WEBVTT 00:00:00.000 --> 00:00:03.140 Right now, AI models can actually boot up and 00:00:03.140 --> 00:00:06.400 beat Pokemon autonomously. Yeah, it totally sounds 00:00:06.400 --> 00:00:08.859 like a really entertaining, harmless party trick. 00:00:09.119 --> 00:00:11.740 But then you read today's MIT Future Tech Research 00:00:11.740 --> 00:00:14.800 Report. Right. Those experts warn this exact 00:00:14.800 --> 00:00:18.179 level of autonomy brings severe dangers. They 00:00:18.179 --> 00:00:21.140 see a 10 % chance of triggering global societal 00:00:21.140 --> 00:00:24.000 catastrophe. I mean, we're talking about $100 00:00:24.000 --> 00:00:29.809 billion damages by 2030. Beat! Welcome to today's 00:00:29.809 --> 00:00:32.570 deep dive. I am incredibly glad you're joining 00:00:32.570 --> 00:00:35.899 us. We truly have a lot of Complex, fascinating 00:00:35.899 --> 00:00:38.299 ground to cover today. So we are unpacking an 00:00:38.299 --> 00:00:41.479 avalanche of massive AI capability jumps right 00:00:41.479 --> 00:00:43.979 now. Anthropic just dropped their highly anticipated 00:00:43.979 --> 00:00:46.500 Claude Fable 5 model today. And we really need 00:00:46.500 --> 00:00:48.420 to examine the structural collision course this 00:00:48.420 --> 00:00:50.259 creates. Yeah, these blistering capabilities 00:00:50.259 --> 00:00:52.799 are hitting global security and hardware supply 00:00:52.799 --> 00:00:55.219 chains. They're also crashing right into our 00:00:55.219 --> 00:00:57.859 necessary safety guardrails. So let's start with 00:00:57.859 --> 00:01:00.140 the sheer raw power hitting the market right 00:01:00.140 --> 00:01:03.000 now. The newest models arriving today are just 00:01:03.000 --> 00:01:06.120 absolutely wild. Anthropic released Claude Fable 00:01:06.120 --> 00:01:09.060 5 directly to the public recently. It is the 00:01:09.060 --> 00:01:12.599 first major model from their new mythos family. 00:01:13.019 --> 00:01:15.500 Fable 5 is absolutely crushing the benchmark 00:01:15.500 --> 00:01:18.219 charts across the board. It shows truly exceptional 00:01:18.219 --> 00:01:21.340 results in complex knowledge and vision tasks. 00:01:21.659 --> 00:01:23.739 It is kind of like stacking Lego blocks of data. 00:01:24.120 --> 00:01:26.900 That is a perfect way to visualize its multimodal 00:01:26.900 --> 00:01:29.359 architecture. You just snap a visual screenshot 00:01:29.359 --> 00:01:32.700 onto a basic text prompt. And the AI just builds 00:01:32.700 --> 00:01:34.700 the entire functional code right from there. 00:01:34.840 --> 00:01:37.340 It does not see images and text as separate languages. 00:01:37.659 --> 00:01:40.299 It blends them together into a unified understanding 00:01:40.299 --> 00:01:42.799 of the problem. Which perfectly explains how 00:01:42.799 --> 00:01:45.019 it can rebuild entire web apps from scratch. 00:01:45.400 --> 00:01:47.879 Yeah, and it also explains the Pokemon Fire Red 00:01:47.879 --> 00:01:50.260 achievement. The AI completed the entire game 00:01:50.260 --> 00:01:53.079 without any human intervention at all. Just think 00:01:53.079 --> 00:01:55.219 about this. deep long term planning required 00:01:55.219 --> 00:01:57.599 for that. You have to manage inventory and navigate 00:01:57.599 --> 00:02:00.819 highly complex spatial maps. The agent handles 00:02:00.819 --> 00:02:03.340 all of those sequential decisions completely 00:02:03.340 --> 00:02:06.299 autonomously. It proves these agents can handle 00:02:06.299 --> 00:02:08.439 long term enterprise software workflows now. 00:02:08.659 --> 00:02:12.080 But this incredible autonomous capability definitely 00:02:12.080 --> 00:02:16.139 comes with a steep price. It costs $10 per million 00:02:16.139 --> 00:02:18.819 input tokens right now. Tokens are just the fundamental 00:02:18.819 --> 00:02:21.479 puzzle pieces of language that AI processes. 