WEBVTT 00:00:00.000 --> 00:00:02.040 An ancient institution warns about the dangers 00:00:02.040 --> 00:00:05.500 of a tiny elite controlling the future. Beat. 00:00:05.900 --> 00:00:08.960 Meanwhile, a machine just cracked a 56 -year 00:00:08.960 --> 00:00:11.339 -old math problem for the price of a fancy dinner. 00:00:11.539 --> 00:00:13.599 Two completely different worlds colliding, you 00:00:13.599 --> 00:00:16.620 know? Let's unpack it. Welcome to today's Deep 00:00:16.620 --> 00:00:18.440 Dive. I'm really glad you are here with us. We 00:00:18.440 --> 00:00:21.820 have a truly fascinating journey ahead today. 00:00:21.980 --> 00:00:24.699 We absolutely do. First, we are going to explore 00:00:24.699 --> 00:00:28.050 a surprising new encyclical. It comes directly 00:00:28.050 --> 00:00:31.530 from Pope Leo XIV. We're looking at AI. power 00:00:31.530 --> 00:00:34.409 and the future of democracy. Right. Then we take 00:00:34.409 --> 00:00:37.250 a rapid -fire tour of today's major AI industry 00:00:37.250 --> 00:00:39.770 moves. We're talking spy agencies buying up chips 00:00:39.770 --> 00:00:42.369 and, you know, workers automating their own jobs. 00:00:42.549 --> 00:00:45.070 And finally, a mind -bending math showdown. And 00:00:45.070 --> 00:00:47.770 AI just proved it can generate entirely new knowledge. 00:00:47.969 --> 00:00:49.810 The three -line for today is the tension of control. 00:00:50.030 --> 00:00:52.409 Who exactly controls this frantic technological 00:00:52.409 --> 00:00:55.210 race? And what is the technology actually becoming 00:00:55.210 --> 00:00:58.049 capable of? It is a massive question. And, well, 00:00:58.090 --> 00:00:59.969 the Vatican just weighed in heavily. Yep. Start 00:00:59.969 --> 00:01:02.530 right there. Pope Leo IV just released a new 00:01:02.530 --> 00:01:05.549 encyclical. It is titled Magnifica Humanitas, 00:01:05.609 --> 00:01:08.409 and it is making serious waves. Oh, for sure. 00:01:08.530 --> 00:01:11.109 I mean, it uses AI as a primary hook, but it 00:01:11.109 --> 00:01:13.670 actually tackles much deeper issues. The rapid 00:01:13.670 --> 00:01:16.510 erosion of democracy, the rise of massive inequality. 00:01:16.609 --> 00:01:19.510 Yeah. Specifically, how tech elites are essentially 00:01:19.510 --> 00:01:22.650 steering the world to benefit themselves. The 00:01:22.650 --> 00:01:25.129 timing of this document is incredibly striking 00:01:25.129 --> 00:01:28.700 to me. The context here is super important. The 00:01:28.700 --> 00:01:31.459 encyclical dropped right after a major political 00:01:31.459 --> 00:01:34.079 shift in Washington. President Trump had just 00:01:34.079 --> 00:01:36.719 delayed a major AI executive order. Just looking 00:01:36.719 --> 00:01:38.799 at the timeline, that order was meant to give 00:01:38.799 --> 00:01:41.659 government oversight, oversight for new powerful 00:01:41.659 --> 00:01:45.040 AI models. Exactly. And billionaires are currently 00:01:45.040 --> 00:01:47.900 pumping hundreds of millions into super PACs. 00:01:47.939 --> 00:01:51.079 They are lobbying specifically to kill AI safety 00:01:51.079 --> 00:01:54.120 regulations. So the pope is reacting to a very 00:01:54.120 --> 00:01:56.159 specific moment in time. He argues something 00:01:56.159 --> 00:01:59.280 very specific here. Tech built by a tiny opaque 00:01:59.280 --> 00:02:02.280 group cannot serve the common good. AI naturally 00:02:02.280 --> 00:02:04.379 amplifies the power of those who already have 00:02:04.379 --> 00:02:06.540 money and data. And they use that data advantage 00:02:06.540 --> 00:02:08.819 very effectively. They use it to manipulate our 00:02:08.819 --> 00:02:10.860 choices. They manipulate what we buy every day 00:02:10.860 --> 