WEBVTT 00:00:00.000 --> 00:00:04.879 Imagine a model, an AI, right, learning from 00:00:04.879 --> 00:00:08.919 nine years of solar data. We're talking millions 00:00:08.919 --> 00:00:12.919 of images. Just so it can tell us when our son 00:00:12.919 --> 00:00:15.259 might, you know, decide to throw a tantrum. Wow. 00:00:15.460 --> 00:00:17.820 And not just predicting, but actually doing better 00:00:17.820 --> 00:00:22.379 than human experts by a pretty big margin. 16%, 00:00:22.379 --> 00:00:26.160 yeah. Yeah, 16%. That's a really profound thought, 00:00:26.219 --> 00:00:28.399 isn't it? The sheer scale of that learning. Whoa. 00:00:28.829 --> 00:00:31.510 It really is something else. And welcome everyone 00:00:31.510 --> 00:00:33.950 to the Deep Dive. Glad to be here. Our mission, 00:00:34.109 --> 00:00:36.770 like always, is to take this, well, stack of 00:00:36.770 --> 00:00:38.329 the latest insights, like the big newsletter 00:00:38.329 --> 00:00:41.090 we just got, and really pull out those key bits 00:00:41.090 --> 00:00:43.189 of knowledge. We want to make sure you're not 00:00:43.189 --> 00:00:45.890 just, you know, hearing the news, but genuinely 00:00:45.890 --> 00:00:48.649 in the know about what's happening on the cutting 00:00:48.649 --> 00:00:51.420 edge of AI. Yeah. Get past the headlines. Exactly. 00:00:51.679 --> 00:00:54.439 So today we've got quite a journey planned. First 00:00:54.439 --> 00:00:57.200 up, we'll really unpack Google's big move with 00:00:57.200 --> 00:00:59.259 Gemini for Home. That's replacing Google Assistant. 00:00:59.500 --> 00:01:02.280 Big change. Huge. Then we're taking a faster 00:01:02.280 --> 00:01:05.319 look, kind of rapid fire, at other critical AI 00:01:05.319 --> 00:01:08.120 developments happening across tech, finance, 00:01:08.500 --> 00:01:11.370 even education. Lots going on there. Always. 00:01:11.569 --> 00:01:14.010 And then finally, we'll circle back to that amazing 00:01:14.010 --> 00:01:17.609 solar AI we just mentioned, Syria, and dig into 00:01:17.609 --> 00:01:20.030 what it really means for us down here on Earth. 00:01:20.230 --> 00:01:22.969 Okay. Sounds good. So let's jump straight into 00:01:22.969 --> 00:01:25.609 that first big story. Google Gemini for Home. 00:01:25.730 --> 00:01:28.549 Right. I mean, for almost 10 years now, Google 00:01:28.549 --> 00:01:31.629 Assistant has been, well, a fixture in a lot 00:01:31.629 --> 00:01:33.909 of smart homes, right? Answering questions, turning 00:01:33.909 --> 00:01:35.890 lights on and off. Yeah, the usual stuff. The 00:01:35.890 --> 00:01:39.219 usual. But now... It's officially being phased 00:01:39.219 --> 00:01:42.219 out. End of an era, almost. It kind of is. And 00:01:42.219 --> 00:01:44.359 in its place, Google is introducing something 00:01:44.359 --> 00:01:46.459 called Gemini for Home. They're pitching it as 00:01:46.459 --> 00:01:50.060 this next generation AI assistant. And honestly, 00:01:50.200 --> 00:01:52.879 listening to the details, it feels like way more 00:01:52.879 --> 00:01:55.120 than just an update. It feels like a fundamental 00:01:55.120 --> 00:01:57.900 shift. How so? Well, Gemini for Home is designed 00:01:57.900 --> 00:02:01.200 to be much smarter, way more conversational. 00:02:01.219 --> 00:02:03.340 And it's really built for those complex, multi 00:02:03.340 --> 00:02:05.420 -step things that are just, you know, part of 00:02:05.420 --> 00:02:07.900 running a busy household. of it less like that 00:02:07.900 --> 00:02:12.979 old smart speaker gimmick and more like a household 00:02:12.979 --> 00:02:16.580 AI co -pilot. A co -pilot. I like that analogy. 00:02:16.740 --> 00:02:19.520 It feels fitting. Yeah. It's like having a truly 00:02:19.520 --> 00:02:22.099 intelligent assistant there, almost anticipating 00:02:22.099 --> 00:02:24.960 what you need before you ask. Okay. So co -pilot 00:02:24.960 --> 00:02:28.979 it is. What are some of the... The standout features 00:02:28.979 --> 00:02:32.039 them. What makes Gemini for Home so different, 00:02:32.180 --> 00:02:34.379 maybe more capable than what we've had before? 