WEBVTT 00:00:00.000 --> 00:00:02.560 Welcome to the Deep Dive. We're here to sift 00:00:02.560 --> 00:00:04.400 through the noise, try and bring some clarity, 00:00:04.519 --> 00:00:06.719 maybe some unexpected insights into the world 00:00:06.719 --> 00:00:09.320 of AI. Yeah, and today we're looking beyond just 00:00:09.320 --> 00:00:11.480 the latest chatbot craze. What if the biggest 00:00:11.480 --> 00:00:14.029 story is actually, well... Who's building the 00:00:14.029 --> 00:00:16.789 whole digital backbone for AI? Exactly. That's 00:00:16.789 --> 00:00:19.530 the core question. We're seeing things like Oracle's 00:00:19.530 --> 00:00:23.050 rumored, what, $300 billion plus deal with open 00:00:23.050 --> 00:00:26.969 AI. And a huge $455 billion backlog they reported. 00:00:27.210 --> 00:00:29.329 It's massive. It really makes you think about 00:00:29.329 --> 00:00:31.809 the foundations being laid. So today we're diving 00:00:31.809 --> 00:00:35.210 deep into this expanding AI universe. We want 00:00:35.210 --> 00:00:37.030 to get past the headlines, look at these foundational 00:00:37.030 --> 00:00:39.789 shifts, and maybe uncover some surprising applications 00:00:39.789 --> 00:00:42.270 along the way. Sounds good. So for this deep 00:00:42.270 --> 00:00:44.229 dive, we've got a bit of a roadmap for you. First 00:00:44.229 --> 00:00:46.189 up, we'll really unpack Oracle's big strategic 00:00:46.189 --> 00:00:49.429 move, especially into healthcare AI, and the 00:00:49.429 --> 00:00:52.130 sheer scale of their financial play there. Then 00:00:52.130 --> 00:00:53.950 we'll do a quick run through of some other interesting 00:00:53.950 --> 00:00:56.009 AI highlights, you know, stuff like clever image 00:00:56.009 --> 00:00:59.289 editing tools, but also a pretty concerning data 00:00:59.289 --> 00:01:01.859 privacy issue. Right. And after that, we'll get 00:01:01.859 --> 00:01:04.319 practical, look at some AI tools you can actually 00:01:04.319 --> 00:01:06.700 use, maybe some smart prompt engineering techniques. 00:01:07.060 --> 00:01:10.040 We'll also touch on some brand new AI tools that 00:01:10.040 --> 00:01:12.379 are already changing how people work. Then it's 00:01:12.379 --> 00:01:14.680 sort of quickfire rounded developments, including 00:01:14.680 --> 00:01:17.819 actually really inspiring personal story I came 00:01:17.819 --> 00:01:20.079 across. Okay. And finally, we'll wrap things 00:01:20.079 --> 00:01:22.739 up by looking at a major challenger in the large 00:01:22.739 --> 00:01:25.159 language model space coming from, well, maybe 00:01:25.159 --> 00:01:27.099 an unexpected part of the world. All right, let's 00:01:27.099 --> 00:01:29.950 dive in. So Oracle. Yep. Legacy player, right? 00:01:29.989 --> 00:01:32.090 Almost 50 years old. Yeah, exactly. But they're 00:01:32.090 --> 00:01:36.109 making some really significant strategic moves 00:01:36.109 --> 00:01:38.810 in AI right now, building out major new ventures. 00:01:39.290 --> 00:01:42.730 And a big piece of that is this new Oracle AI 00:01:42.730 --> 00:01:44.689 Center for Healthcare. That's right. It seems 00:01:44.689 --> 00:01:46.549 like a huge commitment. Basically, this center 00:01:46.549 --> 00:01:48.129 is designed to give healthcare organizations 00:01:48.129 --> 00:01:50.790 the whole package. They've got dedicated teams 00:01:50.790 --> 00:01:54.010 of Oracle experts ready to help hospitals test 00:01:54.010 --> 00:01:57.730 things out, build AI solutions, iterate. So it's 00:01:57.730 --> 00:02:00.000 more than just software. Oh, yeah. Think of it 00:02:00.000 --> 00:02:02.579 like a full toolkit. They provide a library of 00:02:02.579 --> 00:02:05.400 frameworks, best practices, implementation guides, 00:02:05.560 --> 00:02:08.819 and crucially, they help navigate all the tricky 00:02:08.819 --> 00:02:12.099 compliance and data privacy stuff. Which is vital 00:02:12.099 --> 00:02:15.460 in healthcare. Absolutely vital. So the center 00:02:15.460 --> 00:02:18.400 prototypes AI agents, think of them as smart 00:02:18.400 --> 00:02:21.020 digital assistants to help with clinical workflows, 00:02:21.340 --> 00:02:24.039 financial stuff, operations. Streamlining things. 