WEBVTT 00:00:00.000 --> 00:00:02.720 The fundamental shift in AI right now isn't just 00:00:02.720 --> 00:00:05.820 about getting slightly better text output. We 00:00:05.820 --> 00:00:08.679 are watching the technology move from creating 00:00:08.679 --> 00:00:13.259 static art and video to becoming a full, dynamic, 00:00:13.699 --> 00:00:16.399 self -contained simulation engine. And that accountability 00:00:16.399 --> 00:00:19.199 is revolutionary. The concept of world models 00:00:19.199 --> 00:00:22.019 is officially here, and that changes the fundamental 00:00:22.019 --> 00:00:25.000 operating floor for almost every industry, especially 00:00:25.000 --> 00:00:27.300 one massive sector we need to talk about immediately. 00:00:27.760 --> 00:00:30.039 Welcome to the Deep Dive. Our mission today is 00:00:30.039 --> 00:00:31.940 to go straight into the recent source material 00:00:31.940 --> 00:00:34.159 you shared and unpack the acceleration curve 00:00:34.159 --> 00:00:37.280 that is pushing AI rapidly toward 2026. This 00:00:37.280 --> 00:00:39.560 isn't simple iteration. This is a step change 00:00:39.560 --> 00:00:41.840 in capability. Yeah, it's a lot to synthesize 00:00:41.840 --> 00:00:44.140 quickly, but the patterns are sharp. Absolutely. 00:00:44.340 --> 00:00:46.619 We're going to unpack the AI takeover of the 00:00:46.619 --> 00:00:50.240 $190 billion gaming industry to find that powerful 00:00:50.240 --> 00:00:52.820 but sometimes intimidating automation tool known 00:00:52.820 --> 00:00:56.200 as NAN, discuss the strategic implications of 00:00:56.200 --> 00:00:59.130 major acquisitions like NVIDIA and Grok, and 00:00:59.130 --> 00:01:01.630 analyze why trust is quickly becoming the new 00:01:01.630 --> 00:01:05.090 frontier for multi -hour AI agents that are currently 00:01:05.090 --> 00:01:08.170 emerging. OK, let's unpack this, starting with 00:01:08.170 --> 00:01:10.230 the rise of these digital worlds. So if we look 00:01:10.230 --> 00:01:12.170 at the core technical breakthrough first, it's 00:01:12.170 --> 00:01:14.200 the world models. A simple definition for you 00:01:14.200 --> 00:01:18.920 would be AI systems that generate full 3D interactive 00:01:18.920 --> 00:01:21.359 game environments directly from a simple text 00:01:21.359 --> 00:01:22.780 prompt. We're not talking about rendering a picture. 00:01:22.840 --> 00:01:25.359 We're talking about generating a playable space. 00:01:25.680 --> 00:01:27.900 And the scale of this impact is instant, isn't 00:01:27.900 --> 00:01:30.760 it? We're talking about the entire $190 billion 00:01:30.760 --> 00:01:33.379 gaming industry feeling this shift right now. 00:01:33.540 --> 00:01:36.260 That's the key. These tools can replace months 00:01:36.260 --> 00:01:38.959 of highly technical work, things like asset creation, 00:01:39.420 --> 00:01:42.260 texture mapping, with just minutes of AI generation. 00:01:42.349 --> 00:01:45.549 You feed the model something complex, maybe jungle 00:01:45.549 --> 00:01:48.290 temple level with lava traps and hostile AI spiders, 00:01:48.870 --> 00:01:51.409 and it instantly creates a fully rendered playable 00:01:51.409 --> 00:01:54.549 3D space. That speed is just astonishing, and 00:01:54.549 --> 00:01:56.769 we're seeing that efficiency translate directly 00:01:56.769 --> 00:01:59.510 into corporate metrics. The developers of Aliens 00:01:59.510 --> 00:02:02.650 vs. Zombies, Game Gears, they publicly reported 00:02:02.650 --> 00:02:05.629 a four times increase in their development speed 00:02:05.629 --> 00:02:07.989 using these tools. Right, because the AI handles 00:02:07.989 --> 00:02:09.889 the bulk of the tedious work. It creates the 00:02:09.889 --> 00:02:12.789 initial geometry, applies base textures, lighting. 