WEBVTT 00:00:00.000 --> 00:00:02.100 It's strange to think about our current reality. 00:00:02.879 --> 00:00:06.200 Beat. We have robots learning to read physical 00:00:06.200 --> 00:00:08.800 thermometers now. Yeah, which is just incredible. 00:00:09.080 --> 00:00:11.439 Right. But at the exact same moment, executives 00:00:11.439 --> 00:00:14.140 are scheming. Yeah. They leak memos to sabotage 00:00:14.140 --> 00:00:16.780 billion -dollar company valuations. It's a wild 00:00:16.780 --> 00:00:19.699 split screen of technological progress. We are 00:00:19.699 --> 00:00:21.780 living in two completely different realities 00:00:21.780 --> 00:00:25.660 right now. One is pure science, and one is just 00:00:25.660 --> 00:00:29.219 cutthroat business. Welcome to the Deep Dive. 00:00:29.579 --> 00:00:32.020 I am glad you are here with us today. Yeah, it's 00:00:32.020 --> 00:00:33.939 going to be a good one. We are exploring the 00:00:33.939 --> 00:00:37.460 rapidly maturing AI ecosystem, and we are going 00:00:37.460 --> 00:00:40.380 to start at the highest boardroom level. We have 00:00:40.380 --> 00:00:43.780 a massive OpenAI corporate leak to unpack. Oh, 00:00:43.780 --> 00:00:46.520 yeah, a really big one. From there, we will move 00:00:46.520 --> 00:00:49.380 into a major software shift. AI is going from 00:00:49.380 --> 00:00:52.100 simple chat to autonomous employees. Which totally 00:00:52.100 --> 00:00:54.880 changes everything. Exactly. And finally, we 00:00:54.880 --> 00:00:57.140 will look at a very physical evolution. These 00:00:57.140 --> 00:00:58.939 digital agents are breaking out of traditional 00:00:58.939 --> 00:01:01.359 software. They are physically interacting with 00:01:01.359 --> 00:01:04.340 our tangible real world. It's a massive structural 00:01:04.340 --> 00:01:07.060 shift across the board. I mean, every layer of 00:01:07.060 --> 00:01:09.659 the technology stack is basically changing. Let's 00:01:09.659 --> 00:01:11.780 start with the aggressive business maneuvers 00:01:11.780 --> 00:01:13.959 currently happening. Because these corporate 00:01:13.959 --> 00:01:17.709 wars directly shape the tools you use. Right. 00:01:17.829 --> 00:01:20.010 They dictate the market. A leaked internal memo 00:01:20.010 --> 00:01:23.329 from OpenAI just surfaced recently. It comes 00:01:23.329 --> 00:01:26.310 directly from their chief revenue officer. Her 00:01:26.310 --> 00:01:29.409 name is Denise Dresser. She claims Anthropic's 00:01:29.409 --> 00:01:33.870 $30 billion revenue run rate is inflated. She 00:01:33.870 --> 00:01:36.090 basically called their financial success a complete 00:01:36.090 --> 00:01:39.390 mirage. Calling it a mirage is a very aggressive 00:01:39.390 --> 00:01:41.890 stance. That's an incredibly heavy word in corporate 00:01:41.890 --> 00:01:44.790 strategy. It really is. She is... directly attacking 00:01:44.790 --> 00:01:46.969 their core financial narrative there. You don't 00:01:46.969 --> 00:01:49.750 use the word mirage lightly. No, you don't. She 00:01:49.750 --> 00:01:52.290 also labeled Anthropic a single product company 00:01:52.290 --> 00:01:56.359 in a platform war. Ouch. Yeah. She argues that 00:01:56.359 --> 00:02:00.120 OpenAI has a very multi -pronged approach. This 00:02:00.120 --> 00:02:02.640 diversity makes OpenAI a much more resilient 00:02:02.640 --> 00:02:05.219 investment. Right. And the broader context here 00:02:05.219 --> 00:02:07.560 is really fascinating because Anthropic relies 00:02:07.560 --> 00:02:10.719 heavily on a