WEBVTT 00:00:00.000 --> 00:00:02.060 Well, we are just five days out from Google I 00:00:02.060 --> 00:00:05.320 .O., and honestly, the metaphorical dam hasn't 00:00:05.320 --> 00:00:07.639 just broken today. Oh, it is totally washed away. 00:00:07.799 --> 00:00:09.699 I mean, nobody saw this massive wave coming. 00:00:09.740 --> 00:00:11.599 Yeah, it really came out of nowhere. Over the 00:00:11.599 --> 00:00:14.619 last 48 hours, leaks just completely exploded. 00:00:14.859 --> 00:00:18.780 We are seeing sudden Gemini 3 .2 rumors everywhere. 00:00:19.899 --> 00:00:22.579 GPT 5 .6 internal tests are suddenly popping 00:00:22.579 --> 00:00:25.809 up. Mythos version 2 just successfully completed 00:00:25.809 --> 00:00:29.469 complex network breaches, and Codex Mobile screenshots 00:00:29.469 --> 00:00:33.200 are randomly surfacing online. It is wild. Welcome 00:00:33.200 --> 00:00:35.380 to the Deep Dive. I'm very glad you are joining 00:00:35.380 --> 00:00:39.100 us today. Today is Thursday, May 14th, 2026. 00:00:39.259 --> 00:00:41.359 Right. And Google I .O. kicks off in exactly 00:00:41.359 --> 00:00:43.780 five days. Exactly. So we are taking your sources 00:00:43.780 --> 00:00:46.500 to examine all this today. We have four massive 00:00:46.500 --> 00:00:48.820 leaks completely shaking the space. And we really 00:00:48.820 --> 00:00:50.740 need to look at what this acceleration means. 00:00:51.039 --> 00:00:53.520 Because this directly impacts your daily workflow 00:00:53.520 --> 00:00:56.310 and security setups. Okay, let's unpack this. 00:00:56.649 --> 00:00:59.369 The AI race just got incredibly hot again. It 00:00:59.369 --> 00:01:02.030 seemed to happen entirely overnight. Google hasn't 00:01:02.030 --> 00:01:05.590 officially announced Gemini 3 .2 at all. No blog 00:01:05.590 --> 00:01:08.969 posts, no documentation. Right, zero API changelogs. 00:01:08.969 --> 00:01:11.170 But the developer community is completely losing 00:01:11.170 --> 00:01:14.409 its mind anyway. Because developers keep sharing 00:01:14.409 --> 00:01:17.150 these leaked test results. So let us start with 00:01:17.150 --> 00:01:20.329 Google's unannounced Pro models. The earliest 00:01:20.329 --> 00:01:23.950 leaked outputs focus really heavily on SVG generation. 00:01:24.109 --> 00:01:27.150 Yeah, they do. It is a very specific technical 00:01:27.150 --> 00:01:29.569 capability to highlight. And for those wondering, 00:01:29.930 --> 00:01:32.930 an SVG is an image made of math, not pixels, 00:01:33.310 --> 00:01:35.930 staying sharp at any size. That is exactly why 00:01:35.930 --> 00:01:39.150 this specific detail matters so much. An AI generating 00:01:39.150 --> 00:01:41.750 pixel art is basically painting blindly. Right. 00:01:41.790 --> 00:01:44.409 But an AI generating an SVG is actively writing 00:01:44.409 --> 00:01:47.150 code. It has to perfectly plot mathematical coordinates 00:01:47.150 --> 00:01:49.829 and curves. The first test showed a PS5 controller 00:01:49.829 --> 00:01:53.019 design. The result looked... Well, okay, but 00:01:53.019 --> 00:01:55.219 nothing truly mind -blowing. Yeah, the buttons 00:01:55.219 --> 00:01:57.319 did not line up properly at all. The overall 00:01:57.319 --> 00:02:00.180 shape just felt a bit off ergonomically. It understood 00:02:00.180 --> 00:02:02.879 the basic geometry, but not human hands. Exactly. 