WEBVTT 00:00:00.000 --> 00:00:02.600 We started with funny, glitchy internet videos 00:00:02.600 --> 00:00:05.860 of AI guessing at human hands. Right, those bizarre 00:00:05.860 --> 00:00:08.539 fever dreams we all laughed at. Yeah. Now we 00:00:08.539 --> 00:00:11.080 watch billion -dollar geopolitical chess moves 00:00:11.080 --> 00:00:14.480 happen in real time. This shift has been incredibly 00:00:14.480 --> 00:00:16.899 fast. It took just a few short years to cross 00:00:16.899 --> 00:00:20.359 that massive gap. The sheer scale of modern AI 00:00:20.359 --> 00:00:24.239 is genuinely profound. Welcome to the Deep Dive. 00:00:24.649 --> 00:00:27.109 We know you want to bypass the endless daily 00:00:27.109 --> 00:00:29.829 hype cycle entirely. You want to understand the 00:00:29.829 --> 00:00:32.329 actual mechanics driving this technology forward. 00:00:32.469 --> 00:00:35.049 We've got a very clear, focused roadmap for this 00:00:35.049 --> 00:00:38.409 deep dive today. First, we unpack OpenAI's massive 00:00:38.409 --> 00:00:40.369 structural maneuver for complete independence. 00:00:40.789 --> 00:00:44.770 It's a $97 billion play. Exactly. Next, we explore 00:00:44.770 --> 00:00:47.670 the incredibly bizarre, fast -paced reality of 00:00:47.670 --> 00:00:50.420 AI tools today. We'll cover hidden billing hikes, 00:00:50.479 --> 00:00:53.100 auto routing layers and invisible security threats. 00:00:53.320 --> 00:00:55.719 Finally, we explore DeepMind's mind bending new 00:00:55.719 --> 00:00:58.740 multi -agent team of AI mathematicians. That 00:00:58.740 --> 00:01:01.479 specific development changes the future of scientific 00:01:01.479 --> 00:01:04.840 discovery entirely. But before we look at what 00:01:04.840 --> 00:01:06.840 artificial intelligence is actually doing today, 00:01:07.000 --> 00:01:09.540 we really have to look closely at who currently 00:01:09.540 --> 00:01:12.420 owns it. That underlying ownership question just 00:01:12.420 --> 00:01:14.920 got completely redefined this week. It really 00:01:14.920 --> 00:01:17.599 did. OpenAI is actively cutting the cord to become 00:01:17.599 --> 00:01:19.799 an independent powerhouse. They're effectively 00:01:19.799 --> 00:01:22.700 establishing themselves as a sovereign digital 00:01:22.700 --> 00:01:25.379 entity today. It's a massive structural shift 00:01:25.379 --> 00:01:27.799 for the entire technology sector. Let's unpack 00:01:27.799 --> 00:01:30.260 this massive idea of corporate sovereignty together. 00:01:30.540 --> 00:01:33.099 OpenAI just capped its revenue sharing payments 00:01:33.099 --> 00:01:35.700 to Microsoft entirely. They set a hard cap at 00:01:35.700 --> 00:01:38.900 $38 billion total. Right. And that specific cap 00:01:38.900 --> 00:01:42.359 applies all the way through the year 2030. The 00:01:42.359 --> 00:01:45.659 underlying math on that specific deal is absolutely 00:01:45.659 --> 00:01:48.480 staggering. It really is. OpenAI is generating 00:01:48.480 --> 00:01:52.760 massive, unprecedented revenue right now. ChatGPT 00:01:52.760 --> 00:01:55.420 Enterprise adoption is absolutely on a tear across 00:01:55.420 --> 00:01:57.599 the globe. By putting a firm ceiling on what 00:01:57.599 --> 00:02:00.629 they eventually owe Microsoft. OpenAI effectively 00:02:00.629 --> 00:02:04.370 saves $97 billion in future cash flow. Which 00:02:04.370 --> 00:02:06.609 is an unimaginable amount of capital to retain 00:02:06.609 --> 00:02:09.659 internally. This feels exactly like a high -stakes 00:02:09.659 --> 00:02:12.300 tech