00:02:21.840 --> 00:02:24.219 And output tokens will actually cost you $50 00:02:24.219 --> 00:02:26.900 per million. That is exactly double the cost 00:02:26.900 --> 00:02:30.159 of their previous Opus model. Opus 4 .8 was already 00:02:30.159 --> 00:02:32.719 considered a fairly pricey option. Startup founders 00:02:32.719 --> 00:02:36.060 building on Fable 5 will burn cash much faster. 00:02:36.280 --> 00:02:38.520 They really have to route simple tasks to much 00:02:38.520 --> 00:02:41.020 cheaper models. Fable 5 is currently available 00:02:41.020 --> 00:02:44.580 via their developer API system. API is a software 00:02:44.580 --> 00:02:46.919 bridge letting different computer programs talk. 00:02:47.120 --> 00:02:49.800 If you maintain a Pro, Max, or Team subscription, 00:02:49.939 --> 00:02:52.060 though, you actually get it included completely 00:02:52.060 --> 00:02:55.360 free until June 22nd. That gives power users 00:02:55.360 --> 00:02:57.939 a fantastic window to test the capabilities. 00:02:58.300 --> 00:03:00.780 And this rapid capability leap is mirrored across 00:03:00.780 --> 00:03:03.340 the whole ecosystem. We are fully entering the 00:03:03.340 --> 00:03:05.979 agentic era of artificial intelligence. Look 00:03:05.979 --> 00:03:08.120 at this fascinating new tool called KimiWork, 00:03:08.240 --> 00:03:12.360 for example. It deploys up to 300 local AI agents 00:03:12.360 --> 00:03:15.219 simultaneously. Local agents are AI programs 00:03:15.219 --> 00:03:20.740 running directly on your personal hardware. They 00:03:20.740 --> 00:03:25.979 automate complex web tasks and analyze dense 00:03:25.979 --> 00:03:31.860 financial data sets. Then we have another tool 00:03:31.860 --> 00:03:36.099 called Macatron, making major waves today. It 00:03:36.099 --> 00:03:41.939 controls native applications using very simple 00:03:42.000 --> 00:03:45.120 conversational English commands. We're also seeing 00:03:45.120 --> 00:03:49.240 Google aggressively push Gemini 3 .5. Their live 00:03:49.240 --> 00:03:51.939 translate feature handles real -time conversational 00:03:51.939 --> 00:03:54.990 translation. Almost perfectly now. You can translate 00:03:54.990 --> 00:03:58.150 complex, nuanced conversations exactly as they 00:03:58.150 --> 00:04:00.430 happen globally. The business scale behind these 00:04:00.430 --> 00:04:02.689 tools is getting incredibly hard to fathom. A 00:04:02.689 --> 00:04:05.449 company called Lovable just crossed $500 million 00:04:05.449 --> 00:04:08.569 in annualized revenue. They have seen over 50 00:04:08.569 --> 00:04:11.030 million user projects created already. And their 00:04:11.030 --> 00:04:13.810 usage is accelerating to 1 million new projects 00:04:13.810 --> 00:04:16.730 weekly. Whoa, imagine scaling to a billion queries. 00:04:17.129 --> 00:04:19.269 It completely breaks my brain just trying to 00:04:19.269 --> 00:04:22.250 comprehend that volume. It is a totally un - 00:04:22.439 --> 00:04:24.959 unprecedented level of software creation speed. 00:04:25.160 --> 00:04:27.420 You no longer need to hire an entire team of 00:04:27.420 --> 00:04:29.319 software engineers. You just describe what you 00:04:29.319 --> 00:04:32.519 want in simple, plain English. And Lovable spins 00:04:32.519 --> 00:04:35.019 up the entire working application for you instantly. 