WEBVTT 00:00:00.000 --> 00:00:05.379 OK, so imagine you could suddenly just like look 00:00:05.379 --> 00:00:07.860 inside an AI's brain, you know, really trace 00:00:07.860 --> 00:00:11.019 how it's thinking step by step. Or maybe you 00:00:11.019 --> 00:00:13.320 caught that news about AI predicting hurricanes 00:00:13.320 --> 00:00:17.719 like super accurately. Or there's another one 00:00:17.719 --> 00:00:19.120 that somehow can't even tell you what year it 00:00:19.120 --> 00:00:23.339 is. Yeah, that contrast is pretty, pretty stark, 00:00:23.440 --> 00:00:25.320 isn't it? Between the really groundbreaking stuff 00:00:25.320 --> 00:00:29.199 and well. Kind of baffling. Totally. And honestly, 00:00:29.300 --> 00:00:30.920 that pretty much captures what we're going to 00:00:30.920 --> 00:00:33.259 dig into today. Yeah. Welcome to your deep dives. 00:00:33.259 --> 00:00:35.119 We've got the whole stack of sources. You shared 00:00:35.119 --> 00:00:38.259 some research papers, news bits, notes on new 00:00:38.259 --> 00:00:41.119 tools popping up. And it's all about the latest, 00:00:41.259 --> 00:00:44.520 most interesting things happening in AI. It's 00:00:44.520 --> 00:00:46.560 a really great snapshot, actually. It shows where 00:00:46.560 --> 00:00:49.939 the field is making these huge leaps and maybe 00:00:49.939 --> 00:00:51.479 also where it's still kind of finding its feet. 00:00:51.679 --> 00:00:54.600 Yeah, exactly. So our mission here is just to 00:00:54.600 --> 00:00:57.899 unpack all of this with you, pull out the most 00:00:57.899 --> 00:01:00.299 important bits, the surprising stuff, maybe some 00:01:00.299 --> 00:01:04.379 real aha moments, and sort of give you the shortcut 00:01:04.379 --> 00:01:06.120 to understanding what's actually going on in 00:01:06.120 --> 00:01:09.700 this super fast -moving world. This is your custom 00:01:09.700 --> 00:01:12.840 deep dive into your material. Think of us as 00:01:12.840 --> 00:01:15.620 guides, maybe, helping you navigate the info 00:01:15.620 --> 00:01:18.019 you brought to the table. Love that. Yeah. And 00:01:18.019 --> 00:01:19.939 right at the heart of these sources, there's 00:01:19.939 --> 00:01:22.420 this one piece that feels like, you know, the 00:01:22.420 --> 00:01:25.840 main event. It's about anthropic and what they've 00:01:25.840 --> 00:01:28.140 just, well, they've open sourced something pretty 00:01:28.140 --> 00:01:30.180 big. A tool that people are describing as giving 00:01:30.180 --> 00:01:33.620 us X -ray vision into that AI black box everyone 00:01:33.620 --> 00:01:35.819 talks about. Which is huge. Yeah. Because for 00:01:35.819 --> 00:01:37.939 years, everyone's kind of said that box is basically 00:01:37.939 --> 00:01:40.799 sealed shut. So seeing a real move towards cracking 00:01:40.799 --> 00:01:43.890 it open. That's a big deal. It really is. We're 00:01:43.890 --> 00:01:45.769 definitely going to get into that. But your sources 00:01:45.769 --> 00:01:47.370 cover a bunch of other stuff too, right? It's 00:01:47.370 --> 00:01:49.510 not just the super technical breakthroughs. We've 00:01:49.510 --> 00:01:51.829 also got some of the more, let's say, quirky 00:01:51.829 --> 00:01:54.530 applications and just the general industry buzz 00:01:54.530 --> 00:01:56.530 you picked up on. Yeah. And what's fascinating, 00:01:56.750 --> 00:01:59.489 I think, is how these seemingly different pieces 00:01:59.489 --> 00:02:01.930 actually connect. They're not totally random. 00:02:01.969 --> 00:02:03.810 They kind of show these underlying trends, maybe 00:02:03.810 --> 00:02:06.810 some tensions, and really paint a picture of 00:02:06.810 --> 00:02:09.469 where AI is heading right now. Right. It's like 00:02:09.469 --> 00:02:12.490 different sides of the same, I don't know. rapidly 00:02:12.490 --> 00:02:14.750 changing crystal or something. Okay, let's start 00:02:14.750 --> 00:02:16.469 with that big one then, the anthropic noose. 