WEBVTT 00:00:00.000 --> 00:00:03.319 Okay, imagine this. An AI that designs completely 00:00:03.319 --> 00:00:07.160 new cancer drugs. Not just improving old ones, 00:00:07.259 --> 00:00:10.380 but creating molecules out of, well, nothing. 00:00:10.580 --> 00:00:15.419 No prior data, no guides. Just pure generative 00:00:15.419 --> 00:00:17.879 chemistry. How is that even possible? And, you 00:00:17.879 --> 00:00:19.969 know, what does it mean for... where medicine 00:00:19.969 --> 00:00:22.690 is headed. Welcome to the Deep Dive. Today we're 00:00:22.690 --> 00:00:24.690 going to unpack a really compelling collection 00:00:24.690 --> 00:00:28.030 of insights. It's all from a recent newsletter, 00:00:28.149 --> 00:00:30.449 really charting the sharp end of AI development. 00:00:30.910 --> 00:00:32.890 There's a lot to get into. Oh, absolutely. We're 00:00:32.890 --> 00:00:35.210 talking about AI skeptics, longtime skeptics 00:00:35.210 --> 00:00:37.829 suddenly cutting their AGI timelines in half. 00:00:37.950 --> 00:00:40.009 We've got these incredible medical breakthroughs 00:00:40.009 --> 00:00:42.090 that sound like pure sci -fi. And yeah, even 00:00:42.090 --> 00:00:44.990 some pretty wild AI social media arguments. Right. 00:00:45.049 --> 00:00:47.170 So our mission today is to distill all these 00:00:47.170 --> 00:00:48.869 different threads, give you the key takeaways. 00:00:48.969 --> 00:00:51.350 Think of it as maybe a shortcut to understanding 00:00:51.350 --> 00:00:53.869 the big shifts happening in AI right now. We'll 00:00:53.869 --> 00:00:55.250 try and connect the dots for you. Yeah, let's 00:00:55.250 --> 00:00:57.170 unpack it. So we should probably start with what 00:00:57.170 --> 00:00:59.229 feels like a really significant shift in thinking. 00:00:59.409 --> 00:01:01.810 Francois Chalet, you know, the creator of Keras, 00:01:01.890 --> 00:01:04.709 that super popular AI library. He's been a pretty 00:01:04.709 --> 00:01:08.370 vocal AGI skeptic for years. Very measured. Exactly. 00:01:08.549 --> 00:01:12.200 And he just halved his AGI timeline. Slashed 00:01:12.200 --> 00:01:15.079 it from 10 years down to just five. That's that's 00:01:15.079 --> 00:01:17.420 a major change for him. It's huge. And importantly, 00:01:17.540 --> 00:01:19.719 he's saying it's not just because models are 00:01:19.719 --> 00:01:22.819 getting bigger or faster. It's that AI is starting 00:01:22.819 --> 00:01:24.760 to get smarter in a way that actually matters. 00:01:24.840 --> 00:01:27.459 Yeah. Moving beyond just being like he called 00:01:27.459 --> 00:01:30.140 them glorified parrots. He's seeing something 00:01:30.140 --> 00:01:31.620 else emerge. And this is where it gets really 00:01:31.620 --> 00:01:34.579 interesting, I think. Chile's pointing to signs 00:01:34.579 --> 00:01:37.079 of what he calls fluid intelligence in these 00:01:37.079 --> 00:01:39.680 systems. Okay, fluid intelligence. Yeah, so it's 00:01:39.680 --> 00:01:41.680 not just pattern matching or remembering stuff. 00:01:41.780 --> 00:01:44.879 It's the AI's ability to adapt to totally near 00:01:44.879 --> 00:01:48.840 unexpected problems on the fly without needing 00:01:48.840 --> 00:01:51.959 examples or specific training for that situation. 00:01:52.260 --> 00:01:54.719 It's more like spontaneous problem solving. Got 00:01:54.719 --> 00:01:58.000 it. Like imagine you drop an AI into a simple 00:01:58.000 --> 00:02:00.439 video game it's never seen. Zero instructions, 00:02:00.579 --> 00:02:03.420 no labels, just the game. It has to explore, 00:02:03.659 --> 00:02:06.500 poke around, figure out the rules just by trying 00:02:06.500 --> 00:02:09.240 things, and then learn how to win. That's basically 00:02:09.240 --> 00:02:13.599 what his new benchmark, ARC -AGI3, is testing. 