00:04:35.899 --> 00:04:39.220 Two sec silence. It is genuinely magical to watch. 00:04:39.399 --> 00:04:41.959 With Fable 5 dominating, where does that leave 00:04:41.959 --> 00:04:44.959 Claude Opus for a user trying to get work done? 00:04:45.220 --> 00:04:47.600 We actually heard directly from Tariq on the 00:04:47.600 --> 00:04:49.939 Claude Code team about this. Oh, wow. What did 00:04:49.939 --> 00:04:52.160 he specifically say about Opus? Well, he says 00:04:52.160 --> 00:04:54.600 Opus remains the absolute strongest model for 00:04:54.600 --> 00:04:56.879 extended work. So it is better for those really 00:04:56.879 --> 00:05:00.319 long, complex tasks. Right. Opus still handles 00:05:00.319 --> 00:05:03.480 multi -day, sustained, autonomous execution much 00:05:03.480 --> 00:05:05.970 better today. So Opus remains the marathon runner 00:05:05.970 --> 00:05:08.509 while Fable sprints through complex reasoning. 00:05:08.709 --> 00:05:10.529 Exactly. You always need to pick the right tool 00:05:10.529 --> 00:05:12.730 for the job. Which logically brings us to the 00:05:12.730 --> 00:05:14.709 physical reality underneath all this software. 00:05:14.970 --> 00:05:17.850 You cannot magically scale to a million new coding 00:05:17.850 --> 00:05:21.089 projects weekly. It takes immense hardware, massive 00:05:21.089 --> 00:05:23.470 funding, and incredible physical infrastructure 00:05:23.470 --> 00:05:27.370 networks. And we are seeing massive structural 00:05:27.370 --> 00:05:30.069 shifts in the financial markets. The old FAA 00:05:30.069 --> 00:05:32.790 acronym is completely dead at this point. Financial 00:05:32.790 --> 00:05:36.240 analysts are heavily using the new acronym MANGOES 00:05:36.240 --> 00:05:39.199 instead today. That stands for Meta, Anthropic, 00:05:39.519 --> 00:05:43.620 NVIDIA, Google, OpenAI, and SpaceX. It perfectly 00:05:43.620 --> 00:05:46.300 captures the massive financial shift toward artificial 00:05:46.300 --> 00:05:48.500 intelligence dominance. And the sheer volume 00:05:48.500 --> 00:05:50.579 of money flowing into these companies is staggering. 00:05:50.899 --> 00:05:54.319 A startup called Moonshot AI is currently raising 00:05:54.319 --> 00:05:58.379 $2 billion. That represents a massive 50 % jump 00:05:58.379 --> 00:06:00.519 in their overall valuation. They were valued 00:06:00.519 --> 00:06:03.560 at $20 billion just one short month. It is a 00:06:03.560 --> 00:06:06.000 level of rapid capital accumulation we rarely 00:06:06.000 --> 00:06:08.519 ever see. But all that investment money has to 00:06:08.519 --> 00:06:14.439 buy actual physical hardware. They need these 00:06:14.439 --> 00:06:17.139 highly specific AI processing chips for 2028. 00:06:17.560 --> 00:06:20.199 And there is a massive shift happening in global 00:06:20.199 --> 00:06:23.339 chip manufacturing. Google and Nvidia are reportedly 00:06:23.339 --> 00:06:26.019 turning to Intel foundries right now. They are 00:06:26.019 --> 00:06:28.620 actively using Intel as a critical backup to 00:06:28.620 --> 00:06:32.699 TSMC. TSMC has historically dominated the entire 00:06:32.699 --> 00:06:35.879 global semiconductor chip manufacturing market. 00:06:36.379 --> 00:06:39.560 Why are Google and Nvidia suddenly looking at 00:06:39.560 --> 00:06:42.360 Intel for their 2028 chips instead of just sticking 00:06:42.360 --> 00:06:46.579 with TSMC? It largely comes down to intense geopolitical 00:06:46.579 --> 00:06:49.199 realities and sheer volume. Meaning they just 00:06:49.199 --> 00:06:52.600 cannot rely on one single factory anymore. Exactly. 