00:02:24.300 --> 00:02:26.439 Exactly. And they don't just hand over the tech. 00:02:26.580 --> 00:02:29.639 They offer support for training, for change management, 00:02:29.740 --> 00:02:32.000 to make sure this stuff actually gets used effectively. 00:02:32.300 --> 00:02:34.599 So Oracle's basically handing hospitals a playbook, 00:02:34.659 --> 00:02:37.120 the platform, and the people? Pretty much. So 00:02:37.120 --> 00:02:38.900 hospitals don't have to build everything from 00:02:38.900 --> 00:02:41.000 the ground up. It's a really integrated approach. 00:02:41.400 --> 00:02:43.460 That feels like a fundamental shift, especially 00:02:43.460 --> 00:02:45.800 for an industry like healthcare that can be, 00:02:45.840 --> 00:02:49.360 well, cautious about adopting new tech. And the 00:02:49.360 --> 00:02:52.219 money side of this, you mentioned the scale. 00:02:52.379 --> 00:02:55.199 Yeah, it's honestly one of the most eye -popping 00:02:55.199 --> 00:02:57.759 earnings updates I've seen in AI lately. Oracle 00:02:57.759 --> 00:03:01.740 reported a $455 billion backlog. A billion? With 00:03:01.740 --> 00:03:05.219 a B? With a B. And their year -over -year growth, 00:03:05.400 --> 00:03:09.620 up 359%. Just last quarter, they signed... four 00:03:09.620 --> 00:03:14.180 multi -billion dollar cloud AI deals. Wow. And 00:03:14.180 --> 00:03:16.460 then there's that persistent rumor about the 00:03:16.460 --> 00:03:20.659 $300 billion plus contract with OpenAI. So putting 00:03:20.659 --> 00:03:23.479 this together, what really strikes me is the 00:03:23.479 --> 00:03:25.240 bigger picture. Lots of companies are building, 00:03:25.360 --> 00:03:27.659 you know, GPT wrappers or focused on the front 00:03:27.659 --> 00:03:30.770 end chatbots. Oracle seems to be positioning 00:03:30.770 --> 00:03:33.669 itself as the, well, the underlying infrastructure, 00:03:34.030 --> 00:03:36.389 the mission control centers, maybe the digital 00:03:36.389 --> 00:03:39.449 warehouses for AI across whole industries. Yeah, 00:03:39.509 --> 00:03:41.310 less about the shiny interface, more about the 00:03:41.310 --> 00:03:43.090 engine room. Right. That feels like a really 00:03:43.090 --> 00:03:45.870 foundational shift in how AI gets adopted at 00:03:45.870 --> 00:03:48.449 scale. It's an immense play. So if you had to 00:03:48.449 --> 00:03:50.449 boil it down, what's the biggest long term implication 00:03:50.449 --> 00:03:53.719 of this Oracle move into health care AI? I'd 00:03:53.719 --> 00:03:56.219 say it dramatically streamlines AI adoption for 00:03:56.219 --> 00:03:58.340 a traditionally careful industry. It really sets 00:03:58.340 --> 00:03:59.840 a new standard for these kinds of integrated 00:03:59.840 --> 00:04:01.780 solutions. Makes sense. Streamlining adoption, 00:04:02.080 --> 00:04:03.960 setting a standard. Okay. All right. Let's switch 00:04:03.960 --> 00:04:06.340 gears a bit. We're going to do a more rapid -fire 00:04:06.340 --> 00:04:08.819 look at some diverse AI developments that caught 00:04:08.819 --> 00:04:12.139 our eye. First up, something called NanoBanana. 