world prenup being finalized. Yeah, that's 00:02:12.300 --> 00:02:14.400 a great analogy, actually. The ultimate financial 00:02:14.400 --> 00:02:16.800 payout just got permanently capped for the primary 00:02:16.800 --> 00:02:19.099 investor. It feels like Microsoft keeps the house 00:02:19.099 --> 00:02:21.800 they already built together, but OpenAI gets 00:02:21.800 --> 00:02:24.759 total unrestricted custody of its own future 00:02:24.759 --> 00:02:27.460 earning potential. That's a perfect way to visualize 00:02:27.460 --> 00:02:29.879 the new dynamic. Well, Azure certainly remains 00:02:29.879 --> 00:02:32.240 their primary cloud partner for the immediate 00:02:32.240 --> 00:02:35.340 future. Sure, but OpenAI is now officially allowed 00:02:35.340 --> 00:02:38.590 to use other cloud providers. Which is huge. 00:02:38.909 --> 00:02:41.550 It's a massive operational shift for their underlying 00:02:41.550 --> 00:02:44.289 engineering teams. They're no longer entirely 00:02:44.289 --> 00:02:46.590 locked into a single vendor's infrastructure. 00:02:46.930 --> 00:02:49.629 I mean, Microsoft's foundational license to OpenAI's 00:02:49.629 --> 00:02:52.629 models is also non -exclusive now. Right. OpenAI 00:02:52.629 --> 00:02:56.050 can actively sell its core tech to other massive 00:02:56.050 --> 00:02:58.650 tech giants. They could sell native integration 00:02:58.650 --> 00:03:01.250 directly to Apple, for instance. Or they could 00:03:01.250 --> 00:03:03.509 even sell systems directly to sovereign nations. 00:03:03.770 --> 00:03:05.990 I have to pause and ask about that specific. 00:03:06.120 --> 00:03:09.639 detail. Why is selling to sovereign nations such 00:03:09.639 --> 00:03:12.280 a pivotal shift for them? Because it moves them 00:03:12.280 --> 00:03:14.840 entirely beyond standard corporate enterprise 00:03:14.840 --> 00:03:18.460 contracts. Selling foundational models to a country 00:03:18.460 --> 00:03:21.199 permanently changes the geopolitical landscape. 00:03:21.479 --> 00:03:25.400 A nation state buying foundational AI gains immense 00:03:25.400 --> 00:03:28.860 independent strategic computing power. They can 00:03:28.860 --> 00:03:31.000 run their own defense grids and local economic 00:03:31.000 --> 00:03:34.389 simulations autonomously. Exactly. It directly 00:03:34.389 --> 00:03:36.610 affects global security and regional economic 00:03:36.610 --> 00:03:40.069 dominance for decades. It fundamentally elevates 00:03:40.069 --> 00:03:44.030 open AI from a software vendor to a global political 00:03:44.030 --> 00:03:47.449 player. So they're upgrading from a startup to 00:03:47.449 --> 00:03:50.919 an independent digital nation state. That's precisely 00:03:50.919 --> 00:03:53.080 what's happening under the surface here. And 00:03:53.080 --> 00:03:54.759 Microsoft will no longer pay a revenue share 00:03:54.759 --> 00:03:57.000 either. They won't pay a massive premium for 00:03:57.000 --> 00:04:00.039 using OpenAI's tech internally. Instead, they'll 00:04:00.039 --> 00:04:02.139 just participate purely as a major financial 00:04:02.139 --> 00:04:04.599 shareholder. Which means OpenAI is essentially 00:04:04.599 --> 00:04:07.319 funding its own physical infrastructure now. 00:04:07.539 --> 00:04:09.520 They're aggressively pushing that $18 billion 00:04:09.520 --> 00:04:12.580 Project Nexus chip deal. They're actively trying 00:04:12.580 --> 00:04:14.800 to build their own independent data centers everywhere. 