00:02:03.200 --> 00:02:05.540 But the SVG quality improved significantly in 00:02:05.540 --> 00:02:08.000 later tests. Another test showed a pelican riding 00:02:08.000 --> 00:02:10.840 a bicycle. And that specific result looked much 00:02:10.840 --> 00:02:13.990 more coherent visually. The bird had a very clear, 00:02:13.990 --> 00:02:16.949 realistic body shape. It featured far better 00:02:16.949 --> 00:02:19.270 physical proportions. It really understood the 00:02:19.270 --> 00:02:21.610 abstract relationship between bird and machine. 00:02:21.990 --> 00:02:24.129 The bicycle itself was mostly complete this time 00:02:24.129 --> 00:02:26.879 too. No broken parts or completely disconnected 00:02:26.879 --> 00:02:29.300 wheels. But the really interesting part was the 00:02:29.300 --> 00:02:32.419 interactive interface. The model generated a 00:02:32.419 --> 00:02:35.419 full UI right alongside the graphic. Oh, wow. 00:02:35.840 --> 00:02:37.979 Yeah, users could actively customize the graphic 00:02:37.979 --> 00:02:40.759 and inspect curve. They could export files and 00:02:40.759 --> 00:02:43.460 edit specific geometric parts. This suggests 00:02:43.460 --> 00:02:46.020 a huge shift toward interactive vector workflows. 00:02:46.520 --> 00:02:49.340 It is not just about basic static image generation 00:02:49.340 --> 00:02:53.169 anymore. The SPG engine inside Gemini looks incredibly 00:02:53.169 --> 00:02:55.310 robust now. The coordinate shakes are cleaner 00:02:55.310 --> 00:02:57.990 and the layouts feel stable. But the UI generation 00:02:57.990 --> 00:02:59.629 still looks very weak, right? Unfortunately, 00:02:59.810 --> 00:03:02.629 yeah. The interfaces feel extremely flat and 00:03:02.629 --> 00:03:05.210 quite generic. They lack the deep polish of a 00:03:05.210 --> 00:03:07.550 flagship model. One example, use a Web Audio 00:03:07.550 --> 00:03:10.870 API specifically. It also incorporated Tailwind 00:03:10.870 --> 00:03:14.069 CSS into the visual design. Yet the final UI 00:03:14.069 --> 00:03:17.280 still looked very obviously AI generated. Having 00:03:17.280 --> 00:03:20.879 a great SVG engine but weak UI generation is 00:03:20.879 --> 00:03:23.860 like putting a world -class engine in a car with 00:03:23.860 --> 00:03:26.919 no steering wheel. Exactly, or a massive engine 00:03:26.919 --> 00:03:29.979 bolted to a skateboard. The raw computing power 00:03:29.979 --> 00:03:32.639 is definitely under the hood. But if the interactive 00:03:32.639 --> 00:03:35.539 interface remains that clunky... Everyday developers 00:03:35.539 --> 00:03:37.860 cannot steer it into their actual workflows. 00:03:38.360 --> 00:03:40.240 Well, part of this issue could just stem from 00:03:40.240 --> 00:03:42.900 the prompts. If testers focused purely on the 00:03:42.900 --> 00:03:46.000 vector math quality, the interface results would 00:03:46.000 --> 00:03:48.259 naturally look much weaker. That makes sense. 00:03:48.539 --> 00:03:51.300 Based on the leaks, Google clearly prioritized 00:03:51.300 --> 00:03:54.400 mathematical code quality. User experience definitely 00:03:54.400 --> 00:03:56.539 looks like the weak spot right now. But the Gemini 00:03:56.539 --> 00:03:58.919 Pro model surprised people in other ways, too. 00:03:59.439 --> 00:04:02.240 Another leak showed its true creative generation 00:04:02.240 --> 00:04:04.979 power. Right. From one simple prompt, it created 00:04:04.979 --> 00:04:08.180 four distinct robots. And all four robot designs 00:04:08.180 --> 00:04:11.099 look surprisingly good. They were not just lazy, 00:04:11.219 --> 00:04:13.199 simple copies of each other. They had completely 00:04:13.199 --> 00:04:15.280 different colors and structural shapes. Yeah. 00:04:15.479 --> 00:04:17.500 Unique, small design details were built right 00:04:17.500 --> 00:04:20.160 in. This proves it can generate multiple creative 00:04:20.160 --> 00:04:22.660 variations accurately. It does not just retune 00:04:22.660 --> 00:04:25.860 one repetitive, hallucinated output. This is 00:04:25.860 --> 00:04:29.060 incredibly useful for vector asset design workflows. 