00:02:34.979 --> 00:02:38.240 OK, well, the media search is a huge leap forward. 00:02:38.300 --> 00:02:40.280 You know how you're trying to find a show? Maybe 00:02:40.280 --> 00:02:42.300 you describe it kind of vaguely. Oh, yeah. And 00:02:42.300 --> 00:02:44.060 the old assistant just gives you that. Did you 00:02:44.060 --> 00:02:47.340 mean this dead end? All the time. Right. Gemini 00:02:47.340 --> 00:02:49.439 for Home is supposed to understand context and 00:02:49.439 --> 00:02:52.580 intent much better. So it finds exactly what 00:02:52.580 --> 00:02:54.479 you mean across all your different streaming 00:02:54.479 --> 00:02:56.699 services. No more guessing games. I know that 00:02:56.699 --> 00:02:58.960 feeling all too well. Honestly, my family spends 00:02:58.960 --> 00:03:01.219 half our movie night just trying to figure out 00:03:01.219 --> 00:03:04.039 where a show even lives or which service it's 00:03:04.039 --> 00:03:05.900 actually on. It's frustrating. Exactly. It's 00:03:05.900 --> 00:03:07.620 that little friction point, right? It really 00:03:07.620 --> 00:03:09.740 takes the smart out of smart homes sometimes. 00:03:10.520 --> 00:03:12.479 And the smart home control itself is supposed 00:03:12.479 --> 00:03:15.939 to be far deeper, too. It can apparently reason 00:03:15.939 --> 00:03:18.919 through complex commands more naturally. Reason 00:03:18.919 --> 00:03:20.900 through them? Yeah, like connecting multiple 00:03:20.900 --> 00:03:23.300 devices without you having to spell out every 00:03:23.300 --> 00:03:25.900 single step. Imagine just saying, like, get the 00:03:25.900 --> 00:03:28.460 house ready for movie night. Okay. And it maybe 00:03:28.460 --> 00:03:31.419 dims the lights, lowers the blinds, turns on 00:03:31.419 --> 00:03:33.960 the TV to the right input, adjusts the thermostat. 00:03:34.560 --> 00:03:38.460 You know, seamlessly. It moves beyond just simple 00:03:38.460 --> 00:03:42.180 commands to like actual integration based on 00:03:42.180 --> 00:03:44.939 the goal. So it's really anticipating what you 00:03:44.939 --> 00:03:47.319 actually mean, not just reacting to the literal 00:03:47.319 --> 00:03:50.120 keywords you say. That feels like a pretty significant 00:03:50.120 --> 00:03:53.879 leap in understanding intent. It truly is. And 00:03:53.879 --> 00:03:55.699 for families, they're saying it's a potential 00:03:55.699 --> 00:03:58.240 game changer for coordination. You can manage 00:03:58.240 --> 00:04:00.840 complex routines just using plain natural language. 00:04:01.080 --> 00:04:02.639 Makes sense. And here's where it gets really 00:04:02.639 --> 00:04:04.960 interesting, I think. say, hey, Google, let's 00:04:04.960 --> 00:04:07.680 chat. And then you drop the hey, Google wake 00:04:07.680 --> 00:04:11.759 word and just talk. Oh, OK. So like a continuous 00:04:11.759 --> 00:04:14.560 conversation. Exactly. It pulls information from 00:04:14.560 --> 00:04:17.259 search, your streaming platforms, all your connected 00:04:17.259 --> 00:04:20.079 devices all at once. It's a genuine conversational 00:04:20.079 --> 00:04:22.300 mode. It should feel much less like you're talking 00:04:22.300 --> 00:04:26.319 at a machine. So it's going beyond just simple 00:04:26.319 --> 00:04:29.769 trivia, basically handling anything from. complex 00:04:29.769 --> 00:04:33.069 questions to managing your whole home setup just 00:04:33.069 --> 00:04:36.089 through a flowing conversation. That sounds incredibly 00:04:36.089 --> 00:04:38.569 intuitive if it works well. Yeah, if it works 00:04:38.569 --> 00:04:40.449 well. When can people expect to see this actually 00:04:40.449 --> 00:04:42.410 rolling out into their homes? Well, it's starting 00:04:42.410 --> 00:04:44.750 pretty soon. Gemini is set to replace Google 00:04:44.750 --> 00:04:47.170 Assistant on Nest speakers and smart displays. 00:04:47.509 --> 00:04:49.550 Early access is actually beginning in October. 