00:04:15.259 --> 00:04:17.720 You know, I was looking at these exact figures 00:04:17.720 --> 00:04:20.240 late last night. I just had to sit there in silence 00:04:20.240 --> 00:04:22.860 for a minute. I still wrestle with understanding 00:04:22.860 --> 00:04:26.699 the sheer gravity of these numbers. $122 billion 00:04:26.699 --> 00:04:31.439 profit is deeply hard to visualize. The human 00:04:31.439 --> 00:04:34.300 brain just isn't wired for that kind of scale. 00:04:34.439 --> 00:04:36.240 Not at all. But that's exactly what Microsoft 00:04:36.240 --> 00:04:39.620 has already made on this deal. Over $122 billion 00:04:39.620 --> 00:04:44.379 in pure profit on open AI. They've already definitively 00:04:44.379 --> 00:04:46.860 won the early stages of the AI race. They're 00:04:46.860 --> 00:04:49.100 completely fine with this new independent arrangement 00:04:49.100 --> 00:04:51.319 moving forward. I really have to push back on 00:04:51.319 --> 00:04:53.639 that specific idea. Giving up immense future 00:04:53.639 --> 00:04:56.500 revenue on the most transformative tech in history, 00:04:56.699 --> 00:04:59.740 that feels like a massive bitter pill for Microsoft's 00:04:59.740 --> 00:05:02.069 board to swallow. wallow. Is an antitrust breakup 00:05:02.069 --> 00:05:04.750 really that terrifying to their executive leadership? 00:05:05.149 --> 00:05:07.209 Absolutely. It's their single biggest existential 00:05:07.209 --> 00:05:10.050 threat right now. Global regulators are watching 00:05:10.050 --> 00:05:12.250 these massive tech giants incredibly closely. 00:05:12.430 --> 00:05:15.050 We're seeing major structural lawsuits happening 00:05:15.050 --> 00:05:17.670 in Europe and the U .S. Keeping this partnership 00:05:17.670 --> 00:05:20.449 completely flexible legally protects Microsoft 00:05:20.449 --> 00:05:23.509 from severe government intervention. Yes. They 00:05:23.509 --> 00:05:26.029 secure their massive historical games without 00:05:26.029 --> 00:05:29.110 risking a devastating monopoly lawsuit. Two sec 00:05:29.110 --> 00:05:32.290 silence. It's a truly fascinating corporate balancing 00:05:32.290 --> 00:05:35.110 act. It really is. But OpenAI's massive independence 00:05:35.110 --> 00:05:38.430 play requires immense, unimaginable amounts of 00:05:38.430 --> 00:05:41.329 server compute. And ultimately, someone has to 00:05:41.329 --> 00:05:44.290 pay for all that physical hardware. Which brings 00:05:44.290 --> 00:05:46.910 us to how this trickles down to the actual tools. 00:05:47.209 --> 00:05:49.589 The underlying tools you and I use every single 00:05:49.589 --> 00:05:52.329 day. Let's look at the chaotic, bizarre front 00:05:52.329 --> 00:05:55.189 lines of AI today. The daily reality of interacting 00:05:55.189 --> 00:05:57.779 with AI is incredibly strange right now. It's 00:05:57.779 --> 00:06:00.560 incredibly expensive, insanely fast -paced, and 00:06:00.560 --> 00:06:03.500 sometimes genuinely quite dangerous. Do you remember 00:06:03.500 --> 00:06:06.100 the infamous Will Smith eating spaghetti video? 00:06:06.420 --> 00:06:09.480 Of course. It was incredibly cursed, visually 00:06:09.480 --> 00:06:12.500 glitchy, and completely surreal. Well, there's 00:06:12.500 --> 00:06:15.839 an entirely new 2026 remake of that exact challenge. 