00:06:52.660 --> 00:06:55.060 Relying entirely on a single foundry in Taiwan 00:06:55.060 --> 00:06:57.699 is incredibly risky. Right. Diversifying away 00:06:57.699 --> 00:07:00.399 from TSMC is basically a massive supply chain 00:07:00.399 --> 00:07:02.660 insurance policy. That is exactly what it is 00:07:02.660 --> 00:07:05.040 for these massive tech giants. Building a semiconductor 00:07:05.040 --> 00:07:07.879 foundry. It takes many billions of dollars in 00:07:07.879 --> 00:07:10.000 years. You absolutely cannot just spin up a new 00:07:10.000 --> 00:07:12.379 chip factory overnight. It requires massive clean 00:07:12.379 --> 00:07:15.379 rooms and incredibly complex photolithography 00:07:15.379 --> 00:07:18.139 machines today. But Taiwan sits in an incredibly 00:07:18.139 --> 00:07:21.180 precarious geopolitical hotspot right now. If 00:07:21.180 --> 00:07:23.500 those supply chains get disrupted, the entire 00:07:23.500 --> 00:07:26.279 AI industry halts completely. Google and Nvidia 00:07:26.279 --> 00:07:28.879 absolutely cannot afford that massive level of 00:07:28.879 --> 00:07:30.779 existential risk. So they are pouring massive 00:07:30.779 --> 00:07:33.500 resources into Intel's new domestic foundries. 00:07:33.759 --> 00:07:36.060 But the infrastructure... friction is not just 00:07:36.060 --> 00:07:38.439 about the physical semiconductor chips. We also 00:07:38.439 --> 00:07:41.459 have to deeply examine the massive software security 00:07:41.459 --> 00:07:44.519 vulnerabilities. Microsoft just had to surgically 00:07:44.519 --> 00:07:48.000 remove over 70 compromised GitHub repositories. 00:07:48.139 --> 00:07:50.699 Hackers actively slipped highly dangerous malware 00:07:50.699 --> 00:07:53.259 into these specific coding projects. These compromised 00:07:53.259 --> 00:07:56.379 projects were tied directly to automated AI coding 00:07:56.379 --> 00:07:59.120 tools. They specifically targeted automated agent 00:07:59.120 --> 00:08:02.699 platforms like Cloud Code and Gemini CLI. It 00:08:02.699 --> 00:08:05.040 shows how deeply vulnerable these automated development 00:08:05.040 --> 00:08:07.920 environments truly are. When AI agents write 00:08:07.920 --> 00:08:10.699 code, they often pull from open source libraries. 00:08:11.040 --> 00:08:13.819 If a hacker poisons that library, the AI blindly 00:08:13.819 --> 00:08:17.060 pulls the malware. And human developers are trusting 00:08:17.060 --> 00:08:20.079 these AI tools a bit too blindly. It is a massive 00:08:20.079 --> 00:08:22.879 structural security flaw in the modern coding 00:08:22.879 --> 00:08:25.379 ecosystem. This severe friction extends to the 00:08:25.379 --> 00:08:28.019 actual raw coding capabilities as well. A company 00:08:28.019 --> 00:08:30.959 called Cognition just released their new frontier 00:08:30.959 --> 00:08:33.700 code benchmark system. The absolute top frontier 00:08:33.700 --> 00:08:36.440 models currently score just 13 out of 100. That 00:08:36.440 --> 00:08:39.080 explicitly shows real -world enterprise coding 00:08:39.080 --> 00:08:42.039 still hits massive walls of friction. I still 00:08:42.039 --> 00:08:45.039 wrestle with prompt drift myself. Yeah, prompt 00:08:45.039 --> 00:08:47.799 drift is when an AI gradually loses focus on 00:08:47.799 --> 00:08:49.600 your original instructions. You ask the agent 00:08:49.600 --> 00:08:52.620 to fix a specific tiny front -end button issue. 