WEBVTT 00:00:00.000 --> 00:00:03.700 Imagine, if you will, getting this rare glimpse 00:00:03.700 --> 00:00:07.599 into the secret labs where the next really big 00:00:07.599 --> 00:00:10.660 AI model, maybe something like GPT -5, is being 00:00:10.660 --> 00:00:13.539 built, not behind closed doors, but right there, 00:00:13.640 --> 00:00:15.820 kind of unfolding before your eyes. Yeah, it's 00:00:15.820 --> 00:00:17.679 less like your typical product launch and maybe 00:00:17.679 --> 00:00:20.300 more like ship this in July. We're seeing these 00:00:20.300 --> 00:00:23.140 mysterious fragments pop up, hinting at a new 00:00:23.140 --> 00:00:25.239 heavyweight contender. And it really suggests 00:00:25.239 --> 00:00:28.440 a big shift in how these powerful systems are 00:00:28.440 --> 00:00:30.579 being developed worldwide. Welcome to the Deep 00:00:30.579 --> 00:00:32.859 Dive. Today we're taking a journey through some 00:00:32.859 --> 00:00:35.299 really fascinating new sources. Our goal here 00:00:35.299 --> 00:00:38.200 is to unpack the very latest in artificial intelligence. 00:00:38.780 --> 00:00:41.439 Our mission really is to make sense of this complex 00:00:41.439 --> 00:00:43.840 web of innovation that's shaping the field right 00:00:43.840 --> 00:00:45.539 now. Absolutely. We're going to pull back the 00:00:45.539 --> 00:00:48.460 curtain on these enigmatic AI models, explore 00:00:48.460 --> 00:00:51.100 a whole wave of surprisingly practical new tools, 00:00:51.280 --> 00:00:53.399 and then we'll zoom out, look at the bigger picture, 00:00:53.439 --> 00:00:55.700 this global strategic chess match playing out 00:00:55.700 --> 00:00:59.159 between two very different AI philosophies. a 00:00:59.159 --> 00:01:00.859 deep dive into what's happening right now and 00:01:00.859 --> 00:01:02.960 maybe more importantly, what's just coming over 00:01:02.960 --> 00:01:06.019 the horizon. Let's unpack this first piece. Our 00:01:06.019 --> 00:01:09.260 first significant insight comes from a really 00:01:09.260 --> 00:01:11.519 intriguing observation that's been making waves. 00:01:11.959 --> 00:01:14.299 The appearance and then the sudden disappearance 00:01:14.299 --> 00:01:18.920 of six mysterious AI models on LM Arena. That's 00:01:18.920 --> 00:01:21.239 right. These models, they had code names like 00:01:21.239 --> 00:01:24.459 Zenith, Summit. Lobster, Starfish, Nectarine, 00:01:24.579 --> 00:01:28.900 and O3 Alpha. Well, they just appeared. And almost 00:01:28.900 --> 00:01:31.280 immediately, they started crushing some incredibly 00:01:31.280 --> 00:01:33.840 tough coding tasks, showing off capabilities 00:01:33.840 --> 00:01:36.519 we just haven't seen publicly. Then, you know, 00:01:36.540 --> 00:01:39.829 classic open AI style, poof. vanished as quickly 00:01:39.829 --> 00:01:42.189 as they came. It was quite something to watch. 00:01:42.349 --> 00:01:43.989 Right. And here's where it gets really interesting. 00:01:44.129 --> 00:01:46.609 The AI community pretty quickly started theorizing 00:01:46.609 --> 00:01:48.290 that these weren't separate models, but fragments. 00:01:48.989 --> 00:01:51.170 Components, maybe, of the GPT -5 everyone's waiting 00:01:51.170 --> 00:01:53.769 for. Think of them like individual parts being 00:01:53.769 --> 00:01:55.409 tested before the whole thing is assembled. Summit, 00:01:55.469 --> 00:01:57.829 for example. The reports say it generated over 00:01:57.829 --> 00:02:01.209 2 ,300 lines of working Starship UI code on its 00:02:01.209 --> 00:02:03.189 very first attempt. That's not just good. That's 00:02:03.189 --> 00:02:06.049 kind of an unprecedented level of complex output 00:02:06.049 --> 00:02:08.169 on a first go. And it wasn't just Summit either. 