00:02:14.020 --> 00:02:16.740 It's all about how efficiently it acquires new 00:02:16.740 --> 00:02:19.199 skills, that raw learning speed. And humans, 00:02:19.340 --> 00:02:21.539 well, we can usually figure out those kinds of 00:02:21.539 --> 00:02:23.400 simple novel games in less than a minute, right? 00:02:23.460 --> 00:02:26.360 We just kind of get it. Current AI models, still 00:02:26.360 --> 00:02:28.039 really struggling there. Yeah, and they're not 00:02:28.039 --> 00:02:30.189 there yet. Chollet's standard is pretty clear. 00:02:30.370 --> 00:02:32.210 As long as we can come up with problems humans 00:02:32.210 --> 00:02:35.930 can solve, but AI can't, we don't have AGI. It's 00:02:35.930 --> 00:02:38.090 a high bar, sure, but it points to that crucial 00:02:38.090 --> 00:02:40.750 gap. But here's an idea that Dworkesh, another 00:02:40.750 --> 00:02:43.909 AI commentator, really focused on as a potential 00:02:43.909 --> 00:02:47.490 game changer. Collective intelligence. He brought 00:02:47.490 --> 00:02:49.490 up this idea that instead of every AI starting 00:02:49.490 --> 00:02:52.430 fresh, like a, quote, perpetual intern on day 00:02:52.430 --> 00:02:55.909 one, any skill learned by one AI could instantly 00:02:55.909 --> 00:02:58.629 become available to all the other AIs, like instantly 00:02:58.629 --> 00:03:00.550 shared. OK, so connecting that to the bigger 00:03:00.550 --> 00:03:02.810 picture, that implies AI could compound knowledge 00:03:02.810 --> 00:03:05.770 almost exponentially. Learning not as individuals, 00:03:05.810 --> 00:03:09.750 but as one giant connected mind, a hive mind. 00:03:10.009 --> 00:03:13.590 Pretty much. Dworkish put it really simply. That 00:03:13.590 --> 00:03:15.860 would basically be the singularity. It suggests 00:03:15.860 --> 00:03:18.520 this incredibly fast acceleration where knowledge 00:03:18.520 --> 00:03:21.460 just builds and builds across the whole network 00:03:21.460 --> 00:03:25.219 in real time. So what is it about this collective 00:03:25.219 --> 00:03:28.620 learning idea that makes it so potentially transformative 00:03:28.620 --> 00:03:32.560 for AI's progress? Well, it means AI could share 00:03:32.560 --> 00:03:35.800 insights instantly, massively accelerating knowledge 00:03:35.800 --> 00:03:38.439 gains across all systems. Okay, moving on a bit, 00:03:38.539 --> 00:03:40.539 we've seen some other interesting updates and 00:03:40.539 --> 00:03:43.139 notable moments recently that kind of paint a 00:03:43.139 --> 00:03:44.580 picture of the current scene. There was this 00:03:44.580 --> 00:03:47.400 clever chat GPT prompt that apparently went viral. 00:03:47.520 --> 00:03:49.759 Yeah, saw that. It's basically a smart way to 00:03:49.759 --> 00:03:51.860 structure your request to get the AI to think 00:03:51.860 --> 00:03:53.939 more deeply, giving you better, more relevant 00:03:53.939 --> 00:03:56.360 answers. It shows users are getting savvier, 00:03:56.460 --> 00:03:59.479 you know? Definitely. We're also seeing new GPT 00:03:59.479 --> 00:04:02.560 -5 model options pop up, names like auto, fast, 00:04:02.719 --> 00:04:05.099 and thinking. Each seems tuned for different 00:04:05.099 --> 00:04:07.849 kinds of tasks or speeds. Right, and GPT -4 .0 00:04:07.849 --> 00:04:11.069 is back for everyone, but GPT -4 .5, that slightly 00:04:11.069 --> 00:04:13.590 more advanced one, is now only for pro users. 00:04:14.389 --> 00:04:16.649 The landscape keeps shifting, doesn't it? Companies 00:04:16.649 --> 00:04:18.550 trying to figure out how to offer different tiers. 00:04:18.910 --> 00:04:21.990 And with these new models come rate limits. You 00:04:21.990 --> 00:04:25.449 get like 3 ,000 GPT -5 thinking queries a week, 00:04:25.610 --> 00:04:28.170 which sounds like a lot, but maybe not for heavy 00:04:28.170 --> 00:04:31.250 users. But then you look at CloudSonic 4, its 00:04:31.250 --> 00:04:34.649 context window is now massive. One million tokens. 