00:04:29.819 --> 00:04:31.899 Icon designers and mascot creators are going 00:04:31.899 --> 00:04:34.660 to absolutely love this. But the flash variants 00:04:34.660 --> 00:04:37.579 are where things truly heat up. Let's talk about 00:04:37.579 --> 00:04:40.689 that. There are multiple flash variants in active 00:04:40.689 --> 00:04:43.670 internal testing. The leaks mention code names 00:04:43.670 --> 00:04:47.189 Fanta, Sprite, and Cola. These are internal names 00:04:47.189 --> 00:04:49.689 for different experimental model configurations. 00:04:50.230 --> 00:04:52.629 Each variant targets completely different performance 00:04:52.629 --> 00:04:55.050 optimization goals. One might be tuned for pure 00:04:55.050 --> 00:04:57.720 processing speed. Another might focus purely 00:04:57.720 --> 00:05:00.459 on high factual coding accuracy. Right, this 00:05:00.459 --> 00:05:03.019 is standard operating practice at major AI labs. 00:05:03.279 --> 00:05:05.360 They train several versions and basically pick 00:05:05.360 --> 00:05:08.319 the best performer. And the Fanta variant specifically 00:05:08.319 --> 00:05:10.899 showed very competitive benchmark results. It 00:05:10.899 --> 00:05:12.980 is testing incredibly well against the Pro model. 00:05:13.279 --> 00:05:15.680 Fanta is nearly rivaling Pro in many complex 00:05:15.680 --> 00:05:18.079 aspects. Which raises a big question. Will the 00:05:18.079 --> 00:05:20.300 Fanta Flash variant cannibalize the Pro model's 00:05:20.300 --> 00:05:23.420 market if speed wins out? If Google ships a Flash 00:05:23.420 --> 00:05:26.670 variant this strong, It changes everything. Users 00:05:26.670 --> 00:05:29.790 prioritize speed for fast, practical, everyday 00:05:29.790 --> 00:05:32.730 engineering work. It could completely rewrite 00:05:32.730 --> 00:05:35.709 current pricing and tier models. So speed and 00:05:35.709 --> 00:05:38.350 daily efficiency might actually win out over 00:05:38.350 --> 00:05:40.949 raw, complex reasoning power. Right. And that 00:05:40.949 --> 00:05:43.689 is exactly what we are seeing. Pro handles complex 00:05:43.689 --> 00:05:46.029 reasoning and multi -step architectural planning 00:05:46.029 --> 00:05:48.970 tasks. While Flash handles the fast, repetitive 00:05:49.660 --> 00:05:53.180 practical daily coding work. Exactly. But here's 00:05:53.180 --> 00:05:56.139 the really critical thing to consider. If Google 00:05:56.139 --> 00:05:58.720 nails that daily efficiency with their models, 00:05:59.300 --> 00:06:01.139 OpenAI cannot just sit around on their hands 00:06:01.139 --> 00:06:03.500 quietly. Very true. What's fascinating here is 00:06:03.500 --> 00:06:06.740 the sheer development speed. The leaks show OpenAI 00:06:06.740 --> 00:06:09.240 is definitely feeling the heat. While everyone 00:06:09.240 --> 00:06:12.160 watches Google, OpenAI is testing very quietly. 00:06:12.459 --> 00:06:15.079 GPT 5 .6 is already in active internal testing. 