00:02:13.360 and even how we vote. I have to admit something 00:02:13.360 --> 00:02:15.719 here. I still wrestle with knowing what's actually 00:02:15.719 --> 00:02:19.060 true online myself. These deep fakes, they really 00:02:19.060 --> 00:02:22.159 blur reality. You are definitely not alone in 00:02:22.159 --> 00:02:24.300 feeling that way. The Pope actually highlights 00:02:24.300 --> 00:02:27.169 that exact issue. Endless data deepfakes are 00:02:27.169 --> 00:02:30.150 destroying our collective truth. Yeah, we are 00:02:30.150 --> 00:02:33.389 losing our shared ability to agree on basic reality. 00:02:33.550 --> 00:02:36.930 Without a shared reality, democracy really struggles 00:02:36.930 --> 00:02:40.030 to function. So the Pope makes a very explicit 00:02:40.030 --> 00:02:43.030 demand here. He calls for an end to this frantic 00:02:43.030 --> 00:02:45.830 push. He specifically means the push for ever 00:02:45.830 --> 00:02:48.389 larger data sets. He is warning against the blind 00:02:48.389 --> 00:02:51.229 race for more powerful models. The document makes 00:02:51.229 --> 00:02:54.080 a very sharp philosophical point. Just because 00:02:54.080 --> 00:02:57.099 a small group can build insanely powerful tech. 00:02:57.219 --> 00:02:58.819 Right. And they definitely have the capital to 00:02:58.819 --> 00:03:00.439 do it. It doesn't give them the right to govern 00:03:00.439 --> 00:03:02.759 society. It's like building a high speed train, 00:03:02.939 --> 00:03:05.280 but letting the engineers decide what cities 00:03:05.280 --> 00:03:07.960 are allowed to exist. That is a perfect way to 00:03:07.960 --> 00:03:10.259 visualize it. The engineers shouldn't be the 00:03:10.259 --> 00:03:12.840 mayors. And what is fascinating here is the historical 00:03:12.840 --> 00:03:15.659 parallel. It mirrors history almost perfectly. 00:03:15.919 --> 00:03:17.780 You're talking about the Industrial Revolution, 00:03:17.879 --> 00:03:21.319 right? Yes. Back in 1891, the church issued a 00:03:21.319 --> 00:03:23.840 very similar warning. They called out the exact 00:03:23.840 --> 00:03:27.020 same type of exploitation. Back then, it was 00:03:27.020 --> 00:03:30.460 factory owners. Today, it is data centers. Do 00:03:30.460 --> 00:03:33.680 historical warnings like the 1891 decree ever 00:03:33.680 --> 00:03:36.280 actually slow down technological momentum, or 00:03:36.280 --> 00:03:38.800 do they just document it? They rarely stop the 00:03:38.800 --> 00:03:41.539 actual machinery of progress. You can't uninvent 00:03:41.539 --> 00:03:43.639 the steam engine, and you can't uninvent algorithms. 00:03:44.459 --> 00:03:46.919 But they do something equally important. They 00:03:46.919 --> 00:03:49.520 shift the cultural baseline of what society accepts. 00:03:49.840 --> 00:03:52.159 So moral warnings highlight the stakes, even 00:03:52.159 --> 00:03:54.659 if they don't hit the brakes. Exactly. They force 00:03:54.659 --> 00:03:56.719 us to look at the road ahead. They give lawmakers 00:03:56.719 --> 00:03:59.340 a moral vocabulary to eventually pass regulations. 00:03:59.740 --> 00:04:02.580 The Pope warned about this frantic push for dominance. 00:04:02.840 --> 00:04:05.000 Let's look exactly at what that frantic push 00:04:05.000 --> 00:04:06.919 looks like on the ground this week. The corporate 00:04:06.919 --> 00:04:09.039 shifts right now are just staggering. Let's start 00:04:09.039 --> 00:04:10.960 with a company called ClickUp. They just laid 00:04:10.960 --> 00:04:14.379 off 22 % of their workforce. Wait, 22%. Is that 00:04:14.379 --> 00:04:16.980 a direct result of the AI or is the CEO just 00:04:16.980 --> 00:04:19.579 using it as a convenient excuse? It seems directly 00:04:19.579 --> 00:04:22.540 tied to