00:08:52.840 --> 00:08:54.700 And suddenly it tries to rewrite your entire 00:08:54.700 --> 00:08:57.139 back -end database architecture instead. We are 00:08:57.139 --> 00:08:59.679 still very far from having completely autonomous 00:08:59.679 --> 00:09:02.500 software engineers. The agentic era is clearly 00:09:02.500 --> 00:09:05.990 here, but it is deeply. deeply messy. We have 00:09:05.990 --> 00:09:08.730 brilliant flashes of magic mixed with fundamental 00:09:08.730 --> 00:09:11.529 structural failures. It creates a very precarious 00:09:11.529 --> 00:09:13.889 foundation for our critical digital infrastructure 00:09:13.889 --> 00:09:16.409 today. And the mounting geopolitical tensions 00:09:16.409 --> 00:09:19.450 over chips just adds significantly more pressure. 00:09:19.629 --> 00:09:21.950 We are moving incredibly fast on deeply fragile 00:09:21.950 --> 00:09:24.450 underlying foundation systems. Which brings us 00:09:24.450 --> 00:09:27.350 to the biggest structural problem we collectively 00:09:27.350 --> 00:09:30.269 face. The catastrophic risks of this technology 00:09:30.269 --> 00:09:33.490 scaling without proper safety guardrails. Be 00:09:33.490 --> 00:09:36.700 sponsored. Before the break, we analyzed intense 00:09:36.700 --> 00:09:40.320 geopolitical hardware chip supply tensions. And 00:09:40.320 --> 00:09:43.580 we looked at those massive automated GitHub malware 00:09:43.580 --> 00:09:46.200 security breaches. These structural software 00:09:46.200 --> 00:09:49.620 issues lead directly into the broader macro risk 00:09:49.620 --> 00:09:52.120 landscape. We have blistering capability growth 00:09:52.120 --> 00:09:55.879 actively colliding with terrible structural dangers. 00:09:56.399 --> 00:09:59.500 MIT FutureTech just released a deeply unsettling 00:09:59.500 --> 00:10:02.460 new AI blueprint today. They comprehensively 00:10:02.460 --> 00:10:06.000 surveyed 272 leading international AI experts. 00:10:06.220 --> 00:10:08.519 The detailed findings are honestly enough to 00:10:08.519 --> 00:10:10.940 keep you awake tonight. They carefully analyzed 00:10:10.940 --> 00:10:15.100 24 major AI risks currently threatening our modern 00:10:15.100 --> 00:10:17.919 society. Experts critically believe 18 of those 00:10:17.919 --> 00:10:20.820 risks have catastrophic outcome potential. They 00:10:20.820 --> 00:10:23.500 assign a greater than 10 % chance to those catastrophic 00:10:23.500 --> 00:10:26.129 events. That explicitly means... hundred billion 00:10:26.129 --> 00:10:29.190 dollar damages or massive global societal disruption. 00:10:29.509 --> 00:10:31.529 And they predict these massive structural failures 00:10:31.529 --> 00:10:34.330 could easily happen by 2030. That is only a few 00:10:34.330 --> 00:10:36.870 short, highly volatile years away from us today. 00:10:37.029 --> 00:10:39.549 The top catastrophic risks they identified are 00:10:39.549 --> 00:10:42.370 incredibly specific and deeply concerning. They 00:10:42.370 --> 00:10:44.970 deeply worry about dangerous autonomous capabilities 00:10:44.970 --> 00:10:48.129 emerging in these frontier models. They highlight 00:10:48.129 --> 00:10:51.610 severe geopolitical competitive dynamics aggressively 00:10:51.610 --> 00:10:54.919 accelerating the global arms race. And they specifically 00:10:54.919 --> 00:10:58.539 warn about highly weaponized cyber attack capabilities 00:10:58.539 --> 00:11:01.620 rapidly expanding globally. Which links perfectly 00:11:01.620 --> 00:11:03.860 back to that recent Microsoft GitHub malware 00:11:03.860 --> 00:11:06.419 hack. The software vulnerabilities are already 00:11:06.419 --> 00:11:09.240 being actively exploited