00:02:16.530 --> 00:02:18.389 The source talks about this new tool they've 00:02:18.389 --> 00:02:20.129 open sourced and they call the main technique 00:02:20.129 --> 00:02:24.030 circuit tracing. Yes, circuit tracing. And they're 00:02:24.030 --> 00:02:26.129 using something they call attribution graphs 00:02:26.129 --> 00:02:29.210 to do it. This is basically their deep interpretability 00:02:29.210 --> 00:02:31.909 work, understanding how the model works. But 00:02:31.909 --> 00:02:34.110 now it's out there for others to use. Attribution 00:02:34.110 --> 00:02:36.250 graphs. Okay, so what does that actually show 00:02:36.250 --> 00:02:37.889 you? What are you looking at when you use it? 00:02:38.240 --> 00:02:41.159 Well, they're designed to show you the internal 00:02:41.159 --> 00:02:43.659 reasoning steps, right, that a language model 00:02:43.659 --> 00:02:46.120 takes. When you give it a prompt and it spits 00:02:46.120 --> 00:02:48.979 out an answer, these graphs visualize the path 00:02:48.979 --> 00:02:52.080 the information took, which bits of the model 00:02:52.080 --> 00:02:54.300 lit up, and how they actually contributed to 00:02:54.300 --> 00:02:58.020 that specific response. Okay. Wow. So it's not 00:02:58.020 --> 00:03:00.979 just seeing, like, input goes in, answer comes 00:03:00.979 --> 00:03:04.479 out. You're literally watching the AI sort of 00:03:04.479 --> 00:03:07.090 thing. think its way through in a very structured 00:03:07.090 --> 00:03:10.110 way yeah a graph based representation you can 00:03:10.110 --> 00:03:12.150 take a specific bit of the output and trace it 00:03:12.150 --> 00:03:14.270 right back through the model's layers you see 00:03:14.270 --> 00:03:15.909 the sequence of calculations and connections 00:03:15.909 --> 00:03:18.270 that led right to it and the source mentioned 00:03:18.270 --> 00:03:21.180 you can even like mess with it, edit things in 00:03:21.180 --> 00:03:23.099 the graph to see what happens. Precisely. Yeah, 00:03:23.159 --> 00:03:25.300 that's the experimental part. You can spot a 00:03:25.300 --> 00:03:28.159 specific internal feature or like a pathway the 00:03:28.159 --> 00:03:30.840 model used, and then you can tweak it or even 00:03:30.840 --> 00:03:33.560 switch it off and see the direct cause and effect 00:03:33.560 --> 00:03:35.639 on the final answer. It lets you investigate 00:03:35.639 --> 00:03:37.780 really deeply. And this isn't just for Anthropic's 00:03:37.780 --> 00:03:40.319 own models, is it? The source said it works with 00:03:40.319 --> 00:03:42.460 some big open models too. Yeah, that's right. 00:03:42.840 --> 00:03:45.080 They specifically mentioned support for models 00:03:45.080 --> 00:03:47.960 like Gemma and Llama. And importantly, they've 00:03:47.960 --> 00:03:50.520 put out demo notebooks and even some graphs they 00:03:50.520 --> 00:03:52.360 haven't fully figured out themselves yet, kind 00:03:52.360 --> 00:03:54.319 of inviting the whole research community to jump 00:03:54.319 --> 00:03:56.699 in and explore. See, this is where it gets... 00:03:57.180 --> 00:03:59.819 I think really significant because for years, 00:03:59.919 --> 00:04:02.939 the biggest knock against advanced AI, especially 00:04:02.939 --> 00:04:05.599 these huge language models, has been that black 00:04:05.599 --> 00:04:08.680 box issue. We train them. They do amazing things, 00:04:08.800 --> 00:04:11.680 but we don't really get why they make certain 00:04:11.680 --> 00:04:13.400 choices or how they get to their conclusions. 00:04:13.659 --> 00:04:16.379 Exactly. And this release feels like a major 00:04:16.379 --> 00:04:19.399 step towards cracking that open. It's a very 00:04:19.399 --> 00:04:22.060 different vibe compared to labs that tend to 00:04:22.060 --> 00:04:24.560 keep these kinds of internal tools super secret. 