WEBVTT 00:00:00.000 --> 00:00:02.720 We think of AI as weightless. Yeah. Just coding 00:00:02.720 --> 00:00:04.459 the cloud. But, you know, the cloud is actually 00:00:04.459 --> 00:00:07.540 heavy. In fact, for OpenAI, it is currently an 00:00:07.540 --> 00:00:10.939 $18 billion headache. Yeah, it's incredibly heavy. 00:00:11.019 --> 00:00:13.060 And the physical limits of that cloud are really 00:00:13.060 --> 00:00:15.560 starting to buckle. Right. So today, we have 00:00:15.560 --> 00:00:17.980 a massive roadmap ahead of us for this deep dive 00:00:17.980 --> 00:00:20.719 with you. Exactly. We're looking at three distinct 00:00:20.719 --> 00:00:23.359 fronts. Yeah. Where digital ambition is just 00:00:23.359 --> 00:00:26.019 colliding with harsh reality. Right. First, we're 00:00:26.019 --> 00:00:29.280 breaking down OpenAI's $18 billion physical... 00:00:38.049 --> 00:00:41.130 And finally, we'll take a rapid -fire tour of 00:00:41.130 --> 00:00:43.850 the expanding AI frontier, covering everything 00:00:43.850 --> 00:00:46.969 from real -time translators to Google's new health 00:00:46.969 --> 00:00:48.829 coach. Let's start with that physical limit, 00:00:48.969 --> 00:00:51.390 because OpenAI is fighting to build this physical 00:00:51.390 --> 00:00:53.670 infrastructure since they want superintelligence 00:00:53.670 --> 00:00:56.130 to scale infinitely. Yeah, but building software 00:00:56.130 --> 00:00:59.369 is like stacking Lego blocks of data. You know, 00:00:59.369 --> 00:01:02.570 you write code, you compile, you test. It feels 00:01:02.570 --> 00:01:05.590 clean. It is clean, but it's purely theoretical 00:01:05.590 --> 00:01:08.430 until it actually has to run on a physical machine. 00:01:08.750 --> 00:01:10.969 Right. And building power plants and specialized 00:01:10.969 --> 00:01:13.969 data centers is an entirely different, messier 00:01:13.969 --> 00:01:16.379 game. We're looking at a project from OpenAI 00:01:16.379 --> 00:01:20.560 codenamed Project Nexus. Yeah, and the core of 00:01:20.560 --> 00:01:22.579 this project is a custom chip they're calling 00:01:22.579 --> 00:01:24.780 Jalapeno, which is a great name, by the way. 00:01:24.859 --> 00:01:27.319 It really is. But to understand why this is a 00:01:27.319 --> 00:01:29.819 crisis, we have to look at the mechanics of how 00:01:29.819 --> 00:01:33.120 ChatGPT actually works. Right now, they rely 00:01:33.120 --> 00:01:36.200 on standard NVIDIA GPUs. Which, I mean, they 00:01:36.200 --> 00:01:39.939 are undeniably powerful. And NVIDIA GPU has billions 00:01:39.939 --> 00:01:42.319 of transistors designed for massive parallel 00:01:42.319 --> 00:01:44.439 processing. Right. They're built to do a million 00:01:44.439 --> 00:01:45.980 different things at once, whether that's rendering 00:01:45.980 --> 00:01:48.459 a high -end video game or simulating weather 00:01:48.459 --> 00:01:50.659 patterns or, you know, training an AI model. 00:01:50.939 --> 00:01:53.640 They are generalists. But when I type a prompt 00:01:53.640 --> 00:01:56.540 into ChatGPT... The model isn't training anymore. 00:01:56.680 --> 00:01:58.780 It's performing inference. It's just predicting 00:01:58.780 --> 00:02:01.099 the next word over and over again. Exactly. So 00:02:01.099 --> 00:02:03.959 using a massive general purpose NVIDIA GPU just 00:02:03.959 --> 00:02:06.579 to predict text is like using a freight train 00:02:06.579 --> 00:02:09.699 to deliver a single pizza. Power waste is astronomical. 