single core product offering. But 00:02:10.719 --> 00:02:13.219 OpenAI is trying to build an entire operating 00:02:13.219 --> 00:02:15.520 system. They want to be the foundational layer 00:02:15.520 --> 00:02:18.439 for everything. But here is the truly surprising 00:02:18.439 --> 00:02:21.759 part of this leak. Dresser actually offered a 00:02:21.759 --> 00:02:25.240 candid critique of OpenAI itself. Yeah, which 00:02:25.240 --> 00:02:27.639 you almost never see. Right. She criticized their 00:02:27.639 --> 00:02:29.860 highly publicized partnership with Microsoft 00:02:29.860 --> 00:02:32.439 specifically. That is a fascinating admission 00:02:32.439 --> 00:02:35.219 from a key executive. You just rarely see them 00:02:35.219 --> 00:02:37.740 critique their biggest financial backer. Dresser 00:02:37.740 --> 00:02:40.000 admitted the Microsoft deal has been quite limiting. 00:02:40.219 --> 00:02:43.240 It restricts their enterprise growth due to Azure 00:02:43.240 --> 00:02:46.780 -linked bottlenecks. Azure is Microsoft's massive 00:02:46.780 --> 00:02:50.039 cloud computing platform infrastructure. So if 00:02:50.039 --> 00:02:52.199 Azure struggles, OpenAI essentially struggles 00:02:52.199 --> 00:02:54.719 right alongside them. Exactly. They are inherently 00:02:54.719 --> 00:02:56.919 tied to Microsoft's physical server capacity. 00:02:57.199 --> 00:02:59.120 Which explains the strategic deal they made in 00:02:59.120 --> 00:03:01.919 February. They brought OpenAI models directly 00:03:01.919 --> 00:03:04.580 over to Amazon Bedrock. Right. And the demand 00:03:04.580 --> 00:03:06.900 has been totally staggering since they made that 00:03:06.900 --> 00:03:09.379 move. It really looks like OpenAI is using Amazon 00:03:09.379 --> 00:03:12.080 specifically. They want to bypass Microsoft's 00:03:12.080 --> 00:03:14.550 constraints entirely. Yeah, to capture a much 00:03:14.550 --> 00:03:18.289 wider enterprise market quickly, which is a wild 00:03:18.289 --> 00:03:20.590 corporate power move, honestly. It really is. 00:03:20.689 --> 00:03:23.050 They are playing the big cloud providers against 00:03:23.050 --> 00:03:25.009 each other. They are ensuring they don't get 00:03:25.009 --> 00:03:27.789 bottlenecked by one partner. She also slammed 00:03:27.789 --> 00:03:30.569 Anthropic's current internal compute shortage 00:03:30.569 --> 00:03:34.550 aggressively. Dresser called their server constraints 00:03:34.550 --> 00:03:38.389 a massive strategic failure. Wow. Yeah. This 00:03:38.389 --> 00:03:40.789 shortage forces Anthropic to heavily throttle 00:03:40.789 --> 00:03:44.530 user access. Meanwhile, OpenAI just continues 00:03:44.530 --> 00:03:47.030 to scale its operations globally. Well, compute 00:03:47.030 --> 00:03:49.569 power is the ultimate currency in this industry. 00:03:49.669 --> 00:03:52.310 If you cannot give users access, you lose them 00:03:52.310 --> 00:03:55.229 instantly. Right. A mass evaluation means absolutely 00:03:55.229 --> 00:03:57.509 nothing without the server space. It is like 00:03:57.509 --> 00:04:00.150 uncoupling your train from the main engine. You 00:04:00.150 --> 00:04:02.330 are going 100 miles an hour. You just have to 00:04:02.330 --> 00:04:05.669 hope the new engine catches you. Neat. But I 00:04:05.669 --> 00:04:07.330 still wrestle with filtering out the corporate 00:04:07.330 --> 00:04:09.710 spin myself when reading these leaked memos. 