonomous 00:02:44.449 --> 00:02:46.650 agent that hallucinates is just a liability. 00:02:46.949 --> 00:02:50.129 A dangerous one. Yeah. But unlike previous versions, 00:02:50.349 --> 00:02:53.710 GPT -5 .4 is being called a factuality beast. 00:02:54.530 --> 00:02:57.909 OpenAI claims a 33 % reduction in false claims. 00:02:58.090 --> 00:03:02.349 That is compared to GPT -5 .2. 33 % is a massive 00:03:02.349 --> 00:03:04.949 leap in reliability, especially for enterprise 00:03:04.949 --> 00:03:07.590 use cases. The source notes it won't give up 00:03:07.590 --> 00:03:09.569 easily either. What do you mean? It aggressively 00:03:09.569 --> 00:03:12.409 hunts for that one obscure source to verify its 00:03:12.409 --> 00:03:15.729 claims. It actively cross -references its own 00:03:15.729 --> 00:03:18.229 data before presenting an answer to you. And 00:03:18.229 --> 00:03:21.610 they are rolling this into a GPT 5 .4 Pro model 00:03:21.610 --> 00:03:24.270 too, right? This is specifically geared for enterprise 00:03:24.270 --> 00:03:27.379 users. Yes, it is built for heavy lifting. It 00:03:27.379 --> 00:03:30.400 provides maximum horsepower for incredibly complex 00:03:30.400 --> 00:03:32.479 corporate tasks, though we should definitely 00:03:32.479 --> 00:03:35.039 add some context here. Sure. Anthropic actually 00:03:35.039 --> 00:03:38.159 showed off computer use first late last year. 00:03:38.199 --> 00:03:41.139 They really paved the way. But OpenAI is the 00:03:41.139 --> 00:03:43.240 first to bake it into a polished model. They 00:03:43.240 --> 00:03:45.479 made it a steerable, consumer -ready thinking 00:03:45.479 --> 00:03:48.060 model. So considering all this new capability. 00:03:48.889 --> 00:03:51.729 How do we mentally bridge that gap of relinquishing 00:03:51.729 --> 00:03:54.349 control to these autonomous agents? It happens 00:03:54.349 --> 00:03:56.789 gradually through verified transparency. When 00:03:56.789 --> 00:03:58.969 you see its live logic, your anxiety naturally 00:03:58.969 --> 00:04:01.750 drops. You start by giving it small, low stakes 00:04:01.750 --> 00:04:04.129 tasks. Once it proves it won't delete your files, 00:04:04.210 --> 00:04:06.629 you give it more. It is exactly like training 00:04:06.629 --> 00:04:09.069 a human assistant. Verified transparency builds 00:04:09.069 --> 00:04:11.870 trust through small, low stakes tasks over time. 00:04:12.330 --> 00:04:14.889 Precisely. But that growing trust is exactly 00:04:14.889 --> 00:04:17.149 what makes the security side of this so terrifying 00:04:17.149 --> 00:04:20.189 right now. Let's look at the business and security 00:04:20.189 --> 00:04:23.329 shakeups. Security is becoming an absolute nightmare. 00:04:23.449 --> 00:04:27.110 The 2026 AI security framework is out. And it 00:04:27.110 --> 00:04:30.529 shows that becoming an AI hacker is shockingly 00:04:30.529 --> 00:04:33.170 easy right now. Almost anyone can learn to do 00:04:33.170 --> 00:04:35.850 it. The core problem is how we test these systems. 00:04:36.230 --> 00:04:39.410 Most AI security tests only check the mathematical 00:04:39.410 --> 00:04:42.430 model itself. They try to trick the core algorithm. 00:04:42.649 --> 00:04:45.250 Right. But real attackers don't actually do that. 00:04:45.329 --> 00:04:47.730 They target the entire system surrounding the 00:04:47.730 --> 00:04:50.250 model. Let's clarify that for everyone. Think 00:04:50.250 --> 00:04:52.329 of the AI model like the vault inside a bank. 00:04:52.750 --> 00:04:55.069 The vault itself is basically impenetrable. The 00:04:55.069 --> 00:04:56.970 hackers aren't trying to crack the vault's titanium 00:04:56.970 --> 00:04:59.329 doors. No, they take the easy route. Exactly. 