00:04:34.910 --> 00:04:37.389 And just to clarify for listeners, a token is 00:04:37.389 --> 00:04:39.610 basically like a word or even just part of a 00:04:39.610 --> 00:04:42.569 word that the AI processes. So a million tokens 00:04:42.569 --> 00:04:45.329 means Claude can read and understand incredibly 00:04:45.329 --> 00:04:48.670 long texts, think a whole novel or huge code 00:04:48.670 --> 00:04:51.730 bases all at once. Which is amazing. But it comes 00:04:51.730 --> 00:04:53.649 with higher fees. And right now it's only for 00:04:53.649 --> 00:04:55.269 their top tier users, the ones spending over 00:04:55.269 --> 00:04:57.889 $400. So that power isn't cheap or universally 00:04:57.889 --> 00:04:59.889 available just yet. You know, beyond the tech 00:04:59.889 --> 00:05:01.910 specs, what's also kind of fascinating is the, 00:05:01.949 --> 00:05:04.490 well, the social drama between AIs. We saw the 00:05:04.490 --> 00:05:07.610 Grok bot on X basically calling Elon Musk a hypocrite. 00:05:07.689 --> 00:05:10.769 Oh, yeah. Didn't see that one. And then ChatGPT's 00:05:10.769 --> 00:05:13.269 own social account jumped in and started getting 00:05:13.269 --> 00:05:16.009 kind of snarky with Grok. It was a public back 00:05:16.009 --> 00:05:18.490 and forth. It's weirdly human, almost like watching 00:05:18.490 --> 00:05:21.800 digital rivals emerge. Ah, wow. It's definitely 00:05:21.800 --> 00:05:23.379 a reminder that these things can be unpredictable, 00:05:23.620 --> 00:05:26.740 right? And speaking of unpredictable and warnings, 00:05:27.240 --> 00:05:31.100 there was that really serious incident. A man 00:05:31.100 --> 00:05:33.680 ended up hospitalized after following diet advice 00:05:33.680 --> 00:05:36.339 from chat GPT. Yeah, that's concerning. It really 00:05:36.339 --> 00:05:39.139 hammers home the point. You absolutely must verify 00:05:39.139 --> 00:05:41.959 AI outputs, especially, you know, when it involves 00:05:41.959 --> 00:05:43.879 your health or finances or any big decision. 00:05:44.040 --> 00:05:46.860 It's a powerful tool, but it's not an expert 00:05:46.860 --> 00:05:49.120 you can blindly trust. Yeah, I mean, I still 00:05:49.120 --> 00:05:51.139 find myself having to consciously double -check 00:05:51.139 --> 00:05:53.100 things sometimes. It's just so easy to trust 00:05:53.100 --> 00:05:54.920 that confident -sounding answer, especially when 00:05:54.920 --> 00:05:57.180 you're in a hurry. But yeah, that story is a 00:05:57.180 --> 00:05:59.860 stark reminder. Vigilance is key. And on the 00:05:59.860 --> 00:06:03.139 money side of things, the AI chip startup Revos, 00:06:03.240 --> 00:06:06.139 already valued over $2 billion, is apparently 00:06:06.139 --> 00:06:08.639 looking for a big chunk of new funding, like 00:06:08.639 --> 00:06:11.120 $400 -500 million. To compete with NVIDIA, right? 00:06:11.629 --> 00:06:13.529 Exactly. They've already got big names like Intel 00:06:13.529 --> 00:06:16.149 Capital and MediaTek backing them. It shows the 00:06:16.149 --> 00:06:18.230 chip race is seriously heating up. Everyone wants 00:06:18.230 --> 00:06:20.470 a piece of that action because compute power 00:06:20.470 --> 00:06:23.569 is, well, it's the foundation for all this AI 00:06:23.569 --> 00:06:25.610 progress. So thinking about that diet advice 00:06:25.610 --> 00:06:29.009 incident, why is that specific case so crucial 00:06:29.009 --> 00:06:31.029 for us to remember when we interact with AI? 00:06:31.310 --> 00:06:33.970 It just really highlights the need for human 00:06:33.970 --> 00:06:36.930 oversight and checking the facts, especially 00:06:36.930 --> 00:06:39.310 for anything important. Okay, let's touch briefly 00:06:39.310 --> 00:06:41.649 on the sheer amount of money flowing into AI 00:06:41.649 --> 00:06:44.750 right now. The AI grant report for early August 00:06:44.750 --> 00:06:48.430 showed some absolutely huge funding rounds. Investment 00:06:48.430 --> 00:06:50.990 is just pouring in. Seriously, pouring in. OpenAI, 00:06:51.209 --> 00:06:54.529 for instance, secured a massive $8 .3 billion 00:06:54.529 --> 00:06:57.310 in funding. That's staggering. It's not just 00:06:57.310 --> 00:06:59.670 investment. It's a huge signal of belief in their 00:06:59.670 --> 00:07:02.110 long -term AGI goals. Gives them a lot of runway 00:07:02.110 --> 00:07:04.089 to pursue those goals without worrying too much 00:07:04.089 --> 00:07:06.310 about immediate profits. For sure. And then their 00:07:06.310 --> 00:07:11.120 open source rival, Mr. Which is also significant. 00:07:11.279 --> 00:07:13.560 It shows there's still huge appetite and capital 00:07:13.560 --> 00:07:15.779 for different approaches, even with a dominant 00:07:15.779 --> 00:07:18.639 player like OpenAI out there. It suggests a healthy, 00:07:18.720 --> 00:07:21.620 competitive ecosystem. Yeah, these are enormous 00:07:21.620 --> 00:07:25.079 numbers. They really show deep, continuing confidence 00:07:25.079 --> 00:07:28.079 in the whole AI field. The financial world is 00:07:28.079 --> 00:07:30.959 betting big that AI is going to reshape basically 00:07:30.959 --> 00:07:35.120 everything. This kind of cash fuels R &D incredibly 00:07:35.120 --> 00:07:37.839 fast. What does this level of investment really 00:07:37.839 --> 00:07:40.800 signal about where AI is headed? It signals a 00:07:40.800 --> 00:07:43.879 very strong belief in AI's continued growth with 00:07:43.879 --> 00:07:46.139 the major players pushing hard. All right, let's 00:07:46.139 --> 00:07:48.759 do a quick round of other notable AI bits and 00:07:48.759 --> 00:07:50.839 pieces that popped up. Google's Gemini added 00:07:50.839 --> 00:07:53.319 some nice features. Temporary chats so they don't 00:07:53.319 --> 00:07:55.980 save to your history. Good for privacy. And more 00:07:55.980 --> 00:07:58.000 personalization options. Yeah, letting it adapt 00:07:58.000 --> 00:08:00.060 more to you over time, user experience stuff. 00:08:00.620 --> 00:08:03.319 Always good. And there was a change at Elon Musk's 00:08:03.319 --> 00:08:05.839 XAI. One of the co -founders, Igor Babushkin, 00:08:05.920 --> 00:08:08.680 left the company. Oh, interesting. Key people 00:08:08.680 --> 00:08:11.139 moving around in these AI startups often mean, 00:08:11.240 --> 00:08:14.220 well, something's shifting internally. It's such 00:08:14.220 --> 00:08:16.759 a competitive space for talent. OpenAI also put 00:08:16.759 --> 00:08:19.439 out a clarification. GBT -5 thinking's context 00:08:19.439 --> 00:08:22.379 window isn't 32 ,000 tokens like some thought. 00:08:22.660 --> 00:08:26.399 It's actually 196 ,000 tokens. Oh, that's way 00:08:26.399 --> 00:08:29.009 bigger. Like more than five times bigger. Yeah, 00:08:29.069 --> 00:08:31.589 a huge difference. And again, that context window 00:08:31.589 --> 00:08:34.230 is how much info the AI can hold in mind for 00:08:34.230 --> 00:08:38.250 one conversation or task. So 196K means it can 00:08:38.250 --> 00:08:41.370 handle much longer, more complex stuff, documents, 00:08:41.570 --> 00:08:44.370 deep chats, analyzing code. It's a big step up 00:08:44.370 --> 00:08:46.830 for professional use. And Skywork released something 00:08:46.830 --> 00:08:49.669 called Matrix Game 2 .0. They're describing it 00:08:49.669 --> 00:08:52.309 as being like Genie 3, but for interactive video. 