00:04:09.750 --> 00:04:12.810 Oh, totally. It's... incredibly difficult to 00:04:12.810 --> 00:04:15.689 separate fact from fiction. The spin is heavily 00:04:15.689 --> 00:04:18.050 engineered to sway public perception. You always 00:04:18.050 --> 00:04:19.970 have to ask who benefits from the leak. Right. 00:04:20.029 --> 00:04:22.610 Either OpenAI's internal security is an absolute 00:04:22.610 --> 00:04:27.769 disaster right now, or this supposed leak is 00:04:27.769 --> 00:04:31.410 a highly calculated strike. OpenAI is effectively 00:04:31.410 --> 00:04:35.240 writing its S1 filing in public. An S -1 is the 00:04:35.240 --> 00:04:37.939 formal document to go public. You need a highly 00:04:37.939 --> 00:04:40.839 compelling narrative to sell to investors. Exactly. 00:04:40.899 --> 00:04:43.399 It's the story that justifies your multi -billion 00:04:43.399 --> 00:04:45.680 dollar valuation. So Dresser is strategically 00:04:45.680 --> 00:04:49.180 framing Anthropic as a fear -based company. She 00:04:49.180 --> 00:04:51.860 paints them as overly focused on rigid safety 00:04:51.860 --> 00:04:55.100 constraints. Meanwhile, she frames OpenAI as 00:04:55.100 --> 00:04:57.339 the positive message growth engine. In lead up 00:04:57.339 --> 00:05:00.500 to a public debut, narrative is everything. Investors 00:05:00.500 --> 00:05:02.560 are buying the vision just as much as the tech. 00:05:03.040 --> 00:05:05.660 If investors start doubting Anthropic's $30 billion 00:05:05.660 --> 00:05:08.480 financial figure today, it could shift billions 00:05:08.480 --> 00:05:11.399 in valuation toward OpenAI overnight. The financial 00:05:11.399 --> 00:05:13.740 stakes are astronomically high at this moment. 00:05:13.860 --> 00:05:16.420 A single -linked memo can literally alter market 00:05:16.420 --> 00:05:18.279 dynamics. We also see these platform tensions 00:05:18.279 --> 00:05:20.100 with Apple recently. Apple threatened it could 00:05:20.100 --> 00:05:22.420 remove Grok from the App Store. Right, over the 00:05:22.420 --> 00:05:24.699 deepfakes. Yeah, this was over concerns about 00:05:24.699 --> 00:05:27.839 generating highly sexualized deepfakes. This 00:05:27.839 --> 00:05:30.600 issue was raised in a January letter to U .S. 00:05:30.600 --> 00:05:33.399 senators. We're just looking at the facts of 00:05:33.399 --> 00:05:35.959 that tension here. Platform control is truly 00:05:35.959 --> 00:05:39.120 the ultimate corporate battleground today. Gatekeepers 00:05:39.120 --> 00:05:41.740 like Apple hold immense power over distribution 00:05:41.740 --> 00:05:44.740 channels. They essentially dictate what software 00:05:44.740 --> 00:05:48.019 reaches billions of consumers. So are these platform 00:05:48.019 --> 00:05:51.699 wars ultimately hurting the end user? The fight 00:05:51.699 --> 00:05:54.160 for dominance pushes companies to innovate rapidly, 00:05:54.379 --> 00:05:57.139 but when they lock you into their specific ecosystem, 00:05:57.560 --> 00:06:00.939 it hurts. It creates massive roadblocks for actually 00:06:00.939 --> 00:06:03.319 getting work done. So competition forces innovation, 00:06:03.600 --> 00:06:06.620 but platform locks create friction. Exactly. 00:06:06.620 --> 00:06:08.459 It's a very delicate balance for the industry. 00:06:08.879 --> 00:06:11.500 That friction pushes us right into our next phase 00:06:11.500 --> 00:06:13.939 because these companies are fighting to justify 00:06:13.939 --> 00:06:16.120 massive valuations. Right. They have to evolve 00:06:16.120 --> 00:06:18.540 their products from basic chat bots. They become 00:06:18.540 --> 00:06:21.279 actual digital workers inside our systems. They 00:06:21.279 --> 00:06:23.620 need to integrate deeply into our daily software. 