WEBVTT 00:00:00.000 --> 00:00:03.020 So the longstanding rumors about Llama were completely 00:00:03.020 --> 00:00:06.540 wrong. Meta's flagship open source model might 00:00:06.540 --> 00:00:09.679 actually be dead. Yeah, that's crazy. But something 00:00:09.679 --> 00:00:12.179 else entirely just woke up instead. We're stepping 00:00:12.179 --> 00:00:14.560 into a radically different era right now. You 00:00:14.560 --> 00:00:16.980 know, AI models don't just chat with you anymore. 00:00:17.120 --> 00:00:19.940 Right. They actually pause to contemplate their 00:00:19.940 --> 00:00:22.140 next move. And sometimes they just work autonomously 00:00:22.140 --> 00:00:25.039 for. eight straight hours welcome to today's 00:00:25.039 --> 00:00:26.920 deep dive we're really glad you joined us today 00:00:26.920 --> 00:00:29.640 we have a massive fascinating journey mapped 00:00:29.640 --> 00:00:32.939 out for you we're unpacking meta's huge pivot 00:00:32.939 --> 00:00:35.840 away from open source they built a highly guarded 00:00:35.840 --> 00:00:38.200 closed source personal assistant it's officially 00:00:38.200 --> 00:00:40.780 called new spark then we're going to explore 00:00:40.780 --> 00:00:44.380 the very messy reality ahead uh ai agents are 00:00:44.380 --> 00:00:46.579 currently operating out in the wild yeah and 00:00:46.579 --> 00:00:49.380 the growing pains are Honestly, pretty terrifying. 00:00:49.600 --> 00:00:51.439 Finally, we'll see how the open source community 00:00:51.439 --> 00:00:54.500 is striking back. They built an absolute endurance 00:00:54.500 --> 00:00:57.240 coding behemoth recently. It's a brand new model 00:00:57.240 --> 00:01:00.520 called GLM 5 .1. Let's start by looking closely 00:01:00.520 --> 00:01:02.700 at Meta's situation. Their current redemption 00:01:02.700 --> 00:01:05.980 arc is truly fascinating to watch. Yeah, 2025 00:01:05.980 --> 00:01:09.400 was incredibly rough for their AI division. Llama 00:01:09.400 --> 00:01:12.640 4 Maverick was widely seen as a massive disappointment. 00:01:13.099 --> 00:01:15.019 It was just a very middling release for them 00:01:15.019 --> 00:01:18.069 overall. desperately needed a major win to stay 00:01:18.069 --> 00:01:20.409 relevant. So they went out and spent an absolute 00:01:20.409 --> 00:01:23.250 fortune. They hired Alexander Wang away from 00:01:23.250 --> 00:01:26.189 Scale AI. That was a huge, incredibly aggressive 00:01:26.189 --> 00:01:29.090 talent acquisition. And now they've returned 00:01:29.090 --> 00:01:32.409 to the arena with MuseSpark. What's fascinating 00:01:32.409 --> 00:01:35.090 is that MuseSpark is entirely closed source. 00:01:35.430 --> 00:01:38.549 That's a massive fundamental shift in their corporate 00:01:38.549 --> 00:01:41.049 philosophy. For years, Meta was the absolute 00:01:41.049 --> 00:01:43.609 darling of the open AI movement. Now they're 00:01:43.609 --> 00:01:45.629 suddenly closing all the doors and locking the 00:01:45.629 --> 00:01:47.849 gates. They completely evolved how they approach 00:01:47.849 --> 00:01:50.049 problem solving. Instead of relying on just one 00:01:50.049 --> 00:01:53.310 massive digital brain, Meta is now heavily leaning 00:01:53.310 --> 00:01:55.650 on an orchestration layer. Well, let me define 00:01:55.650 --> 00:01:58.010 that concept for you really quickly. It's a system 00:01:58.010 --> 00:02:00.329 managing multiple AI models working together. 