their internal tech adoption. They rolled 00:04:22.540 --> 00:04:25.680 out 3 ,000 internal AI agents just before the 00:04:25.680 --> 00:04:28.699 layoffs. The CEO claims the future belongs to 00:04:28.699 --> 00:04:31.500 employees who automate their own jobs. Automate 00:04:31.500 --> 00:04:34.439 your job or lose your job entirely. That is a 00:04:34.439 --> 00:04:37.540 brutal new reality for average workers. It is 00:04:37.540 --> 00:04:40.449 a ruthless efficiency engine at work. Meanwhile, 00:04:40.709 --> 00:04:43.230 the stakes are rising on a national level. The 00:04:43.230 --> 00:04:45.350 White House just approved a massive spending 00:04:45.350 --> 00:04:49.110 bill. They approved $9 billion for U .S. spy 00:04:49.110 --> 00:04:52.850 agencies, buying up advanced AI chips at an incredible 00:04:52.850 --> 00:04:55.829 rate. Officials are reportedly terrified of falling 00:04:55.829 --> 00:04:58.389 behind globally. They believe they need to deploy 00:04:58.389 --> 00:05:01.610 frontier AI systems as quickly as possible. Frontier 00:05:01.610 --> 00:05:04.050 AI systems, basically the most advanced models 00:05:04.050 --> 00:05:06.430 pushing the absolute limits. Yeah. The government 00:05:06.430 --> 00:05:08.470 sees hoarding these chips as a core security 00:05:08.470 --> 00:05:10.939 imperative now. At the same time, the broader 00:05:10.939 --> 00:05:13.459 development speed is dizzying. Look at Google's 00:05:13.459 --> 00:05:16.019 AI Studio. They let users build native Android 00:05:16.019 --> 00:05:17.980 apps. You can do it directly on their platform 00:05:17.980 --> 00:05:20.680 for free. And the volume of output is just insane. 00:05:20.839 --> 00:05:23.480 Over 250 ,000 apps were built in a single week. 00:05:23.680 --> 00:05:26.899 A quarter of a million apps in seven days. That 00:05:26.899 --> 00:05:29.519 changes the entire software landscape completely. 00:05:29.779 --> 00:05:32.980 It is like stacking Lego blocks of data. It really 00:05:32.980 --> 00:05:35.709 democratizes code. But it also floods the zone. 00:05:36.009 --> 00:05:38.290 And the models themselves are evolving incredibly 00:05:38.290 --> 00:05:41.610 fast, too. Leaks suggest the next Grok model 00:05:41.610 --> 00:05:44.750 is very close. They are calling it the 1 .5T 00:05:44.750 --> 00:05:47.750 V9 medium model. It might drop in just two to 00:05:47.750 --> 00:05:49.649 three weeks. Right. And the most interesting 00:05:49.649 --> 00:05:51.350 part of the leak involves the training data. 00:05:51.910 --> 00:05:54.490 Cursor data is heavily added into the mix. Let's 00:05:54.490 --> 00:05:56.769 pause on that. Why does cursor data matter so 00:05:56.769 --> 00:05:58.920 much? Well, Cursor is currently one of the most 00:05:58.920 --> 00:06:02.019 popular AI coding editors. Developers use it 00:06:02.019 --> 00:06:04.680 every day to write software. If Grok is craning 00:06:04.680 --> 00:06:07.360 on how developers interact with Cursor, its coding 00:06:07.360 --> 00:06:09.759 capabilities will skyrocket. I mean, it is learning 00:06:09.759 --> 00:06:12.660 how we code while we are coding. That is a wild 00:06:12.660 --> 00:06:15.000 feedback loop. We are also seeing practical daily 00:06:15.000 --> 00:06:17.540 advice emerge. Yes. The creator of Cloud Code 00:06:17.540 --> 00:06:20.779 shared a major tip for daily users. People are 00:06:20.779 --> 00:06:22.620 always looking for ways to get better results. 00:06:22.939 --> 00:06:24.860 What was the number one piece of advice? It is 00:06:24.860 --> 00:06:27.500 surprisingly simple. But it changes everything. 00:06:27.879 --> 00:06:31.220 You just explicitly tell the AI to review its 00:06:31.220 --> 00:06:33.899 own code for bugs before it shows you the output. 