00:04:12.460 --> 00:04:15.199 nano banana yeah it's this tool for image edits 00:04:15.199 --> 00:04:18.120 and it's apparently exploding in popularity they've 00:04:18.120 --> 00:04:20.459 even put out official prompt templates which 00:04:20.459 --> 00:04:22.620 makes getting those really high quality edits 00:04:22.620 --> 00:04:24.980 you know much more accessible to everyone interesting 00:04:24.980 --> 00:04:28.459 accessibility seems to be a theme i saw something 00:04:28.459 --> 00:04:31.220 similar an entrepreneur put out like a 17 -minute 00:04:31.220 --> 00:04:34.180 tutorial showing how to build these really slick 00:04:34.180 --> 00:04:38.750 high -end motion websites completely using AI 00:04:38.750 --> 00:04:41.290 tools. Oh, wow. It just shows how fast these 00:04:41.290 --> 00:04:43.029 powerful tools are getting into the hands of 00:04:43.029 --> 00:04:45.029 people who aren't necessarily expert coders. 00:04:45.129 --> 00:04:46.689 Definitely. Now, speaking of powerful tools, 00:04:46.870 --> 00:04:49.350 Adobe's making a move here, too. They've launched 00:04:49.350 --> 00:04:52.170 what they're calling a squad of AI agents, kind 00:04:52.170 --> 00:04:55.199 of branded as agent force, but with. Photoshop's 00:04:55.199 --> 00:04:58.639 digital DNA baked in. AgentForce. So what do 00:04:58.639 --> 00:05:01.019 these agents do? They're designed to automate 00:05:01.019 --> 00:05:03.620 workflows in marketing, data analysis, customer 00:05:03.620 --> 00:05:05.759 support. It feels like a pretty big step for 00:05:05.759 --> 00:05:08.259 automating stuff at the enterprise level. But, 00:05:08.360 --> 00:05:10.860 you know, this rapid progress also throws up 00:05:10.860 --> 00:05:12.920 some really important ethical questions, doesn't 00:05:12.920 --> 00:05:16.209 it? Especially around data. Like what? Well, 00:05:16.310 --> 00:05:19.170 there were reports that something like 16 million 00:05:19.170 --> 00:05:22.569 YouTube videos got scraped to train AI models 00:05:22.569 --> 00:05:26.149 from Meta and Microsoft, apparently without explicit 00:05:26.149 --> 00:05:29.750 consent. 16 million. Wow. Including what kind 00:05:29.750 --> 00:05:32.949 of videos? Even things like no film school videos, 00:05:33.189 --> 00:05:35.170 you know, that high quality, pretty specialized 00:05:35.170 --> 00:05:37.110 content. They were reportedly scooped up, too. 00:05:37.209 --> 00:05:39.670 There's even a tool now where creators can check 00:05:39.670 --> 00:05:42.470 if their stuff was used. Huh. That really makes 00:05:42.470 --> 00:05:44.579 you think about like. ownership in the digital 00:05:44.579 --> 00:05:47.459 commons in this whole new AI age. It does. And 00:05:47.459 --> 00:05:50.500 on the job front, any news there? Yeah, actually, 00:05:50.600 --> 00:05:52.899 an AI politics professor offered some predictions 00:05:52.899 --> 00:05:56.579 about which white -collar jobs might be safest 00:05:56.579 --> 00:05:58.699 from automation in the near future. Okay, what's 00:05:58.699 --> 00:06:02.060 the takeaway? Generally, roles that need a lot 00:06:02.060 --> 00:06:04.600 of human touch or complex creative problem solving, 00:06:04.779 --> 00:06:06.959 those seem more likely to stick around or at 00:06:06.959 --> 00:06:09.360 least evolve alongside AI. Makes sense. I have 00:06:09.360 --> 00:06:11.519 to share maybe a slightly vulnerable admission 00:06:11.519 --> 00:06:14.610 here. Earlier... Anthropix Cloud had that brief 00:06:14.610 --> 00:06:18.029 downtime. It wasn't long, but quite a few developers 00:06:18.029 --> 00:06:21.209 I know were, let's just say, not thrilled about 00:06:21.209 --> 00:06:23.589 having to suddenly code entirely by themselves 00:06:23.589 --> 00:06:26.370 again. Ah, yeah. I can imagine. I mean, I still 00:06:26.370 --> 00:06:29.610 wrestle with prompt drift myself sometimes, getting 00:06:29.610 --> 00:06:31.769 the AI to consistently do what I want. So, yeah, 00:06:31.870 --> 00:06:35.089 the panic when your main tool just goes dark. 00:06:35.149 --> 00:06:37.329 I get that. It's just a little reminder of how 00:06:37.329 --> 00:06:39.970 quickly we're becoming reliant on these tools. 