00:06:15.220 --> 00:06:16.819 Yeah, the first development checkpoints were 00:06:16.819 --> 00:06:18.930 evaluated very recently to internal code. code 00:06:18.930 --> 00:06:20.889 names showed up in the recent leaf. Amber alpha 00:06:20.889 --> 00:06:23.189 and beacon alpha, right? Those are the ones. 00:06:23.569 --> 00:06:25.550 These are likely two advanced variants being 00:06:25.550 --> 00:06:27.709 heavily tested. They want to see which gives 00:06:27.709 --> 00:06:30.730 better complex benchmark results. They run advanced 00:06:30.730 --> 00:06:33.509 candidates in parallel and pick one. We might 00:06:33.509 --> 00:06:36.180 see these internal candidates appear. very quietly 00:06:36.180 --> 00:06:38.519 soon. They often show up on LM Arena testing 00:06:38.519 --> 00:06:40.920 first. They just run public blind testing without 00:06:40.920 --> 00:06:43.540 telling anyone beforehand. The gap between major 00:06:43.540 --> 00:06:47.139 releases is shrinking drastically. I mean, GPT 00:06:47.139 --> 00:06:49.879 5 .5 just launched a few short weeks ago. And 00:06:49.879 --> 00:06:52.420 it was a massive structural success for the company. 00:06:52.740 --> 00:06:54.879 Reviewers said it outperformed Anthropix Opus 00:06:54.879 --> 00:06:58.079 easily. That momentum gives OpenAI strong reason 00:06:58.079 --> 00:07:00.709 to push forward. especially with Google about 00:07:00.709 --> 00:07:03.850 to show new flagship models. If OpenAI waits 00:07:03.850 --> 00:07:07.089 too long, they lose the narrative entirely. Some 00:07:07.089 --> 00:07:09.350 reliable sources actually suggest a release by 00:07:09.350 --> 00:07:11.779 next month. Wow. That would be one of the shortest 00:07:11.779 --> 00:07:14.060 release gaps ever. One major reason for this 00:07:14.060 --> 00:07:17.060 acceleration is quite profound. AI itself now 00:07:17.060 --> 00:07:19.819 actively helps build new AI models. Right. Models 00:07:19.819 --> 00:07:22.339 constantly assist with complex code generation 00:07:22.339 --> 00:07:25.279 tasks now. They handle vast data preparation 00:07:25.279 --> 00:07:27.699 and evaluation during training. They optimize 00:07:27.699 --> 00:07:30.279 the actual architectural training process itself 00:07:30.279 --> 00:07:33.160 continuously. They completely remove the human 00:07:33.160 --> 00:07:36.509 bottleneck. from data collection. The model generates 00:07:36.509 --> 00:07:39.310 millions of complex edge case coding problems. 00:07:39.610 --> 00:07:41.930 And then it autonomously solves those exact same 00:07:41.930 --> 00:07:45.149 coding problems. And finally, an internal reward 00:07:45.149 --> 00:07:47.850 model grades the final solution. You know, Ibeat, 00:07:48.370 --> 00:07:51.379 it is a very humbling reality. I still wrestle 00:07:51.379 --> 00:07:54.379 with prompt drift myself on a daily basis. Yet 00:07:54.379 --> 00:07:56.500 these models are already debugging their own 00:07:56.500 --> 00:07:58.699 training code. It really makes you realize how 00:07:58.699 --> 00:08:01.639 quickly we become bottlenecks. Models generate 00:08:01.639 --> 00:08:04.420 synthetic data way more efficiently than human 00:08:04.420 --> 00:08:06.959 teams. They instantly spot critical processing 00:08:06.959 --> 00:08:09.699 bottlenecks earlier than us. Every lab knows 00:08:09.699 --> 00:08:12.560 slowing down means losing crucial ground. Absolutely. 00:08:12.860 --> 00:08:15.180 So since AI is building AI right now, are we 00:08:15.180 --> 00:08:17.360 going to see massive capability leaps or just 00:08:17.360 --> 00:08:19.660 better integration? We might not see those massive 00:08:19.660 --> 00:08:22.240 historical capability jumps anymore. Early years 00:08:22.240 --> 00:08:24.879 were defined by those massive intelligence leaps. 