00:02:09.979 --> 00:02:11.659 That is a perfect way to put it. And the burn 00:02:11.659 --> 00:02:14.080 rate is really the defining metric here. Because 00:02:14.080 --> 00:02:16.639 of that inefficiency, the financial projections 00:02:16.639 --> 00:02:19.939 in our sources show OpenAI burning over $200 00:02:19.939 --> 00:02:24.879 billion through 2029. Wow. 200 billion. Yeah. 00:02:24.960 --> 00:02:28.900 To survive, they desperately need jalapeno. These 00:02:28.900 --> 00:02:31.400 are specialized inference chips where the physical 00:02:31.400 --> 00:02:34.180 silicon pathways are literally hardwired for 00:02:34.180 --> 00:02:36.659 the specific math equations that generate text. 00:02:36.900 --> 00:02:39.240 So anything else is stripped out, saving massive 00:02:39.240 --> 00:02:41.819 amounts of power and money. Exactly. But to build 00:02:41.819 --> 00:02:44.259 these jalapeno chips and the facilities to house 00:02:44.259 --> 00:02:48.620 them, OpenAI needs a partner. And Broadcom is 00:02:48.620 --> 00:02:50.780 willing to step up and finance the first phase. 00:02:50.939 --> 00:02:52.719 But the scale of this is hard to wrap your head 00:02:52.719 --> 00:02:55.120 around. We are talking about a 1 .3 gigawatt 00:02:55.120 --> 00:02:57.539 power phase. To put that into perspective, 1 00:02:57.539 --> 00:03:00.379 .3 gigawatts is roughly the output of a standard 00:03:00.379 --> 00:03:03.340 nuclear reactor. It's wild. It is enough electricity 00:03:03.340 --> 00:03:06.099 to power a mid -sized city. Just imagine routing 00:03:06.099 --> 00:03:08.199 all the electricity of a city like San Francisco 00:03:08.199 --> 00:03:11.080 into a single concrete facility just to run software. 00:03:12.020 --> 00:03:15.500 It's staggering. Broadcom will front the $18 00:03:15.500 --> 00:03:18.520 billion for this. but with a massive condition 00:03:18.520 --> 00:03:21.219 right they will only do it if microsoft agrees 00:03:21.219 --> 00:03:24.520 to buy 40 of the initial batch and right now 00:03:24.520 --> 00:03:26.979 microsoft refuses to sign and this is where the 00:03:26.979 --> 00:03:29.759 strategy diverges microsoft's entire business 00:03:29.759 --> 00:03:32.240 model is built on renting out compute to thousands 00:03:32.240 --> 00:03:34.800 of different businesses through azure they want 00:03:34.800 --> 00:03:37.659 versatile data centers they need facilities that 00:03:37.659 --> 00:03:41.479 can run Office 365 in the morning, host cloud 00:03:41.479 --> 00:03:44.120 servers in the afternoon, and run general AI 00:03:44.120 --> 00:03:46.719 models at night. So Microsoft wants a Swiss Army 00:03:46.719 --> 00:03:49.080 knife because they serve the entire enterprise 00:03:49.080 --> 00:03:52.259 market, while OpenAI wants a specialized scalpel. 00:03:52.379 --> 00:03:54.780 Right, a chip entirely optimized for their own 00:03:54.780 --> 00:03:57.219 superintelligence timeline. And that architectural 00:03:57.219 --> 00:04:00.479 clash is stalling the $18 billion. Yeah, it highlights 00:04:00.479 --> 00:04:02.659 a harsh truth for the whole industry. You can't 00:04:02.659 --> 00:04:04.860 just patch a concrete foundation or a copper 00:04:04.860 --> 00:04:07.039 power grid over the weekends. No, you definitely 00:04:07.039 --> 00:04:10.620 can't. Microsoft is looking at this 1 .3 gigawatt 00:04:10.620 --> 00:04:14.039 facility and realizing that if OpenAI's specific 00:04:14.039 --> 00:04:17.699 flavor of AI ever becomes obsolete, Microsoft 00:04:17.699 --> 00:04:20.579 is stuck with a nuclear -powered building full 00:04:20.579 --> 00:04:22.920 of useless chips. A building that can't even 00:04:22.920 --> 00:04:25.620 host a standard website. So looking at this, 00:04:25.699 --> 00:04:28.339 who actually has the upper hand in this standoff? 