00:04:49.889 --> 00:04:52.589 Okay, this fall. Yeah. And there will be both 00:04:52.589 --> 00:04:55.910 free and paid tiers, which is an interesting 00:04:55.910 --> 00:04:58.790 detail to watch. Subscription AI in the home. 00:04:59.160 --> 00:05:02.000 Could be. But the real significance here, I think, 00:05:02.000 --> 00:05:04.040 is what this means for the broader market. This 00:05:04.040 --> 00:05:07.139 shift really puts competitors like Alexa and 00:05:07.139 --> 00:05:11.040 Siri in a, well, a tough spot. If Gemini becomes 00:05:11.040 --> 00:05:14.639 the household sort of default AI brain, the one 00:05:14.639 --> 00:05:16.519 that really understands and connects everything, 00:05:16.740 --> 00:05:20.569 the entire category could shift overnight. It 00:05:20.569 --> 00:05:23.089 forces everyone else to step up their game significantly. 00:05:23.329 --> 00:05:26.629 Right. Okay, so let's crystallize that. How fundamentally 00:05:26.629 --> 00:05:29.509 does this change our interaction with home technology? 00:05:29.730 --> 00:05:33.839 Yeah, in short. It makes AI a truly conversational, 00:05:33.839 --> 00:05:37.240 intelligent home co -pilot. Okay, so now let's 00:05:37.240 --> 00:05:40.639 pivot a bit from the living room to the wider 00:05:40.639 --> 00:05:43.120 and, frankly, incredibly dynamic world of artificial 00:05:43.120 --> 00:05:45.319 intelligence. Moving so fast. It really is. It's 00:05:45.319 --> 00:05:47.300 such a fast -moving space, and we're seeing these 00:05:47.300 --> 00:05:49.560 rapid developments just, well, across the board. 00:05:49.699 --> 00:05:51.779 It feels like every single day brings a new headline, 00:05:51.980 --> 00:05:54.560 sometimes several. It's hard to keep up. What 00:05:54.560 --> 00:05:56.040 are some of the most striking highlights from 00:05:56.040 --> 00:05:58.269 the sources we looked at today? beyond the smart 00:05:58.269 --> 00:05:59.870 home stuff okay well let's start with coding 00:05:59.870 --> 00:06:02.230 agents this is pretty interesting a founder john 00:06:02.230 --> 00:06:06.389 rush recently tested get this 61 different viral 00:06:06.389 --> 00:06:11.220 ai coding agents 61 yeah He created this really 00:06:11.220 --> 00:06:13.699 comprehensive list, put demos and notes together, 00:06:13.860 --> 00:06:16.980 and it apparently got over 1 .1 million views 00:06:16.980 --> 00:06:19.620 in less than a day. That's just wild. The sheer 00:06:19.620 --> 00:06:22.240 volume of interest in these tools is incredible. 00:06:22.519 --> 00:06:26.199 It is. And just for anyone unfamiliar, AI coding 00:06:26.199 --> 00:06:28.860 agents are basically specialized AI programs. 00:06:29.160 --> 00:06:31.720 They can write code, help you write code, even 00:06:31.720 --> 00:06:34.480 debug human code sometimes. Like digital assistants 00:06:34.480 --> 00:06:37.060 for developers? Exactly. Like digital apprentices 00:06:37.060 --> 00:06:39.569 almost. That is a massive response. his list. 00:06:39.670 --> 00:06:41.949 You know, I have to admit, I still wrestle with 00:06:41.949 --> 00:06:44.209 prompt drift myself sometimes. Oh, yeah. Yeah. 00:06:44.230 --> 00:06:46.189 You know, where the AI starts to wander off from 00:06:46.189 --> 00:06:48.670 the initial instructions or loses context halfway 00:06:48.670 --> 00:06:51.589 through. Trying to get really consistent output 00:06:51.589 --> 00:06:54.750 from these tools can be tricky. I know what you 00:06:54.750 --> 00:06:56.689 mean. So the idea of having a comprehensive, 00:06:56.930 --> 00:07:01.189 maybe vetted list of coding agents sounds incredibly 00:07:01.189 --> 00:07:03.300 helpful for a lot of people. myself included. 00:07:03.459 --> 00:07:05.839 Absolutely. And moving into higher education 00:07:05.839 --> 00:07:08.620 for a sec, Anthropic just announced two pretty 00:07:08.620 --> 00:07:11.319 major moves. First, they're forming a higher 00:07:11.319 --> 00:07:13.959 education advisory board with leaders from top 00:07:13.959 --> 00:07:17.040 universities. Second, they've also launched three 00:07:17.040 --> 00:07:20.920 new completely free AI fluency courses. Oh, free. 00:07:21.100 --> 00:07:23.519 That's great. Yeah. Available for anyone to access. 