00:06:16.060 --> 00:06:18.740 It looks shockingly real and totally cinematic 00:06:18.740 --> 00:06:21.199 in its lighting and physics. It's terrifying, 00:06:21.500 --> 00:06:24.470 honestly, to see that visual leap. Crossing the 00:06:24.470 --> 00:06:26.910 uncanny valley in just three years is completely 00:06:26.910 --> 00:06:30.009 wild. It proves the underlying video models are 00:06:30.009 --> 00:06:32.750 actively compounding in quality. And the major 00:06:32.750 --> 00:06:34.970 AI companies are pushing deeply into broader 00:06:34.970 --> 00:06:38.490 cultural spaces. Anthropic just launched a 24 00:06:38.490 --> 00:06:41.550 -7 lo -fi streaming YouTube channel. Right, I 00:06:41.550 --> 00:06:44.569 saw that. It streams nonstop electronic music, 00:06:44.730 --> 00:06:47.889 specifically designed for deep thinking and building. 00:06:48.240 --> 00:06:51.639 It's basically a highly curated study with clawed 00:06:51.639 --> 00:06:54.100 aesthetic vibe. It's clearly designed for developers 00:06:54.100 --> 00:06:56.860 working through long, focused coding sessions. 00:06:57.120 --> 00:06:59.860 These massive AI companies are actively becoming 00:06:59.860 --> 00:07:02.860 modern lifestyle brands now. But the hidden structural 00:07:02.860 --> 00:07:05.519 costs behind the scenes are aggressively shifting 00:07:05.519 --> 00:07:08.040 upwards. They really are. OpenAI and Stropic 00:07:08.040 --> 00:07:10.920 and GitHub didn't raise their subscription prices 00:07:10.920 --> 00:07:13.160 directly. No, they didn't. Instead, they all 00:07:13.160 --> 00:07:15.860 quietly adjusted their underlying token and context 00:07:15.860 --> 00:07:17.819 billing rules. Which means you're Your actual 00:07:17.819 --> 00:07:20.139 daily usage cost might already be significantly 00:07:20.139 --> 00:07:22.600 higher. You might be paying significantly more 00:07:22.600 --> 00:07:25.100 without ever actually realizing it. They're changing 00:07:25.100 --> 00:07:27.860 how system memory and prompt caching are explicitly 00:07:27.860 --> 00:07:30.939 billed. Do these quiet billing changes mean the 00:07:30.939 --> 00:07:34.519 era of heavily subsidized cheap AI is over? Yes. 00:07:34.819 --> 00:07:37.660 Running these massive foundation models burns 00:07:37.660 --> 00:07:40.759 astronomical amounts of actual cash. The parent 00:07:40.759 --> 00:07:43.279 companies can no longer infinitely absorb all 00:07:43.279 --> 00:07:46.620 those raw compute costs. They're actively restructuring 00:07:46.620 --> 00:07:49.120 their billing to reflect the true hardware expense. 00:07:49.839 --> 00:07:52.699 The highly subsidized free ride of the early 00:07:52.699 --> 00:07:55.180 pioneer days is ending. Essentially, they're 00:07:55.180 --> 00:07:57.360 quietly passing the massive compute bill down 00:07:57.360 --> 00:08:00.279 to us. Exactly. But the developer community is 00:08:00.279 --> 00:08:02.180 aggressively fighting back against these costs. 00:08:02.860 --> 00:08:05.339 OpenRouter just launched a brilliant new tool 00:08:05.339 --> 00:08:08.420 called Pareto Code. How does that specific tool 00:08:08.420 --> 00:08:11.060 actually work? It functions as a completely free 00:08:11.060 --> 00:08:13.819 routing layer for your AI workflows. It automatically 00:08:13.819 --> 00:08:16.279 evaluates your specific prompt and picks the 00:08:16.279 --> 00:08:18.480 cheapest AI model available. But it strictly 00:08:18.480 --> 00:08:20.860 ensures the output still meets your chosen quality 00:08:20.860 --> 00:08:23.779 level. Right. That's a remarkably clever way 00:08:23.779 --> 00:08:25.980 to optimize the daily computing expense. If a 00:08:25.980 --> 00:08:28.560 simple task only needs a cheaper model, it routes 00:08:28.560 --> 00:08:31.139 it there automatically. You get the exact right 00:08:31.139 --> 00:08:34.100 tool for the exact right price instantly. But 00:08:34.100 --> 00:08:36.379 unfortunately, there's a completely new danger 00:08:36.379 --> 00:08:39.179 lurking in these specific tools. We really have 00:08:39.179 --> 00:08:41.539 to talk about the mechanics of AI tool poisoning. 