00:02:00.799 --> 00:02:02.420 Think of it kind of like a corporate board of 00:02:02.420 --> 00:02:04.459 directors. Right. You don't just rely on one 00:02:04.459 --> 00:02:07.120 single CEO for everything. You have a researcher, 00:02:07.239 --> 00:02:09.759 an analyst, and a strategist collaborating. Exactly. 00:02:09.759 --> 00:02:12.419 It's like stacking Lego blocks of data to build 00:02:12.419 --> 00:02:15.479 something stronger. MuseSpark features a brand 00:02:15.479 --> 00:02:18.900 new contemplating mode. It runs an entire squad 00:02:18.900 --> 00:02:22.120 of specialized AI agents. They run in parallel 00:02:22.120 --> 00:02:24.759 to solve incredibly complex problems. They can 00:02:24.759 --> 00:02:27.439 reason through dense scientific queries. They 00:02:27.439 --> 00:02:30.460 easily tackle heavy legal queries, too. The model 00:02:30.460 --> 00:02:33.550 recently scored a 52 on a major benchmark. That's 00:02:33.550 --> 00:02:35.569 on the Artificial Analysis Intelligence Index. 00:02:35.849 --> 00:02:38.330 That score places it firmly in the global top 00:02:38.330 --> 00:02:40.689 five. It sits right alongside heavyweights like 00:02:40.689 --> 00:02:44.969 GPT -5 .4 and Gemini 3 .1 Pro. Meta also took 00:02:44.969 --> 00:02:47.689 the training data incredibly seriously this time. 00:02:47.770 --> 00:02:49.930 They didn't just scrape random information off 00:02:49.930 --> 00:02:52.069 the internet. Yeah, they actually consulted over 00:02:52.069 --> 00:02:54.650 1 ,000 practicing physicians. They carefully 00:02:54.650 --> 00:02:57.330 curated the medical training data to ensure ultimate 00:02:57.330 --> 00:03:00.169 accuracy. The engineering team is claiming something 00:03:00.169 --> 00:03:03.500 genuinely incredible here. They say they've achieved 00:03:03.500 --> 00:03:06.340 Lama 4 -level performance. But they managed to 00:03:06.340 --> 00:03:09.580 use 10 times less compute power. Let's talk about 00:03:09.580 --> 00:03:11.639 why that matters for a second. It allows for 00:03:11.639 --> 00:03:13.900 incredibly fast inference speeds. And it makes 00:03:13.900 --> 00:03:15.879 that inference much cheaper on mobile devices. 00:03:16.300 --> 00:03:18.659 Mark Zuckerberg still hopes to open source future 00:03:18.659 --> 00:03:22.120 versions eventually. But right now, Meta is desperately 00:03:22.120 --> 00:03:24.520 looking for direct revenue. They're staring at 00:03:24.520 --> 00:03:28.949 a $135 billion capital expenditure. They need 00:03:28.949 --> 00:03:31.530 to turn that massive investment into actual cash 00:03:31.530 --> 00:03:34.789 soon. This is basically their big kid table moment. 00:03:34.930 --> 00:03:37.229 They're no longer satisfied just providing the 00:03:37.229 --> 00:03:39.189 underlying tools. They aren't just helping other 00:03:39.189 --> 00:03:41.689 developers build cool apps anymore. They're trying 00:03:41.689 --> 00:03:43.669 to build the ultimate personal assistant themselves. 00:03:45.120 --> 00:03:47.560 Beat. OK, I need to understand this better. Is 00:03:47.560 --> 00:03:50.180 this 10 times compute efficiency actually the 00:03:50.180 --> 00:03:52.560 real game changer here? Absolutely. It means 00:03:52.560 --> 00:03:55.860 incredibly powerful models can run directly on 00:03:55.860 --> 00:03:58.580 edge devices. You don't need a giant server farm 00:03:58.580 --> 00:04:01.300 for every single query. So less compute means 00:04:01.300 --> 00:04:03.580 powerful AI runs directly on your phone. Right. 00:04:03.699 --> 00:04:06.280 And that completely rewires the hardware ecosystem. 