in the wild today. But 00:11:09.240 --> 00:11:12.980 the MIT study also revealed a massive, terrifying 00:11:12.980 --> 00:11:16.240 societal vulnerability gap. The people most deeply 00:11:16.240 --> 00:11:18.899 vulnerable to these catastrophic failures are 00:11:18.899 --> 00:11:21.500 everyday users. The general public absorbs the 00:11:21.500 --> 00:11:23.659 massive financial damage when vital... systems 00:11:23.659 --> 00:11:26.309 break. But the general purpose AI developers 00:11:26.309 --> 00:11:28.470 are the ones structurally responsible today. 00:11:28.649 --> 00:11:31.090 Government regulators also hold massive inherent 00:11:31.090 --> 00:11:33.309 responsibility for preventing these specific 00:11:33.309 --> 00:11:35.970 catastrophes. The critical crosshairs are currently 00:11:35.970 --> 00:11:39.009 aimed at information, finance and national security. 00:11:39.370 --> 00:11:42.149 Think about autonomous AI getting deeply embedded 00:11:42.149 --> 00:11:44.450 in critical financial decision making systems. 00:11:44.690 --> 00:11:47.149 If these AI tools get embedded in the financial 00:11:47.149 --> 00:11:49.870 sector and fail, who is actually left holding 00:11:49.870 --> 00:11:52.409 the bag? Well, the MIT study clearly shows the 00:11:52.409 --> 00:11:54.789 general public takes the massive hit. Wait, really? 00:11:54.950 --> 00:11:57.750 The public takes the hit? Yeah. While developers 00:11:57.750 --> 00:12:00.350 and regulators are responsible, they lack financial 00:12:00.350 --> 00:12:03.990 incentives to slow down. So every day people 00:12:03.990 --> 00:12:06.899 suffer. But tech developers and government regulators 00:12:06.899 --> 00:12:10.539 hold the actual responsibility. That is the tragic, 00:12:10.720 --> 00:12:14.100 undeniable reality of this massive, glaring accountability 00:12:14.100 --> 00:12:17.059 gap. The underlying structural dilemma here is 00:12:17.059 --> 00:12:19.919 incredibly difficult to solve quickly. Any tech 00:12:19.919 --> 00:12:22.100 company that deliberately slows down for safety 00:12:22.100 --> 00:12:25.100 takes a massive hit. They immediately lose massive 00:12:25.100 --> 00:12:27.759 competitive ground in a violently red hot market. 00:12:28.039 --> 00:12:30.659 Absent strict external government rules, slowing 00:12:30.659 --> 00:12:33.120 down is essentially financial suicide today. 00:12:33.529 --> 00:12:35.690 Developers have deep structural and financial 00:12:35.690 --> 00:12:38.629 reasons not to prioritize vital risk mitigation. 00:12:38.990 --> 00:12:41.590 This perfectly explains Anthropic's highly specific 00:12:41.590 --> 00:12:44.210 approach to the recent Fable release. Ampropic 00:12:44.210 --> 00:12:46.870 purposefully put a massive safety guardrail system 00:12:46.870 --> 00:12:49.789 on Claude Fable 5. They have loudly warned everyone 00:12:49.789 --> 00:12:52.570 about severe recursive self -improvement dangers 00:12:52.570 --> 00:12:55.429 very recently. An AI system continuously making 00:12:55.429 --> 00:12:57.730 itself smarter without human help. So if you 00:12:57.730 --> 00:13:00.169 prompt Fable 5 with highly sensitive or dangerous 00:13:00.169 --> 00:13:03.250 topics. The model will actually actively hard 00:13:03.250 --> 00:13:06.129 block your specific request right away. It automatically 00:13:06.129 --> 00:13:08.129 routes your session straight back to the safer 00:13:08.129 --> 00:13:11.690 Opus model. To legally use Fable 5, you also 00:13:11.690 --> 00:13:15.070 face a mandatory data retention policy. You must 00:13:15.070 --> 00:13:18.250 