00:06:23.879 --> 00:06:26.480 A simple chat window isn't enough to justify 00:06:26.480 --> 00:06:29.439 a massive subscription. Microsoft is considering 00:06:29.439 --> 00:06:32.600 a truly massive shift right now. They may start 00:06:32.600 --> 00:06:34.759 charging for AI agents like human employees. 00:06:34.959 --> 00:06:37.120 They would give them logins, seats, accounts, 00:06:37.259 --> 00:06:40.040 and permissions. You'd onboard an AI just like 00:06:40.040 --> 00:06:42.680 a new hire. We should clarify what that terminology 00:06:42.680 --> 00:06:45.779 actually means here. AI agents are basically 00:06:45.779 --> 00:06:48.660 programs that act independently to complete multi 00:06:48.660 --> 00:06:50.899 -step tasks for you. That is the core definition 00:06:50.899 --> 00:06:53.779 of these AI agents. Perfect. And companies could 00:06:53.779 --> 00:06:56.699 literally pay per digital worker. They would 00:06:56.699 --> 00:06:58.779 manage them in their traditional HR software. 00:06:59.079 --> 00:07:01.980 Paying for software per seat. Like a human employee. 00:07:02.800 --> 00:07:05.920 That fundamentally changes the economics of running 00:07:05.920 --> 00:07:08.279 a modern business. You are essentially hiring 00:07:08.279 --> 00:07:10.860 a digital staff member now. You are not just 00:07:10.860 --> 00:07:13.720 buying a static software tool anymore. You are 00:07:13.720 --> 00:07:16.660 buying labor, not just a bunch of code. Microsoft 00:07:16.660 --> 00:07:19.579 is working on yet another open -claw -like digital 00:07:19.579 --> 00:07:22.980 agent. The goal is a highly secure, always -on 00:07:22.980 --> 00:07:25.779 assistant tool. It handles multi -step tasks 00:07:25.779 --> 00:07:28.899 with much stronger enterprise controls. Enterprise 00:07:28.899 --> 00:07:31.259 security is the biggest hurdle for widespread 00:07:31.259 --> 00:07:34.779 adoption. Companies need absolute certainty before 00:07:34.779 --> 00:07:38.579 granting agents system access. Exactly. An autonomous 00:07:38.579 --> 00:07:42.220 agent with a login is a huge security risk. A 00:07:42.220 --> 00:07:44.899 mysterious leak claims Anthropic is planning 00:07:44.899 --> 00:07:48.100 something similar. Early hints suggest a major 00:07:48.100 --> 00:07:50.720 shift toward built -in workflows. Everyone is 00:07:50.720 --> 00:07:53.000 trying to escape the basic chat window. They 00:07:53.000 --> 00:07:55.000 want the AI to do the work in the background. 00:07:55.300 --> 00:07:57.540 We see these workflow tools expanding everywhere 00:07:57.540 --> 00:08:00.740 right now. Google added new AI skills directly 00:08:00.740 --> 00:08:03.240 to their Chrome browser. Yeah, that's a big deal. 00:08:03.379 --> 00:08:05.399 You can save your absolute favorite workflows 00:08:05.399 --> 00:08:08.220 very easily. Then you can run them on any website 00:08:08.220 --> 00:08:10.939 later. Gemini Personal Intelligence also just 00:08:10.939 --> 00:08:13.600 expanded operations into India. So the global 00:08:13.600 --> 00:08:15.800 footprint of these tools is growing incredibly 00:08:15.800 --> 00:08:21.589 fast. is actually going viral currently. Yeah. 00:08:21.730 --> 00:08:24.449 It is heavily inspired by the developer Andrej 00:08:24.449 --> 00:08:26.889 Karpathy. Oh, I saw that. Yeah. It fundamentally 00:08:26.889 --> 00:08:29.629 controls how your coding AI behaves locally. 