00:04:06.639 --> 00:04:08.919 That actually brings us right into our next major 00:04:08.919 --> 00:04:12.939 topic. MuSpark uses that squad of parallel agents. 00:04:13.120 --> 00:04:15.400 This reflects a much broader shift happening 00:04:15.400 --> 00:04:17.980 across the tech industry. AI isn't just answering 00:04:17.980 --> 00:04:20.399 questions, it's taking real action now. But the 00:04:20.399 --> 00:04:23.240 broader ecosystem is experiencing some severe 00:04:23.240 --> 00:04:26.100 growing pains. It's honestly getting incredibly 00:04:26.100 --> 00:04:29.240 messy and chaotic out there. Yeah. Let's look 00:04:29.240 --> 00:04:32.100 at some of the major risks first. Anthropic has 00:04:32.100 --> 00:04:34.939 a brand new safety testing model called Mythos. 00:04:35.120 --> 00:04:38.279 It's already exposed thousands of severe system 00:04:38.279 --> 00:04:40.240 vulnerabilities. We aren't just talking about 00:04:40.240 --> 00:04:42.600 minor bugs here. We're talking about critical 00:04:42.600 --> 00:04:46.220 browser flaws and major operating system vulnerabilities 00:04:46.220 --> 00:04:49.560 that hackers could easily exploit. Yeah. Anthropic 00:04:49.560 --> 00:04:51.420 actually had to launch something called Project 00:04:51.420 --> 00:04:54.959 Glasswing. This creates a quiet period for major 00:04:54.959 --> 00:04:57.379 tech companies. It lets them prepare quietly 00:04:57.379 --> 00:04:59.939 before the flaws become public. They desperately 00:04:59.939 --> 00:05:02.160 need that time to patch these systems. Then we 00:05:02.160 --> 00:05:04.339 have the everyday errors affecting normal people. 00:05:04.740 --> 00:05:07.879 Google's AI overviews look really fast and highly 00:05:07.879 --> 00:05:10.660 convenient. But a new industry study just came 00:05:10.660 --> 00:05:13.019 out recently. It shows they give blatantly incorrect 00:05:13.019 --> 00:05:15.500 answers. They're failing about 10 % of the time. 00:05:15.620 --> 00:05:18.079 That's a dangerously high failure rate for an 00:05:18.079 --> 00:05:20.379 authoritative search engine. If you're trusting 00:05:20.379 --> 00:05:23.800 an AI agent with your daily tasks, a 10 % failure 00:05:23.800 --> 00:05:26.100 rate means frequent, unpredictable disasters. 00:05:26.720 --> 00:05:28.990 It's a harsh reminder. that you always need to 00:05:28.990 --> 00:05:32.029 verify. A quick verification step saves you from 00:05:32.029 --> 00:05:34.569 massive trouble later. We've also seen some really 00:05:34.569 --> 00:05:38.209 wild user hacks recently. Someone found a highly 00:05:38.209 --> 00:05:41.449 controversial trick for anthropic systems. It 00:05:41.449 --> 00:05:43.569 basically speeds up Claude's processing times 00:05:43.569 --> 00:05:46.649 significantly. The method went super viral across 00:05:46.649 --> 00:05:49.649 developer forums online. People absolutely love 00:05:49.649 --> 00:05:52.850 getting those incredibly fast results, but... 00:05:53.050 --> 00:05:55.629 The ethics discussion around this is getting 00:05:55.629 --> 00:05:58.449 really heated. It's a very controversial workaround 00:05:58.449 --> 00:06:00.850 that bypasses normal safety checks. You're basically 00:06:00.850 --> 00:06:03.750 trading thoughtful safety guardrails for raw 00:06:03.750 --> 00:06:06.579 speed. The Wild West of autonomous agents is 00:06:06.579 --> 00:06:09.019 officially here. A new tech startup called Poke 00:06:09.019 --> 00:06:11.579 just dropped a product. They released a new AI 00:06:11.579 --> 00:06:13.800 agent for everyday consumers. It actually works 00:06:13.800 --> 00:06:16.360 directly inside your personal messaging apps. 00:06:16.540 --> 00:06:18.839 It operates seamlessly inside Apple's iMessage. 