00:07:23.660 --> 00:07:25.980 So that seems like a big push for making foundational 00:07:25.980 --> 00:07:28.839 AI concepts more accessible and better understood. 00:07:29.060 --> 00:07:31.379 Free courses are always fantastic for broadening 00:07:31.379 --> 00:07:34.120 access to about critical knowledge like this. 00:07:34.199 --> 00:07:36.399 What else is making waves out there? Well, Google 00:07:36.399 --> 00:07:38.759 also shared some really excellent use cases for 00:07:38.759 --> 00:07:43.120 their Gemma 3270M model. Ah, the compact one. 00:07:43.220 --> 00:07:45.300 Exactly. Remember, that's a remarkably small 00:07:45.300 --> 00:07:48.060 and hyper -efficient AI. It's all about packing 00:07:48.060 --> 00:07:50.899 powerful AI into smaller packages. Which is important. 00:07:51.079 --> 00:07:53.620 Why? Well, it's crucial for running AI directly 00:07:53.620 --> 00:07:56.779 on devices, phones, laptops, maybe even cars, 00:07:56.879 --> 00:07:59.279 instead of always relying on the cloud. It expands 00:07:59.279 --> 00:08:01.699 accessibility and reduces the need for that massive 00:08:01.699 --> 00:08:04.040 cloud infrastructure for a lot of tasks. This 00:08:04.040 --> 00:08:07.279 whole on -device AI idea is a major step towards 00:08:07.279 --> 00:08:10.220 making AI more widespread and personalized. Right. 00:08:10.300 --> 00:08:12.480 That efficiency piece is key for wider adoption 00:08:12.480 --> 00:08:14.759 and probably enables totally new capabilities, 00:08:15.000 --> 00:08:18.600 too. It truly is. Then there's the geopolitical 00:08:18.600 --> 00:08:20.879 angle, which is always simmering. NVIDIA might 00:08:20.879 --> 00:08:24.379 be looking at losing billions. China has reportedly 00:08:24.379 --> 00:08:27.879 told its tech giants like Alibaba and ByteDance 00:08:27.879 --> 00:08:31.899 to ditch Nvidia's H20 chips. Those are the watered 00:08:31.899 --> 00:08:33.899 down ones, right? The ones designed to comply 00:08:33.899 --> 00:08:37.019 with U .S. export rules. Exactly. The less powerful 00:08:37.019 --> 00:08:39.480 versions Nvidia made specifically for the Chinese 00:08:39.480 --> 00:08:42.320 market due to U .S. restrictions. But it seems 00:08:42.320 --> 00:08:45.980 China now views even those as, well, maybe an 00:08:45.980 --> 00:08:48.220 insulting concession they're not willing to make 00:08:48.220 --> 00:08:50.840 anymore, according to the reports. Wow. It just 00:08:50.840 --> 00:08:53.000 shows how sensitive the whole global... tech 00:08:53.000 --> 00:08:55.600 landscape is and how political statements can 00:08:55.600 --> 00:08:58.080 have these huge ripple effects on business. Indeed. 00:08:58.279 --> 00:09:00.360 The strategic implications of something like 00:09:00.360 --> 00:09:02.419 that are enormous. It impacts supply chains, 00:09:02.679 --> 00:09:05.679 national tech strategies, everything. They really 00:09:05.679 --> 00:09:07.600 are. And, you know, big tech just keeps making 00:09:07.600 --> 00:09:10.580 these massive deals. Meta just signed a $10 billion 00:09:10.580 --> 00:09:13.620 six -year agreement to run its AI operations 00:09:13.620 --> 00:09:16.620 on Google Cloud. $10 billion on a competitor's 00:09:16.620 --> 00:09:18.980 cloud. Interestingly, Meta's stock dipped a bit 00:09:18.980 --> 00:09:21.000 after the news while Google saw a little bump. 00:09:21.399 --> 00:09:23.519 This isn't just about a big check changing hands. 00:09:23.720 --> 00:09:26.679 It signals a really critical strategic choice, 00:09:26.860 --> 00:09:29.519 right? Even tech giants with vast infrastructure 00:09:29.519 --> 00:09:32.100 of their own are recognizing the specialized, 00:09:32.440 --> 00:09:35.120 frankly, enormous scale needed for advanced AI. 00:09:35.500 --> 00:09:38.159 They're choosing to partner, even with rivals. 