00:06:34.079 --> 00:06:36.480 Just asking it to double check its work. Yeah, 00:06:36.540 --> 00:06:39.180 that simple second pass catches a huge percentage 00:06:39.180 --> 00:06:41.660 of errors. These tools are becoming our everyday 00:06:41.660 --> 00:06:44.259 digital companions, but they still have some 00:06:44.259 --> 00:06:47.500 very weird, unsettling quirks. Like Google's 00:06:47.500 --> 00:06:50.279 new AI search feature right now, it has been 00:06:50.279 --> 00:06:52.439 acting a little bit strange lately. If you try 00:06:52.439 --> 00:06:55.800 searching for certain command words, words like 00:06:55.800 --> 00:06:59.500 ignore or stop. It breaks the illusion of a standard 00:06:59.500 --> 00:07:01.920 search engine, starts talking back to you directly. 00:07:02.079 --> 00:07:04.600 It responds just like a conversational assistant 00:07:04.600 --> 00:07:06.959 would. It reminds you there is an agent behind 00:07:06.959 --> 00:07:08.939 the search bar. Google says they are currently 00:07:08.939 --> 00:07:11.579 working on a fix for it. Meanwhile, hardware 00:07:11.579 --> 00:07:14.459 is shifting its focus entirely right now. Apple 00:07:14.459 --> 00:07:17.920 is working on rumored new AI glasses. But they 00:07:17.920 --> 00:07:19.620 are taking a very different approach from their 00:07:19.620 --> 00:07:22.500 competitors. They may skip flashy augmented reality 00:07:22.500 --> 00:07:25.040 screens completely. Right. They are focusing 00:07:25.040 --> 00:07:27.459 on a very lightweight design instead. Simple 00:07:27.459 --> 00:07:30.420 gesture controls for the user. And prioritizing 00:07:30.420 --> 00:07:33.339 a much smarter, context -aware Siri built right 00:07:33.339 --> 00:07:37.180 in. A Siri built for actual everyday use. How 00:07:37.180 --> 00:07:40.339 do we reconcile this? Spy agencies hoard AI chips 00:07:40.339 --> 00:07:43.360 for global dominance, while average workers must 00:07:43.360 --> 00:07:46.139 automate their desk jobs or get fired. It really 00:07:46.139 --> 00:07:49.240 highlights how AI is a dual use extreme right 00:07:49.240 --> 00:07:52.279 now. It is a powerful weapon of statecraft at 00:07:52.279 --> 00:07:55.019 the very top and a ruthless efficiency engine 00:07:55.019 --> 00:07:57.980 at the very bottom. It's a two front race. National 00:07:57.980 --> 00:08:00.379 security at the top. Personal automation at the 00:08:00.379 --> 00:08:03.079 bottom. Precisely. The squeeze is happening to 00:08:03.079 --> 00:08:05.180 society from both sides simultaneously. If our 00:08:05.180 --> 00:08:07.319 national security, our daily livelihoods depend 00:08:07.319 --> 00:08:09.899 on these frontier models, just how smart are 00:08:09.899 --> 00:08:12.120 they actually getting? Let's look at a massive 00:08:12.120 --> 00:08:13.959 recent breakthrough, a breakthrough that proves 00:08:13.959 --> 00:08:16.019 they aren't just memorizing data anymore. We 00:08:16.019 --> 00:08:17.879 will dive into that right after this short break. 00:08:18.019 --> 00:08:21.540 Mid -roll sponsor break. We are back. Let's talk 00:08:21.540 --> 00:08:23.480 about the math showdown. This is where things 00:08:23.480 --> 00:08:26.399 get truly futuristic. Google DeepMind is officially 00:08:26.399 --> 00:08:29.019 firing back at OpenAI. This all started when 00:08:29.019 --> 00:08:32.879 OpenAI made a very large public claim. They claim 00:08:32.879 --> 00:08:35.759 their system managed to disprove an 80 -year 00:08:35.759 --> 00:08:38.620 -old math conjecture. So Google DeepMind just 00:08:38.620 --> 00:08:41.059 answered with something undeniably huge. Their 00:08:41.059 --> 00:08:43.860 new system is called Alpha Proof Nexus. And Alpha 00:08:43.860 --> 00:08:46.759 Proof Nexus just solved nine open -air disk problems. 