00:06:19.019 --> 00:06:21.779 It works in standard SMS and Telegram, too. You 00:06:21.779 --> 00:06:24.500 can automate complex tasks just by sending a 00:06:24.500 --> 00:06:26.920 simple text. People are calling it an open -claw 00:06:26.920 --> 00:06:29.079 model for everyone. Right. Well, wait, I have 00:06:29.079 --> 00:06:31.139 to push back on the excitement here. Yeah. Google 00:06:31.139 --> 00:06:33.699 is actively hallucinating 10 % of the time. Yeah. 00:06:33.800 --> 00:06:36.180 Anthropic's own models are finding thousands 00:06:36.180 --> 00:06:39.339 of critical system vulnerabilities. Aren't hyper 00:06:39.339 --> 00:06:41.720 -accessible agents like poker a massive security 00:06:41.720 --> 00:06:44.720 risk? Especially when we hand them over to the 00:06:44.720 --> 00:06:47.439 average user. It's an undeniably massive risk 00:06:47.439 --> 00:06:50.199 for everyday consumers. We're giving these early 00:06:50.199 --> 00:06:52.879 -stage agents direct, unfettered access. They 00:06:52.879 --> 00:06:55.240 can take real -world actions on our behalf. If 00:06:55.240 --> 00:06:57.120 they hallucinate a command, the consequences 00:06:57.120 --> 00:07:00.949 are real. Two -sec silence. How does Amazon AWS 00:07:00.949 --> 00:07:03.730 fit into this chaotic landscape? They're financially 00:07:03.730 --> 00:07:06.290 backing both OpenAI and Anthropic simultaneously. 00:07:07.029 --> 00:07:09.089 What's their underlying strategy in this messy 00:07:09.089 --> 00:07:11.990 ecosystem? AWS publicly says this platform conflict 00:07:11.990 --> 00:07:14.470 is totally normal. They argue that a multi -model 00:07:14.470 --> 00:07:17.149 strategy is actually a massive benefit. It helps 00:07:17.149 --> 00:07:19.709 users choose the absolute best AI for each specific 00:07:19.709 --> 00:07:22.649 task. AWS lets you pick the perfect tool for 00:07:22.649 --> 00:07:25.250 every unique job. Sponsor. Mid -roll sponsor 00:07:25.250 --> 00:07:27.990 break. Let's shift our focus over to the open 00:07:27.990 --> 00:07:30.470 source world now. Meta is aggressively building 00:07:30.470 --> 00:07:32.310 closed source moats right now. They're trying 00:07:32.310 --> 00:07:33.970 to own the consumer personal assistant market 00:07:33.970 --> 00:07:36.970 entirely. But the open source community is absolutely 00:07:36.970 --> 00:07:40.110 striking back. While closed labs build walled 00:07:40.110 --> 00:07:42.790 gardens, open source developers build marathon 00:07:42.790 --> 00:07:46.209 runners. A group called ZAI just released something 00:07:46.209 --> 00:07:49.110 genuinely massive. It's a brand new open source 00:07:49.110 --> 00:07:53.250 model called GLM 5 .1. People are already calling 00:07:53.250 --> 00:07:55.959 it the new open source code king. Traditionally, 00:07:56.079 --> 00:07:59.100 AI models are essentially just really fast sprinters. 00:07:59.199 --> 00:08:02.220 They give a really great, quick answer to a single 00:08:02.220 --> 00:08:05.379 prompt. But long, sustained tasks are a massive 00:08:05.379 --> 00:08:07.680 problem for them. I have to admit something here. 00:08:07.800 --> 00:08:10.360 I still wrestle with prompt drift myself. If 00:08:10.360 --> 00:08:12.160 I'm working on a coding task that takes more 00:08:12.160 --> 00:08:14.459 than 10 minutes, the back and forth usually gets 00:08:14.459 --> 00:08:17.100 incredibly messy. The models usually just lose 00:08:17.100 --> 00:08:19.459 the original context entirely. That's the exact... 