00:06:40.110 --> 00:06:42.769 For sure. Beat. Oh, and some quick financial 00:06:42.769 --> 00:06:45.389 news replet, the AI coding startup. They just 00:06:45.389 --> 00:06:47.990 closed a big funding round. Oh, yeah. How big? 00:06:48.170 --> 00:06:50.610 They raised $250 million, putting their valuation 00:06:50.610 --> 00:06:54.029 at $3 billion now. And get this, their annual 00:06:54.029 --> 00:06:56.209 revenue apparently jumped from like $2 .8 million 00:06:56.209 --> 00:07:00.370 to $150 million. Whoa. That's quite a jump. Yeah, 00:07:00.389 --> 00:07:02.389 it just shows the massive demand in that developer 00:07:02.389 --> 00:07:05.009 tool space. Google's AI Futures Fund was one 00:07:05.009 --> 00:07:06.629 of the investors, interestingly. Okay, so back 00:07:06.629 --> 00:07:08.430 to that YouTube scraping issue for a second. 00:07:08.529 --> 00:07:10.269 What does that incident fundamentally reveal 00:07:10.269 --> 00:07:12.740 about? data rights and training these AI models. 00:07:13.000 --> 00:07:16.360 I think it just reveals a really urgent and very 00:07:16.360 --> 00:07:20.040 complex legal and ethical challenge. Yeah. It's 00:07:20.040 --> 00:07:23.160 all about consent and intellectual property in 00:07:23.160 --> 00:07:26.420 this new AI era. Urgent and complex. Yeah, that 00:07:26.420 --> 00:07:29.660 sounds about right. Okay, let's unpack this a 00:07:29.660 --> 00:07:31.420 bit more. Let's pivot now towards the practical 00:07:31.420 --> 00:07:34.160 side. You know, the tools, the techniques we 00:07:34.160 --> 00:07:36.339 can actually use to leverage AI right now. Good 00:07:36.339 --> 00:07:38.600 idea. Like what? Well, Google just rolled out 00:07:38.600 --> 00:07:41.240 a new AI image editor. It's powered by their 00:07:41.240 --> 00:07:44.220 Gemini AI. And this thing lets you pretty much 00:07:44.220 --> 00:07:47.000 alter reality using text commands. Alter reality. 00:07:47.199 --> 00:07:49.519 How so? You can tell it to change the clothes 00:07:49.519 --> 00:07:51.839 someone's wearing in a photo or seamlessly merge 00:07:51.839 --> 00:07:54.339 different photos together, even build whole new 00:07:54.339 --> 00:07:57.740 imaginary worlds just with text in seconds. Wow. 00:07:57.759 --> 00:07:59.579 OK, that makes you wonder, right? What does that 00:07:59.579 --> 00:08:02.509 mean for like? professional photo editing? Is 00:08:02.509 --> 00:08:04.709 it becoming less about the technical fiddling 00:08:04.709 --> 00:08:06.790 and more about just having the vision? That's 00:08:06.790 --> 00:08:08.730 the question, isn't it? And speaking of vision, 00:08:08.810 --> 00:08:10.829 there's also this new framework people are talking 00:08:10.829 --> 00:08:13.709 about, an AI framework for pinpointing gaps in 00:08:13.709 --> 00:08:16.149 digital markets. Okay, pinpointing gaps. How 00:08:16.149 --> 00:08:18.759 does that work? The idea is to validate demand 00:08:18.759 --> 00:08:20.860 for a product or service before you actually 00:08:20.860 --> 00:08:23.540 build it. So this framework teaches you how to 00:08:23.540 --> 00:08:27.420 use really specific targeted AI prompts. These 00:08:27.420 --> 00:08:29.839 prompts analyze market signals, forum discussions, 00:08:30.139 --> 00:08:32.559 search trends, reviews, that kind of stuff to 00:08:32.559 --> 00:08:35.259 help you find potentially profitable ideas where 00:08:35.259 --> 00:08:38.740 there's maybe low supply right now. So using 00:08:38.740 --> 00:08:41.440 AI to find overlooked opportunities with data. 00:08:41.909 --> 00:08:44.529 That's smart. Yeah, exactly. Finding those gaps 00:08:44.529 --> 00:08:47.350 with data -driven precision. And this leads nicely 00:08:47.350 --> 00:08:50.149 into the whole concept of agentic AI, which is 00:08:50.149 --> 00:08:52.490 really starting to take shape. Agentic AI. We 00:08:52.490 --> 00:08:54.769 