00:09:38.419 --> 00:09:41.379 So the market might be seeing it as Google gaining 00:09:41.379 --> 00:09:45.399 an edge in this AI infrastructure arms race. 00:09:45.580 --> 00:09:47.889 That seems to be the read, yeah. Almost like 00:09:47.889 --> 00:09:50.529 meta is becoming reliant on a competitor for 00:09:50.529 --> 00:09:52.870 something core to its future strategy. That is 00:09:52.870 --> 00:09:55.450 a fascinating dynamic. Wow. Precisely. And the 00:09:55.450 --> 00:09:57.980 fundraising. Oh, man. The fundraising frenzy 00:09:57.980 --> 00:09:59.899 just continues. Anthropic is reportedly raising 00:09:59.899 --> 00:10:03.080 $10 billion now. $10 billion? Didn't they just 00:10:03.080 --> 00:10:05.139 raise a huge round? They did. They're apparently 00:10:05.139 --> 00:10:07.059 doubling their initial goals for this round. 00:10:07.200 --> 00:10:10.220 This could push its valuation to, well, an astounding 00:10:10.220 --> 00:10:14.100 $170 billion. Whoa. Yeah. With major investments 00:10:14.100 --> 00:10:16.279 already coming in from Iconic Capital and other 00:10:16.279 --> 00:10:18.919 big VCs, the amount of capital pouring into this 00:10:18.919 --> 00:10:21.580 AI space is just incredible. It's a staggering 00:10:21.580 --> 00:10:24.789 figure. Truly reflects the markets. maybe hunger 00:10:24.789 --> 00:10:27.909 is the right word, for AI right now. We've also 00:10:27.909 --> 00:10:30.850 seen a few other quick but pretty notable things 00:10:30.850 --> 00:10:34.350 pop up. Like what? Well, Google's AI mode in 00:10:34.350 --> 00:10:36.909 search is now global. Yeah. Rolled out to 180 00:10:36.909 --> 00:10:39.929 countries. That really makes AI search almost 00:10:39.929 --> 00:10:42.250 ubiquitous for information retrieval worldwide. 00:10:42.669 --> 00:10:45.830 Right. Part of daily life now. Yeah. And OpenAI 00:10:45.830 --> 00:10:48.029 executives are apparently claiming that their 00:10:48.029 --> 00:10:50.929 next model, maybe GPT -5 Pro, might even be able 00:10:50.929 --> 00:10:53.529 to prove new interesting mathematical theorems. 00:10:53.590 --> 00:10:55.889 Prove new math. That's what they're hinting at, 00:10:55.950 --> 00:10:59.110 which, if true, is. That's not just mind -boggling. 00:10:59.110 --> 00:11:01.269 It potentially moves AI beyond just mimicking 00:11:01.269 --> 00:11:03.960 patterns into true original discovery. That's 00:11:03.960 --> 00:11:06.100 a profound leap. Yeah, it could redefine the 00:11:06.100 --> 00:11:08.039 role of human intellect and feels like pure math. 00:11:08.220 --> 00:11:11.360 No. A huge potential shift. And on a completely 00:11:11.360 --> 00:11:13.299 different note, but still AI, China actually 00:11:13.299 --> 00:11:16.080 launched its Wukong AI on its space station during 00:11:16.080 --> 00:11:18.700 some recent spacewalk. AI in orbit during a spacewalk. 00:11:18.759 --> 00:11:20.720 Yep, bringing AI into space in a very direct 00:11:20.720 --> 00:11:23.840 way. Meanwhile, back on Earth, Meta, after that 00:11:23.840 --> 00:11:26.340 big hiring spree for AI talent, you know, poaching 00:11:26.340 --> 00:11:29.700 people, is now reportedly pausing its AI hiring, 00:11:29.899 --> 00:11:33.539 which could signal maybe a broader market recalibration 00:11:33.539 --> 00:11:35.639 or maybe they just got everyone they needed for 00:11:35.639 --> 00:11:38.539 now. Hard to say. And finally, Sam Ullman himself, 00:11:38.860 --> 00:11:41.759 while, you know, successfully selling open AI 00:11:41.759 --> 00:11:44.299 shares at a valuation around, what, $500 billion? 