00:04:25.069 --> 00:04:27.790 you know, proprietary. Anthropix seems to be 00:04:27.790 --> 00:04:30.189 making a statement here like transparency is 00:04:30.189 --> 00:04:33.810 key, which is probably also a competitive thing, 00:04:33.829 --> 00:04:36.110 too. It's a huge signal. And the source toss 00:04:36.110 --> 00:04:39.930 out this really interesting idea. What if? down 00:04:39.930 --> 00:04:42.290 the road, we could actually edit the model's 00:04:42.290 --> 00:04:44.990 reasoning the same way we edit prompts now. That 00:04:44.990 --> 00:04:47.009 does raise a big question. If you can clearly 00:04:47.009 --> 00:04:49.870 trace, say, a flawed line of reasoning or see 00:04:49.870 --> 00:04:52.189 a really good one, could you potentially go right 00:04:52.189 --> 00:04:54.129 into the model's internal workings and correct 00:04:54.129 --> 00:04:56.689 or reinforce those patterns? This tool feels 00:04:56.689 --> 00:04:58.870 like a first step towards maybe having that kind 00:04:58.870 --> 00:05:02.110 of capability someday. Yeah, that's definitely 00:05:02.110 --> 00:05:05.329 something to think about. Okay, so shifting gears 00:05:05.329 --> 00:05:10.420 a bit. From seeing inside the AI mind to predicting 00:05:10.420 --> 00:05:13.300 hurricanes. Your sources also brought up NASA's 00:05:13.300 --> 00:05:17.180 GAIA model. Ah, yes, GAIA. This is a fantastic 00:05:17.180 --> 00:05:20.620 example of these big foundation models. The really 00:05:20.620 --> 00:05:23.959 flexible, powerful ones moving completely beyond 00:05:23.959 --> 00:05:27.060 just language and text. GAIA stands for Geospatial 00:05:27.060 --> 00:05:29.519 Artificial Intelligence for Atmospheres. Right, 00:05:29.600 --> 00:05:32.079 the atmosphere. So it's like an AI weather model. 00:05:33.300 --> 00:05:35.379 different somehow. Much, much more than just 00:05:35.379 --> 00:05:37.560 a standard weather model, really. It's a generative 00:05:37.560 --> 00:05:40.000 AI model built just for our planet's atmosphere. 00:05:40.180 --> 00:05:42.019 It was trained on something like 25 years of 00:05:42.019 --> 00:05:44.300 Earth data. It can do things like predict hurricanes. 00:05:44.519 --> 00:05:46.800 Yeah. But also spot wildfires, estimate rainfall, 00:05:47.100 --> 00:05:50.220 potentially all in real time. 25 years of data. 00:05:50.379 --> 00:05:52.079 Yeah. Wow. That sounds like a massive project. 00:05:52.180 --> 00:05:53.500 What makes this different from, you know, the 00:05:53.500 --> 00:05:55.240 super complex weather models we already have? 00:05:55.360 --> 00:05:57.839 A couple of key things based on the source. One 00:05:57.839 --> 00:06:00.439 is the detail and speed. It apparently has a 00:06:00.439 --> 00:06:02.819 four kilometer spatial resolution that's really 00:06:02.819 --> 00:06:05.740 fine grained and updates every 30 minutes, which 00:06:05.740 --> 00:06:08.360 is incredibly detailed for real time use over 00:06:08.360 --> 00:06:11.240 big areas. But it's also it's generative power. 00:06:11.360 --> 00:06:14.139 It can intelligently like patch up big holes 00:06:14.139 --> 00:06:17.579 in satellite data or reconstruct full high res 00:06:17.579 --> 00:06:20.279 weather maps, even if the input is noisy or incomplete. 00:06:20.970 --> 00:06:22.930 Whoa, that's amazing for places where you might 00:06:22.930 --> 00:06:24.889 not have perfect sensor coverage or, you know, 00:06:24.910 --> 00:06:27.250 if data drops out. Exactly. And it's designed 00:06:27.250 --> 00:06:29.689 to be, you know, a foundation, something other 00:06:29.689 --> 00:06:32.209 labs and organizations can build on top of. It 00:06:32.209 --> 00:06:34.670 really shows these powerful AI methods aren't 00:06:34.670 --> 00:06:36.829 just for writing code or text anymore. They're 00:06:36.829 --> 00:06:39.490 starting to tackle really complex physical systems. 