00:02:08.430 --> 00:02:11.069 Zenith and O3 Alpha. Yeah. They seem to lead 00:02:11.069 --> 00:02:13.689 at reasoning and general coding stuff. Lobster 00:02:13.689 --> 00:02:16.610 and Starfish felt maybe lighter, possibly open 00:02:16.610 --> 00:02:18.870 source variants, some people speculated. The 00:02:18.870 --> 00:02:20.770 main theory, which is pretty cool, is that these 00:02:20.770 --> 00:02:24.370 are pieces of a mixture of expert system, AOE. 00:02:24.969 --> 00:02:27.349 It's this powerful way to build AI. Imagine like 00:02:27.349 --> 00:02:29.650 stacking specialized Lego blocks of data and 00:02:29.650 --> 00:02:32.210 algorithms. Each block is good at one thing. 00:02:32.610 --> 00:02:34.509 Combine them and you get a much more capable, 00:02:34.669 --> 00:02:36.629 efficient system. And the idea is they'll eventually 00:02:36.629 --> 00:02:39.949 fuse into, well, GPT -5. It's a clever way to 00:02:39.949 --> 00:02:44.669 scale up. This modular strategy. It seems genuinely 00:02:44.669 --> 00:02:47.210 ingenious. OpenAI is essentially running live 00:02:47.210 --> 00:02:50.669 A -B tests on future AI cognition, right? They're 00:02:50.669 --> 00:02:52.490 watching how these pieces perform in the wild, 00:02:52.650 --> 00:02:54.669 getting tons of data, fine -tuning them. It's 00:02:54.669 --> 00:02:56.830 even reportedly outpacing CloudSonic 4, which 00:02:56.830 --> 00:02:58.409 for a long time was seen as the coding champ. 00:02:58.610 --> 00:03:00.750 And sure, Grok 4 might be keeper, but these fragments, 00:03:00.930 --> 00:03:03.050 they really suggest GPT -5 is shaving up to be 00:03:03.050 --> 00:03:05.210 the heavyweight in raw capability. And what's 00:03:05.210 --> 00:03:07.310 fascinating here is that this Moe approach, this 00:03:07.310 --> 00:03:11.169 modular style. it probably isn't just for GPT 00:03:11.169 --> 00:03:13.129 -5. It's likely going to stick around for future 00:03:13.129 --> 00:03:15.409 versions, maybe even through GPT -8, believe 00:03:15.409 --> 00:03:18.870 it or not. So when GPT -5 officially drops, it 00:03:18.870 --> 00:03:21.150 won't feel like some big surprise reveal. It'll 00:03:21.150 --> 00:03:23.189 be more like, oh yeah, we saw the pieces of this 00:03:23.189 --> 00:03:26.210 already. And connecting this back, it really 00:03:26.210 --> 00:03:29.770 is Shipmas in July. We're watching open AI build 00:03:29.770 --> 00:03:33.389 its next big thing live in public. Instead of 00:03:33.389 --> 00:03:35.650 some staged event, they're kind of inviting the 00:03:35.650 --> 00:03:38.270 whole internet, the whole AI community to watch 00:03:38.270 --> 00:03:40.229 and even participate through these public tests. 00:03:40.449 --> 00:03:43.250 Whoa. Just imagine the scale of that. Scaling 00:03:43.250 --> 00:03:45.889 to, what, a billion queries? Seeing these pieces 00:03:45.889 --> 00:03:47.530 actually come together in real time is quite 00:03:47.530 --> 00:03:49.409 something. So what would you say is the biggest 00:03:49.409 --> 00:03:52.060 takeaway from seeing GPT -5? built out in the 00:03:52.060 --> 00:03:54.560 open like this. AI development is becoming open. 00:03:54.659 --> 00:03:56.860 It's like a real time public beta speeding up 00:03:56.860 --> 00:03:58.919 progress and getting everyone involved. OK, so 00:03:58.919 --> 00:04:01.199 that's the high end frontier. But what does this 00:04:01.199 --> 00:04:03.900 