00:08:29.790 --> 00:08:31.870 Think of it like a strict constitution for your 00:08:31.870 --> 00:08:34.470 AI. Before it writes a single line of code, it 00:08:34.470 --> 00:08:37.429 checks this text file. It ensures it isn't breaking 00:08:37.429 --> 00:08:40.669 your specific local rules. It forces much clearer 00:08:40.669 --> 00:08:44.340 logic and reduces coding mistakes. It also prevents 00:08:44.340 --> 00:08:47.360 the AI from making unintended file edits. It's 00:08:47.360 --> 00:08:49.519 about giving the human much tighter control. 00:08:50.019 --> 00:08:53.059 The AI needs boundaries to be truly effective 00:08:53.059 --> 00:08:56.399 for you. We are seeing a new wave of empowered 00:08:56.399 --> 00:09:00.200 builders everywhere. New tools treat AI as highly 00:09:00.200 --> 00:09:03.019 independent software workers. Rather than listing 00:09:03.019 --> 00:09:05.740 them all out separately today, let's look at 00:09:05.740 --> 00:09:08.379 how they create a unified ecosystem. Figma for 00:09:08.379 --> 00:09:10.320 agents is a perfect example of this. Yeah, that 00:09:10.320 --> 00:09:12.500 one is fascinating. You can design directly on 00:09:12.500 --> 00:09:14.840 the standard Figma canvas. You guide agents with 00:09:14.840 --> 00:09:17.419 context about your team's intent. It is like 00:09:17.419 --> 00:09:20.360 stacking Lego blocks of data visually. You are 00:09:20.360 --> 00:09:23.159 giving the AI visual parameters to work within. 00:09:23.419 --> 00:09:25.820 Then there is a tool called CatDoze version 4. 00:09:26.100 --> 00:09:28.899 It is essentially a powerful no -code application 00:09:28.899 --> 00:09:31.379 builder tool. Right. It runs in the cloud and 00:09:31.379 --> 00:09:33.480 has its own computer. You can literally close 00:09:33.480 --> 00:09:36.840 your laptop browser tab completely. And the system 00:09:36.840 --> 00:09:38.899 just keeps working in the background. That is 00:09:38.899 --> 00:09:41.519 the real shift toward autonomous digital employees 00:09:41.519 --> 00:09:44.159 right there. It handles the tedious labor while 00:09:44.159 --> 00:09:47.120 you do other things. Soft AI co -builder is another 00:09:47.120 --> 00:09:49.919 incredibly powerful new tool. It builds custom 00:09:49.919 --> 00:09:52.700 portals and complex internal operation systems. 00:09:52.940 --> 00:09:54.919 You just describe what you need to the system, 00:09:55.000 --> 00:09:58.000 then you refine it with prompts or edit visually. 00:09:58.490 --> 00:10:01.029 These tools democratize software creation for 00:10:01.029 --> 00:10:03.830 non -technical business users. You don't need 00:10:03.830 --> 00:10:06.049 an engineering degree to build complex applications 00:10:06.049 --> 00:10:09.470 anymore. Caveman is another tool that optimizes 00:10:09.470 --> 00:10:13.350 AI token usage. It cuts 75 % of Claude's output 00:10:13.350 --> 00:10:16.730 tokens easily. Which is huge. It manages to do 00:10:16.730 --> 00:10:19.529 this without losing any factual accuracy. That 00:10:19.529 --> 00:10:21.850 represents a massive cost saving for heavy users. 00:10:22.549 --> 00:10:24.850 Efficiency is just as important as raw computing 00:10:24.850 --> 00:10:28.110 power today. It is a simple one -line install 00:10:28.110 --> 00:10:30.950 for popular coding tools. It works with quad 00:10:30.950 --> 00:10:34.769 code, cursor, windsurf, and copilot. AI is becoming 00:10:34.769 --> 00:10:37.149 the foundational layer of modern work. It is 00:10:37.149 --> 00:10:39.389 weaving itself into every tool we actively use. 00:10:39.529 --> 00:10:42.190 So if an AI has its own seat, who takes the blame 00:10:42.190 --> 00:10:45.009 when it fails? That is a very complicated legal 00:10:45.009 --> 00:10:49.070 and operational issue. Legally, the system absolutely 00:10:49.070 --> 00:10:52.389 requires a responsible human overseer. Right. 