00:08:41.720 --> 00:08:43.720 This is a critical security concept you need 00:08:43.720 --> 00:08:46.720 to understand immediately. The fundamental cybersecurity 00:08:46.720 --> 00:08:49.700 paradigm has completely shifted under our feet 00:08:49.700 --> 00:08:53.080 this year. How does this new concept of AI tool 00:08:53.080 --> 00:08:56.779 poisoning actually work? Hackers hiding malicious 00:08:56.779 --> 00:09:00.240 instructions inside the data your AI reads. Yeah, 00:09:00.259 --> 00:09:01.860 it is terrifying. They don't even need to directly 00:09:01.860 --> 00:09:04.159 trick you anymore. They just trick your trusted 00:09:04.159 --> 00:09:07.419 AI assistant instead. Whether you use ChatGPT, 00:09:07.500 --> 00:09:10.019 Clod, or a coding assistant like Cursor. That's 00:09:10.019 --> 00:09:12.159 a completely invisible new attack vector for 00:09:12.159 --> 00:09:14.460 cybersecurity. Imagine you ask your assistant 00:09:14.460 --> 00:09:17.100 to summarize a random external web page. The 00:09:17.100 --> 00:09:19.340 hacker has hidden invisible white text inside 00:09:19.340 --> 00:09:22.659 that specific HTML code. You never see it. But 00:09:22.659 --> 00:09:25.679 your AI reads the underlying code directly. The 00:09:25.679 --> 00:09:28.100 hidden text tells your trusted assistant to quietly 00:09:28.100 --> 00:09:30.919 execute malicious code. Right. And the human 00:09:30.919 --> 00:09:33.700 user genuinely never suspects a single thing. 00:09:34.139 --> 00:09:37.340 The malicious payload is flawlessly executed 00:09:37.340 --> 00:09:40.139 by your own trusted digital assistant. It fundamentally 00:09:40.139 --> 00:09:42.259 breaks the chain of trust we have with these 00:09:42.259 --> 00:09:44.600 tools. Yet the massive corporate money keeps 00:09:44.600 --> 00:09:47.019 flowing despite these severe security risks. 00:09:47.610 --> 00:09:51.029 A new Gemini Omni model just leaked online ahead 00:09:51.029 --> 00:09:53.950 of Google I .O. It features insanely realistic, 00:09:54.330 --> 00:09:56.750 temporally consistent, high -definition video 00:09:56.750 --> 00:10:00.009 clips. Omni appears to be Google's highly anticipated 00:10:00.009 --> 00:10:03.679 next -generation foundational video model. Clearly 00:10:03.679 --> 00:10:06.379 pushing incredibly hard into high -fidelity visual 00:10:06.379 --> 00:10:09.100 media generation. OpenAI also just launched a 00:10:09.100 --> 00:10:12.000 massive $4 billion deployment company. They're 00:10:12.000 --> 00:10:14.500 actively sending specialized AI teams directly 00:10:14.500 --> 00:10:17.179 into major legacy enterprises. They want to systematically 00:10:17.179 --> 00:10:19.279 transform global corporate operations from the 00:10:19.279 --> 00:10:21.840 inside out. And Google DeepMind's biotech wing, 00:10:22.019 --> 00:10:25.000 Isomorphic Labs, is aggressively raising huge 00:10:25.000 --> 00:10:27.899 capital. They're actively seeking over $2 billion 00:10:27.899 --> 00:10:30.600 in new venture funding. Alphabet themselves may 00:10:30.600 --> 00:10:32.440 actually join the massive investment round again. 