hear that term a lot. We do. It's captured in 00:08:54.769 --> 00:08:57.490 this idea of your digital colleague is here, 00:08:57.549 --> 00:09:00.190 run your workflow on autopilot. So agentic AI, 00:09:00.409 --> 00:09:02.970 just in plain English, what is it? It's basically 00:09:02.970 --> 00:09:05.970 AI that can string tasks together, can make decisions 00:09:05.970 --> 00:09:08.529 along the way, and really act more like, well, 00:09:08.629 --> 00:09:10.649 a digital teammate. It's not just doing one simple 00:09:10.649 --> 00:09:12.879 command. Got it. So it can handle more complex 00:09:12.879 --> 00:09:15.259 stuff. Exactly. You could delegate things like, 00:09:15.279 --> 00:09:19.200 say, doing competitive analysis or managing invoices, 00:09:19.320 --> 00:09:21.440 maybe even drafting marketing campaign outlines. 00:09:21.580 --> 00:09:25.220 Apparently, there are now like eight proven systems 00:09:25.220 --> 00:09:27.740 you can use to set this kind of thing up effectively. 00:09:28.039 --> 00:09:30.460 OK, that's pretty powerful. So thinking about 00:09:30.460 --> 00:09:33.679 that market gap analysis again, what's the single 00:09:33.679 --> 00:09:37.289 biggest advantage of using AI for that? I'd say 00:09:37.289 --> 00:09:39.730 it lets you validate demand really early on. 00:09:39.809 --> 00:09:42.610 It significantly de -risks starting something 00:09:42.610 --> 00:09:44.490 new before you pour in a bunch of resources. 00:09:44.909 --> 00:09:47.539 Early validation. De -risking. Makes total sense. 00:09:47.740 --> 00:09:49.879 So let's talk about some specific new tools then. 00:09:49.960 --> 00:09:52.200 Things that are actually out there changing workflows. 00:09:52.639 --> 00:09:55.960 Yeah, definitely. Like Seedream 4 .0 is one. 00:09:55.980 --> 00:09:59.259 It's generating really stunning 4K images, apparently 00:09:59.259 --> 00:10:02.379 with much faster inference times. Faster inference, 00:10:02.679 --> 00:10:05.179 meaning the images pop up quicker. Exactly. So 00:10:05.179 --> 00:10:06.960 if you need high -res images, you get them faster. 00:10:07.139 --> 00:10:09.679 That speeds up creative work quite a bit. Okay. 00:10:10.139 --> 00:10:12.399 What else? There's Keyvid. This sounds pretty 00:10:12.399 --> 00:10:14.659 useful. It helps you find specific scenes, objects, 00:10:14.899 --> 00:10:17.139 even facial expressions within your video files. 00:10:17.480 --> 00:10:20.360 Oh, wow. For video editors or creators, that 00:10:20.360 --> 00:10:22.320 must be huge instead of scrubbing through hours 00:10:22.320 --> 00:10:24.440 of footage. Right. Just tell the AI what you're 00:10:24.440 --> 00:10:27.419 looking for. Huge time saver. Then there's NoForm 00:10:27.419 --> 00:10:29.720 AI. This one's focused on websites. Oh, what 00:10:29.720 --> 00:10:32.419 does it do? It's designed to turn website visitors 00:10:32.419 --> 00:10:35.419 into qualified leads and apparently helps drive 00:10:35.419 --> 00:10:39.679 sales. like 247 acts as an always -on assistant 00:10:39.679 --> 00:10:43.519 capturing interest interesting and one more honey 00:10:43.519 --> 00:10:47.139 one image 2 .1 this one creates 2k images but 00:10:47.139 --> 00:10:50.549 the key thing is supposedly really precise text 00:10:50.549 --> 00:10:52.870 rendering within the image. Ah, that's been a 00:10:52.870 --> 00:10:54.590 weakness for a lot of image generators, getting 00:10:54.590 --> 00:10:57.049 text right. Exactly. So this could be a big step 00:10:57.049 --> 00:10:59.889 forward for making AI -generated images with 00:10:59.889 --> 00:11:02.190 text much more usable for, you know, presentations, 00:11:02.389 --> 00:11:04.809 ads, stuff like that. So looking at these tools, 00:11:05.070 --> 00:11:08.730 Seedream, Keyvid, NoForm, HoneyWanImage, what's 00:11:08.730 --> 00:11:10.789 the