00:08:19.629 --> 00:08:23.050 limitation that GLM 5 .1 officially solves. It's 00:08:23.050 --> 00:08:25.410 specifically designed for long autonomous coding 00:08:25.410 --> 00:08:27.769 sessions. It can run continuously for up to eight 00:08:27.769 --> 00:08:29.569 hours straight. Let's look closely at the benchmark 00:08:29.569 --> 00:08:33.049 scores. It scored an impressive 58 .4 on SWE 00:08:33.049 --> 00:08:34.990 Bench Pro. Let's make sure we define that metric 00:08:34.990 --> 00:08:36.769 for everyone. It's a rigorous test measuring 00:08:36.769 --> 00:08:39.990 how AI fixes software bugs. Yeah, and that specific 00:08:39.990 --> 00:08:42.629 score is incredibly significant. It moves significantly 00:08:42.629 --> 00:08:45.350 ahead of recent proprietary models. It consistently 00:08:45.350 --> 00:08:48.460 beats the latest models from OpenAI. It beats 00:08:48.460 --> 00:08:50.379 Anthropic's leading models, too. It just keeps 00:08:50.379 --> 00:08:53.120 working tirelessly on complex engineering problems. 00:08:53.240 --> 00:08:55.620 It genuinely doesn't lose its direction over 00:08:55.620 --> 00:08:58.659 time. The fundamental key here is a process called 00:08:58.659 --> 00:09:01.210 continuous optimization. It can actually write 00:09:01.210 --> 00:09:03.970 and run its own software tests. It identifies 00:09:03.970 --> 00:09:06.690 complex software issues entirely autonomously. 00:09:06.850 --> 00:09:09.590 It constantly refines the underlying solution 00:09:09.590 --> 00:09:12.490 through many automated tool calls. It actually 00:09:12.490 --> 00:09:15.070 keeps a persistent memory of its earlier decisions. 00:09:15.330 --> 00:09:17.870 It systematically adjusts its broader strategy 00:09:17.870 --> 00:09:20.909 after encountering failed attempts. It's mimicking 00:09:20.909 --> 00:09:23.330 how a human engineer actually solves problems. 00:09:23.549 --> 00:09:26.389 I watched the large project build demo they released 00:09:26.389 --> 00:09:28.830 yesterday. It created a completely functional 00:09:29.559 --> 00:09:32.960 Linux desktop web app. Wow. It built a working 00:09:32.960 --> 00:09:35.860 terminal from scratch. It built simple, playable 00:09:35.860 --> 00:09:39.720 video games natively, all within a single, continuous, 00:09:39.899 --> 00:09:43.419 uninterrupted workflow. Whoa. Imagine scaling 00:09:43.419 --> 00:09:45.679 to a billion queries. It's completely mind -blowing 00:09:45.679 --> 00:09:47.899 to think about. It's also proving to be highly 00:09:47.899 --> 00:09:51.100 efficient under the hood. We saw the recent kernel 00:09:51.100 --> 00:09:54.019 bench level 3 workloads. It achieved a massive 00:09:54.019 --> 00:09:57.860 3 .6x GPU speedup. It's optimizing the core hardware 00:09:57.860 --> 00:10:01.039 logic as it writes code. It's honestly as efficient 00:10:01.039 --> 00:10:04.379 as it is capable. It also possesses amazing creative 00:10:04.379 --> 00:10:06.919 flexibility. It isn't just relying on cold, raw 00:10:06.919 --> 00:10:09.620 logic. It took second place overall in the global 00:10:09.620 --> 00:10:11.940 design arena. It performed nearly as well as 00:10:11.940 --> 00:10:16.019 Cloud Opus 4 .6 in creative tasks. GLM 5 .1 is 00:10:16.019 --> 00:10:18.159 essentially a massive warning shot across the 00:10:18.159 --> 00:10:20.799 bow. It's aimed directly at every closed -source 00:10:20.799 --> 00:10:23.460 lab operating today. If an open -source model 00:10:23.460 --> 00:10:26.460 seamlessly handles eight -hour workflows, if 00:10:26.460 --> 00:10:29.700 it consistently tops the SWE Bench Pro leaderboard. 