all mean for what AI can actually do like right 00:04:03.900 --> 00:04:06.680 now? in our daily lives. Let's shift gears a 00:04:06.680 --> 00:04:08.539 bit and dive into some of the more immediate 00:04:08.539 --> 00:04:10.780 practical highlights our sources picked up on. 00:04:10.939 --> 00:04:12.699 Yeah, it's pretty wild to see these things emerge, 00:04:12.900 --> 00:04:16.920 like OpenAI's chat GPT agent. It was shown just 00:04:16.920 --> 00:04:19.639 smoothly clicking through Cloudflare's CaptiCCH 00:04:19.639 --> 00:04:22.120 checkbox, you know, the I'm not a robot thing. 00:04:22.420 --> 00:04:25.019 And the coolest part, it apparently narrates 00:04:25.019 --> 00:04:27.660 what it's doing, explains its own success as 00:04:27.660 --> 00:04:30.779 it browses. That's a small step, maybe, but significant 00:04:30.779 --> 00:04:33.199 for autonomous agents. That narration piece. 00:04:33.459 --> 00:04:35.889 Yeah. It really does hint at a deeper level of 00:04:35.889 --> 00:04:37.949 understanding, doesn't it? We also saw this viral, 00:04:38.050 --> 00:04:40.449 like, 45 -minute tutorial showing how to set 00:04:40.449 --> 00:04:43.110 up Claude Code to automate some surprisingly 00:04:43.110 --> 00:04:46.170 complex tasks. And then there's that story, really 00:04:46.170 --> 00:04:48.209 fascinating, about someone building an entire 00:04:48.209 --> 00:04:50.889 AI business from scratch in just two hours. And 00:04:50.889 --> 00:04:52.389 it actually worked. It really points towards 00:04:52.389 --> 00:04:54.889 a future where, you know, solopreneurship, small 00:04:54.889 --> 00:04:57.629 business, it's getting way more automated, lowering 00:04:57.629 --> 00:05:01.129 the barriers. And building on that work impact 00:05:01.129 --> 00:05:04.149 idea, there's this new study from Anthropic. 00:05:04.230 --> 00:05:07.310 It breaks down how AI is affecting over 700 different 00:05:07.310 --> 00:05:09.949 jobs. And it asks that question everyone's kind 00:05:09.949 --> 00:05:12.329 of wrestling with. Is AI going to replace you 00:05:12.329 --> 00:05:15.250 or will it just augment your work, help you collaborate? 00:05:15.769 --> 00:05:17.629 Definitely food for thought for all of us. Interestingly, 00:05:17.930 --> 00:05:20.389 Claude's been so popular, they've hit some scaling 00:05:20.389 --> 00:05:22.750 issues. They just put in two new usage caps. 00:05:22.930 --> 00:05:25.089 There's a total weekly cap and a specific one 00:05:25.089 --> 00:05:27.389 just for Claude Opus 4, their top model. Now, 00:05:27.389 --> 00:05:29.569 they say this only hits less than 5 % of users, 00:05:29.790 --> 00:05:32.129 but, well, it's often the most active users, 00:05:32.329 --> 00:05:34.569 the power users. And they tend to be the loudest 00:05:34.569 --> 00:05:36.250 about it. It just shows the incredible demand. 00:05:36.529 --> 00:05:38.069 Oh, and check this out. This one's kind of mind 00:05:38.069 --> 00:05:41.430 -bending. Hunyon World 1 .0. It's a new system 00:05:41.430 --> 00:05:44.329 that creates entire editable 3D worlds from just 00:05:44.329 --> 00:05:47.079 a cent. or a picture. We're not just talking 00:05:47.079 --> 00:05:48.860 videos here. These are actual environments you 00:05:48.860 --> 00:05:52.399 can virtually walk around in. That's a huge jump 00:05:52.399 --> 00:05:54.819 for content creation. And on the business side, 00:05:55.060 --> 00:05:57.680 sources highlighted MicroOne. They're a scale 00:05:57.680 --> 00:06:00.720 AI rival, apparently raising funds now at a pretty 00:06:00.720 --> 00:06:03.500 impressive $500 million valuation. They've seen 00:06:03.500 --> 00:06:06.040 huge growth revenue up five times this year alone, 00:06:06.240 --> 00:06:09.680 from $10 million to $50 million. And they expect 00:06:09.680 --> 00:06:13.290 to hit $100 million by September. That kind of 00:06:13.290 --> 00:06:15.470 speed really tells you about the hunger for AI 00:06:15.470 --> 00:06:17.850 infrastructure and services. Yeah, and practical 00:06:17.850 --> 00:06:20.129 problems are getting AI solutions, too. Like, 00:06:20.129 --> 00:06:22.069 is your AI agent always forgetting what you talked 00:06:22.069 --> 00:06:24.509 about earlier? Yeah. You can now build real long 00:06:24.509 --> 00:06:26.649 -term memory using ZEPP's knowledge graphs. It 00:06:26.649 --> 00:06:29.410 helps the AI keep context, learn over time, and 00:06:29.410 --> 00:06:31.509 keeps the API cost low, which is a big plus. 00:06:32.060 --> 00:06:34.459 You know, I still wrestle with prompt drift myself 00:06:34.459 --> 00:06:36.459 sometimes where the conversation just goes off 00:06:36.459 --> 00:06:38.579 track. So yeah, these long -term memory solutions, 00:06:38.699 --> 00:06:41.060 they seem absolutely key for anyone really using 00:06:41.060 --> 00:06:43.500 these tools seriously. Totally. And if you're 00:06:43.500 --> 00:06:45.860 looking for automation that's maybe more powerful 00:06:45.860 --> 00:06:48.620 than Zapier, there's a great guide on using NEN. 00:06:48.620 --> 00:06:51.259 It's open source to build your own personal intelligence 00:06:51.259 --> 00:06:54.079 agent. We also saw a guide on blending, like 00:06:54.079 --> 00:06:56.959 McKinsey consulting methods with AI, turning 00:06:56.959 --> 00:06:59.180 insights into actual strategies clients might 00:06:59.180 --> 00:07:02.220 use. And just a few quick tool mentions. Wordwriter 00:07:02.220 --> 00:07:04.500 says it writes 200 plus pages of research with 00:07:04.500 --> 00:07:08.420 references. Copycat automates web tasks. Startup 00:07:08.420 --> 00:07:11.300 Sonar tries to find hidden startup ideas on Reddit. 00:07:11.459 --> 00:07:14.120 And Free Image Bulk helps download lots of images 00:07:14.120 --> 00:07:16.279 from websites. That's a lot of different tools 00:07:16.279 --> 00:07:18.079 and abilities there. What feels like the common 00:07:18.079 --> 00:07:20.240 thread running through all these diverse applications? 00:07:20.699 --> 00:07:23.560 AI is getting incredibly specialized. It's automating 00:07:23.560 --> 00:07:27.269 really complex human tasks everywhere. OK, so 00:07:27.269 --> 00:07:29.509 beyond the individual tools, beyond the models 00:07:29.509 --> 00:07:32.029 themselves, there's this much larger strategic 00:07:32.029 --> 00:07:34.670 game being played globally. Our sources point 00:07:34.670 --> 00:07:37.509 to two very distinct, almost opposing visions 00:07:37.509 --> 00:07:39.850 for where AI is headed. Yeah, it's a fascinating 00:07:39.850 --> 00:07:43.209 contrast. Just days after the U .S. put out its 00:07:43.209 --> 00:07:46.449 AI action plan, which mostly emphasized deregulation, 00:07:46.689 --> 00:07:50.009 a race to global dominance led by private companies. 