common thread here? What's really striking 00:11:10.789 --> 00:11:14.000 you? I think it's that they offer these very 00:11:14.000 --> 00:11:16.340 specialized, quite powerful capabilities. They're 00:11:16.340 --> 00:11:19.259 automating or seriously enhancing specific tasks 00:11:19.259 --> 00:11:22.039 that used to be really time consuming. Specialized, 00:11:22.039 --> 00:11:24.340 powerful enhancements. Got it. Mid -roll sponsor 00:11:24.340 --> 00:11:26.740 read. Okay, let's jump back into some quickfire 00:11:26.740 --> 00:11:30.179 updates from the AI world. And, hey, we're hearing 00:11:30.179 --> 00:11:32.759 that OpenAI and Oracle rumor again. The $300 00:11:32.759 --> 00:11:35.820 billion one. Yeah, reportedly signing that massive 00:11:35.820 --> 00:11:39.419 cloud computing deal. It really ties back to 00:11:39.419 --> 00:11:40.879 what we were saying at the start about Oracle 00:11:40.879 --> 00:11:42.879 building that foundational infrastructure. The 00:11:42.879 --> 00:11:46.919 scale is just, well, mind -boggling if it's true. 00:11:47.120 --> 00:11:50.220 Definitely. And Sam Altman, OpenAI CEO, he also 00:11:50.220 --> 00:11:52.340 laid out some of his latest predictions for AI 00:11:52.340 --> 00:11:55.519 recently, shared a kind of roadmap for OpenAI. 00:11:55.779 --> 00:11:57.419 Always interesting to hear his long -term view, 00:11:57.539 --> 00:12:01.279 yeah. What else? Google's AI Max model is apparently 00:12:01.279 --> 00:12:03.600 going global now, and they're including these 00:12:03.600 --> 00:12:07.210 handy one -click experiments. Oh, nice. So making 00:12:07.210 --> 00:12:10.450 powerful AI more accessible worldwide, lowering 00:12:10.450 --> 00:12:12.210 the barrier for developers to play around with 00:12:12.210 --> 00:12:14.149 the cutting edge stuff. Seems like it. And Meta's 00:12:14.149 --> 00:12:15.950 making moves, too. They're planning to pay, I 00:12:15.950 --> 00:12:19.690 think it was $140 million to use AI from a company 00:12:19.690 --> 00:12:22.450 called Black Forest Labs specifically for image 00:12:22.450 --> 00:12:25.429 generation. So buying specialized tech instead 00:12:25.429 --> 00:12:27.210 of building everything themselves. Looks like 00:12:27.210 --> 00:12:29.710 it. Maybe a strategic play for niche expertise 00:12:29.710 --> 00:12:33.029 seems to be a trend. Yeah, makes sense. Okay, 00:12:33.090 --> 00:12:35.289 this next one. This one really struck me. It's 00:12:35.289 --> 00:12:36.750 kind of a moment of wonder, I guess. Oh, yeah. 00:12:36.909 --> 00:12:39.570 Tell me. It's a story about a cancer patient 00:12:39.570 --> 00:12:44.429 who basically built his own personalized AI medical 00:12:44.429 --> 00:12:47.350 team using tools that are pretty widely accessible. 00:12:47.610 --> 00:12:51.220 Wow. Built his own AI medical team. What does 00:12:51.220 --> 00:12:53.860 that even mean? I think it means using AI to 00:12:53.860 --> 00:12:57.139 research, track symptoms, maybe analyze treatment 00:12:57.139 --> 00:12:59.659 options in a really personalized way. Just, whoa, 00:12:59.879 --> 00:13:03.379 imagine scaling that kind of personalized, empowering 00:13:03.379 --> 00:13:06.139 approach for millions of people facing health 00:13:06.139 --> 00:13:07.899 challenges. Yeah, that's incredible. It really 00:13:07.899 --> 00:13:10.299 speaks to the potential human impact, doesn't 00:13:10.299 --> 00:13:12.919 it? So what does that story, the cancer patient 00:13:12.919 --> 00:13:15.080 building his own AI team, what does that really 00:13:15.080 --> 00:13:18.240 tell us about AI's potential? I think it profoundly 00:13:18.240 --> 00:13:21.559 illustrates AI's capacity, its capacity for highly 00:13:21.559 --> 00:13:24.919 personalized, accessible and deeply empowering 00:13:24.919 --> 00:13:27.399 applications right in the hands of individuals. 