00:10:52.450 --> 00:10:55.570 You cannot fire an AI for making a critical error. 00:10:55.769 --> 00:10:58.149 You must trace the failure back to the human 00:10:58.149 --> 00:11:00.590 deployer. Accountability stays with the human 00:11:00.590 --> 00:11:03.690 manager, even for digital staff. Yeah, the human 00:11:03.690 --> 00:11:07.289 is always the ultimate safety net. Sponsor. We 00:11:07.289 --> 00:11:09.350 just discussed the rapid rise of workflow autonomy. 00:11:09.769 --> 00:11:12.470 If an AI can run while you close the PAB, What 00:11:12.470 --> 00:11:14.269 happens when it gets physical eyes and hands? 00:11:14.450 --> 00:11:16.149 It breaks out of the digital screen entirely 00:11:16.149 --> 00:11:19.549 then. It starts interacting with the messy, unpredictable 00:11:19.549 --> 00:11:23.950 real world. Exactly. Gemini Robotics ER 1 .6 00:11:23.950 --> 00:11:26.750 is Google's answer to this. Most robots are great 00:11:26.750 --> 00:11:29.070 at following a pre -written script. But they 00:11:29.070 --> 00:11:31.330 are terrible at understanding a dynamic, changing 00:11:31.330 --> 00:11:33.470 environment. They struggle when the physical 00:11:33.470 --> 00:11:35.710 world does something unexpected. If a coffee 00:11:35.710 --> 00:11:37.710 cup is moved two inches, they fail completely. 00:11:38.330 --> 00:11:41.350 Gemini 1 .6 addresses this massive physical reasoning 00:11:41.350 --> 00:11:44.490 gap directly. It is a total overhaul of how a 00:11:44.490 --> 00:11:47.110 robot perceives. Yeah. It allows the machine 00:11:47.110 --> 00:11:50.210 to verify its own physical work. This is a massive 00:11:50.210 --> 00:11:53.549 milestone for physical AI systems. They are finally 00:11:53.549 --> 00:11:57.470 moving past rigid, pre -programmed robotic movements. 00:11:57.789 --> 00:12:00.090 Let's examine instrument reading as a prime example 00:12:00.090 --> 00:12:02.929 here. This is a major win for the industrial 00:12:02.929 --> 00:12:05.250 inspection field. Oh, absolutely. The model can 00:12:05.250 --> 00:12:07.610 now interpret complex circular pressure gauges. 00:12:08.059 --> 00:12:10.179 It can accurately read traditional thermometers 00:12:10.179 --> 00:12:12.980 and basic digital displays. Reading a physical 00:12:12.980 --> 00:12:16.960 dial is notoriously difficult for machines. Lighting, 00:12:16.960 --> 00:12:19.200 angles, and reflections usually confuse basic 00:12:19.200 --> 00:12:22.039 computer vision. The real world is just full 00:12:22.039 --> 00:12:25.700 of messy visual noise. The system actually zooms 00:12:25.700 --> 00:12:28.100 in on the physical instrument. It runs custom 00:12:28.100 --> 00:12:31.539 code to estimate the specific tick marks. It 00:12:31.539 --> 00:12:34.200 uses its broad world knowledge to evaluate things. 00:12:34.360 --> 00:12:36.960 It inherently knows if the recorded pressure 00:12:36.960 --> 00:12:39.629 is too high. This process relies heavily on a 00:12:39.629 --> 00:12:42.169 concept called agentic vision. It is fundamentally 00:12:42.169 --> 00:12:44.169 different from traditional image recognition 00:12:44.169 --> 00:12:46.769 software. Agentic vision is truly the secret 00:12:46.769 --> 00:12:49.490 sauce here today. The model successfully combines 00:12:49.490 --> 00:12:52.309 visual reasoning with real -time code execution. 