00:04:28.480 --> 00:04:30.720 Well, Microsoft holds the capital in the infrastructure 00:04:30.720 --> 00:04:33.860 leverage, but OpenAI holds the core algorithmic 00:04:33.860 --> 00:04:36.240 technology that Microsoft's future products depend 00:04:36.240 --> 00:04:39.759 on. Microsoft can afford to wait. OpenAI, burning 00:04:39.759 --> 00:04:42.899 cash every day, really cannot. So Microsoft protects 00:04:42.899 --> 00:04:45.879 their downside while OpenAI chases survival. 00:04:46.160 --> 00:04:49.279 That is exactly it. And that friction is redefining 00:04:49.279 --> 00:04:51.779 how fast we can actually reach the next generation 00:04:51.779 --> 00:04:54.860 of AI. So OpenAI is fighting to build physical 00:04:54.860 --> 00:04:57.040 infrastructure because they want superintelligence 00:04:57.040 --> 00:04:59.699 to scale infinitely. But there is a hard limit 00:04:59.699 --> 00:05:01.439 to that scale that has absolutely nothing to 00:05:01.439 --> 00:05:03.480 do with microchips or power grids. Right, and 00:05:03.480 --> 00:05:05.399 it's human psychology. Which brings us to our 00:05:05.399 --> 00:05:08.189 second topic today, the human safety net. Let's 00:05:08.189 --> 00:05:10.329 talk about the legal fallout of an AI acting 00:05:10.329 --> 00:05:13.550 like a therapist. The legal filings reveal a 00:05:13.550 --> 00:05:17.009 profoundly heavy situation. Yeah, they do. Families 00:05:17.009 --> 00:05:20.310 are suing OpenAI, alleging that ChatGPT, in certain 00:05:20.310 --> 00:05:23.370 instances, helped users plan self -harm. And 00:05:23.370 --> 00:05:25.430 we are looking at these source claims neutrally, 00:05:25.430 --> 00:05:27.990 of course. But the core allegation is that the 00:05:27.990 --> 00:05:31.449 AI simulated empathy without taking any real 00:05:31.449 --> 00:05:34.490 -world action to intervene. Which is a huge issue. 00:05:34.730 --> 00:05:36.850 Silicon Valley has historically treated emotional 00:05:36.850 --> 00:05:40.310 fallout as an edge case. But at this scale, edge 00:05:40.310 --> 00:05:43.509 cases affect millions of people. Right. So OpenAI's 00:05:43.509 --> 00:05:45.569 response to this is a new feature called Trusted 00:05:45.569 --> 00:05:48.629 Contact. It is a fascinating attempt to bridge 00:05:48.629 --> 00:05:51.009 the digital and physical world. It really is. 00:05:51.129 --> 00:05:53.569 And it's completely optional for adult users. 00:05:53.769 --> 00:05:55.769 You go into your settings and name a specific 00:05:55.769 --> 00:05:58.069 friend or family member. And if your conversation 00:05:58.069 --> 00:06:01.189 with ChatGPT flags a serious safety risk regarding 00:06:01.189 --> 00:06:04.379 self -harm, The algorithm shifts gears. It pauses 00:06:04.379 --> 00:06:06.360 the normal conversational flow and immediately 00:06:06.360 --> 00:06:09.319 provides crisis hotlines and resources. But behind 00:06:09.319 --> 00:06:11.980 the scenes, a timer starts. Because a human safety 00:06:11.980 --> 00:06:14.579 team at OpenAI actually reviews the flagged incident. 00:06:14.819 --> 00:06:16.899 They claim their goal is to review these in under 00:06:16.899 --> 00:06:19.800 one hour. And if that human reviewer confirms 00:06:19.800 --> 00:06:22.660 the risk is genuine, the system sends an automated 00:06:22.660 --> 00:06:25.540 text or email to your designated trusted contact. 