00:02:13.150 --> 00:02:15.469 This frees up human designers to focus on artistic 00:02:15.469 --> 00:02:18.810 refinement, unique assets, and the game experience 00:02:18.810 --> 00:02:21.430 itself. It goes beyond environments too. We also 00:02:21.430 --> 00:02:23.810 saw big players integrating AI for character 00:02:23.810 --> 00:02:26.370 depths, that Fortnite example, an AI -driven 00:02:26.370 --> 00:02:29.210 Darth Vader NPC. That was a fascinating collaborative 00:02:29.210 --> 00:02:31.349 build between Google, Eleven Labs for the Voice, 00:02:31.409 --> 00:02:34.310 and Disney. And this tech isn't locked away in 00:02:34.310 --> 00:02:37.229 some lab. We saw Runway launching their own world 00:02:37.229 --> 00:02:39.949 model back in December, and World Labs released 00:02:39.949 --> 00:02:43.199 a system called Marv - It means game engines 00:02:43.199 --> 00:02:45.360 are rapidly becoming conversational. You just 00:02:45.360 --> 00:02:47.659 talk to them and they build. But here's the tension, 00:02:47.860 --> 00:02:49.680 right? While a corporate side is celebrating 00:02:49.680 --> 00:02:52.520 these massive efficiency gains, not everyone 00:02:52.520 --> 00:02:56.400 is cheering. Exactly. Six European game developer 00:02:56.400 --> 00:02:58.919 unions recently raised significant concerns. 00:02:59.520 --> 00:03:01.580 They're saying that AI adoption is being forced 00:03:01.580 --> 00:03:04.199 onto production teams, and they argue it's actually 00:03:04.199 --> 00:03:06.699 worsening working conditions. So the tension 00:03:06.699 --> 00:03:09.479 is cultural and technical. Beyond job security, 00:03:10.000 --> 00:03:11.759 what is the core worry developers are talking 00:03:11.759 --> 00:03:14.199 about? It centers on what they call AI slop. 00:03:14.939 --> 00:03:17.099 The concern is that quality control just drops 00:03:17.099 --> 00:03:19.379 off a cliff when everything is generated so quickly. 00:03:19.879 --> 00:03:22.479 You get a model churning out huge volumes of 00:03:22.479 --> 00:03:25.360 content that meets the basic prompt, a jungle, 00:03:25.599 --> 00:03:28.860 a trap, a spider. But it lacks refinement, artistic 00:03:28.860 --> 00:03:32.310 intention, and originality. So if 2023 was the 00:03:32.310 --> 00:03:36.569 year of AI art and 2024 was AI video, then 2025 00:03:36.569 --> 00:03:39.169 and 2026 are definitely shaping up to be the 00:03:39.169 --> 00:03:42.590 year of AI as the full simulation engine. Beyond 00:03:42.590 --> 00:03:44.409 the jobs concern, what's the biggest technical 00:03:44.409 --> 00:03:47.050 challenge to this rapid adoption of world models? 00:03:47.310 --> 00:03:49.710 Maintaining quality control over that flood of 00:03:49.710 --> 00:03:52.169 generated content is the immediate limiting factor. 00:03:52.349 --> 00:03:54.710 That makes perfect sense. So we've talked about 00:03:54.710 --> 00:03:57.469 AI simulating worlds, but the real challenge 00:03:57.469 --> 00:04:00.289 begins when that AI starts taking action inside 00:04:00.289 --> 00:04:02.949 your workflow. That shifts our focus from simulation 00:04:02.949 --> 00:04:05.969 to agents and automation. Yes. We're moving into 00:04:05.969 --> 00:04:08.930 AI agents, automated systems that perform multiple 00:04:08.930 --> 00:04:11.610 steps across different applications without needing 00:04:11.610 --> 00:04:13.409 constant human oversight once they're kicked 00:04:13.409 --> 00:04:15.729 