00:13:27.639 --> 00:13:29.960 Personalized, accessible, empowering. That's 00:13:29.960 --> 00:13:33.009 powerful. Okay, let's shift focus now, geographically 00:13:33.009 --> 00:13:35.090 maybe. We need to talk about a major development 00:13:35.090 --> 00:13:37.250 from China. You really don't want to sleep on 00:13:37.250 --> 00:13:39.509 their large language model game. Right, you're 00:13:39.509 --> 00:13:42.190 talking about Baidu. Exactly. Baidu's Ernie X1 00:13:42.190 --> 00:13:44.769 .1. It just launched at their Wave Summit 2025. 00:13:45.470 --> 00:13:48.590 And this looks like a direct, very serious competitor 00:13:48.590 --> 00:13:52.409 to models like GPT -5, Gemini 2 .5 Pro. We're 00:13:52.409 --> 00:13:55.559 seeing a real global race heat up. And it's not 00:13:55.559 --> 00:13:57.840 just the model itself. There's a whole ecosystem 00:13:57.840 --> 00:13:59.940 Baidu's built around it. Tell me about that. 00:14:00.159 --> 00:14:01.740 Well, they rolled out what they call a full stack 00:14:01.740 --> 00:14:04.559 developer ecosystem. They open sourced a model 00:14:04.559 --> 00:14:08.399 with 128K context window, which is huge. And 00:14:08.399 --> 00:14:11.320 they've given their AI coding assistant significant 00:14:11.320 --> 00:14:14.519 new superpowers, as they put it. So lots of tools 00:14:14.519 --> 00:14:16.659 for developers. Yeah, this whole paddle ecosystem, 00:14:16.860 --> 00:14:18.980 as they call it. It gives developers basically 00:14:18.980 --> 00:14:21.299 everything they need to build and deploy advanced 00:14:21.299 --> 00:14:25.450 AI stuff using Ernie. And the model itself, or 00:14:25.450 --> 00:14:28.389 NEX 1 .1, got a big upgrade in its reasoning 00:14:28.389 --> 00:14:30.269 abilities. Okay, reasoning upgrade. What kind 00:14:30.269 --> 00:14:32.169 of improvements are we talking? The numbers are 00:14:32.169 --> 00:14:34.250 pretty specific. Factuality is apparently up 00:14:34.250 --> 00:14:37.149 almost 35%. Instruction following improved by 00:14:37.149 --> 00:14:40.230 12 .5%. And it's agentic reasoning, that ability 00:14:40.230 --> 00:14:42.649 to plan and execute multiple steps, is up nearly 00:14:42.649 --> 00:14:45.669 10%. Those are solid jumps, especially that agentic 00:14:45.669 --> 00:14:48.230 part. Definitely. And just to reinforce that 00:14:48.230 --> 00:14:50.750 term, agentic, like we discussed earlier, it 00:14:50.750 --> 00:14:52.769 means the AI isn't just answering one question. 00:14:53.279 --> 00:14:56.100 It can chain tasks, make decisions along the 00:14:56.100 --> 00:14:58.679 way, act like that intelligent digital teammate. 00:14:58.779 --> 00:15:01.179 That's crucial for complex real world stuff. 00:15:01.360 --> 00:15:03.120 And how does it stack up against the competition? 00:15:03.840 --> 00:15:05.899 Well, the benchmarks look pretty striking. They're 00:15:05.899 --> 00:15:10.299 saying Ernie X 1 .1 beats DeepSeek R10528, which 00:15:10.299 --> 00:15:13.220 is a capable model itself. And crucially, it 00:15:13.220 --> 00:15:16.559 apparently matches GPT -5 and Gemini 2 .5 Pro 00:15:16.559 --> 00:15:19.360 on several key benchmarks. Matches them. Okay, 00:15:19.419 --> 00:15:21.440 that's a serious claim. It is. It shows it's 00:15:21.440 --> 00:15:24.240 really advanced and ready to compete. And here's 00:15:24.240 --> 00:15:26.159 maybe the kicker for people who want to use it. 00:15:26.399 --> 00:15:28.320 What's that? There's no wait list. Oh, really? 00:15:28.519 --> 00:15:30.799 Yeah. Ernie X 1 .1 is apparently already available 00:15:30.799 --> 00:15:33.440 on their Ernie bot web platform. So if you're 00:15:33.440 --> 00:15:36.379 building agents or assistants or apps. It's production 00:15:36.379 --> 00:15:38.860 ready right now. That immediate access is a big 00:15:38.860 --> 00:15:40.960 advantage. No kidding. And the scale of their 00:15:40.960 --> 00:15:43.460 developer community. It's quietly massive. They're 00:15:43.460 --> 00:15:47.960 reporting 23 .33 million developers and 750 ,000 00:15:47.960 --> 00:15:50.580 enterprise users already using their paddle ecosystem. 