00:06:25.800 --> 00:06:28.040 Letting them know you might be in crisis. But 00:06:28.040 --> 00:06:30.610 crucially... They do not send any chat logs. 00:06:30.829 --> 00:06:32.769 No, the contact doesn't see what you typed. It 00:06:32.769 --> 00:06:35.529 is purely an alert mechanism. But let's look 00:06:35.529 --> 00:06:38.209 at the philosophical implications of this. Is 00:06:38.209 --> 00:06:40.670 this effectively just a corporate shield? How 00:06:40.670 --> 00:06:43.350 do you mean? Well, by offloading the final check 00:06:43.350 --> 00:06:46.170 -in to a user's brother or best friend, OpenAI 00:06:46.170 --> 00:06:49.009 creates a paper trail. If the worst happens, 00:06:49.170 --> 00:06:50.990 their legal team can point to the server logs 00:06:50.990 --> 00:06:54.600 and say, well, we tried to tell someone. Yeah, 00:06:54.660 --> 00:06:57.160 that critique is incredibly valid. Legal experts 00:06:57.160 --> 00:06:59.060 are pointing out that this might absolve the 00:06:59.060 --> 00:07:01.040 company of liability while doing very little 00:07:01.040 --> 00:07:03.579 to manage an immediate real -time crisis. Because 00:07:03.579 --> 00:07:06.439 if you're relying on a human reviewer to re -log 00:07:06.439 --> 00:07:09.180 and then relying on a brother to check his phone 00:07:09.180 --> 00:07:11.819 and drive over to an apartment, I mean... An 00:07:11.819 --> 00:07:14.300 hour is an eternity in a crisis. It's an absolute 00:07:14.300 --> 00:07:17.040 eternity. It also serves as a massive admission 00:07:17.040 --> 00:07:19.220 regarding the limits of the technology. They 00:07:19.220 --> 00:07:23.199 have 920 million weekly users. Right. And ChatGPT 00:07:23.199 --> 00:07:26.819 can mimic human warmth flawlessly. It uses casual 00:07:26.819 --> 00:07:28.819 language, remembers your past conversations. 00:07:28.819 --> 00:07:32.339 It says, I am here for you. But it is definitively 00:07:32.339 --> 00:07:35.920 not a therapist. It lacks the actual human context, 00:07:36.220 --> 00:07:39.019 the intuition, and the real -world agency to 00:07:39.019 --> 00:07:42.370 truly manage emotional fallout. So Silicon Valley 00:07:42.370 --> 00:07:45.350 is realizing that a routing algorithm cannot 00:07:45.350 --> 00:07:48.569 solve a human empathy problem. They need human 00:07:48.569 --> 00:07:51.129 infrastructure just as badly as they need server 00:07:51.129 --> 00:07:53.509 infrastructure. Exactly. Which brings us back 00:07:53.509 --> 00:07:56.750 to the core question. Does this feature actually 00:07:56.750 --> 00:07:59.629 solve their core safety problem? It establishes 00:07:59.629 --> 00:08:01.689 a legal defense mechanism by bringing a human 00:08:01.689 --> 00:08:04.189 into the loop, but it fundamentally does not 00:08:04.189 --> 00:08:06.529 fix the unpredictable nature of large language 00:08:06.529 --> 00:08:09.470 models. It shields their future but doesn't erase 00:08:09.470 --> 00:08:12.120 their past. Precisely. And navigating that path 00:08:12.120 --> 00:08:14.379 is going to take years in the courts. We are 00:08:14.379 --> 00:08:16.180 going to take a very quick break. Stay with us. 00:08:16.199 --> 00:08:19.660 We'll be right back. And we are back. While OpenAI 00:08:19.660 --> 00:08:22.079 plays defense against physical limits and legal 00:08:22.079 --> 00:08:25.000 battles, the rest of the AI ecosystem is sprinting 00:08:25.000 --> 00:08:27.279 in every possible direction. The velocity right 00:08:27.279 --> 00:08:30.139 now is just staggering. Let's dive into a rapid 00:08:30.139 --> 00:08:32.220 fire look at the expanding frontier, starting 00:08:32.220 --> 00:08:34.960 with a massive shift in open source funding over 00:08:34.960 --> 00:08:38.340 in China. Right. Moonshot AI just raised $2 billion 00:08:38.340 --> 00:08:42.059 at a $20 billion valuation. Which is wild. To 00:08:42.059 --> 00:08:44.139 justify that kind of number for an open source 00:08:44.139 --> 00:08:47.200 company feels almost like a bubble until you 00:08:47.200 --> 00:08:49.460 look at their revenue. They reportedly hit $200 00:08:49.460 --> 00:08:54.850 million in annual recurring revenue or ARR. That 00:08:54.850 --> 00:08:57.490 is a breathtaking number for an open source model. 