00:07:50.839 --> 00:07:53.579 China presented its own plan. And China's plan. 00:07:53.759 --> 00:07:56.959 It focuses heavily on global cooperation. Things 00:07:56.959 --> 00:07:59.720 like joint AI research, pushing for open data 00:07:59.720 --> 00:08:02.800 sets and model sharing, working towards shared 00:08:02.800 --> 00:08:05.360 computing resources, really trying to broaden 00:08:05.360 --> 00:08:08.120 access. Right. They're also pushing hard on education 00:08:08.120 --> 00:08:11.180 and access. Specifically, they're championing 00:08:11.180 --> 00:08:13.560 AI upskilling for developing nations, trying 00:08:13.560 --> 00:08:15.920 to give them a real seat at the table. And importantly, 00:08:16.000 --> 00:08:17.680 they're calling for risk management frameworks, 00:08:17.839 --> 00:08:20.569 AI ethics policies. developed with help from 00:08:20.569 --> 00:08:22.930 the UN. It's a very different governance model 00:08:22.930 --> 00:08:25.509 they're proposing. Their plan even encourages 00:08:25.509 --> 00:08:27.949 the global developer community to co -develop 00:08:27.949 --> 00:08:30.949 AI tools together, promoting open source instead 00:08:30.949 --> 00:08:33.029 of relying on what they call Western black box 00:08:33.029 --> 00:08:35.490 APIs. You know, those proprietary systems where 00:08:35.490 --> 00:08:37.190 you don't know what's inside, controlled by just 00:08:37.190 --> 00:08:39.950 a few big companies. It's a clear call for something 00:08:39.950 --> 00:08:41.990 more decentralized. So you can see it taking 00:08:41.990 --> 00:08:44.429 shape. Two really different philosophies. Yeah. 00:08:44.490 --> 00:08:46.870 The U .S. approach seems, well, faster maybe. 00:08:47.320 --> 00:08:50.059 Richer in terms of private money, but more closed. 00:08:50.279 --> 00:08:53.200 It leans heavily on giants like OpenAI, Meta, 00:08:53.240 --> 00:08:56.179 Google, kind of a Silicon Valley model. And then 00:08:56.179 --> 00:08:59.019 China's vision is broader, shared, and explicitly 00:08:59.019 --> 00:09:02.320 open. Beijing is actively pitching itself as 00:09:02.320 --> 00:09:05.080 the AI partner for the global south. Engaging 00:09:05.080 --> 00:09:07.500 with regions the U .S. hasn't really prioritized 00:09:07.500 --> 00:09:09.620 in the same way. It really suggests that these 00:09:09.620 --> 00:09:12.720 open models, shared data sets, they're being 00:09:12.720 --> 00:09:15.159 positioned as a direct alternative to the Western 00:09:15.159 --> 00:09:18.330 AI silos. Which naturally leads to the big question. 00:09:18.570 --> 00:09:20.769 How might these two very different philosophies 00:09:20.769 --> 00:09:23.169 shape the future? Technology, geopolitics, all 00:09:23.169 --> 00:09:25.169 of it. We might really be heading towards two 00:09:25.169 --> 00:09:28.289 distinct global AI ecosystems. Two sex islands. 00:09:28.870 --> 00:09:30.090 All right, to kind of round things out, let's 00:09:30.090 --> 00:09:31.690 just run through a few quick hits. These really 00:09:31.690 --> 00:09:34.789 highlight how pervasive AI is becoming just everywhere. 00:09:35.029 --> 00:09:37.409 Okay, get this one. A viral AI generated video. 00:09:37.570 --> 00:09:40.179 Cat riding a leopard. Got over 100 million views. 00:09:40.299 --> 00:09:42.740 It's just fun, sure, but it shows how engaging 00:09:42.740 --> 00:09:45.860 and easy this tech is becoming for, like, creative 00:09:45.860 --> 00:09:48.600 stuff. Microsoft Edge. It's basically an AI browser 00:09:48.600 --> 00:09:51.120 now. They fully launched co -pilot mode, built 00:09:51.120 --> 00:09:53.419 right in. Your web browser is now an intelligent 00:09:53.419 --> 00:09:55.899 assistant. And on the hardware side, huge news. 00:09:55.980 --> 00:09:59.580 Tesla just signed a massive $16 .5 billion deal 00:09:59.580 --> 00:10:04.090 with Samsung for AI chips. That's not just big 00:10:04.090 --> 00:10:07.129 money. It signals the insane demand for specialized 00:10:07.129 --> 00:10:09.730 hardware as these models keep growing. Yeah, 00:10:09.870 --> 00:10:12.029 and related to that, tech giants like Google 00:10:12.029 --> 00:10:14.350 and Meta, they're investing heavily in nuclear 00:10:14.350 --> 00:10:16.929 and hydropower, basically building their own 00:10:16.929 --> 00:10:19.169 energy grids to power their massive data centers. 