00:12:52.629 --> 00:12:54.970 Yeah. It literally performs complex math on what 00:12:54.970 --> 00:12:58.870 it currently sees. Whoa. Imagine robots literally 00:12:58.870 --> 00:13:02.590 writing and executing code in real time just 00:13:02.590 --> 00:13:04.850 to understand the physical object they're looking 00:13:04.850 --> 00:13:08.700 at. That is an unbelievable leap in robotic cognitive 00:13:08.700 --> 00:13:11.419 ability. It really is. It is a fundamentally 00:13:11.419 --> 00:13:14.559 new paradigm for machine perception. It bridges 00:13:14.559 --> 00:13:17.779 the gap between digital logic and physical reality. 00:13:18.580 --> 00:13:21.759 Two sec silence. Let's discuss its multi -view 00:13:21.759 --> 00:13:24.799 success detection capabilities now. A robot with 00:13:24.799 --> 00:13:27.500 a wrist cam and an overhead camera view. It can 00:13:27.500 --> 00:13:29.519 finally triangulate its operational success. 00:13:30.190 --> 00:13:32.690 Very reliably. Right. And it uses multiple camera 00:13:32.690 --> 00:13:35.190 angles to verify a completed task. It doesn't 00:13:35.190 --> 00:13:38.070 just assume it did the job correctly. It synthesizes 00:13:38.070 --> 00:13:40.169 multiple data streams to confirm the physical 00:13:40.169 --> 00:13:43.090 reality. It can verify that a blue pen is actually 00:13:43.090 --> 00:13:45.409 in a holder. Yeah. It could do this even if one 00:13:45.409 --> 00:13:47.490 view is entirely blocked. It understands object 00:13:47.490 --> 00:13:50.340 permanence in a very human -like way. It knows 00:13:50.340 --> 00:13:52.659 an object exists even when completely hidden. 00:13:52.799 --> 00:13:55.059 It can now handle incredibly complex relational 00:13:55.059 --> 00:13:57.779 logic commands. You can instruct it to perform 00:13:57.779 --> 00:14:00.580 highly specific spatial tasks. Right. For example, 00:14:00.580 --> 00:14:02.460 moving every object that can fit inside the blue 00:14:02.460 --> 00:14:05.799 cup. That level of spatial awareness is incredibly 00:14:05.799 --> 00:14:08.879 difficult to program manually. The system has 00:14:08.879 --> 00:14:11.679 to understand volume, shape, and physical relationships 00:14:11.679 --> 00:14:14.799 intuitively. The collaboration with Boston Dynamics 00:14:14.799 --> 00:14:17.600 is the real signal here. They are integrating 00:14:17.600 --> 00:14:21.779 Gemini ER 1 .6 into the orbit platform. By doing 00:14:21.779 --> 00:14:24.519 this, Google has effectively provided a prefrontal 00:14:24.519 --> 00:14:27.460 cortex. It acts as a brain for the world's most 00:14:27.460 --> 00:14:29.879 advanced hardware. Boston Dynamics makes some 00:14:29.879 --> 00:14:32.320 of the most capable robotic bodies out there. 00:14:32.539 --> 00:14:34.980 Now those agile bodies finally have a dynamic 00:14:34.980 --> 00:14:37.309 brain to match. We should also pivot slightly 00:14:37.309 --> 00:14:40.250 to the microscopic physical world. Amazon just 00:14:40.250 --> 00:14:42.850 recently launched their new Amazon Biodiscovery 00:14:42.850 --> 00:14:45.090 platform. This is a major development for the 00:14:45.090 --> 00:14:47.470 global pharmaceutical industry. It takes this 00:14:47.470 --> 00:14:50.049 autonomous reasoning down to the cellular level. 