00:08:47.210 --> 00:08:49.309 For those who might not know, what exactly is 00:08:49.309 --> 00:08:52.070 an Erdos problem? Paul Erdos was a legendary 00:08:52.070 --> 00:08:55.509 eccentric mathematician. He left behind hundreds 00:08:55.509 --> 00:08:58.570 of unsolved problems in number theory and combinatorics. 00:08:58.750 --> 00:09:01.230 They are notoriously tricky puzzles that require 00:09:01.230 --> 00:09:03.649 deep logical leaps. An alpha -proof nexus just 00:09:03.649 --> 00:09:06.289 solved nine of them. Two of these specific problems 00:09:06.289 --> 00:09:08.750 have stumped human mathematicians for 56 years 00:09:08.750 --> 00:09:11.070 straight. It didn't stop there, either. It also 00:09:11.070 --> 00:09:13.649 recently solved 44 open conjectures from the 00:09:13.649 --> 00:09:16.679 online encyclopedia of integer sequences. The 00:09:16.679 --> 00:09:18.779 mechanism behind this breakthrough is truly fascinating 00:09:18.779 --> 00:09:21.860 to me. It relies on a very specific loop between 00:09:21.860 --> 00:09:24.759 two different systems. The AI system itself writes 00:09:24.759 --> 00:09:27.340 the actual mathematical proof. It generates the 00:09:27.340 --> 00:09:29.620 logical steps. Then a completely separate system 00:09:29.620 --> 00:09:32.840 called Lean steps in to check it. Lean is essentially 00:09:32.840 --> 00:09:35.600 a programming language designed to verify mathematical 00:09:35.600 --> 00:09:38.879 proofs. It checks the math for any potential 00:09:38.879 --> 00:09:42.360 logical errors in real time. Right. And here's 00:09:42.360 --> 00:09:45.659 the magic part. They repeat this exact loop continuously. 00:09:46.279 --> 00:09:49.740 The AI generates, Lean checks, the AI refines. 00:09:49.840 --> 00:09:51.899 They run it over and over again until the proof 00:09:51.899 --> 00:09:55.159 is completely flawless. Two secs silence. Whoa. 00:09:56.080 --> 00:09:59.159 Imagine an AI grading its own math homework with 00:09:59.159 --> 00:10:02.059 100 % certainty, cracking a 56 -year -old puzzle 00:10:02.059 --> 00:10:04.240 for a few hundred bucks. The cost efficiency 00:10:04.240 --> 00:10:06.980 is truly mind -blowing. Solving these historic 00:10:06.980 --> 00:10:09.500 math problems used to take human lifetimes. It 00:10:09.500 --> 00:10:11.600 took massive grants and years of intense focus. 00:10:12.090 --> 00:10:14.090 Now, each problem costs only a few hundred dollars 00:10:14.090 --> 00:10:16.769 in compute power. It completely flips the economics 00:10:16.769 --> 00:10:19.169 of scientific discovery. But it is important 00:10:19.169 --> 00:10:21.110 to note it does have a current hard limitation. 00:10:21.470 --> 00:10:23.750 Yes, it hits a wall on certain types of abstract 00:10:23.750 --> 00:10:27.049 problems. AlphaProof absolutely crushed all of 00:10:27.049 --> 00:10:29.509 these existing logical puzzles. But it currently 00:10:29.509 --> 00:10:31.429 struggles with inventing entirely new mathematical 00:10:31.429 --> 00:10:34.029 constructions from scratch. It optimizes existing 00:10:34.029 --> 00:10:36.330 frameworks beautifully, but it isn't dreaming 00:10:36.330 --> 00:10:38.950 up new branches of math. It's like a grandmaster 00:10:38.950 --> 00:10:41.370 who can play millions of chess games to find 00:10:41.370 --> 00:10:43.830 the perfect move, but can't invent a new board 00:10:43.830 --> 00:10:46.389 game. That is exactly the right way to view its 00:10:46.389 --> 00:10:48.929 current limits. It masters the established rules 00:10:48.929 --> 00:10:51.730 of the universe perfectly well, but it cannot 00:10:51.730 --> 00:10:54.830 invent brand new math concepts on its own just 00:10:54.830 --> 00:10:57.389 