00:08:52.490 --> 00:08:55.429 Okay, so that suggests maybe AI generating interactive 00:08:55.429 --> 00:08:57.620 video content. Like experiences that change based 00:08:57.620 --> 00:08:59.399 on what you do. Sounds like it. Could be cool 00:08:59.399 --> 00:09:02.220 for gaming or education maybe. Creating more 00:09:02.220 --> 00:09:03.960 dynamic, personalized stuff. Definitely one to 00:09:03.960 --> 00:09:06.580 watch. And finally, Google is putting a massive 00:09:06.580 --> 00:09:09.740 $9 billion into expanding its AI infrastructure 00:09:09.740 --> 00:09:12.379 down in Oklahoma. $9 billion. Yeah. That's a 00:09:12.379 --> 00:09:14.600 huge investment in just the physical hardware 00:09:14.600 --> 00:09:17.379 needed for AI. Shows their long -term commitment, 00:09:17.500 --> 00:09:19.700 but also just the immense scale and, frankly, 00:09:19.759 --> 00:09:22.580 energy required for these advanced models. So 00:09:22.580 --> 00:09:25.309 back to that context window point. What's the 00:09:25.309 --> 00:09:27.769 real implication of having a much larger one 00:09:27.769 --> 00:09:30.529 for models like GPT -5 thinking? It basically 00:09:30.529 --> 00:09:34.210 means the AI can handle much longer, more complex 00:09:34.210 --> 00:09:37.070 conversations and process larger documents effectively. 00:09:37.470 --> 00:09:39.610 Okay, now for something that really stood out, 00:09:39.690 --> 00:09:43.830 genuinely groundbreaking stuff. Korean scientists 00:09:43.830 --> 00:09:47.919 have unveiled a new AI model. One that designs 00:09:47.919 --> 00:09:51.159 cancer drugs entirely from scratch. Wow. From 00:09:51.159 --> 00:09:53.480 scratch. From zero. No starting data needed. 00:09:53.720 --> 00:09:55.720 Okay. Tell me more. This is BIND, right? It's 00:09:55.720 --> 00:09:58.639 a diffusion model. Exactly. BIND. And yeah, it's 00:09:58.639 --> 00:10:00.399 a diffusion model. Just quickly, a diffusion 00:10:00.399 --> 00:10:02.700 model is kind of like an AI artist starting with 00:10:02.700 --> 00:10:05.039 random noise and gradually refining it into a 00:10:05.039 --> 00:10:07.480 coherent image. Except here, it's refining noise 00:10:07.480 --> 00:10:09.899 into potential drug molecules. Okay. But the 00:10:09.899 --> 00:10:11.759 key thing is, it's not just making drug discovery 00:10:11.759 --> 00:10:14.259 faster. It seems to be fundamentally rewriting 00:10:14.259 --> 00:10:16.620 the whole process. How so? Because most... AI 00:10:16.620 --> 00:10:18.960 drug discovery now, it starts with known drugs, 00:10:19.039 --> 00:10:21.200 right? And tries to improve them. Precisely. 00:10:21.460 --> 00:10:24.500 Most current models are like optimizers. They 00:10:24.500 --> 00:10:27.419 take existing molecules, tweak them a bit, predict 00:10:27.419 --> 00:10:30.240 if the tweak is good, test it virtually, and 00:10:30.240 --> 00:10:33.100 repeat. It's iterative, building on what we already 00:10:33.100 --> 00:10:36.559 know. But Bindy just... skips that apparently 00:10:36.559 --> 00:10:39.379 yeah it generates completely new molecules no 00:10:39.379 --> 00:10:42.559 templates no examples needed just raw generative 00:10:42.559 --> 00:10:45.419 chemistry and it designs the molecule and how 00:10:45.419 --> 00:10:48.820 it binds to its target all in one step whoa okay 00:10:48.820 --> 00:10:51.320 so it's like designing the key in the specific 00:10:51.320 --> 00:10:54.600 lock it fits perfectly matched at the exact same 00:10:54.600 --> 00:10:56.440 time instead of having a key and trying to find 00:10:56.440 --> 00:10:58.970 a lock That's a great analogy, yeah. And it gets 00:10:58.970 --> 00:11:01.029 better. It optimizes for multiple things at once, 00:11:01.090 --> 00:11:03.610 like is it effective? Is it safe? Will it dissolve 00:11:03.610 --> 00:11:05.990 properly? Stuff that's usually hard to balance. 