00:11:44.679 --> 00:11:47.080 Half a trillion, yeah. Has also apparently warned 00:11:47.080 --> 00:11:50.299 about an AI bubble. Which for listeners tracking 00:11:50.299 --> 00:11:52.580 the market raises that critical question, doesn't 00:11:52.580 --> 00:11:56.059 it? Are we seeing genuine sustainable growth 00:11:56.059 --> 00:11:58.840 here or is it getting frothy? Yeah. Driven by 00:11:58.840 --> 00:12:01.620 speculation, maybe like past tech booms. It's 00:12:01.620 --> 00:12:03.799 definitely a lot to chew on. So pulling it all 00:12:03.799 --> 00:12:06.679 together, what's the unifying theme you see across 00:12:06.679 --> 00:12:09.419 all these really diverse AI developments? I think 00:12:09.419 --> 00:12:12.120 it has to be AI's relentless, almost pervasive 00:12:12.120 --> 00:12:14.200 integration into nearly every facet of life. 00:12:14.279 --> 00:12:17.610 It's just everywhere. Sponsor. Okay, so earlier, 00:12:17.809 --> 00:12:20.029 right at the start, we touched on this really 00:12:20.029 --> 00:12:22.909 astounding AI development from NASA and IBM. 00:12:23.269 --> 00:12:25.809 Surya. Surya, exactly. Now, let's really dive 00:12:25.809 --> 00:12:28.429 into that. This solar foundation model that literally 00:12:28.429 --> 00:12:30.850 learned to read the sun. Yeah, this is a truly 00:12:30.850 --> 00:12:33.649 fascinating project. It's marrying, you know, 00:12:33.649 --> 00:12:36.169 cutting edge AI with really crucial space science. 00:12:36.570 --> 00:12:38.669 Historically, sure, we've had tools to monitor 00:12:38.669 --> 00:12:40.929 the sun. Right, satellites, telescopes. Exactly. 00:12:41.090 --> 00:12:44.649 But apparently none have been as fast or as accurate 00:12:44.649 --> 00:12:47.669 as Surya seems to be at recognizing those really 00:12:47.669 --> 00:12:50.429 early signs of danger coming from our star. And 00:12:50.429 --> 00:12:52.070 that's the core problem it's tackling, right? 00:12:52.110 --> 00:12:54.169 Because space weather isn't just an abstract 00:12:54.169 --> 00:12:57.049 thing. It can have serious impacts here on Earth. 00:12:57.190 --> 00:12:59.799 Oh, absolutely. Disrupting communication. Messing 00:12:59.799 --> 00:13:02.139 with power grids, even knocking out satellite 00:13:02.139 --> 00:13:05.879 navigation. Yeah, GPS. GPS, exactly. So this 00:13:05.879 --> 00:13:09.100 solution, Surya, they're calling it a solar foundation 00:13:09.100 --> 00:13:11.000 model. Maybe we should quickly explain what a 00:13:11.000 --> 00:13:13.000 foundation model is in this context. Yeah. Good 00:13:13.000 --> 00:13:15.419 idea. So a foundation model is essentially a 00:13:15.419 --> 00:13:17.840 very large AI model. It's trained on a really 00:13:17.840 --> 00:13:20.899 broad range of data, often unlabeled data. Okay. 00:13:20.960 --> 00:13:23.539 And that broad training makes it highly adaptable 00:13:23.539 --> 00:13:26.539 for many different, more specific tasks later 00:13:26.539 --> 00:13:29.470 on. Think of it like a... Like a highly educated 00:13:29.470 --> 00:13:32.429 generalist AI that can then specialize quickly 00:13:32.429 --> 00:13:35.909 in various fields. Got it. So Surya was trained 00:13:35.909 --> 00:13:39.169 as this generalist, but specifically focused 00:13:39.169 --> 00:13:41.590 on the sun to become a specialist. Yeah. How 00:13:41.590 --> 00:13:43.769 exactly did they train it to be so effective? 00:13:43.929 --> 00:13:47.059 What was the data? They fed it an absolutely 00:13:47.059 --> 00:13:50.039 immense amount of data. Millions of images gathered 00:13:50.039 --> 00:13:53.820 over nine years from NASA's Solar Dynamics Observatory, 00:13:53.840 --> 00:13:56.779 or SDO. Nine years of constant observation. Yeah, 00:13:56.860 --> 00:13:59.019 it's an incredible volume. And through that, 00:13:59.240 --> 00:14:01.759 Surya basically learned to track everything, 00:14:01.960 --> 00:14:04.500 from emerging sunspots, you know, the little 00:14:04.500 --> 00:14:07.139 dark patches, to tracking solar wind speeds. 00:14:07.240 --> 00:14:10.120 And crucially, it processes multiple wavelengths 00:14:10.120 --> 00:14:12.879 of light simultaneously. Ah, so not just visible 00:14:12.879 --> 00:14:15.519 light. Right. Different wavelengths reveal different 00:14:15.519 --> 00:14:18.080 things happening on the sun. This allows it to 00:14:18.080 --> 00:14:20.340 spot tiny surface changes, things that humans 00:14:20.340 --> 00:14:23.139 often miss, or that current automated systems 00:14:23.139 --> 00:14:25.440 just aren't sensitive enough to pick up reliably. 