yet. Even with that limitation, the significance 00:10:57.389 --> 00:11:00.889 of this milestone is profound. For a very long 00:11:00.889 --> 00:11:03.129 time, skeptics have made a specific argument 00:11:03.129 --> 00:11:05.830 about AI. They said AI could only regurgitate 00:11:05.830 --> 00:11:08.139 its training data. They called large language 00:11:08.139 --> 00:11:11.620 models very fancy autocomplete machines. They 00:11:11.620 --> 00:11:13.980 argued the models were just guessing the next 00:11:13.980 --> 00:11:17.139 likely word based on statistics. But disproving 00:11:17.139 --> 00:11:19.279 and solving decades old math problems changes 00:11:19.279 --> 00:11:22.799 everything. It is the literal definition of generating 00:11:22.799 --> 00:11:25.220 brand new knowledge. It proves the system is 00:11:25.220 --> 00:11:27.990 capable of genuine reasoning. Google showed us 00:11:27.990 --> 00:11:30.909 a very clear working blueprint here. It is a 00:11:30.909 --> 00:11:33.149 blueprint for the future of all scientific discovery. 00:11:33.450 --> 00:11:35.710 Does this finally end the autocomplete debate 00:11:35.710 --> 00:11:38.289 entirely, whether AI is just an autocomplete 00:11:38.289 --> 00:11:41.250 or something actually capable of reasoning? It 00:11:41.250 --> 00:11:44.789 absolutely should end that debate for good. Solving 00:11:44.789 --> 00:11:47.490 a previously unsolved mathematical problem proves 00:11:47.490 --> 00:11:50.250 a critical point. The system is reaching far 00:11:50.250 --> 00:11:53.289 beyond its initial training data. It is tapping 00:11:53.289 --> 00:11:57.269 into genuine, verifiable logic. Right. It's generating 00:11:57.269 --> 00:11:59.769 genuine new knowledge, not just remixing its 00:11:59.769 --> 00:12:02.669 training data. Exactly. The ceiling of the capability 00:12:02.669 --> 00:12:05.580 is rising incredibly fast right now. Let's bring 00:12:05.580 --> 00:12:08.000 this whole journey full circle today. We started 00:12:08.000 --> 00:12:11.259 with deeply human moral concerns from the Vatican. 00:12:11.539 --> 00:12:14.440 The Pope warned about morality, political power 00:12:14.440 --> 00:12:17.039 and massive inequality. He was deeply worried 00:12:17.039 --> 00:12:19.419 about who exactly wields this new technology. 00:12:19.759 --> 00:12:22.059 He warned about the tiny elite controlling the 00:12:22.059 --> 00:12:24.259 trajectory of society. And we contrasted that 00:12:24.259 --> 00:12:27.399 human warning with cold, undeniable machine capability. 00:12:27.720 --> 00:12:30.220 Systems like Alpha Proof Nexus are rewriting 00:12:30.220 --> 00:12:32.740 human knowledge boundaries as we speak. The technology 00:12:32.740 --> 00:12:35.960 is undeniably crossing a major... It is moving 00:12:35.960 --> 00:12:38.799 from a tool of convenience to an engine of discovery. 00:12:39.080 --> 00:12:41.299 Which makes the question of control more critical 00:12:41.299 --> 00:12:43.799 than ever. Who steers this ship matters deeply 00:12:43.799 --> 00:12:46.399 to all of us. It really does. The technological 00:12:46.399 --> 00:12:49.100 race is simply moving much faster than our global 00:12:49.100 --> 00:12:51.360 regulations. I want to leave you with a lingering 00:12:51.360 --> 00:12:55.299 question today. If we can build an AI that flawlessly 00:12:55.299 --> 00:12:58.600 grades its own logic in complex math, how long 00:12:58.600 --> 00:13:00.720 until we ask it to logically grade the messy, 00:13:00.820 --> 00:13:03.200 unequal societal systems the Pope is so worried 00:13:03.200 --> 00:13:05.559 about? And would we even listen to its answer? 00:13:05.700 --> 00:13:08.480 Thank you for joining our deep dive today, Otiro 00:13:08.480 --> 00:13:08.679 Music.