00:08:25.120 --> 00:08:28.060 Now, updates will focus heavily on pure processing 00:08:28.060 --> 00:08:30.399 speed. They will focus on smoother workflows 00:08:30.399 --> 00:08:33.820 and deep personalization. Right. Smoother workflows 00:08:33.820 --> 00:08:36.600 and integration matter far more now than chasing 00:08:36.600 --> 00:08:39.340 massive capability leaps. And that concept of 00:08:39.340 --> 00:08:41.320 smoother workflows is actually very important. 00:08:41.200 --> 00:08:45.080 It connects directly to highly complex, high 00:08:45.080 --> 00:08:47.480 -stakes digital environments. Which brings us 00:08:47.480 --> 00:08:50.899 to the intense cybersecurity sector today. This 00:08:50.899 --> 00:08:53.440 specific space does not get enough mainstream 00:08:53.440 --> 00:08:56.429 attention at all. It really doesn't. And the 00:08:56.429 --> 00:08:58.870 cybersecurity AI crown is shifting incredibly 00:08:58.870 --> 00:09:01.970 fast lately. We started with Claude Opus 4 .6, 00:09:02.110 --> 00:09:04.629 leading everyone. Then the original Mythos preview 00:09:04.629 --> 00:09:07.190 appeared on the scene. It stayed slightly behind 00:09:07.190 --> 00:09:10.769 the Opus model initially. But then GPT 5 .5 Cyber 00:09:10.769 --> 00:09:14.320 closed that performance gap. It was tuned specifically 00:09:14.320 --> 00:09:17.100 for complex security network workflows. Right, 00:09:17.200 --> 00:09:19.259 and now Mythos Preview Version 2 is officially 00:09:19.259 --> 00:09:21.700 arrived. Some elite developers are currently 00:09:21.700 --> 00:09:24.220 pronouncing it as Mythos. It just received a 00:09:24.220 --> 00:09:26.379 major upgrade that is turning industry heads. 00:09:27.039 --> 00:09:29.639 Early benchmark results show a clear capability 00:09:29.639 --> 00:09:33.100 jump immediately. Institutions like the AI Security 00:09:33.100 --> 00:09:36.039 Institute are actively testing it. It has confidently 00:09:36.039 --> 00:09:39.279 jumped back into the absolute lead. This constant 00:09:39.279 --> 00:09:42.519 back and forth really shows the development speed. 00:09:42.620 --> 00:09:46.460 The competitive cybersecurity AI race is incredibly 00:09:46.460 --> 00:09:49.120 intense right now. The most striking test involves 00:09:49.120 --> 00:09:52.980 a complex network simulation. They use a 32 -step 00:09:52.980 --> 00:09:55.259 corporate network attack simulation. And this 00:09:55.259 --> 00:09:57.860 is definitely not a simple theoretical toy benchmark. 00:09:58.259 --> 00:10:01.240 The model needs to identify obscure digital system 00:10:01.240 --> 00:10:04.379 vulnerabilities. It has to actually chain multiple 00:10:04.379 --> 00:10:08.179 obscure software exploits together. It must execute 00:10:08.179 --> 00:10:11.259 lateral movement smoothly through the whole network. 00:10:11.480 --> 00:10:13.639 moving from a compromised printer straight to 00:10:13.639 --> 00:10:17.029 an admin laptop. It has to complete a full, complex 00:10:17.029 --> 00:10:19.750 attack path. A human cybersecurity expert needs 00:10:19.750 --> 00:10:22.870 roughly 20 hours total for this. But this new 00:10:22.870 --> 00:10:25.330 mythos preview completed the entire simulation 00:10:25.330 --> 00:10:27.909 rapidly. Yeah, it took only 6 to 10 digital attempts 00:10:27.909 --> 00:10:30.450 total. This is a dramatic difference in operational 00:10:30.450 --> 00:10:33.590 security capability. Whoa, Pete. Imagine simulating 00:10:33.590 --> 00:10:36.389 a 32 -step network breach in just a few tries. 