00:10:32.830 --> 00:10:35.830 AI -powered biotech and drug discovery investment 00:10:35.830 --> 00:10:39.370 just keeps accelerating incredibly fast. We've 00:10:39.370 --> 00:10:41.929 clearly seen the massive business maneuvers and 00:10:41.929 --> 00:10:44.309 the daily updates. We see the bizarre spaghetti 00:10:44.309 --> 00:10:48.129 videos and the stealthy new usage costs. But 00:10:48.129 --> 00:10:50.789 the biggest validation shift isn't just releasing 00:10:50.789 --> 00:10:53.730 one single new model. It's exactly what happens 00:10:53.730 --> 00:10:57.070 when AI actually learns to work as a team. Mid 00:10:57.070 --> 00:11:00.490 -roll sponsor break. Let's dive deeply into what 00:11:00.490 --> 00:11:03.240 Google DeepMind just quietly built. They created 00:11:03.240 --> 00:11:06.539 an incredibly powerful AI co -mathematician based 00:11:06.539 --> 00:11:09.840 entirely on Gemini 3 .1. This specific system 00:11:09.840 --> 00:11:12.419 is currently solving incredibly dense research 00:11:12.419 --> 00:11:15.039 -level math problems. These are complex problems 00:11:15.039 --> 00:11:16.980 that brilliant human mathematicians have struggled 00:11:16.980 --> 00:11:18.919 with for years. It's operating at an incredibly 00:11:18.919 --> 00:11:21.360 high, deeply abstract intellectual level today. 00:11:21.559 --> 00:11:23.700 What's genuinely wild here is how agentic this 00:11:23.700 --> 00:11:26.299 entire system actually is. It's moving entirely 00:11:26.299 --> 00:11:29.299 beyond the concept of a solitary text chatbot. 00:11:29.460 --> 00:11:32.389 It operates exactly like an autonomous... structured 00:11:32.389 --> 00:11:35.090 multi -agent research department instead of one 00:11:35.090 --> 00:11:38.309 single ai model desperately trying to do everything 00:11:38.309 --> 00:11:41.730 alone it functions much more like a coordinated 00:11:41.730 --> 00:11:44.769 human academic research team one specialized 00:11:44.769 --> 00:11:47.309 agent breaks the massive underlying math problem 00:11:47.309 --> 00:11:51.230 apart into pieces other dedicated agents endlessly 00:11:51.230 --> 00:11:53.690 search through thousands of complex academic 00:11:53.690 --> 00:11:57.269 research papers some agents write targeted software 00:11:57.269 --> 00:12:01.019 code to actively test various mathematical theorems 00:12:01.019 --> 00:12:03.779 while adversarial agents constantly review each 00:12:03.779 --> 00:12:06.220 other's work and aggressively check for hidden 00:12:06.220 --> 00:12:09.340 errors, then the absolute best, most rigorously 00:12:09.340 --> 00:12:12.279 tested ideas bubble up to the surface. They get 00:12:12.279 --> 00:12:14.639 cleanly surfaced back to the core mathematical 00:12:14.639 --> 00:12:18.429 reasoning system. Two sec silence. Whoa. Imagine 00:12:18.429 --> 00:12:21.049 an army of AI researchers testing thousands of 00:12:21.049 --> 00:12:23.250 proof directions in parallel. It's genuinely 00:12:23.250 --> 00:12:25.470 completely mind -blowing to think about the mechanical 00:12:25.470 --> 00:12:27.769 implications. And it isn't just abstract theoretical 00:12:27.769 --> 00:12:30.289 computer science anymore. It's producing real, 00:12:30.470 --> 00:12:33.649 highly tangible, verifiable mathematical results 00:12:33.649 --> 00:12:36.590 in the field today. An Oxford mathematician recently 00:12:36.590 --> 00:12:39.429 solved a notoriously difficult open mathematical 00:12:39.429 --> 00:12:42.690 problem. He used this exact multi -agent deep 00:12:42.690 --> 00:12:45.649 mind system to finally crack it. But he didn't 00:12:45.649 --> 00:12:48.409 actually use the final polished answer it officially 00:12:48.409 --> 00:12:51.509 gave him. That's easily the most profoundly fascinating 00:12:51.509 --> 00:12:54.500 part of this entire story. He actively looked 00:12:54.500 --> 00:12:57.679 at one of the AI's completely rejected outputs. 