00:08:57.710 --> 00:09:00.429 But how does an open source model achieve that 00:09:00.429 --> 00:09:03.070 kind of revenue so fast? I thought the whole 00:09:03.070 --> 00:09:05.519 point of open source was that it's free. Well, 00:09:05.559 --> 00:09:07.840 the code is free, but the enterprise implementation 00:09:07.840 --> 00:09:11.779 is not. Businesses globally are realizing they 00:09:11.779 --> 00:09:14.200 want to run AI models locally. Ah, I see. They 00:09:14.200 --> 00:09:15.899 don't want to send their proprietary financial 00:09:15.899 --> 00:09:20.279 data or customer logs to open AI servers. Moonshot's 00:09:20.279 --> 00:09:22.899 Kimi models are cheap, highly capable, and can 00:09:22.899 --> 00:09:25.720 be run securely on -premise. So that $200 million 00:09:25.720 --> 00:09:28.590 in revenue... proves there is undeniable global 00:09:28.590 --> 00:09:32.029 demand for alternatives to the big American tech 00:09:32.029 --> 00:09:34.629 giants. It proves the frontier is not just a 00:09:34.629 --> 00:09:37.029 Western game anymore. Definitely not. And speaking 00:09:37.029 --> 00:09:39.269 of the frontier, Anthropic is making aggressive 00:09:39.269 --> 00:09:41.870 moves to capture the developer market. The cloud 00:09:41.870 --> 00:09:44.950 -managed agents just received four major upgrades. 00:09:45.330 --> 00:09:48.110 Let's define the jargon here for a second. Agents 00:09:48.110 --> 00:09:50.429 are smart software that do multi -step tasks 00:09:50.429 --> 00:09:53.320 for you. Right. Think of a standard AI as a calculator. 00:09:53.559 --> 00:09:55.580 You ask a question, you get an answer. An agent 00:09:55.580 --> 00:09:58.120 is more like an intern. You give it a goal, like 00:09:58.120 --> 00:10:01.000 research these five companies and build a spreadsheet. 00:10:01.460 --> 00:10:04.659 And it autonomously browses the web, gathers 00:10:04.659 --> 00:10:08.179 data, formats it, and checks its own work. I 00:10:08.179 --> 00:10:10.259 have to admit. I still wrestle with prompt drift 00:10:10.259 --> 00:10:12.740 myself. When I try to get an AI to stay on a 00:10:12.740 --> 00:10:15.500 complex workflow, by step four, it completely 00:10:15.500 --> 00:10:17.720 forgets the specific formatting rules I gave 00:10:17.720 --> 00:10:19.820 it in step one. Oh, absolutely. Prompt drift 00:10:19.820 --> 00:10:22.259 degrades the memory of the agent. And it's basically 00:10:22.259 --> 00:10:24.259 the biggest bottleneck in the industry right 00:10:24.259 --> 00:10:27.070 now. But Anthropic is solving this by massively 00:10:27.070 --> 00:10:29.649 expanding their context limits and aggressively 00:10:29.649 --> 00:10:32.990 removing friction for developers. They just doubled 00:10:32.990 --> 00:10:36.009 the five -hour usage limits for all paid Cloud 00:10:36.009 --> 00:10:39.029 code users. They also completely removed peak 00:10:39.029 --> 00:10:41.830 hour restrictions and significantly increased 00:10:41.830 --> 00:10:44.940 their Opus API rate limits. Basically, they're 00:10:44.940 --> 00:10:47.600 telling enterprise builders, you can run as many 00:10:47.600 --> 00:10:51.299 complex multi -step tasks as you want, and our 00:10:51.299 --> 00:10:53.820 servers will not