00:10:29.879 --> 00:10:32.159 The traditional moat around proprietary coding 00:10:32.159 --> 00:10:35.720 models is officially drying up. Beat. But I really 00:10:35.720 --> 00:10:37.960 have to push back on this narrative, too. If 00:10:37.960 --> 00:10:39.940 the technological moat is drying up around the 00:10:39.940 --> 00:10:42.139 models themselves, doesn't the moat just shift 00:10:42.139 --> 00:10:44.240 entirely somewhere else? Right. Doesn't it shift 00:10:44.240 --> 00:10:46.840 to whoever has the massive compute required to 00:10:46.840 --> 00:10:49.000 run these intensive eight -hour sessions at scale? 00:10:49.200 --> 00:10:51.620 That's the harsh economic reality of this space 00:10:51.620 --> 00:10:54.240 right now. The models themselves are rapidly 00:10:54.240 --> 00:10:56.620 becoming democratized. But the sheer compute 00:10:56.620 --> 00:10:59.580 required to run them is astronomical. The real 00:10:59.580 --> 00:11:01.860 power is just shifting directly to the hardware 00:11:01.860 --> 00:11:04.379 owners. What does continuous optimization actually 00:11:04.379 --> 00:11:07.299 look like in practice for these models? It means 00:11:07.299 --> 00:11:10.600 the AI constantly evaluates its own work. It 00:11:10.600 --> 00:11:13.259 learns from its mistakes mid -task and tries 00:11:13.259 --> 00:11:16.039 new approaches automatically. It basically fixes 00:11:16.039 --> 00:11:18.019 its own errors while it's still working. Right. 00:11:18.120 --> 00:11:21.299 And it never gets tired or frustrated. Let's 00:11:21.299 --> 00:11:23.100 take a look at the emerging microtrends now. 00:11:24.039 --> 00:11:28.220 GLM 5 .1 clearly proves AI is mastering deep 00:11:28.220 --> 00:11:31.340 logic. over long time horizons. That advanced 00:11:31.340 --> 00:11:34.100 capability is rapidly trickling down everywhere 00:11:34.100 --> 00:11:36.879 else. It's fundamentally altering physical hardware 00:11:36.879 --> 00:11:39.039 development. It's hitting highly specific daily 00:11:39.039 --> 00:11:41.399 tools you use on your computer. The robotics 00:11:41.399 --> 00:11:44.179 sector is seeing absolutely huge financial investments. 00:11:44.440 --> 00:11:46.399 Look at what's happening with D Robotics over 00:11:46.399 --> 00:11:48.919 in China. They recently secured an additional 00:11:48.919 --> 00:11:52.659 $150 million in funding. That massive influx 00:11:52.659 --> 00:11:55.440 raises their total funding substantially. They're 00:11:55.440 --> 00:11:58.899 sitting at $270 million in total capital now. 00:11:59.039 --> 00:12:01.220 This deeply strengthens their leadership role 00:12:01.220 --> 00:12:04.519 in AI robotics. Prominent venture investors are 00:12:04.519 --> 00:12:07.320 backing their hardware integration heavily. Video 00:12:07.320 --> 00:12:09.779 generation capabilities are also shifting incredibly 00:12:09.779 --> 00:12:12.960 fast right now. A surprise generative model just 00:12:12.960 --> 00:12:14.980 dropped out of nowhere. It's bizarrely called 00:12:14.980 --> 00:12:18.360 Happy Horse 1 .0. It abruptly took the number 00:12:18.360 --> 00:12:21.070 one overall spot. It's currently leading the 00:12:21.070 --> 00:12:23.529 artificial analysis video generation rankings. 