00:06:39.670 --> 00:06:42.069 So, like, could cities or maybe disaster relief 00:06:42.069 --> 00:06:45.560 groups actually use something like this? The 00:06:45.560 --> 00:06:47.759 source suggested that within maybe the next couple 00:06:47.759 --> 00:06:50.779 of years, GAA -like models could realistically 00:06:50.779 --> 00:06:53.360 be powering real -time dashboards. You know, 00:06:53.360 --> 00:06:55.040 for critical infrastructure, disaster response, 00:06:55.220 --> 00:06:57.019 it seems to be moving pretty fast from research 00:06:57.019 --> 00:06:59.500 to potential real -world, maybe even life -saving 00:06:59.500 --> 00:07:03.360 uses. Wow. Okay. So we have this absolutely groundbreaking 00:07:03.360 --> 00:07:07.079 stuff, seeing inside AI predicting massive storms 00:07:07.079 --> 00:07:10.750 with incredible detail, and then... Your sources 00:07:10.750 --> 00:07:13.509 also had that little snippet about Google's AI 00:07:13.509 --> 00:07:15.970 assistant apparently telling people the current 00:07:15.970 --> 00:07:20.189 year is 2024. Yes, that little detail was interesting, 00:07:20.290 --> 00:07:22.269 wasn't it? It definitely highlights the current 00:07:22.269 --> 00:07:24.829 state of things that even while we're making 00:07:24.829 --> 00:07:28.550 these massive leaps like GAIA, really basic factual 00:07:28.550 --> 00:07:32.050 slip ups can still. It is kind of humbling, right? 00:07:32.129 --> 00:07:34.610 Like any kid knows the year, but a cutting edge 00:07:34.610 --> 00:07:37.490 AI occasionally gets it wrong. It certainly brings 00:07:37.490 --> 00:07:39.629 up that important point about reliability and, 00:07:39.670 --> 00:07:41.329 you know, how much we can really trust these 00:07:41.329 --> 00:07:43.850 tools for simple, checkable facts, especially 00:07:43.850 --> 00:07:45.930 when they might be baked into systems we depend 00:07:45.930 --> 00:07:48.290 on. The source even tied it back to some critics 00:07:48.290 --> 00:07:50.629 who worry about relying on AI for the kind of 00:07:50.629 --> 00:07:52.910 factual info we used to just get from the Web. 00:07:53.009 --> 00:07:55.410 Right. And speaking of Google. Your sources also 00:07:55.410 --> 00:07:58.850 mentioned those kind of weirdly realistic, sometimes 00:07:58.850 --> 00:08:01.389 unsettling videos from VO3. People were sharing 00:08:01.389 --> 00:08:03.730 them showing impossible challenges or physics 00:08:03.730 --> 00:08:06.990 being broken. Yes, the sheer ability and realism 00:08:06.990 --> 00:08:10.449 in AI video generation is just advancing incredibly 00:08:10.449 --> 00:08:13.449 fast. Those examples really drive that home. 00:08:13.949 --> 00:08:15.529 And then there was Odyssey, which showed off 00:08:15.529 --> 00:08:18.009 something called an interactive video AI model. 00:08:18.189 --> 00:08:20.350 Right. They described it like stepping into a 00:08:20.350 --> 00:08:22.790 playable movie. That's a different angle, again, 00:08:22.829 --> 00:08:25.110 taking AI video creation beyond just making a 00:08:25.110 --> 00:08:28.029 fixed clip towards potentially creating dynamic, 00:08:28.029 --> 00:08:31.430 responsive visuals that you can actually influence 00:08:31.430 --> 00:08:34.950 somehow. That's a whole other thing for like 00:08:34.950 --> 00:08:36.909 content creation or maybe training simulations 00:08:36.909 --> 00:08:39.450 or something. And then you have tools like Opera's 00:08:39.450 --> 00:08:42.429 new browser, Neon. They're calling it an AI browser 00:08:42.429 --> 00:08:45.350 that can act on your behalf. Yeah, that sounds 00:08:45.350 --> 00:08:47.690 like more of an interface shift. An AI browser 00:08:47.690 --> 00:08:50.549 that can chat, sure, but also fill forms, maybe 00:08:50.549 --> 00:08:52.649 book trips. The source even suggested it could 00:08:52.649 --> 00:08:55.110 build symbol apps for you. The browser itself 00:08:55.110 --> 00:08:57.409 becomes, you know, an AI assistant layered on 00:08:57.409 --> 00:08:59.090 top of everything. So instead of just finding 00:08:59.090 --> 00:09:01.049 the flight info, you could potentially just tell 00:09:01.049 --> 00:09:03.789 it, hey, book me the cheapest flight next Tuesday. 