off. And the tool that often shows people just 00:04:15.729 --> 00:04:17.850 how deep this rabbit hole goes and sometimes 00:04:17.850 --> 00:04:20.730 scares them a bit is N -AID. It does look complex 00:04:20.730 --> 00:04:23.329 because it is powerful. NAN is one of the most 00:04:23.329 --> 00:04:25.970 powerful no -code tools for building these real 00:04:25.970 --> 00:04:28.670 multi -step AI agents. So for the listener trying 00:04:28.670 --> 00:04:30.870 to grasp its power, how does it actually work? 00:04:31.029 --> 00:04:33.589 It's like stacking Lego blocks of data and logic. 00:04:34.029 --> 00:04:36.850 You pull a trigger block, say new email and Gmail, 00:04:37.290 --> 00:04:40.129 then you chain a logic block, analyze with GPT, 00:04:40.410 --> 00:04:43.389 and then an action block, post a summary to Notion. 00:04:43.769 --> 00:04:46.189 It allows for full automation. auto posting, 00:04:46.449 --> 00:04:49.930 auto replying, auto writing. And crucially, it 00:04:49.930 --> 00:04:52.370 connects AI to the services you use every day, 00:04:52.490 --> 00:04:55.290 like Sheets, Discord, Telegram. It's the central 00:04:55.290 --> 00:04:57.449 nervous system. But you can't master something 00:04:57.449 --> 00:04:59.509 that powerful instantly. I think a lot of us 00:04:59.509 --> 00:05:01.449 feel that. I mean, I'll admit, I still wrestle 00:05:01.449 --> 00:05:04.029 with prompt drift myself, even in simple automations, 00:05:04.110 --> 00:05:06.970 when I try to get an AI to maintain context over 00:05:06.970 --> 00:05:09.449 just a few steps. And that's a necessary admission, 00:05:09.730 --> 00:05:12.470 because expertise is built slowly. The goal isn't 00:05:12.470 --> 00:05:14.970 to master NNN1 sitting, but to make sure your 00:05:14.970 --> 00:05:17.589 brain already gets the underlying logic. If you 00:05:17.589 --> 00:05:20.189 understand how data moves from A to B, then when 00:05:20.189 --> 00:05:22.470 you move to advanced agents later, the ones that 00:05:22.470 --> 00:05:25.069 run for hours, it all becomes much easier. So 00:05:25.069 --> 00:05:28.170 if NNN is so powerful, why do experts emphasize 00:05:28.170 --> 00:05:30.490 building that slow foundational understanding 00:05:30.490 --> 00:05:33.829 first? Understanding the underlying logic prevents 00:05:33.829 --> 00:05:36.689 critical cascading errors once you move into 00:05:36.689 --> 00:05:39.920 deep automation. Let's pivot now to the broader 00:05:39.920 --> 00:05:42.040 financial picture, because the money tells us 00:05:42.040 --> 00:05:44.379 just how existential this race has become. We 00:05:44.379 --> 00:05:47.060 saw a massive acquisition. Nvidia bought Grok 00:05:47.060 --> 00:05:50.220 for $20 billion. That valuation is incredible. 00:05:50.519 --> 00:05:53.500 It nearly tripled Grok's previous $7 billion 00:05:53.500 --> 00:05:56.680 number. Why such a huge jump for chip technology? 00:05:56.980 --> 00:05:59.139 This is the definitive answer to where the arms 00:05:59.139 --> 00:06:02.639 race is focused. Nvidia clearly saw Grok's AI 00:06:02.639 --> 00:06:05.819 chip technology. their LPU architecture as a 00:06:05.819 --> 00:06:08.579 real, fundamental, existential threat that needed 00:06:08.579 --> 00:06:11.100 to be absorbed immediately, no matter the cost. 00:06:11.220 --> 00:06:13.220 And the threat wasn't based on size, it was speed, 00:06:13.339 --> 00:06:16.160 right? Precisely. Grok specialized in low latency 00:06:16.160 --> 00:06:19.040 inference. While traditional GPUs are great for 00:06:19.040 --> 00:06:21.500 training models, the heavy lifting Grok's tech 00:06:21.500 --> 00:06:24.300 was built for lightning -fast deployment. For 00:06:24.300 --> 00:06:26.259 these emerging multi -hour agents we're discussing, 00:06:26.459 --> 00:06:28.519 that low latency is crucial for decision -making. 