00:10:36.610 --> 00:10:38.909 The magnitude of that technical achievement is 00:10:38.909 --> 00:10:41.629 honestly staggering. A 32 -step attack chain 00:10:41.629 --> 00:10:44.320 requires deep architectural understanding. It 00:10:44.320 --> 00:10:47.200 must know exactly which obscure exploits actually 00:10:47.200 --> 00:10:50.279 work. It strings everything together in the correct 00:10:50.279 --> 00:10:52.860 logical sequence. The fact that an AI does this 00:10:52.860 --> 00:10:55.879 reliably really matters. It is a massive structural 00:10:55.879 --> 00:10:58.539 deal for the entire industry. But does giving 00:10:58.539 --> 00:11:02.139 an AI the ability to chain exploits hand a dangerous 00:11:02.139 --> 00:11:05.000 weapon straight to bad actors? That is where 00:11:05.000 --> 00:11:07.220 things become significantly more complicated 00:11:07.220 --> 00:11:10.559 ethically. The exact same model defending a network 00:11:10.559 --> 00:11:13.200 can attack it. The potential risks of widespread 00:11:13.200 --> 00:11:16.340 misuse are incredibly high. That is exactly why 00:11:16.340 --> 00:11:19.539 software releases are carefully staged. Governments 00:11:19.539 --> 00:11:21.759 and security institutions usually get private 00:11:21.759 --> 00:11:24.220 access first. They have crucial time to prepare 00:11:24.220 --> 00:11:26.799 necessary security defenses. They actively do 00:11:26.799 --> 00:11:29.179 this before tools become widely available. Exactly. 00:11:29.500 --> 00:11:31.460 Defense and attack grow together. Early access 00:11:31.460 --> 00:11:33.600 for human defenders is absolutely crucial. But 00:11:33.600 --> 00:11:36.899 this immense autonomous computing power is expanding 00:11:36.899 --> 00:11:39.919 very rapidly elsewhere, too. It is not staying 00:11:39.919 --> 00:11:41.960 securely locked inside corporate data centers 00:11:41.960 --> 00:11:43.980 forever. Right. It is moving directly to your 00:11:43.980 --> 00:11:46.440 personal mobile phone. recent leaked screenshots 00:11:46.440 --> 00:11:49.039 started showing OpenAI's codecs running locally. 00:11:49.340 --> 00:11:51.899 It is operating directly from a standard mobile 00:11:51.899 --> 00:11:54.639 device. This immediately caught the attention 00:11:54.639 --> 00:11:57.879 of many online developers. It does not look like 00:11:57.879 --> 00:12:00.360 a highly polished app yet. It seems much closer 00:12:00.360 --> 00:12:02.919 to a remote coding workflow. Yeah, they are using 00:12:02.919 --> 00:12:05.460 a mobile CLI right now. Let's quickly define 00:12:05.460 --> 00:12:08.600 that. A mobile CLI is a text -based command line 00:12:08.600 --> 00:12:10.759 to control remote computers from your phone. 00:12:11.080 --> 00:12:13.820 That is a perfect, concise summary of the tool. 00:12:14.190 --> 00:12:16.970 You can manage complex coding tasks away from 00:12:16.970 --> 00:12:19.129 your desk. You can comfortably review outputs 00:12:19.129 --> 00:12:22.110 and interact directly with codecs. Earlier leaks 00:12:22.110 --> 00:12:24.690 hinted at much deeper workflow integrations coming 00:12:24.690 --> 00:12:27.750 too. Tools like NotionHQ are heavily rumored 00:12:27.750 --> 00:12:30.429 right here. You could plan, document, and code 00:12:30.429 --> 00:12:33.450 in one single place. There is also specific mention 00:12:33.450 --> 00:12:35.889 of remote control features. You can trigger heavy 00:12:35.889 --> 00:12:38.289 codecs compiling tasks directly from your phone. 