formally agree to a strict 30 -day data retention 00:13:18.250 --> 00:13:22.269 window. Anthropic explicitly promises not to 00:13:22.269 --> 00:13:24.750 train future models on that retained user data. 00:13:24.990 --> 00:13:27.690 But they absolutely need the logs to actively 00:13:27.690 --> 00:13:31.049 monitor for dangerous emergent behavior. Meanwhile, 00:13:31.230 --> 00:13:33.230 the completely unrestricted, deeply powerful 00:13:33.230 --> 00:13:36.129 version is called Claude Mythos 5. But Anthropic 00:13:36.129 --> 00:13:38.629 intentionally locked that specific model away 00:13:38.629 --> 00:13:41.210 behind heavily closed doors. It is currently 00:13:41.210 --> 00:13:44.110 only available for highly approved, heavily vetted 00:13:44.110 --> 00:13:46.649 security partners now. They are actively taking 00:13:46.649 --> 00:13:49.350 a massive, deliberate financial hit to ensure 00:13:49.350 --> 00:13:51.590 global safety. They essentially built the fastest 00:13:51.590 --> 00:13:54.549 car but permanently locked the top gear. Because 00:13:54.549 --> 00:13:56.669 they clearly know the regulatory brakes on the 00:13:56.669 --> 00:13:59.159 broader industry are failing. It is a fascinating, 00:13:59.360 --> 00:14:02.340 rare, real -world example of prioritizing structural 00:14:02.340 --> 00:14:05.200 safety first. But they are largely acting completely 00:14:05.200 --> 00:14:08.000 alone in this highly specific, cautious approach. 00:14:08.340 --> 00:14:11.120 Most corporate competitors are just blindly racing 00:14:11.120 --> 00:14:14.279 toward maximum capability and profit. And Fropic 00:14:14.279 --> 00:14:16.740 has always actively branded themselves as the 00:14:16.740 --> 00:14:20.570 premier safety -first AI lab. They famously split 00:14:20.570 --> 00:14:23.889 off from OpenAI years ago over fundamental core 00:14:23.889 --> 00:14:26.909 safety disagreements. So releasing a highly capable 00:14:26.909 --> 00:14:30.289 autonomous model like Fable 5 is deeply risky. 00:14:30.450 --> 00:14:32.509 They truly had to thread a nearly impossible 00:14:32.509 --> 00:14:35.570 corporate and ethical needle here. The mandatory 00:14:35.570 --> 00:14:38.230 30 -day enterprise data retention policy is a 00:14:38.230 --> 00:14:40.970 truly perfect example. Most massive enterprise 00:14:40.970 --> 00:14:43.509 users absolutely hate corporate data retention 00:14:43.509 --> 00:14:45.809 policies today. They desperately want complete 00:14:45.809 --> 00:14:48.490 privacy and zero data logging for internal... 00:14:48.590 --> 00:14:50.690 corporate security. But Anthropic stubbornly 00:14:50.690 --> 00:14:53.450 insists on keeping those specific user logs for 00:14:53.450 --> 00:14:56.850 30 days. They need to actively monitor for extremely 00:14:56.850 --> 00:14:59.429 dangerous emergent capabilities developing online. 00:14:59.879 --> 00:15:01.600 They are specifically looking for terrifying 00:15:01.600 --> 00:15:03.960 signs of biological weapons research queries 00:15:03.960 --> 00:15:06.480 happening. Or highly automated, deeply destructive 00:15:06.480 --> 00:15:09.100 cyber warfare scripts being actively generated 00:15:09.100 --> 00:15:11.559 by their models. Think about the sheer terrifying 00:15:11.559 --> 00:15:15.340 scale of $100 billion in damages. That isn't 00:15:15.340 --> 00:15:18.220 just a tiny, easily ignored blip in the global 00:15:18.220 --> 00:15:20.899 financial market. That represents massive global 00:15:20.899 --> 00:15:24.399 job losses and entirely decimated corporate pension 00:15:24.399 --> 00:15:27.340 funds. And a 10 % chance is practically playing 00:15:27.340 --> 00:15:29.279 Russian roulette with the economy. You would 00:15:29.279 --> 00:15:31.700 never voluntarily board an airplane with a 10 00:15:31.700 --> 00:15:34.039 % crash rate. But we are currently strapping 00:15:34.039 --> 00:15:37.299 our entire global economy to this exact rocket. 