00:06:28.680 --> 00:06:31.379 So, Nvidia didn't just buy a company. They bought 00:06:31.379 --> 00:06:33.899 an insurance policy against the disruptive architecture. 00:06:34.339 --> 00:06:36.139 Shifting gears a bit to the culture around all 00:06:36.139 --> 00:06:39.180 this. Even as the tech gets faster, public trust 00:06:39.180 --> 00:06:43.439 remains extremely volatile. We saw some strange 00:06:43.439 --> 00:06:45.899 news from the U .S. government. We did. The U 00:06:45.899 --> 00:06:47.579 .S. Department of Homeland Security released 00:06:47.579 --> 00:06:51.500 an AI -generated video of Santa Claus acting 00:06:51.500 --> 00:06:53.639 as a deportation agent part of a naughty list 00:06:53.639 --> 00:06:56.540 campaign. And the public reaction was not good. 00:06:56.699 --> 00:07:00.300 People widely deem the video disgusting, especially 00:07:00.300 --> 00:07:03.259 using a figure like Santa for that kind of message. 00:07:03.439 --> 00:07:05.339 And there was a fascinating historical irony 00:07:05.339 --> 00:07:07.639 to it. St. Nicholas was originally from Turkey, 00:07:08.120 --> 00:07:10.160 which led some people to call it a cultural cell 00:07:10.160 --> 00:07:12.399 phone. It just shows how fragile public trust 00:07:12.399 --> 00:07:14.910 is here. On a much lighter note, though, we also 00:07:14.910 --> 00:07:18.370 saw smaller, helpful integrations. Notebook LM 00:07:18.370 --> 00:07:21.949 teased new British voices for 2026. And Google 00:07:21.949 --> 00:07:24.730 shared useful tips on how to maximize their slide 00:07:24.730 --> 00:07:27.490 deck feature. Right, that balance is key. And 00:07:27.490 --> 00:07:29.269 addressing that concern of over -reliance, we 00:07:29.269 --> 00:07:31.629 saw a Nobel Prize -winning physicist sharing 00:07:31.629 --> 00:07:34.750 strategies on how to leverage AI without completely 00:07:34.750 --> 00:07:37.050 outsourcing your critical thinking. That's an 00:07:37.050 --> 00:07:39.350 essential skill for every learner today. So what 00:07:39.350 --> 00:07:42.050 key insight does that massive grok acquisition 00:07:42.050 --> 00:07:45.129 give us about the state of the AI arms race? 00:07:45.750 --> 00:07:48.230 Companies view superior chip technology for inference 00:07:48.230 --> 00:07:51.889 speed as an existential threat, justifying enormous 00:07:51.889 --> 00:07:54.490 valuations to eliminate competition. And that 00:07:54.490 --> 00:07:58.110 brings us to the core story of 2025, acceleration. 00:07:58.610 --> 00:08:00.730 The breakthrough isn't just that the AI is smarter, 00:08:00.990 --> 00:08:03.689 it's about how fast the old limits on task length 00:08:03.689 --> 00:08:05.750 are breaking. Yeah, consider where we were just 00:08:05.750 --> 00:08:09.189 12 months ago. Gemerite 1 .5 was new, multimodal 00:08:09.189 --> 00:08:11.790 understanding is clumsy, and agent -style projects 00:08:11.790 --> 00:08:14.310 were just starting to appear. But the critical 00:08:14.310 --> 00:08:16.790 metric, the one that drives real organizational 00:08:16.790 --> 00:08:19.129 change, has been task length. The amount of time 00:08:19.129 --> 00:08:21.389 and complexity an AI can handle autonomously 00:08:21.389 --> 00:08:24.930 is now doubling every seven months. Whoa. I mean, 00:08:24.990 --> 00:08:27.329 just imagine scaling that to a billion queries 00:08:27.329 --> 00:08:30.269 that don't need human input. That compounding 00:08:30.269 --> 00:08:32.629 speed is what's different now. When models could 00:08:32.629 --> 00:08:35.149 only handle short tasks, the human was obsessed 00:08:35.149 --> 00:08:37.789 with the prompt quality. But now, these models 00:08:37.789 --> 00:08:40.629 can actually think for hours across full repositories. 