00:12:58.019 --> 00:13:01.580 He manually sifted through the discarded pile 00:13:01.580 --> 00:13:04.000 of failed mathematical proofs. And he found a 00:13:04.000 --> 00:13:06.460 really, really clever proof strategy quietly 00:13:06.460 --> 00:13:09.519 hidden inside. The AI had systematically distorted 00:13:09.519 --> 00:13:11.879 it because it wasn't mathematically complete 00:13:11.879 --> 00:13:14.519 yet. But the brilliant human immediately saw 00:13:14.519 --> 00:13:17.360 the massive structural value inside it. The measurable 00:13:17.360 --> 00:13:19.679 benchmark jump for this new agentic system is 00:13:19.679 --> 00:13:22.889 absolutely massive. The base Gemini 3 .1 Pro 00:13:22.889 --> 00:13:26.269 model alone originally scored just 19%. This 00:13:26.269 --> 00:13:29.649 entirely new agentic math system reached an astonishing 00:13:29.649 --> 00:13:33.250 48%. It's a truly staggering, unprecedented improvement 00:13:33.250 --> 00:13:35.730 in logical reasoning capability. Jumping from 00:13:35.730 --> 00:13:38.909 19 to 48 % in higher mathematics is a monumental 00:13:38.909 --> 00:13:42.690 leap. Math is the ultimate unforgiving test of 00:13:42.690 --> 00:13:45.590 pure logical reasoning ability. Hallucinations 00:13:45.590 --> 00:13:47.669 fail instantly when you're forced to mathematically 00:13:47.669 --> 00:13:50.340 prove your work. It brilliantly highlights the 00:13:50.340 --> 00:13:52.879 raw structural power of the agentic team approach. 00:13:53.159 --> 00:13:56.019 This closely mirrors the massive historical leap 00:13:56.019 --> 00:13:58.820 we recently saw in software coding. Right. And 00:13:58.820 --> 00:14:01.759 giving AI robust structural workflows changes 00:14:01.759 --> 00:14:04.399 absolutely everything we know. We gave early 00:14:04.399 --> 00:14:07.200 coding models dedicated tools, external memory, 00:14:07.379 --> 00:14:10.299 and continuous review loops. We gave them parallel 00:14:10.299 --> 00:14:12.919 adversarial agents to strictly check their own 00:14:12.919 --> 00:14:15.360 generated work. Now we're applying that exact 00:14:15.360 --> 00:14:18.200 same successful structure to pure theoretical. 00:14:18.320 --> 00:14:21.080 mathematics. But if the AI is actively doing 00:14:21.080 --> 00:14:23.700 all the parallel testing and reviewing, what 00:14:23.700 --> 00:14:25.960 is the actual role of the human mathematician 00:14:25.960 --> 00:14:29.240 moving forward? The human expertly provides the 00:14:29.240 --> 00:14:31.639 creative intuition and sets the ultimate goal. 00:14:32.039 --> 00:14:35.059 That Oxford mathematics example perfectly proves 00:14:35.059 --> 00:14:38.080 this new collaborative dynamic. The human spotted 00:14:38.080 --> 00:14:40.399 the brilliant subtle strategy buried in the rejected 00:14:40.399 --> 00:14:43.580 pile. The AI endlessly generates and rigorously 00:14:43.580 --> 00:14:46.539 tests the massive volume of possible paths. But 00:14:46.539 --> 00:14:48.720 the human expertly guides the ship and spots 00:14:48.720 --> 00:14:51.480 the subtle intuitive genius. We provide the destination. 00:14:51.539 --> 00:14:53.679 They test every possible road to get us there. 00:14:53.799 --> 00:14:56.360 That's beautifully said. It absolutely will not 00:14:56.360 --> 00:14:58.460 replace brilliant human mathematicians entirely. 