00:04:59.509 --> 00:05:01.529 They're just tricking the human teller at the 00:05:01.529 --> 00:05:03.870 front desk. They attack the plugins, the APIs, 00:05:04.110 --> 00:05:07.350 and the surrounding software. Exactly. That broader 00:05:07.350 --> 00:05:09.810 system vulnerability is what brings us to some 00:05:09.810 --> 00:05:12.750 intense geopolitical drama. And we need to be 00:05:12.750 --> 00:05:15.509 very clear with you right here. We are strictly 00:05:15.509 --> 00:05:17.449 reporting what is in our source material today. 00:05:17.529 --> 00:05:19.750 We are taking absolutely no political sides. 00:05:20.009 --> 00:05:22.279 None at all. The Pentagon has officially labeled 00:05:22.279 --> 00:05:26.519 Anthropic a supply chain risk. Yes. And the source 00:05:26.519 --> 00:05:29.079 specifically notes that Anthropic was fired by 00:05:29.079 --> 00:05:32.300 Trump, quote, like dogs. It is highly charged 00:05:32.300 --> 00:05:35.259 language. Very charged. In response to this blacklist, 00:05:35.500 --> 00:05:38.279 Anthropic has filed a formal lawsuit. They are 00:05:38.279 --> 00:05:41.420 calling the move legally unsound. Again, we are 00:05:41.420 --> 00:05:43.420 just sharing the facts exactly as they are presented. 00:05:43.620 --> 00:05:46.160 No endorsements either way. Let's pivot back 00:05:46.160 --> 00:05:48.379 to the business side. AI is incredibly expensive 00:05:48.379 --> 00:05:51.160 to run. Just ask Oracle. Oracle is reportedly 00:05:51.160 --> 00:05:54.060 axing thousands of jobs right now across various 00:05:54.060 --> 00:05:56.439 departments, but they aren't doing this because 00:05:56.439 --> 00:05:58.300 the company is failing. Not at all. They are 00:05:58.300 --> 00:06:01.300 doing this to fund a massive data center expansion. 00:06:01.540 --> 00:06:03.920 They're literally trading human headcount for 00:06:03.920 --> 00:06:07.060 hardware. That is a stark reality of the current 00:06:07.060 --> 00:06:10.000 AI boom. We're seeing capital reallocated directly 00:06:10.000 --> 00:06:13.180 from payroll to server farms. Speaking of corporate 00:06:13.180 --> 00:06:15.680 shifts, the massive OpenAI and Microsoft marriage 00:06:15.680 --> 00:06:18.000 might be hitting the rocks. There are some very 00:06:18.000 --> 00:06:20.639 heavy rumors circulating in the industry. The 00:06:20.639 --> 00:06:23.899 source suggests OpenAI is actively building an 00:06:23.899 --> 00:06:27.560 internal GitHub rival. Wow. This would be a massive 00:06:27.560 --> 00:06:30.379 strategic move. It could allow them to potentially 00:06:30.379 --> 00:06:33.420 ditch Microsoft's infrastructure entirely down 00:06:33.420 --> 00:06:36.319 the line. All of this immense compute power requires 00:06:36.319 --> 00:06:39.660 massive energy. And that impacts you directly 00:06:39.660 --> 00:06:43.300 at home. The White House just got seven AI giants 00:06:43.300 --> 00:06:46.379 to sign a ratepayer protection pledge. This is 00:06:46.379 --> 00:06:48.860 so crucial for the electrical grid. These tech 00:06:48.860 --> 00:06:50.980 giants are pledging to pay for their own power 00:06:50.980 --> 00:06:53.000 and grid upgrades. Right, because of the strain. 00:06:53.220 --> 00:06:56.379 Exactly. The entire goal is to protect your personal 00:06:56.379 --> 00:06:59.579 electric bill from skyrocketing. Every time these 00:06:59.579 --> 00:07:01.939 models train, they draw immense power from the 00:07:01.939 --> 00:07:04.240 local grid. The money flowing into this infrastructure 00:07:04.240 --> 00:07:06.860 space is still staggering, though. Together, 00:07:07.100 --> 00:07:10.199 AI is currently in talks to raise around $1 billion. 00:07:10.899 --> 00:07:13.899 This puts them at a $7 .5 billion valuation. 