00:12:23.730 --> 00:12:26.110 It even managed to beat the highly anticipated 00:12:26.110 --> 00:12:30.179 C -Dens 2 .0 model. A full developer reveal is 00:12:30.179 --> 00:12:32.120 supposedly coming very soon. Then we have these 00:12:32.120 --> 00:12:35.200 incredibly powerful daily empowered tools. A 00:12:35.200 --> 00:12:37.659 new project called CareerOps is incredibly interesting 00:12:37.659 --> 00:12:40.500 to explore. It essentially turns cloud code into 00:12:40.500 --> 00:12:43.340 an open source automation pipeline. It's a fully 00:12:43.340 --> 00:12:46.139 autonomous AI job search pipeline. The early 00:12:46.139 --> 00:12:48.340 stats on it are honestly quite impressive. It 00:12:48.340 --> 00:12:51.480 has successfully passed 631 technical evaluations. 00:12:51.879 --> 00:12:54.340 It autonomously sent out 68 job applications 00:12:54.340 --> 00:12:56.980 already. It currently has over 9 ,000 GitHub 00:12:56.980 --> 00:12:59.379 stars from developers. It's literally hunting 00:12:59.379 --> 00:13:02.399 for jobs while you sleep. Velo is another amazing, 00:13:02.500 --> 00:13:05.720 highly specific daily tool. It takes your messy, 00:13:05.840 --> 00:13:08.759 raw screen recordings. It magically turns them 00:13:08.759 --> 00:13:11.899 into incredibly watch -worthy videos. You get 00:13:11.899 --> 00:13:14.340 ready to share instructional videos powered entirely 00:13:14.340 --> 00:13:17.320 by AI editing. You can easily share anything 00:13:17.320 --> 00:13:20.220 as polished video messages. A tool called Flint 00:13:20.220 --> 00:13:22.360 is doing something very similar for web design. 00:13:22.720 --> 00:13:24.720 It helps marketing teams scale their landing 00:13:24.720 --> 00:13:27.279 pages effortlessly. You can launch beautifully 00:13:27.279 --> 00:13:30.340 on brand pages instantly. You can spin them up 00:13:30.340 --> 00:13:33.279 for every single campaign or ad prospect. It 00:13:33.279 --> 00:13:35.779 removes the entire bottleneck of front end development. 00:13:36.019 --> 00:13:38.100 We also have emerging tools designed specifically 00:13:38.100 --> 00:13:41.000 for maintaining focus. Google Chrome recently 00:13:41.000 --> 00:13:43.960 introduced vertical tabs. It neatly organizes 00:13:43.960 --> 00:13:46.320 your open tab side by side on your screen. It 00:13:46.320 --> 00:13:49.279 makes complex multitasking much easier to handle 00:13:49.279 --> 00:13:51.879 visually. It actively promotes deep focus and 00:13:52.009 --> 00:13:54.269 sustained productivity. A utility called Lookaway 00:13:54.269 --> 00:13:56.789 is another really great one. It's a highly intelligent, 00:13:57.009 --> 00:14:00.370 context -aware, smart break reminder. It actively 00:14:00.370 --> 00:14:03.250 helps reduce your daily eye strain. It significantly 00:14:03.250 --> 00:14:06.690 reduces overall screen fatigue during long work 00:14:06.690 --> 00:14:09.629 sessions. And your breaks never interrupt you 00:14:09.629 --> 00:14:12.029 randomly while you're typing. I really want you 00:14:12.029 --> 00:14:13.970 to think about how this connects to you. Your 00:14:13.970 --> 00:14:16.870 daily professional life is changing at breakneck 00:14:16.870 --> 00:14:20.559 speed. These disparate digital tools stack together 00:14:20.559 --> 00:14:23.620 beautifully. This modern stack essentially turns 00:14:23.620 --> 00:14:27.120 a single person into an agency. You become a 00:14:27.120 --> 00:14:29.879 full -scale creative and operational agency overnight. 