00:14:25.799 --> 00:14:28.279 So it's not just seeing what we see, but it's 00:14:28.279 --> 00:14:30.980 quite literally seeing more detail across more 00:14:30.980 --> 00:14:34.000 spectrums and processing it faster than we ever 00:14:34.000 --> 00:14:36.179 could manually. Precisely. And what can it do 00:14:36.179 --> 00:14:37.960 once it's learned all this? What are its main 00:14:37.960 --> 00:14:41.019 capabilities? Well, its primary capability and 00:14:41.019 --> 00:14:44.620 the really vital one is prediction. It can predict 00:14:44.620 --> 00:14:48.220 when a solar eruption like a flare or a coronal 00:14:48.220 --> 00:14:50.419 mass ejection might blast out towards Earth. 00:14:50.580 --> 00:14:53.659 Okay. And the key thing is it's already outperforming 00:14:53.659 --> 00:14:56.659 the existing predictive models by a remarkable 00:14:56.659 --> 00:15:00.980 16%. 16 % better prediction. Yeah. That's a significant 00:15:00.980 --> 00:15:03.539 leap in accuracy for something that's so crucial 00:15:03.539 --> 00:15:06.220 to protecting our infrastructure and frankly, 00:15:06.240 --> 00:15:08.950 our safety down here. Outperforming. by 16 % 00:15:08.950 --> 00:15:11.129 is huge when you're talking about potentially 00:15:11.129 --> 00:15:13.710 damaging space weather events that could impact 00:15:13.710 --> 00:15:16.549 global systems. But what I find truly exciting 00:15:16.549 --> 00:15:18.669 about this beyond the science is the accessibility 00:15:18.669 --> 00:15:21.929 aspect. Oh, absolutely. This is maybe the coolest 00:15:21.929 --> 00:15:25.409 part for the wider community. NASA has open sourced 00:15:25.409 --> 00:15:27.370 the entire Syria model. They put it up on Hugging 00:15:27.370 --> 00:15:29.409 Face. Hugging Face, yeah, the AI platform. Exactly. 00:15:29.649 --> 00:15:31.830 So this means any researcher, any student, even 00:15:31.830 --> 00:15:34.250 a private company working on space weather, they 00:15:34.250 --> 00:15:36.549 can access and use Syria for their own projects. 00:15:36.870 --> 00:15:39.889 Wow. It truly democratizes this incredible scientific 00:15:39.889 --> 00:15:42.330 tool. It should really help accelerate further 00:15:42.330 --> 00:15:44.830 innovation in the field. That's such a powerful 00:15:44.830 --> 00:15:47.830 move. Making sure its impact can be felt far 00:15:47.830 --> 00:15:50.669 and wide, not just with... in NASA or IBM, you 00:15:50.669 --> 00:15:52.990 know, it really makes you realize with climate 00:15:52.990 --> 00:15:55.470 change already putting so much strain on Earth's 00:15:55.470 --> 00:15:58.009 systems, knowing when our nearest star plans 00:15:58.009 --> 00:16:01.830 to, well, throw a tantrum, as we said, that might 00:16:01.830 --> 00:16:04.070 just be the most important forecast of all. It 00:16:04.070 --> 00:16:06.610 directly impacts our planet's resilience, our 00:16:06.610 --> 00:16:09.500 ability to prepare. So thinking about the direct 00:16:09.500 --> 00:16:11.799 impact then, how does this specific scientific 00:16:11.799 --> 00:16:14.679 AI innovation really affect us, you know, day 00:16:14.679 --> 00:16:17.299 to day on Earth? Well, fundamentally, it crucially 00:16:17.299 --> 00:16:19.980 improves forecasts for space weather, which helps 00:16:19.980 --> 00:16:22.580 protect our planet's vital infrastructure communications 00:16:22.580 --> 00:16:26.899 power. Okay, so let's try to synthesize our whole 00:16:26.899 --> 00:16:29.659 deep dive today. We've covered some truly transformative 00:16:29.659 --> 00:16:32.360 ground, really highlighting how pervasive and 00:16:32.360 --> 00:16:35.259 powerful AI's evolution is becoming. Yeah, it 00:16:35.259 --> 00:16:37.100 feels like three main themes emerged. Right. 