00:08:40.889 --> 00:08:42.850 Which is why the problem set for organizations 00:08:42.850 --> 00:08:46.070 has shifted entirely. It's no longer about maximizing 00:08:46.070 --> 00:08:48.870 the quality of one single prompt. It's about 00:08:48.870 --> 00:08:51.970 managing three new concepts. Planning, state, 00:08:52.350 --> 00:08:55.389 and trust. Exactly. When an agent runs for hours, 00:08:55.629 --> 00:08:58.490 it needs a reliable internal memory. We can define 00:08:58.490 --> 00:09:00.990 planning as its to -do list, the sequence of 00:09:00.990 --> 00:09:04.090 steps. State is its internal clipboard. It has 00:09:04.090 --> 00:09:06.210 to remember what step it's on, what it just did. 00:09:06.269 --> 00:09:08.789 And trust, that third piece is the audit trail. 00:09:09.190 --> 00:09:11.210 That's the proof of what the agent did, why it 00:09:11.210 --> 00:09:13.590 made those decisions over that long time frame 00:09:13.590 --> 00:09:16.210 that's mandatory for any regulated industry. 00:09:16.409 --> 00:09:19.389 And if this curve continues into 2026, AI will 00:09:19.389 --> 00:09:22.210 be able to reliably run entire high -value workflows 00:09:22.210 --> 00:09:24.669 end -to -end. That's why the biggest organizations 00:09:24.669 --> 00:09:27.269 are redesigning their entire AI strategies around 00:09:27.269 --> 00:09:30.129 this new capability. So if AI task length is 00:09:30.129 --> 00:09:32.429 the critical variable, what issue defines the 00:09:32.429 --> 00:09:34.769 success or failure of these long -running automated 00:09:34.769 --> 00:09:37.370 systems? The ability to prove what the agent 00:09:37.370 --> 00:09:40.230 did, ensuring reliable planning and state management, 00:09:40.590 --> 00:09:43.009 is now absolutely paramount. And that really 00:09:43.009 --> 00:09:46.190 summarizes the core movement. AI is accelerating 00:09:46.190 --> 00:09:49.230 from helpful assistant to autonomous workflow 00:09:49.230 --> 00:09:52.309 executor and simulation designer. The key lesson 00:09:52.309 --> 00:09:54.590 for you, the learner, is to shift your focus. 00:09:55.190 --> 00:09:57.610 Stop optimizing only for writing better prompts. 00:09:57.990 --> 00:10:00.149 You have to start focusing on the knowledge of 00:10:00.149 --> 00:10:03.049 better planning and trust systems for these powerful 00:10:03.049 --> 00:10:05.490 new agents. And as a final thought to mull over, 00:10:05.669 --> 00:10:07.929 connect that concept task length doubling every 00:10:07.929 --> 00:10:09.909 seven months back to your own professional life. 00:10:10.370 --> 00:10:12.470 If big organizations are already redesigning 00:10:12.470 --> 00:10:14.909 full workflows based on multi -hour autonomous 00:10:14.909 --> 00:10:18.389 tasks, ask yourself this. What common multi -hour 00:10:18.389 --> 00:10:20.389 task or research effort in your work life will 00:10:20.389 --> 00:10:22.470 be the first to be run end -to -end by an agent 00:10:22.470 --> 00:10:24.950 in the next 12 months? Something worth exploring 00:10:24.950 --> 00:10:26.970 tonight. Thank you for joining us for the DAPE 00:10:26.970 --> 00:10:29.029 Dive. We'll be back soon with more insights from 00:10:29.029 --> 00:10:29.970 the cutting edge of change.