00:12:38.440 --> 00:12:40.700 Because the heavy processing still runs safely 00:12:40.700 --> 00:12:43.299 on a remote server. Rohan Burma from the Core 00:12:43.299 --> 00:12:46.019 Codex team was actually mentioned. He was specifically 00:12:46.019 --> 00:12:48.320 cited as a primary source for this leak. Which 00:12:48.320 --> 00:12:51.080 adds major industry credibility to these current 00:12:51.080 --> 00:12:53.679 screenshots. This definitely does not look like 00:12:53.679 --> 00:12:55.960 random internet speculation. If Codex arrives 00:12:55.960 --> 00:12:58.940 on mobile, it completely changes developer workflows. 00:12:59.519 --> 00:13:01.720 Right now, most professional tools need a heavy 00:13:01.720 --> 00:13:04.399 desktop IDE. Mobile access lets you actively 00:13:04.399 --> 00:13:07.509 review code from anywhere. You can approve massive 00:13:07.509 --> 00:13:09.970 pull requests while on the go. It's not about 00:13:09.970 --> 00:13:12.450 typing syntax with your thumbs. It's like carrying 00:13:12.450 --> 00:13:14.710 a senior developer in your pocket as a remote 00:13:14.710 --> 00:13:17.289 control. That is exactly how it will psychologically 00:13:17.289 --> 00:13:19.570 function for people. You could be sitting on 00:13:19.570 --> 00:13:21.710 a busy train commute. Or you could be sitting 00:13:21.710 --> 00:13:24.509 quietly in a long meeting. You can still keep 00:13:24.509 --> 00:13:26.629 your complex coding projects moving forward. 00:13:26.830 --> 00:13:29.970 You open your phone and check the project documentation. 00:13:30.389 --> 00:13:33.009 You trigger Codex to actively review the new 00:13:33.009 --> 00:13:36.500 logic. Codex finds a bug, fixes it. and pushes 00:13:36.500 --> 00:13:39.240 the commits. Will developers actually want to 00:13:39.240 --> 00:13:41.379 manage heavy processing tasks away from their 00:13:41.379 --> 00:13:43.399 desks? They'll definitely not do heavy typing 00:13:43.399 --> 00:13:46.019 on a mobile screen, but they will urgently want 00:13:46.019 --> 00:13:47.919 to unblock their teammates. They will want to 00:13:47.919 --> 00:13:50.679 review AI -generated code snippets very quickly. 00:13:50.980 --> 00:13:53.000 No, it adds flexibility to keep projects moving 00:13:53.000 --> 00:13:55.759 anywhere without replacing the desktop entirely. 00:13:57.440 --> 00:13:59.740 Mid -roll sponsor break. So what does this all 00:13:59.740 --> 00:14:01.740 mean? If we connect this to the bigger picture 00:14:01.740 --> 00:14:05.450 today... A very clear, undeniable historical 00:14:05.450 --> 00:14:08.169 pattern is showing up. All four leak stories 00:14:08.169 --> 00:14:10.889 point to the exact same trend. AI development 00:14:10.889 --> 00:14:13.590 labs are releasing massive models faster than 00:14:13.590 --> 00:14:15.870 ever. The technical time between major versions 00:14:15.870 --> 00:14:18.600 is completely collapsing now. Internal testing 00:14:18.600 --> 00:14:21.379 cycles overlap heavily with public release cycles. 00:14:21.919 --> 00:14:24.340 Significant leaks happen before official corporate 00:14:24.340 --> 00:14:27.100 announcements way more often. Two massive market 00:14:27.100 --> 00:14:30.139 forces constantly drive this rapid acceleration. 00:14:30.679 --> 00:14:33.620 Advanced AI tools actively make software development 00:14:33.620 --> 00:14:36.100 much faster. An extreme competitive corporate 00:14:36.100 --> 00:14:38.679 pressure drives the entire industry forward. 