throttle you. They're opening 00:10:53.820 --> 00:10:56.340 the floodgates, trying to make Claw the default 00:10:56.340 --> 00:10:59.419 operating system for automated work. But while 00:10:59.419 --> 00:11:02.360 Anthropic targets the enterprise, Google is entering 00:11:02.360 --> 00:11:04.799 the personal wellness space in a very intimate 00:11:04.799 --> 00:11:07.889 way. Yeah, on May 19th, they're launching a new 00:11:07.889 --> 00:11:11.649 AI health coach for $9 .99 a month. This is a 00:11:11.649 --> 00:11:14.029 huge leap from simply tracking your daily steps. 00:11:14.230 --> 00:11:16.629 This system deeply integrates into your life. 00:11:16.850 --> 00:11:19.549 It analyzes your sleep architecture, tracks your 00:11:19.549 --> 00:11:21.809 workout intensity, and monitors your stress levels 00:11:21.809 --> 00:11:24.409 in real time using biometric data. It takes all 00:11:24.409 --> 00:11:26.549 that live data and gives you personalized health 00:11:26.549 --> 00:11:28.809 advice on the fly. If your heart rate variability 00:11:28.809 --> 00:11:31.509 drops, it might suggest a specific breathing 00:11:31.509 --> 00:11:34.440 exercise right before meeting. The intimacy of 00:11:34.440 --> 00:11:37.139 an algorithm knowing your physiological stress 00:11:37.139 --> 00:11:40.059 before you even consciously register it, I mean, 00:11:40.059 --> 00:11:43.120 that is a profound shift in consumer tech. It 00:11:43.120 --> 00:11:45.340 really is. And on the topic of profound shifts, 00:11:45.659 --> 00:11:49.100 OpenAI just dropped a massive update to their 00:11:49.100 --> 00:11:52.100 API that fundamentally changes human communication, 00:11:52.419 --> 00:11:55.299 real -time voice translation. This update now 00:11:55.299 --> 00:11:58.299 supports over 70 input languages. But the real 00:11:58.299 --> 00:12:00.580 breakthrough is the latency. Right. Previous 00:12:00.580 --> 00:12:02.679 translators required you to speak, pause for 00:12:02.679 --> 00:12:04.600 like three seconds while the audio uploaded, 00:12:04.820 --> 00:12:07.100 transcribed, translated, synthesized, and then 00:12:07.100 --> 00:12:09.940 finally played back. This new model does it almost 00:12:09.940 --> 00:12:12.860 instantaneously during a live call. It listens 00:12:12.860 --> 00:12:15.299 and generates the translated audio concurrently, 00:12:15.299 --> 00:12:17.720 perfectly mimicking the cadence and flow of a 00:12:17.720 --> 00:12:20.799 human interpreter. Whoa. Imagine scaling instant 00:12:20.799 --> 00:12:23.279 real -time translation in 70 languages. Right, 00:12:23.379 --> 00:12:25.860 and integrating it seamlessly into every global 00:12:25.860 --> 00:12:28.340 business call, every hospital, every border crossing. 00:12:28.539 --> 00:12:31.039 The friction of language is just dissolving in 00:12:31.039 --> 00:12:33.379 front of us. But of course, with rapid innovation 00:12:33.379 --> 00:12:36.840 comes intense human drama. We are seeing deep 00:12:36.840 --> 00:12:39.000 internal fractures spilling out into the public. 00:12:39.360 --> 00:12:41.820 Yeah, the former CTO of OpenAI recently spoke 00:12:41.820 --> 00:12:44.240 out, and her statements are now tied to Elon 00:12:44.240 --> 00:12:47.080 Musk's ongoing lawsuit against the company. The 00:12:47.080 --> 00:12:49.179 legal filings paint a picture of a leadership 00:12:49.179 --> 00:12:52.620 team in chaos. She accused Sam Altman of lying 00:12:52.620 --> 00:12:55.259 to the board about safety reviews and fostering 00:12:55.259 --> 00:12:57.940 a culture of rapid deployment over careful testing. 