00:14:30.159 --> 00:14:32.899 You essentially have a tireless robotic coder. 00:14:32.940 --> 00:14:35.039 You have a professional video editor working 00:14:35.039 --> 00:14:37.559 for free. You have an expert web designer on 00:14:37.559 --> 00:14:39.720 call. All of them are running simultaneously 00:14:39.720 --> 00:14:43.330 in the background. Two -second silence. OK, I 00:14:43.330 --> 00:14:45.169 have to ask about these bizarre naming conventions. 00:14:45.450 --> 00:14:47.669 Why are breakthrough models called silly things 00:14:47.669 --> 00:14:50.470 like Happy Horse 1 .0? It's basically researchers 00:14:50.470 --> 00:14:52.769 making a loud counterculture statement. They 00:14:52.769 --> 00:14:55.210 really want to reject that boring corporate tech 00:14:55.210 --> 00:14:57.990 branding and open source. Silly names just reject 00:14:57.990 --> 00:15:00.149 boring corporate tech labels and open source. 00:15:00.330 --> 00:15:02.909 Yeah, it definitely keeps that rebellious open 00:15:02.909 --> 00:15:05.809 source spirit alive. Let's briefly recap the 00:15:05.809 --> 00:15:09.220 really big idea from today's deep dive. The era 00:15:09.220 --> 00:15:12.120 of the simple single turn chat bot is completely 00:15:12.120 --> 00:15:14.960 over. We're rapidly entering an age of parallel 00:15:14.960 --> 00:15:18.460 agent coordination. We're seeing real long horizon 00:15:18.460 --> 00:15:21.679 autonomy becoming the new standard. Meta is aggressively 00:15:21.679 --> 00:15:24.460 closing its doors to the open community. They're 00:15:24.460 --> 00:15:27.019 frantically building the ultimate consumer personal 00:15:27.019 --> 00:15:29.860 assistant. New Spark is definitely their critical 00:15:29.860 --> 00:15:32.899 big kid table moment. But the broader open source 00:15:32.899 --> 00:15:35.899 world is fighting back incredibly hard. Massive 00:15:35.899 --> 00:15:38.700 models like GLM 5 .1 are changing the underlying 00:15:38.700 --> 00:15:41.500 math of development. They're rapidly democratizing 00:15:41.500 --> 00:15:43.759 complex software engineering itself. We really 00:15:43.759 --> 00:15:46.379 covered a lot of important ground today, from 00:15:46.379 --> 00:15:48.860 orchestration layers to eight -hour autonomous 00:15:48.860 --> 00:15:51.440 AI sessions. From thousands of critical system 00:15:51.440 --> 00:15:54.100 vulnerabilities to fully autonomous job hunters. 00:15:54.340 --> 00:15:56.879 The technological landscape is shifting constantly 00:15:56.879 --> 00:15:59.179 under our feet. It's messy, it's chaotic, and 00:15:59.179 --> 00:16:01.240 it's moving incredibly fast. I want to leave 00:16:01.240 --> 00:16:03.759 you with one final thought today. We started 00:16:03.759 --> 00:16:06.559 by talking about an AI that pauses to actually 00:16:06.559 --> 00:16:09.299 think. An AI that works autonomously for eight 00:16:09.299 --> 00:16:13.220 hours straight without fatigue. Beat deep. If 00:16:13.220 --> 00:16:15.779 an open source AI can continuously test and refine 00:16:15.779 --> 00:16:17.860 its own code. Right. If it can do this for eight 00:16:17.860 --> 00:16:20.379 hours without losing focus, what does the human's 00:16:20.379 --> 00:16:23.279 role in the loop look like tomorrow? Are we eventually 00:16:23.279 --> 00:16:25.559 becoming mere supervisors of abundant digital 00:16:25.559 --> 00:16:28.000 labor? Nice. Thank you for joining us on this 00:16:28.000 --> 00:16:29.399 deep dive. IUTRO Music.