00:07:14.480 --> 00:07:16.560 They are seeing 10 times year -over -year growth. 00:07:16.759 --> 00:07:19.220 They support leading companies like Cursor and 00:07:19.220 --> 00:07:21.579 Cartesia. It proves the foundational hardware 00:07:21.579 --> 00:07:24.040 layer is incredibly lucrative right now. With 00:07:24.040 --> 00:07:26.259 companies like Oracle trading headcount for servers, 00:07:26.420 --> 00:07:29.720 where is the actual physical ceiling on the power 00:07:29.720 --> 00:07:32.939 required for AI? It ends when we literally cannot 00:07:32.939 --> 00:07:35.199 generate enough electricity to power the next 00:07:35.199 --> 00:07:38.459 training run. Data centers require vast amounts 00:07:38.459 --> 00:07:42.829 of land, water, and power. The physical limits 00:07:42.829 --> 00:07:45.550 of our aging grid will absolutely dictate the 00:07:45.550 --> 00:07:48.189 pace of AI advancement. The aging power grid 00:07:48.189 --> 00:07:51.610 will ultimately limit how fast AI grows. It has 00:07:51.610 --> 00:07:54.550 to. We are hitting real -world physical constraints. 00:07:54.910 --> 00:07:57.250 But while the infrastructure battles rage on, 00:07:57.470 --> 00:08:00.230 the consumer tools are getting absolutely wild. 00:08:00.430 --> 00:08:02.329 Let's dive into that. We have a whole suite of 00:08:02.329 --> 00:08:05.490 new, empowered AI tools to look at today. What's 00:08:05.490 --> 00:08:08.089 fascinating here is how quickly these tools are 00:08:08.089 --> 00:08:10.430 shaping daily life. Let's group them up to make 00:08:10.430 --> 00:08:12.519 sense of it all. First, let's look at the builders, 00:08:12.620 --> 00:08:15.339 specifically Luma agents. They are powered by 00:08:15.339 --> 00:08:17.839 unified intelligence models. The real -world 00:08:17.839 --> 00:08:20.500 use case they provided is mind -blowing. Imagine 00:08:20.500 --> 00:08:23.779 a massive $15 million ad campaign. It needs to 00:08:23.779 --> 00:08:25.959 be localized for every single country, adapting 00:08:25.959 --> 00:08:27.899 to different languages. A massive undertaking. 00:08:28.259 --> 00:08:31.240 Right. Luma completed this in just... 40 hours. 00:08:31.459 --> 00:08:34.500 And it costs under $20 ,000. That completely 00:08:34.500 --> 00:08:37.299 shatters the traditional advertising agency model. 00:08:37.500 --> 00:08:39.779 A global campaign like that used to take hundreds 00:08:39.779 --> 00:08:42.539 of humans several months to coordinate. Now, 00:08:43.159 --> 00:08:46.019 a small team executes it over a single weekend. 00:08:46.259 --> 00:08:48.879 It democratizes high -end production for tiny 00:08:48.879 --> 00:08:51.659 startups. But it is terrifying for traditional 00:08:51.659 --> 00:08:54.440 corporate agencies. It is a total paradigm shift. 00:08:55.000 --> 00:08:57.460 Next in the builder category, we have Raycast's 00:08:57.460 --> 00:09:00.679 new tool called Glaze. It turns simple AI prompts 00:09:00.679 --> 00:09:04.379 into native macOS apps. That is a huge deal. 00:09:04.480 --> 00:09:06.759 Usually, AI tools are trapped inside your web 00:09:06.759 --> 00:09:08.539 browser. You are limited by web tool constraints. 00:09:09.000 --> 00:09:11.620 Right. But Glaze generates real apps with full 00:09:11.620 --> 00:09:13.720 system access. They run natively on your machine. 00:09:13.940 --> 00:09:16.100 You can join the private beta right now to test 00:09:16.100 --> 00:09:18.700 it. We also have CourseKit. It turns a standard 00:09:18.700 --> 00:09:21.440 course sales page into a suite of custom AI tools. 