00:14:39.080 --> 00:14:41.740 No single lab can afford to look completely behind. 00:14:41.940 --> 00:14:44.500 We are looking at a massive volatile pressure 00:14:44.500 --> 00:14:48.269 cooker. OpenAI aggressively pushes 5 .6 because 00:14:48.269 --> 00:14:50.889 of Google I .O. Google rushes Gemini development 00:14:50.889 --> 00:14:53.549 because Anthropic's Mythos is gaining ground. 00:14:53.789 --> 00:14:55.909 Anthropic accelerates Mythos testing because 00:14:55.909 --> 00:14:59.049 of GPT -Cyber's incredible performance. Every 00:14:59.049 --> 00:15:01.789 strategic move by one lab creates intense market 00:15:01.789 --> 00:15:03.929 pressure. It forces the other major competitors 00:15:03.929 --> 00:15:06.750 to sprint even faster. That is exactly why we 00:15:06.750 --> 00:15:09.210 see this massive flood of leaks. The broader 00:15:09.210 --> 00:15:11.809 AI space moves way faster than people can track. 00:15:12.309 --> 00:15:14.750 The real practical lesson here is about managing 00:15:14.750 --> 00:15:18.299 the pacing. The technical field heavily rewards 00:15:18.299 --> 00:15:21.139 people who stay consistently informed. You must 00:15:21.139 --> 00:15:24.080 actively act on what actually matters most. Right. 00:15:24.120 --> 00:15:27.100 Do not treat every single internet leak as gospel. 00:15:27.519 --> 00:15:30.080 You should patiently wait for the official software 00:15:30.080 --> 00:15:32.500 releases. Compare the real technical outputs 00:15:32.500 --> 00:15:35.559 when they are finally public. Pay close, careful 00:15:35.559 --> 00:15:38.500 attention to actual real -world software deployments. 00:15:38.919 --> 00:15:41.639 Track the quiet, unannounced deployments on LM 00:15:41.639 --> 00:15:44.720 Arena closely. The most meaningful software improvements 00:15:44.720 --> 00:15:47.559 are often very quiet. They just secretly make 00:15:47.559 --> 00:15:50.519 your daily work much faster. They smoothly integrate 00:15:50.519 --> 00:15:52.419 into your workflow without you even noticing. 00:15:52.600 --> 00:15:54.759 Two secs silence. Here's where it gets really 00:15:54.759 --> 00:15:57.179 interesting for the future. Google I .O. officially 00:15:57.179 --> 00:15:59.980 happens on May the 19th. You need to keep a very 00:15:59.980 --> 00:16:03.240 close eye on it. We must actively see if the 00:16:03.240 --> 00:16:06.039 pro versus flash reality actually matches the 00:16:06.039 --> 00:16:08.679 leaks. We desperately need to see if flash truly 00:16:08.679 --> 00:16:10.639 wins out. I want to leave you with one final 00:16:10.639 --> 00:16:13.399 thought today. We know AI is successfully debugging 00:16:13.399 --> 00:16:16.460 its own training code. It is autonomously accelerating 00:16:16.460 --> 00:16:19.320 the software development loop drastically. What 00:16:19.320 --> 00:16:21.320 happens when these models begin designing the 00:16:21.320 --> 00:16:23.659 next generation of physical hardware and chips 00:16:23.659 --> 00:16:26.980 optimized purely for themselves? That is a truly 00:16:26.980 --> 00:16:29.580 staggering philosophical concept to consider. 00:16:29.919 --> 00:16:31.980 They could autonomously design the next generation 00:16:31.980 --> 00:16:34.220 of silicon chips. They could completely remove 00:16:34.220 --> 00:16:36.299 the human bottleneck from the physical timeline. 00:16:37.000 --> 00:16:39.419 It is something to seriously ponder on your own. 00:16:39.659 --> 00:16:40.120 Stay sharp.