00:15:37.500 --> 00:15:39.500 And the ambitious people building the massive 00:15:39.500 --> 00:15:42.639 rocket currently have no real brakes. Two secs 00:15:42.639 --> 00:15:45.059 silence. If we logically pull back and look at 00:15:45.059 --> 00:15:47.340 the broader big picture today. We are living 00:15:47.340 --> 00:15:50.840 in a deeply profound jarring split screen reality 00:15:50.840 --> 00:15:54.159 right now. On one side we have. deeply magical, 00:15:54.440 --> 00:15:57.299 fully autonomous software coding agents. They 00:15:57.299 --> 00:16:00.019 are completely rebuilding complex web applications 00:16:00.019 --> 00:16:02.779 entirely from visual scratch today. We have a 00:16:02.779 --> 00:16:06.100 violently red -hot mangoes economy that is basically 00:16:06.100 --> 00:16:08.529 printing infinite money. But on the other darker 00:16:08.529 --> 00:16:11.389 side, we face a deeply terrifying structural 00:16:11.389 --> 00:16:14.350 gap. We have a massive, currently unaddressed 00:16:14.350 --> 00:16:16.929 structural safety and global accountability gap. 00:16:17.029 --> 00:16:19.470 Actively slowing down to implement crucial safety 00:16:19.470 --> 00:16:22.149 guardrails takes immense corporate courage today. 00:16:22.470 --> 00:16:25.350 Anthropic, courageously restricted Mythos 5, 00:16:25.590 --> 00:16:28.830 but willingly took a massive financial hit. This 00:16:28.830 --> 00:16:31.269 dangerous competitive dynamic leaves our most 00:16:31.269 --> 00:16:34.250 critical financial and national security sectors 00:16:34.250 --> 00:16:37.370 vulnerable. Completely vulnerable to massive, 00:16:37.409 --> 00:16:40.269 highly catastrophic global risks by the year 00:16:40.269 --> 00:16:43.169 2030. The fundamental corporate incentive structures 00:16:43.169 --> 00:16:46.250 are completely misaligned with long -term human 00:16:46.250 --> 00:16:48.909 survival today. Which ultimately leaves you with 00:16:48.909 --> 00:16:51.490 a deeply unsettling, highly provocative question 00:16:51.490 --> 00:16:54.330 to ponder. Taking a severe competitive financial 00:16:54.330 --> 00:16:56.970 hit is currently the only reward for safety. 00:16:57.169 --> 00:16:59.769 And external federal government regulations are 00:16:59.769 --> 00:17:02.190 historically moving far too slowly today. Will 00:17:02.190 --> 00:17:04.829 it necessarily take the first $100 billion global 00:17:04.829 --> 00:17:07.930 financial market crash? Or perhaps a massive, 00:17:07.970 --> 00:17:10.630 deeply weaponized global political disinformation 00:17:10.630 --> 00:17:13.710 event soon? To finally force the entire global 00:17:13.710 --> 00:17:16.190 AI industry's brake pedal straight to the floor. 00:17:16.410 --> 00:17:19.029 It is a truly terrifying, sobering thought to 00:17:19.029 --> 00:17:21.539 leave hanging in the air. Thank you so much for 00:17:21.539 --> 00:17:23.960 taking this fascinating deep dive with us today. 00:17:24.119 --> 00:17:26.400 We always deeply appreciate you thoughtfully 00:17:26.400 --> 00:17:28.900 exploring these complex technical topics with 00:17:28.900 --> 00:17:32.440 us. Stay intensely curious, and we will definitely 00:17:32.440 --> 00:17:33.700 see you on the next one.