00:15:50.899 --> 00:15:54.139 Wow. That's a huge base. It really is. Yeah. 00:15:54.200 --> 00:15:58.360 So, yeah, Ernie X 1 .1 feels like a main character 00:15:58.360 --> 00:16:00.919 in the global AI story now. It's not just hype. 00:16:01.289 --> 00:16:04.009 Okay, so bring that all together. What makes 00:16:04.009 --> 00:16:07.850 Ernie X 1 .1 such a significant and maybe immediate 00:16:07.850 --> 00:16:11.929 competitor on the world stage? I'd say it's the 00:16:11.929 --> 00:16:15.029 combination. It's advanced reasoning, that broad 00:16:15.029 --> 00:16:17.669 developer ecosystem supporting it, and the fact 00:16:17.669 --> 00:16:19.610 that it's immediately available and production 00:16:19.610 --> 00:16:22.049 ready. That makes it a really formidable global 00:16:22.049 --> 00:16:24.509 challenger right now. Advanced reasoning, broad 00:16:24.509 --> 00:16:28.019 ecosystem, immediate availability. Got it. OK, 00:16:28.120 --> 00:16:30.000 let's try and synthesize this whole deep dive 00:16:30.000 --> 00:16:32.000 then. We've covered a lot of ground. We really 00:16:32.000 --> 00:16:33.899 have. We started with Oracle kind of building 00:16:33.899 --> 00:16:36.779 the foundational warehouse for AI across industries. 00:16:37.080 --> 00:16:39.059 Then we zipped through a bunch of rapid fire 00:16:39.059 --> 00:16:41.360 innovations, some simplifying everyday tasks, 00:16:41.460 --> 00:16:43.500 some raising tricky questions. Right. We touched 00:16:43.500 --> 00:16:45.720 on practical tools like for market analysis or 00:16:45.720 --> 00:16:48.120 image editing. And then we landed on these powerful 00:16:48.120 --> 00:16:50.860 global challengers like Baidu's Ernie. What's 00:16:50.860 --> 00:16:53.059 really striking looking back is just how many 00:16:53.059 --> 00:16:55.639 different fronts AI is advancing on all at once. 00:16:55.840 --> 00:16:58.080 Totally. You've got the massive infrastructure 00:16:58.080 --> 00:17:00.539 build out. You've got these incredibly specific 00:17:00.539 --> 00:17:03.460 applications popping up. And then this intense 00:17:03.460 --> 00:17:05.920 global competition driving things forward. It's 00:17:05.920 --> 00:17:09.039 definitely not just about one shiny new toy anymore. 00:17:09.160 --> 00:17:11.720 It feels like this huge interconnected ecosystem 00:17:11.720 --> 00:17:15.039 that's just growing and transforming right in 00:17:15.039 --> 00:17:18.299 front of us. Yeah, moving fast. So here's a final 00:17:18.299 --> 00:17:20.160 thought, maybe something for you to consider. 00:17:20.829 --> 00:17:24.589 As AI becomes more agentic, as it starts taking 00:17:24.589 --> 00:17:27.029 on more complex tasks, making more decisions, 00:17:27.349 --> 00:17:30.250 what are the new skills we are going to need? 00:17:30.490 --> 00:17:33.829 How do we learn to collaborate effectively with 00:17:33.829 --> 00:17:36.390 these emerging digital teammates? That's a great 00:17:36.390 --> 00:17:38.369 question. Definitely something to mull over. 00:17:38.549 --> 00:17:41.029 And we'd encourage you, our listeners, keep doing 00:17:41.029 --> 00:17:43.450 your own deep dives. Maybe explore prompt engineering 00:17:43.450 --> 00:17:46.309 more or look into how these agentic systems are 00:17:46.309 --> 00:17:48.069 actually being built. There's always more to 00:17:48.069 --> 00:17:50.380 learn. Absolutely. Well, thanks for joining us 00:17:50.380 --> 00:17:52.180 on the deep dive. Yeah. Thanks for listening. 00:17:52.240 --> 00:17:54.259 We'll catch you next time. Out to your own music.