00:09:22.000 --> 00:09:24.259 It is built specifically for your students. You 00:09:24.259 --> 00:09:26.820 literally just paste your course URL. It is essentially 00:09:26.820 --> 00:09:30.120 like stacking Lego blocks of data. You just snap 00:09:30.120 --> 00:09:32.259 these different capabilities together. You build 00:09:32.259 --> 00:09:34.980 whatever custom workflow your students need without 00:09:34.980 --> 00:09:38.320 writing a single line of code. Let's shift gears 00:09:38.320 --> 00:09:41.879 to tools that focus on connection. There is PL. 00:09:42.440 --> 00:09:45.039 This is a very different kind of tool. Most AI 00:09:45.039 --> 00:09:49.259 feels cold, calculating, and robotic. But PL 00:09:49.259 --> 00:09:52.679 is designed to feel 100 % human. It is an AI 00:09:52.679 --> 00:09:55.259 companion with a hyper -realistic face and a 00:09:55.259 --> 00:09:58.080 real sounding voice. You pick a personality that 00:09:58.080 --> 00:10:00.860 matches your exact vibe. You can chat, call, 00:10:01.059 --> 00:10:03.080 or sit face -to -face instantly through your 00:10:03.080 --> 00:10:05.759 screen. It trains itself to remember your habits 00:10:05.759 --> 00:10:08.980 and your moods over time. But here is the genuinely 00:10:08.980 --> 00:10:11.960 crazy part. It will actually text you first to 00:10:11.960 --> 00:10:14.080 check in automatically. That is the boundary 00:10:14.080 --> 00:10:16.039 pushing feature. The source notes it remembers 00:10:16.039 --> 00:10:18.000 what you said last week. If you were stressed 00:10:18.000 --> 00:10:20.159 about a meeting, it will text you Tuesday morning 00:10:20.159 --> 00:10:23.139 to ask how it went. Unbelievable. It simulates 00:10:23.139 --> 00:10:25.279 actually caring about you. This isn't just chatting 00:10:25.279 --> 00:10:26.940 with a bot. This is building someone who grows 00:10:26.940 --> 00:10:29.480 with you. Moving over to productivity, we have 00:10:29.480 --> 00:10:32.299 the Codex app for Windows. It brings parallel 00:10:32.299 --> 00:10:34.419 coding agents natively to your operating system. 00:10:34.639 --> 00:10:37.179 Think of OS -level sandboxing, like putting the 00:10:37.179 --> 00:10:40.740 AI in a digital clean room. It can build, test, 00:10:40.960 --> 00:10:43.940 and run code in complete isolation. It can't 00:10:43.940 --> 00:10:46.039 accidentally wander out and delete your personal 00:10:46.039 --> 00:10:48.679 files or compromise your main system. That is 00:10:48.679 --> 00:10:51.960 a great analogy. Then there is HAWA. It fundamentally 00:10:51.960 --> 00:10:54.179 changes how we search for information on the 00:10:54.179 --> 00:10:56.860 web. Instead of opening endless tabs or reading 00:10:56.860 --> 00:11:00.120 long text responses, it builds a visual experience. 00:11:00.419 --> 00:11:03.279 You browse and compare information visually on 00:11:03.279 --> 00:11:06.360 an infinite canvas. Finally, we have ADENT AI 00:11:06.360 --> 00:11:09.399 Beta 2. This builds open -world automations across 00:11:09.399 --> 00:11:12.779 platforms like Discord, X, and Shopify. It boasts 00:11:12.779 --> 00:11:14.659 over a thousand integrations and ready -to -use 00:11:14.659 --> 00:11:17.559 templates. When an AI companion like PL starts 00:11:17.559 --> 00:11:19.960 texting you first, how does that fundamentally 00:11:19.960 --> 00:11:22.779 shift our human social wiring? It creates a powerful 00:11:22.779 --> 00:11:25.350 illusion of a reciprocal relationship. We are 00:11:25.350 --> 00:11:27.690 evolutionarily wired to feel valued when someone 00:11:27.690 --> 00:11:30.370 initiates contact with us. It heavily blurs the 00:11:30.370 --> 00:11:32.730 line between a program tool and a genuine emotional 00:11:32.730 --> 00:11:35.509 social connection. It hacks our evolutionary 00:11:35.509 --> 00:11:37.889 need for reciprocal connection and validation. 