00:09:03.889 --> 00:09:05.809 Yeah. And I might just do it. That seems to be 00:09:05.809 --> 00:09:07.690 the direction they're looking at. Yeah. Your 00:09:07.690 --> 00:09:10.059 sources also caught some of that. broader industry 00:09:10.059 --> 00:09:13.200 pulse, like Grammarly, the AI writing helper, 00:09:13.379 --> 00:09:15.879 getting a billion dollars in funding. That's 00:09:15.879 --> 00:09:18.320 huge. A billion dollars. Yeah, that is serious 00:09:18.320 --> 00:09:20.779 cash flowing into AI productivity tools. And 00:09:20.779 --> 00:09:23.139 some other quick hits, Perplexity, having a tool 00:09:23.139 --> 00:09:25.419 to make spreadsheets from conversations, the 00:09:25.419 --> 00:09:27.860 Netflix co -founder joining Anthropix board. 00:09:28.379 --> 00:09:30.679 It all just shows the whole ecosystem is buzzing, 00:09:30.799 --> 00:09:33.299 you know, evolving really fast, technically, 00:09:33.500 --> 00:09:35.679 commercially, even in terms of who's leading 00:09:35.679 --> 00:09:38.039 things. Oh, and there were mentions of layoffs 00:09:38.039 --> 00:09:40.799 at business. insider, too, maybe hinting at wider 00:09:40.799 --> 00:09:43.720 economic stuff or how AI might start shaking 00:09:43.720 --> 00:09:45.620 up certain jobs. Yeah, there's definitely a lot 00:09:45.620 --> 00:09:47.320 going on on all these different fronts at the 00:09:47.320 --> 00:09:50.059 same time. It's kind of dizzying. Yeah. So, OK, 00:09:50.179 --> 00:09:52.539 we've covered quite a bit here from trying to 00:09:52.539 --> 00:09:55.200 peek inside the AI's mind with Anthropic's new 00:09:55.200 --> 00:09:57.940 tool to predicting massive weather events with 00:09:57.940 --> 00:10:02.919 NASA's GAIA model to. The surprising little glitch 00:10:02.919 --> 00:10:05.200 of a major AI forgetting what year it is and 00:10:05.200 --> 00:10:07.379 all these new ways AI is popping up in video 00:10:07.379 --> 00:10:10.000 and even how we browse the web. Yeah. And if 00:10:10.000 --> 00:10:13.200 we try to like connect the dots here, see the 00:10:13.200 --> 00:10:16.639 bigger picture. What this collection of news 00:10:16.639 --> 00:10:19.179 really signals is the incredible, almost kind 00:10:19.179 --> 00:10:22.240 of jarring speed of AI evolution right now. You 00:10:22.240 --> 00:10:24.620 see this fundamental push for transparency, for 00:10:24.620 --> 00:10:27.639 understanding with Anthropic. But that's happening 00:10:27.639 --> 00:10:29.580 at the same time as we're building these incredibly 00:10:29.580 --> 00:10:32.440 powerful foundation models for really complex 00:10:32.440 --> 00:10:34.639 physical systems like the atmosphere with GAIA. 00:10:34.879 --> 00:10:36.639 And then you have that really stark contrast, 00:10:36.779 --> 00:10:38.840 right, between that super cutting edge stuff, 00:10:38.919 --> 00:10:42.149 the fact that, you know. Some basic reliability 00:10:42.149 --> 00:10:44.809 things like an AI getting the year wrong are 00:10:44.809 --> 00:10:46.830 still cropping up. Exactly. It shows the tech 00:10:46.830 --> 00:10:49.509 is both incredibly advanced and still kind of 00:10:49.509 --> 00:10:53.100 immature. Or maybe just unpredictable in some 00:10:53.100 --> 00:10:56.039 ways. It also clearly demonstrates how fast AI's 00:10:56.039 --> 00:10:57.759 application space is