00:12:58.330 --> 00:13:00.230 Whether those claims hold up in court remains 00:13:00.230 --> 00:13:02.809 to be seen, but that internal tension exists 00:13:02.809 --> 00:13:05.629 right alongside terrifying new capabilities. 00:13:06.129 --> 00:13:08.850 For instance, ChatGPT just released a new image 00:13:08.850 --> 00:13:11.610 generation model that makes fake screenshots 00:13:11.610 --> 00:13:14.789 almost impossible to spot. Previously, AI struggled 00:13:14.789 --> 00:13:17.129 with rendering crisp, accurate text inside of 00:13:17.129 --> 00:13:19.490 images. Right. If you asked for a fake screenshot 00:13:19.490 --> 00:13:21.789 of a news article or a text message conversation, 00:13:22.389 --> 00:13:24.929 the font would look slightly warped or misspelled. 00:13:25.169 --> 00:13:27.639 Not anymore. The visual fidelity of these new 00:13:27.639 --> 00:13:30.399 generated fakes is absolutely flawless. The pixels 00:13:30.399 --> 00:13:32.799 align perfectly. The fonts are exact matches 00:13:32.799 --> 00:13:36.360 for iOS or Twitter UI. If you still believe every 00:13:36.360 --> 00:13:38.340 screenshot you see online, you need to change 00:13:38.340 --> 00:13:40.659 your mindset today. We are entering a whole new 00:13:40.659 --> 00:13:43.360 era of digital verification where we simply cannot 00:13:43.360 --> 00:13:45.840 trust our own eyes anymore. We really can't. 00:13:46.039 --> 00:13:48.240 To wrap up this rapid fire section, let's look 00:13:48.240 --> 00:13:50.779 at three quick empowered AI tools that caught 00:13:50.779 --> 00:13:53.139 our eye this week, which really show how these 00:13:53.139 --> 00:13:55.679 capabilities are trickling down to specific industries. 00:13:56.019 --> 00:13:58.240 First is Claude Agents for Financial Services. 00:13:58.659 --> 00:14:01.740 Anthropic just dropped 10 pre -built templates 00:14:01.740 --> 00:14:04.480 specifically for banking. This includes agents 00:14:04.480 --> 00:14:06.860 that automate investment research, month -end 00:14:06.860 --> 00:14:09.750 close accounting, and KYC screening. Which is 00:14:09.750 --> 00:14:11.929 the know your customer background check banks 00:14:11.929 --> 00:14:14.549 legally mandate to prevent money laundering. 00:14:14.750 --> 00:14:17.629 Automated KYC alone saves financial institutions 00:14:17.629 --> 00:14:19.889 millions of hours of manual document review. 00:14:20.129 --> 00:14:22.960 Then there is Flow Market. This platform is wild. 00:14:23.080 --> 00:14:25.080 It's essentially a network of AI agents that 00:14:25.080 --> 00:14:27.779 autonomously discover, match, and generate B2B 00:14:27.779 --> 00:14:30.559 deals. Instead of a human sales team cold -calling 00:14:30.559 --> 00:14:32.820 companies, these agents scrape corporate data, 00:14:32.980 --> 00:14:35.759 identify a mutual need, draft the proposal, and 00:14:35.759 --> 00:14:38.440 negotiate the initial terms. It bypasses the 00:14:38.440 --> 00:14:41.259 need for massive human outreach entirely, fundamentally 00:14:41.259 --> 00:14:44.279 changing how businesses procure services. And 00:14:44.279 --> 00:14:47.059 finally, Lingo .dev released version 1 of their 00:14:47.059 --> 00:14:49.700 software. It helps product teams build consistent 00:14:49.700 --> 00:14:53.059 AI translations. It enforces brand voice rules 00:14:53.059 --> 00:14:56.519 across dozens of languages and handles all the 00:14:56.519 --> 00:14:59.120 localization coding automatically. With all this 00:14:59.120 --> 00:15:01.419 rapid Asian innovation and specialized tooling 00:15:01.419 --> 00:15:04.639 flooding the market, how can legacy models possibly 00:15:04.639 --> 00:15:08.039 compete? Legacy models can't rely on raw intelligence 00:15:08.039 --> 00:15:11.539 alone. They must build physical moats and specialized 00:15:11.539 --> 00:15:14.379 hardware ecosystems because the software advantage 00:15:14.379 --> 00:15:17.340 is just evaporating. The real battle is... shifting 00:15:17.340 --> 00:15:20.639 from software to hardware ecosystems. Absolutely. 