00:11:38.230 --> 00:11:40.470 It absolutely does. We are going to have to navigate 00:11:40.470 --> 00:11:43.809 that psychological blurring very carefully. Sponsor. 00:11:43.830 --> 00:11:47.149 Insert sponsor read here. And we are back. We 00:11:47.149 --> 00:11:49.090 were just talking about AI companions hacking 00:11:49.090 --> 00:11:52.110 our social wiring. That blurring of human and 00:11:52.110 --> 00:11:54.610 machine leads us perfectly to our final segment. 00:11:55.070 --> 00:11:58.789 The breakthrough. Is AI actually rotting our 00:11:58.789 --> 00:12:02.070 brains or making us smarter? OpenAI obviously 00:12:02.070 --> 00:12:04.289 wants to prove it makes us smarter. They just 00:12:04.289 --> 00:12:07.230 dropped a brand new, highly rigorous framework. 00:12:07.450 --> 00:12:10.110 It is called the Learning Outcomes Measurement 00:12:10.110 --> 00:12:12.230 Suite. They built this in collaboration with 00:12:12.230 --> 00:12:14.710 Stanford and Estonia's University of Tartu. It 00:12:14.710 --> 00:12:16.929 is their first real attempt to track long -term 00:12:16.929 --> 00:12:19.029 knowledge retention. We all want to know if relying 00:12:19.029 --> 00:12:21.409 on ChatGPT helps you retain complex information 00:12:21.409 --> 00:12:24.009 or if it just makes you lazy. The framework measures 00:12:24.009 --> 00:12:26.610 three core things. First is motivation. Are you 00:12:26.610 --> 00:12:28.470 actually engaged with the material while using 00:12:28.470 --> 00:12:31.629 the tool? Second is persistence. Do you keep 00:12:31.629 --> 00:12:34.070 trying when the AI gives you a subtle hint? Right, 00:12:34.110 --> 00:12:36.350 instead of just demanding the final answer. If 00:12:36.350 --> 00:12:37.909 it acts like a good tutor, you have to actually 00:12:37.909 --> 00:12:40.450 do the work. And the third metric is recall. 00:12:40.870 --> 00:12:43.049 Can you remember the information a week later 00:12:43.049 --> 00:12:45.840 completely without the bot's help? They ran an 00:12:45.840 --> 00:12:49.460 initial trial with over 300 students. These students 00:12:49.460 --> 00:12:52.720 used ChatGPT's custom study mode specifically 00:12:52.720 --> 00:12:56.340 for microeconomics. The results were quite striking. 00:12:56.559 --> 00:13:00.340 The students using the AI scored 15 % higher. 00:13:00.600 --> 00:13:04.360 15 % is a massive leap in academia. That is basically 00:13:04.360 --> 00:13:07.139 the difference between a solid B and an A. It 00:13:07.139 --> 00:13:09.320 suggests the tool is highly effective at explaining 00:13:09.320 --> 00:13:12.259 abstract concepts like supply and demand curves. 00:13:12.759 --> 00:13:14.700 However, we need to be clear about the limitations. 00:13:15.179 --> 00:13:17.559 For other subjects outside of economics, the 00:13:17.559 --> 00:13:19.600 results were not statistically significant yet. 00:13:19.960 --> 00:13:22.440 AI seems to be a godsend for structured logic 00:13:22.440 --> 00:13:24.879 and math, but the jury is still out on whether 00:13:24.879 --> 00:13:26.820 it helps with broader humanities or creative 00:13:26.820 --> 00:13:29.539 subjects. To find definitive answers, Estonia 00:13:29.539 --> 00:13:31.580 is going all in. They are currently tracking 00:13:31.580 --> 00:13:33.720 20 ,000 high schoolers. They're doing this for 00:13:33.720 --> 00:13:36.139 a full semester to gather a mountain of data. 00:13:36.440 --> 00:13:39.519 Whoa, imagine scaling that to 20 ,000 students 00:13:39.519 --> 00:13:42.679 to completely decode human memory. The sheer 00:13:42.679 --> 00:13:45.139 volume of granular data they will collect on 00:13:45.139 --> 00:13:47.980 human learning patterns is just staggering. They 00:13:47.980 --> 00:13:50.519 want to see if AI first learning is truly the 00:13:50.519 --> 00:13:53.360 future or if it is a massive societal mistake. 