00:11:06.169 --> 00:11:08.330 That's huge. And here's the really wild part, 00:11:08.409 --> 00:11:10.889 I think. When buying finds a design that works, 00:11:11.049 --> 00:11:13.750 it learns from that success. It recycles the 00:11:13.750 --> 00:11:16.110 successful strategies it used into future attempts, 00:11:16.289 --> 00:11:18.309 getting better and better over time. It's got 00:11:18.309 --> 00:11:21.080 this internal improvement loop. Whoa. Just imagine 00:11:21.080 --> 00:11:24.460 scaling that, running it a billion times, generating 00:11:24.460 --> 00:11:27.460 countless possibilities. And because it doesn't 00:11:27.460 --> 00:11:30.659 need existing data, that could open doors for 00:11:30.659 --> 00:11:33.500 treating really rare diseases or mutations we've 00:11:33.500 --> 00:11:35.659 never even seen before. Places where we just 00:11:35.659 --> 00:11:37.600 don't have the data for traditional methods. 00:11:38.139 --> 00:11:41.340 That's genuinely revolutionary. It feels like 00:11:41.340 --> 00:11:43.320 a completely different way of tackling drug discovery. 00:11:43.899 --> 00:11:46.820 So how does Bindy's approach fundamentally change 00:11:46.820 --> 00:11:49.460 the game compared to the older AI methods? Well, 00:11:49.519 --> 00:11:51.440 it creates entirely new drugs without needing 00:11:51.440 --> 00:11:54.100 prior examples, unlike older AIs that mostly 00:11:54.100 --> 00:11:56.759 just modify existing ones. Sponsor read provided 00:11:56.759 --> 00:11:58.710 separately. Okay, let's try and pull this all 00:11:58.710 --> 00:12:00.490 together. What does this deep dive really tell 00:12:00.490 --> 00:12:03.269 us? I think it shows AI isn't just getting incrementally 00:12:03.269 --> 00:12:05.509 better anymore. We're seeing fundamental shifts, 00:12:05.710 --> 00:12:07.789 real step changes. Yeah, definitely. We're seeing 00:12:07.789 --> 00:12:10.029 this move towards actual fluid intelligence, 00:12:10.070 --> 00:12:12.929 like Joel talked about, and this potential for 00:12:12.929 --> 00:12:15.230 exponential collective learning. That's a huge 00:12:15.230 --> 00:12:18.210 concept. It points to AI maybe learning and growing 00:12:18.210 --> 00:12:20.350 at a speed we can barely comprehend. And at the 00:12:20.350 --> 00:12:22.809 same time, we're seeing these incredibly transformative 00:12:22.809 --> 00:12:25.929 applications becoming real, like designing cancer 00:12:25.929 --> 00:12:28.450 drugs from absolute zero. That could change medicine 00:12:28.450 --> 00:12:31.129 forever. These aren't just lab experiments anymore. 00:12:31.210 --> 00:12:33.549 They're starting to happen. It's such a powerful 00:12:33.549 --> 00:12:35.990 reminder, isn't it? AI is developing incredibly 00:12:35.990 --> 00:12:39.450 fast. It brings just immense promise, but also 00:12:39.450 --> 00:12:42.509 this constant need for us to be vigilant, to 00:12:42.509 --> 00:12:44.289 understand what we're building and deploying. 00:12:44.629 --> 00:12:47.450 The excitement is real, but so is the responsibility 00:12:47.450 --> 00:12:49.570 that comes with it. So here's something to think 00:12:49.570 --> 00:12:52.370 about as we wrap up. A question for you to mull 00:12:52.370 --> 00:12:57.210 over. If AI can now learn collectively as a network 00:12:57.210 --> 00:12:59.909 and design truly novel solutions without needing 00:12:59.909 --> 00:13:03.190 any prior data, what kinds of human problems, 00:13:03.370 --> 00:13:05.370 problems we currently think are just impossible 00:13:05.370 --> 00:13:09.120 to solve, might suddenly become solvable? Yeah, 00:13:09.179 --> 00:13:11.019 that's definitely something to chew on. We really 00:13:11.019 --> 00:13:13.019 hope this deep dive helped you connect some of 00:13:13.019 --> 00:13:14.740 these important docs and feel a bit more informed 00:13:14.740 --> 00:13:17.340 about this incredibly fast -moving and honestly 00:13:17.340 --> 00:13:20.360 fascinating world of AI. Yeah, thanks for diving 00:13:20.360 --> 00:13:21.940 deep with us today. We'll catch you on the next 00:13:21.940 --> 00:13:23.620 one. Out Hero Music.