expanding. I mean, it's 00:10:57.759 --> 00:10:59.600 not just about generating text or code anymore, 00:10:59.700 --> 00:11:02.460 not by a long shot. It's moving firmly into physical 00:11:02.460 --> 00:11:05.200 systems like GAIA and to creative media like 00:11:05.200 --> 00:11:07.600 VO and Odyssey. And it's fundamentally changing 00:11:07.600 --> 00:11:10.039 how we interact with information, like with these 00:11:10.039 --> 00:11:13.440 new AI browsers. OK, so what does all this mean 00:11:13.440 --> 00:11:16.580 for you listening right now? Why does knowing 00:11:16.580 --> 00:11:20.019 about circuit tracing or a hurricane model actually 00:11:20.019 --> 00:11:22.480 matter? Well, I think understanding these specific 00:11:22.480 --> 00:11:24.659 points gives you a much better feel for where 00:11:24.659 --> 00:11:26.820 AI is actually heading, you know, beyond just 00:11:26.820 --> 00:11:29.019 the general hype you hear everywhere. Knowing 00:11:29.019 --> 00:11:30.679 about something like circuit tracing, for instance, 00:11:30.840 --> 00:11:33.399 it fundamentally changes that idea of the AI 00:11:33.399 --> 00:11:37.139 black box. Maybe it's becoming less opaque. And 00:11:37.139 --> 00:11:39.019 that's crucial if we want to build trust and 00:11:39.019 --> 00:11:41.710 make sure these things are safe. The GIA model 00:11:41.710 --> 00:11:44.230 shows that AI isn't just changing how we use 00:11:44.230 --> 00:11:46.629 computers. It's starting to potentially change 00:11:46.629 --> 00:11:48.710 our understanding and ability to deal with the 00:11:48.710 --> 00:11:50.809 physical world around us in pretty significant 00:11:50.809 --> 00:11:54.370 ways. And seeing that mix the incredible breakthroughs 00:11:54.370 --> 00:11:57.090 right alongside the basic errors, that gives 00:11:57.090 --> 00:11:59.309 you a more realistic view of where things stand 00:11:59.309 --> 00:12:01.769 today, where it's powerful and where the limitations 00:12:01.769 --> 00:12:04.149 still are. Right. It helps you maybe anticipate 00:12:04.149 --> 00:12:07.250 how AI might show up next in your own life or 00:12:07.250 --> 00:12:10.230 your work or just out in the world. and what 00:12:10.230 --> 00:12:12.049 you can realistically expect from it, or maybe 00:12:12.049 --> 00:12:14.649 what you shouldn't expect from it just yet. Precisely, 00:12:14.649 --> 00:12:18.090 yeah. It's about being informed, grounded, and 00:12:18.090 --> 00:12:20.210 what the actual research and the real -world 00:12:20.210 --> 00:12:22.730 applications are showing us. Yeah, that really 00:12:22.730 --> 00:12:25.149 is something to think about. Okay, so based on 00:12:25.149 --> 00:12:26.610 these sources you brought us, here's a final 00:12:26.610 --> 00:12:29.230 thought to kind of leave you with. We saw with 00:12:29.230 --> 00:12:31.820 Anthropix tool... that we can now start to see 00:12:31.820 --> 00:12:34.879 AI reasoning, right? And they even hinted at 00:12:34.879 --> 00:12:37.200 maybe being able to edit that reasoning directly 00:12:37.200 --> 00:12:40.039 in the future. So what happens when you combine 00:12:40.039 --> 00:12:42.940 the ability to look exactly at why an AI made 00:12:42.940 --> 00:12:46.440 a certain decision with the potential power to 00:12:46.440 --> 00:12:48.940 directly go in and change its internal logic? 00:12:49.220 --> 00:12:51.879 What does that mean for AI safety, for how we 00:12:51.879 --> 00:12:53.740 control these systems, or maybe even for how 00:12:53.740 --> 00:12:55.820 we think about intelligence itself down the line? 00:12:56.639 --> 00:12:58.740 Yeah, definitely a lot to ponder as these capabilities 00:12:58.740 --> 00:13:01.720 keep developing so so quickly. Indeed. Thanks 00:13:01.720 --> 00:13:03.259 again for bringing your sources for this deep 00:13:03.259 --> 00:13:03.500 dive.