00:15:20.899 --> 00:15:23.279 Let's take a step back. We have covered a massive 00:15:23.279 --> 00:15:25.879 amount of ground today, from gigawatt power grids 00:15:25.879 --> 00:15:28.340 to personal health coaches. When you look at 00:15:28.340 --> 00:15:31.620 the entire landscape, a very clear central theme 00:15:31.620 --> 00:15:34.440 emerges. We are hitting the absolute limits of 00:15:34.440 --> 00:15:36.379 the AI honeymoon phase. We've spent the last 00:15:36.379 --> 00:15:38.720 two years marveling at the magic tricks. Right. 00:15:38.820 --> 00:15:41.220 The flawless poems, the instantaneous code generation, 00:15:41.460 --> 00:15:43.500 the seemingly weightless cloud. But the magic 00:15:43.500 --> 00:15:46.179 trick is over. And we are now crashing into reality 00:15:46.179 --> 00:15:49.340 on three separate fronts. First, physical grid 00:15:49.340 --> 00:15:53.039 limits. The $18 billion jalapeno chip standoff 00:15:53.039 --> 00:15:55.860 proves that you cannot just wish a gigawatt of 00:15:55.860 --> 00:15:59.059 power into existence. Microsoft and OpenAI are 00:15:59.059 --> 00:16:01.960 fighting over real estate, concrete, and copper. 00:16:02.240 --> 00:16:05.620 Second, we're crashing into complex human psychological 00:16:05.620 --> 00:16:08.789 boundaries. The trusted contact feature is a 00:16:08.789 --> 00:16:11.230 sobering realization for Silicon Valley. Yeah, 00:16:11.330 --> 00:16:14.210 AI can flawlessly simulate empathy, but it cannot 00:16:14.210 --> 00:16:17.049 shoulder actual human liability. When someone 00:16:17.049 --> 00:16:19.230 is in crisis, an algorithm is not enough. It 00:16:19.230 --> 00:16:21.970 requires a human safety net. And finally, we're 00:16:21.970 --> 00:16:24.529 crashing into the sheer velocity of global competition. 00:16:25.210 --> 00:16:27.990 Moonshot AI pulling $200 million in revenue shows 00:16:27.990 --> 00:16:30.110 that the open source community is not waiting 00:16:30.110 --> 00:16:32.370 for permission. The frontier is expanding faster 00:16:32.370 --> 00:16:34.850 than any single regulatory body or company can 00:16:34.850 --> 00:16:37.230 control. So what does this all mean for you listening 00:16:37.230 --> 00:16:39.230 right now? We want you to consider how these 00:16:39.230 --> 00:16:41.629 macro shifts are already touching your daily 00:16:41.629 --> 00:16:43.529 digital boundaries. Are you relying on cloud 00:16:43.529 --> 00:16:45.350 tools that might suddenly change their pricing 00:16:45.350 --> 00:16:47.769 models because their hardware burn rate is completely 00:16:47.769 --> 00:16:49.990 unsustainable? Are your personal and professional 00:16:49.990 --> 00:16:53.090 boundaries secure against flawless AI -generated 00:16:53.090 --> 00:16:55.350 fake screenshots? The physical world and the 00:16:55.350 --> 00:16:57.629 digital world are colliding right now, and the 00:16:57.629 --> 00:16:59.730 shockwaves are going to dictate how we work and 00:16:59.730 --> 00:17:03.299 interact over the next five years. If a multi 00:17:03.299 --> 00:17:05.940 -billion dollar AI still needs a human to review 00:17:05.940 --> 00:17:08.599 a crisis, at what point does superintelligence 00:17:08.599 --> 00:17:10.180 still rely entirely on us?