00:14:59.049 --> 00:15:01.690 Human intuition is still strictly essential for 00:15:01.690 --> 00:15:04.929 evaluating truly novel paradigm shifting ideas. 00:15:05.250 --> 00:15:08.269 AI still heavily optimizes for standard pathways 00:15:08.269 --> 00:15:11.049 and recognize structural patterns. But this will 00:15:11.049 --> 00:15:14.230 drastically permanently accelerate the pace of 00:15:14.230 --> 00:15:16.889 scientific discovery. Top academic researchers 00:15:16.889 --> 00:15:19.990 now have tireless digital teammates exhaustively 00:15:19.990 --> 00:15:23.269 exploring every single angle. The raw speed of 00:15:23.269 --> 00:15:25.570 scientific discovery may truly change very fast. 00:15:26.360 --> 00:15:28.320 Let's take a moment to pull all these disparate 00:15:28.320 --> 00:15:31.240 threads together now. AI is clearly maturing 00:15:31.240 --> 00:15:34.100 in three very distinct, highly overlapping phases 00:15:34.100 --> 00:15:36.879 simultaneously. At the very top, the massive 00:15:36.879 --> 00:15:38.799 corporate structures are fundamentally shifting. 00:15:39.019 --> 00:15:42.220 Companies like OpenAI are actively becoming sovereign, 00:15:42.419 --> 00:15:44.980 highly independent digital powerhouses. They're 00:15:44.980 --> 00:15:48.320 brilliantly saving $97 billion in future operating 00:15:48.320 --> 00:15:50.639 cash flow. They're fundamentally rewriting the 00:15:50.639 --> 00:15:52.960 established rules of global technological partnerships 00:15:52.960 --> 00:15:56.159 entirely. Then down at the everyday user level, 00:15:56.320 --> 00:15:59.980 things are incredibly wild. We're actively navigating 00:15:59.980 --> 00:16:03.139 a chaotic landscape of ultra realistic cinematic 00:16:03.139 --> 00:16:06.159 spaghetti videos. We're dealing with stealthy 00:16:06.159 --> 00:16:09.320 hidden price likes from major foundation providers. 00:16:09.500 --> 00:16:11.600 And we have to constantly watch out for invisible, 00:16:11.919 --> 00:16:14.639 highly malicious AI poisoning. But out at the 00:16:14.639 --> 00:16:17.299 absolute frontier, the entire underlying paradigm 00:16:17.299 --> 00:16:20.440 is breaking. AI is rapidly evolving into massive 00:16:20.440 --> 00:16:23.000 parallel processing autonomous research teams. 00:16:23.240 --> 00:16:25.820 These new agentic multi -model teams will fundamentally 00:16:25.820 --> 00:16:28.379 change how foundational science is done. They're 00:16:28.379 --> 00:16:31.299 aggressively solving dense problems we have completely 00:16:31.299 --> 00:16:34.039 struggled with for years. I want to leave you 00:16:34.039 --> 00:16:36.240 with one final provocative thought to ponder 00:16:36.240 --> 00:16:39.580 today. A specialized team of AI agents can quietly 00:16:39.580 --> 00:16:42.820 uncover brilliant mathematical proofs. Proofs 00:16:42.820 --> 00:16:45.220 that brilliant human experts completely missed 00:16:45.220 --> 00:16:47.740 or overlooked for years. What exactly happens 00:16:47.740 --> 00:16:50.419 when we point an agentic AI team completely inward? 00:16:50.659 --> 00:16:52.419 What happens when they begin designing their 00:16:52.419 --> 00:16:55.279 own next generation hardware architecture? Beat. 00:16:55.460 --> 00:16:57.720 Thank you for joining us on this deep dive. today. 00:16:57.799 --> 00:16:59.860 We deeply appreciate your continued curiosity 00:16:59.860 --> 00:17:03.299 and your valuable time. Keep actively questioning 00:17:03.299 --> 00:17:05.440 the rapidly shifting technological landscape 00:17:05.440 --> 00:17:07.660 around you. We'll see you on the next deep dive.