00:14:50.190 --> 00:14:52.870 It is specifically designed to accelerate modern 00:14:52.870 --> 00:14:55.870 drug discovery. It can strategically create and 00:14:55.870 --> 00:14:58.970 test thousands of molecule ideas fast. Yeah. 00:14:59.090 --> 00:15:02.429 Early tests successfully generated 300 ,000 unique 00:15:02.429 --> 00:15:05.870 medical antibodies. Generating 300 ,000 antibodies 00:15:05.870 --> 00:15:09.090 represents an unbelievable scientific scale. 00:15:09.330 --> 00:15:11.909 Humans could never physically test variations 00:15:11.909 --> 00:15:14.850 at that specific speed. That scale of physical 00:15:14.850 --> 00:15:17.509 world interaction is truly staggering today. 00:15:17.850 --> 00:15:22.129 It moves AI from digital abstraction into biological 00:15:22.129 --> 00:15:24.789 manipulation capabilities. It changes the entire 00:15:24.789 --> 00:15:27.129 timeline for developing life -saving medications. 00:15:27.629 --> 00:15:30.289 What used to take decades might only take a few 00:15:30.289 --> 00:15:33.139 months now. Does this visual reasoning mean robots 00:15:33.139 --> 00:15:36.639 finally understand context? Well, they don't 00:15:36.639 --> 00:15:38.440 experience understanding the way humans naturally 00:15:38.440 --> 00:15:42.100 do, but they process enough variables to effectively 00:15:42.100 --> 00:15:44.480 mimic situational awareness. They don't feel 00:15:44.480 --> 00:15:46.840 context, but they can finally calculate it. Right. 00:15:46.940 --> 00:15:49.419 It is a mathematical approximation of actual 00:15:49.419 --> 00:15:51.679 human awareness. Let's step back and synthesize 00:15:51.679 --> 00:15:53.980 this entire deep dive. The overarching narrative 00:15:53.980 --> 00:15:57.279 arc today is actually very clear. AI is no longer 00:15:57.279 --> 00:15:59.720 just a clever little party trick. It's certainly 00:15:59.720 --> 00:16:02.340 no longer just a simple chat window. It has completely 00:16:02.340 --> 00:16:04.620 outgrown its original digital boundaries now. 00:16:04.759 --> 00:16:07.720 It is touching every single aspect of our physical 00:16:07.720 --> 00:16:11.039 and corporate world. It is the absolute center 00:16:11.039 --> 00:16:14.580 of ruthless corporate warfare today. Yeah. Companies 00:16:14.580 --> 00:16:17.740 are battling for dominance over... massive platform 00:16:17.740 --> 00:16:20.480 distribution channels. It is actively demanding 00:16:20.480 --> 00:16:23.899 its own employee login in our software. It wants 00:16:23.899 --> 00:16:26.220 a dedicated seat at the modern enterprise table. 00:16:26.340 --> 00:16:28.840 Exactly. And it is finally getting the necessary 00:16:28.840 --> 00:16:32.080 digital prefrontal cortex. It needs this dynamic 00:16:32.080 --> 00:16:35.259 brain to manipulate the tangible physical world. 00:16:35.419 --> 00:16:37.980 It is even beginning to actively manipulate our 00:16:37.980 --> 00:16:40.279 microscopic biology. The evolution from text 00:16:40.279 --> 00:16:42.100 generation to physical action is staggering. 00:16:42.379 --> 00:16:44.740 I mean, if an AI can get its own employee seat 00:16:44.740 --> 00:16:47.360 in a Microsoft database. and a robotic body with 00:16:47.360 --> 00:16:50.159 a prefrontal cortex from Google, at what point 00:16:50.159 --> 00:16:52.399 will our legal definition of a worker have to 00:16:52.399 --> 00:16:54.659 fundamentally change? Thank you for joining us 00:16:54.659 --> 00:16:56.820 on this deep dive. We hope this exploration gave 00:16:56.820 --> 00:16:59.179 you some valuable new perspectives. Out to row 00:16:59.179 --> 00:16:59.399 music.