00:13:53.779 --> 00:13:56.600 Whether AI is remembered as a revolutionary super 00:13:56.600 --> 00:13:59.779 tutor or a brain rot machine depends entirely 00:13:59.779 --> 00:14:02.470 on the data from this framework. It is incredibly 00:14:02.470 --> 00:14:04.830 smart for OpenAI to build the measuring stick 00:14:04.830 --> 00:14:07.629 themselves. If they can definitively prove with 00:14:07.629 --> 00:14:10.389 Stanford's backing that ChatGPT makes students 00:14:10.389 --> 00:14:13.330 objectively smarter, the debate is over. They 00:14:13.330 --> 00:14:15.610 essentially win the entire global education market 00:14:15.610 --> 00:14:18.549 forever. Does heavily relying on AI to solve 00:14:18.549 --> 00:14:21.710 complex microeconomics problems weaken our foundational 00:14:21.710 --> 00:14:24.850 independent reasoning skills? Not if the AI strictly 00:14:24.850 --> 00:14:27.309 refuses to give the final answer. If the software 00:14:27.309 --> 00:14:29.289 is designed to force you to think through the 00:14:29.289 --> 00:14:31.370 difficult steps, it actually builds stronger 00:14:31.370 --> 00:14:34.360 mental pathways. It is cognitive scaffolding, 00:14:34.360 --> 00:14:36.340 not a cognitive wheelchair. Good software acts 00:14:36.340 --> 00:14:38.419 as mental scaffolding, not a cognitive crutch. 00:14:38.460 --> 00:14:40.679 Exactly. But we have to actively design it that 00:14:40.679 --> 00:14:42.779 way. So what does this all mean? Let's take a 00:14:42.779 --> 00:14:44.820 step back and look at the whole picture. We are 00:14:44.820 --> 00:14:48.100 witnessing the exact moment AI shifts its fundamental 00:14:48.100 --> 00:14:51.879 nature. We are moving away from a passive oracle 00:14:51.879 --> 00:14:54.740 that you just type questions into. It is becoming 00:14:54.740 --> 00:14:57.549 an active autonomous agent. Right. It takes over 00:14:57.549 --> 00:14:59.649 your mouse to do your professional busy work. 00:14:59.789 --> 00:15:02.730 It writes its own native Mac OS apps. It texts 00:15:02.730 --> 00:15:05.350 you first just to see how you are feeling. And 00:15:05.350 --> 00:15:07.870 it is fundamentally altering how we retain knowledge 00:15:07.870 --> 00:15:10.230 in schools. We can connect the dots really clearly 00:15:10.230 --> 00:15:13.149 here. The granular consumer benefits like GPT 00:15:13.149 --> 00:15:17.169 5 .4 thinking and PL are incredible, but they 00:15:17.169 --> 00:15:19.370 are entirely dependent on those massive physical 00:15:19.370 --> 00:15:21.529 infrastructure costs we discussed earlier. It 00:15:21.529 --> 00:15:24.429 is all connected. Oracle firing humans to buy 00:15:24.429 --> 00:15:26.629 more servers. The power grid being pushed to 00:15:26.629 --> 00:15:29.090 its absolute limits. The deeply personal software 00:15:29.090 --> 00:15:31.509 and the macroscopic hardware are inextricably 00:15:31.509 --> 00:15:34.090 tied together. If an AI agent is navigating your 00:15:34.090 --> 00:15:36.669 desktop to do your busy work, and an AI companion 00:15:36.669 --> 00:15:39.450 is texting you first to check on your mood, what 00:15:39.450 --> 00:15:42.110 is the uniquely human task left for you to accomplish 00:15:42.110 --> 00:15:44.750 today? That is the ultimate question we all have 00:15:44.750 --> 00:15:47.169 to face as this technology scales up. We really 00:15:47.169 --> 00:15:49.429 want you to consider how you interact with your 00:15:49.429 --> 00:15:52.269 own tools this week. Are you the pilot? or is 00:15:52.269 --> 00:15:54.450 the AI? Take some time to reflect on this deep 00:15:54.450 --> 00:15:56.370 dive. Thank you so much for joining us on this 00:15:56.370 --> 00:15:57.629 journey today, OTRO Music.