00:16:37.220 --> 00:16:39.860 First, we saw AI making this deep integration 00:16:39.860 --> 00:16:42.700 into the home. Gemini for Home isn't just, you 00:16:42.700 --> 00:16:44.580 know, another update. It feels like a fundamental 00:16:44.580 --> 00:16:47.860 shift. The co -pilot idea. Exactly. Moving AI 00:16:47.860 --> 00:16:51.039 from just simple command and response to being 00:16:51.039 --> 00:16:54.590 a true... household co -pilot, maybe intelligently 00:16:54.590 --> 00:16:57.490 anticipating needs. And that challenges all the 00:16:57.490 --> 00:17:00.370 astounding players, Alexa, Siri, and hints at 00:17:00.370 --> 00:17:03.169 this new era of genuinely conversational intelligence 00:17:03.169 --> 00:17:05.809 right in our living spaces. Okay, that was theme 00:17:05.809 --> 00:17:09.069 one. Then theme two was that whirlwind tour of 00:17:09.069 --> 00:17:11.730 AI's explosive growth and all the strategic plays 00:17:11.730 --> 00:17:14.089 happening. Yeah, from those specialized AI coding 00:17:14.089 --> 00:17:16.569 agents getting millions of views. Crazy numbers. 00:17:16.690 --> 00:17:19.190 To multi -billion dollar cloud deals between 00:17:19.190 --> 00:17:22.069 giants. And even those geopolitical shifts driven 00:17:22.069 --> 00:17:25.960 by chip technology. AI is just everywhere now. 00:17:26.099 --> 00:17:28.579 It's evolving at this incredible pace, constantly 00:17:28.579 --> 00:17:30.680 weaving into new industries, new aspects of our 00:17:30.680 --> 00:17:33.039 lives. And that's prompting both massive innovation 00:17:33.039 --> 00:17:35.220 and some serious strategic recalculations across 00:17:35.220 --> 00:17:36.819 the board. And then finally, the third theme 00:17:36.819 --> 00:17:39.400 was seeing AI tackle these really grand challenges, 00:17:39.559 --> 00:17:42.059 like with the Syria project. Right. That demonstrates 00:17:42.059 --> 00:17:45.319 AI's immense power to help solve complex scientific 00:17:45.319 --> 00:17:47.980 problems. It's offering unprecedented accuracy 00:17:47.980 --> 00:17:50.380 in vital areas like space weather forecasting 00:17:50.380 --> 00:17:52.640 in this case. Yeah. It's really about leveraging 00:17:52.640 --> 00:17:55.109 AI for... some of the biggest questions facing 00:17:55.109 --> 00:17:58.170 humanity, pushing the boundaries of what's possible 00:17:58.170 --> 00:18:01.190 in scientific discovery and, well, planetary 00:18:01.190 --> 00:18:03.250 protection. It really has been a fascinating 00:18:03.250 --> 00:18:05.450 journey through the latest developments. And 00:18:05.450 --> 00:18:07.609 honestly, there's so much more to explore within 00:18:07.609 --> 00:18:10.049 each of these topics. Always more to learn. Always. 00:18:10.569 --> 00:18:13.309 If you, the listener, are curious to continue 00:18:13.309 --> 00:18:16.109 your own deep dive after this, I'd highly recommend 00:18:16.109 --> 00:18:19.130 checking out Google's use cases for their compact 00:18:19.130 --> 00:18:22.869 Gemma 3 model. The efficient one. Yeah. Understanding 00:18:22.869 --> 00:18:26.430 how powerful AI can actually run on smaller devices 00:18:26.430 --> 00:18:28.930 is, I think, a really crucial insight into where 00:18:28.930 --> 00:18:31.369 things might be heading. Or, you know, for the 00:18:31.369 --> 00:18:34.250 truly adventurous, go explore the open sourced 00:18:34.250 --> 00:18:37.009 Syria model on Hugging Face yourself. See the 00:18:37.009 --> 00:18:39.170 sun's secrets. Exactly. There's a lot to learn 00:18:39.170 --> 00:18:41.009 there. And these powerful tools are increasingly 00:18:41.009 --> 00:18:43.670 right at our fingertips. So thank you all for 00:18:43.670 --> 00:18:47.509 joining us on this deep dive today. As AI integrates 00:18:47.509 --> 00:18:50.109 more deeply into our homes, into our global systems, 00:18:50.210 --> 00:18:52.049 even into our very understanding of the universe, 00:18:52.250 --> 00:18:54.789 what new questions do we maybe need to start 00:18:54.789 --> 00:18:57.829 asking about its role in shaping our future? 00:18:57.890 --> 00:18:59.109 It's definitely something worth pondering.