00:10:19.549 --> 00:10:22.809 That's a huge, often overlooked sign of the monumental 00:10:22.809 --> 00:10:25.789 energy needs future AI is going to demand. And 00:10:25.789 --> 00:10:28.190 finally, sort of bringing it back to geopolitics, 00:10:28.190 --> 00:10:31.429 a group of 20 security experts urged former President 00:10:31.429 --> 00:10:34.529 Trump to restrict Nvidia H20 chip sales to China, 00:10:34.710 --> 00:10:36.909 pointing directly at the high stakes involved 00:10:36.909 --> 00:10:39.950 in controlling this advanced AI tech. AI is just 00:10:39.950 --> 00:10:42.330 rapidly integrating into pretty much every single 00:10:42.330 --> 00:10:45.049 part of our daily lives. mid -roll sponsor, Red. 00:10:45.230 --> 00:10:47.269 So as we wrap up this deep dive, I think the 00:10:47.269 --> 00:10:49.590 big idea that really stands out is just the incredible 00:10:49.590 --> 00:10:53.330 speed, the velocity, and the multifaceted nature 00:10:53.330 --> 00:10:56.529 of AI development right now. It's moving unbelievably 00:10:56.529 --> 00:10:59.090 fast on so many different fronts all at once. 00:10:59.149 --> 00:11:00.750 Yeah, we're seeing these cutting -edge models, 00:11:00.909 --> 00:11:04.190 like maybe GPT -5, being built and tested right 00:11:04.190 --> 00:11:06.230 out in the open. Yeah. And they're showing capabilities 00:11:06.230 --> 00:11:09.230 that are, frankly, almost hard to believe sometimes. 00:11:10.080 --> 00:11:12.759 And at the exact same time, practical AI tools 00:11:12.759 --> 00:11:15.320 are just exploding, making automation, content 00:11:15.320 --> 00:11:18.580 creation, even starting a business way easier 00:11:18.580 --> 00:11:22.159 and more accessible for regular users, for solopreneurs. 00:11:22.299 --> 00:11:25.159 And then on that global scale, these two fundamentally 00:11:25.159 --> 00:11:28.159 different visions for AI's future, they're really 00:11:28.159 --> 00:11:31.019 solidifying, potentially leading to these distinct 00:11:31.019 --> 00:11:33.620 global ecosystems, each with its own rules, its 00:11:33.620 --> 00:11:35.799 own approach. We've certainly covered a lot today. 00:11:35.899 --> 00:11:37.679 It's probably worth taking a moment to think 00:11:37.679 --> 00:11:39.940 about how these developments might impact your 00:11:39.940 --> 00:11:43.139 own work or maybe just your daily life. The ripples 00:11:43.139 --> 00:11:45.600 are definitely already spreading. And maybe here's 00:11:45.600 --> 00:11:47.860 a provocative thought for you to mull over. As 00:11:47.860 --> 00:11:52.360 AI gets more modular, more distributed. You know, 00:11:52.360 --> 00:11:54.580 all these specialized agents and open source 00:11:54.580 --> 00:11:57.379 models popping up. Well, the whole idea of one 00:11:57.379 --> 00:12:00.919 big AI, like one single super intelligence, become 00:12:00.919 --> 00:12:03.519 less important than having this vast network 00:12:03.519 --> 00:12:05.980 of many interconnected specialized intelligences 00:12:05.980 --> 00:12:08.019 working together. Something to think about. Thank 00:12:08.019 --> 00:12:09.720 you for joining us for this deep dive. Until 00:12:09.720 --> 00:12:11.139 next time, utero music.