WEBVTT 00:00:00.000 --> 00:00:01.980 Welcome to the Deep Dive. We're really glad you're 00:00:01.980 --> 00:00:05.200 joining us today, ready to take a closer look 00:00:05.200 --> 00:00:07.219 at some significant developments happening right 00:00:07.219 --> 00:00:10.759 now in the fast -moving world of AI. Yep, definitely 00:00:10.759 --> 00:00:13.480 fast -moving. We've got this stack of sources 00:00:13.480 --> 00:00:17.280 here. It's a mix of recent news excerpts, a couple 00:00:17.280 --> 00:00:19.660 of market reports, and some quick highlights 00:00:19.660 --> 00:00:22.059 pulled from various places. That's right. And 00:00:22.059 --> 00:00:24.260 our mission for this Deep Dive is basically to 00:00:24.260 --> 00:00:26.980 unpack these pieces of news, get past just the 00:00:26.980 --> 00:00:29.440 headlines. and figure out what's the most important 00:00:29.440 --> 00:00:31.239 stuff buried within them. What does it really 00:00:31.239 --> 00:00:33.960 mean? Exactly. What does it really mean for the 00:00:33.960 --> 00:00:36.740 evolving AI landscape that you're navigating? 00:00:37.119 --> 00:00:39.259 Right. So we're going to dig into the details 00:00:39.259 --> 00:00:42.780 that matter. We'll be talking about OpenAI's 00:00:42.780 --> 00:00:45.579 new model that's quite the brainiac, apparently. 00:00:45.960 --> 00:00:48.219 Some pretty interesting moves Google is making, 00:00:48.259 --> 00:00:51.399 not just in video, but also... kind of surprisingly, 00:00:51.560 --> 00:00:54.179 in government efficiency. And yeah, honestly, 00:00:54.320 --> 00:00:57.159 a bunch of other fascinating tidbits that really 00:00:57.159 --> 00:00:59.039 show us, you know, where things are heading and 00:00:59.039 --> 00:01:01.500 the sheer breadth of AI's impact right now. It's 00:01:01.500 --> 00:01:04.939 pretty wild. Okay, let's unpack this stack. First 00:01:04.939 --> 00:01:08.040 up, the big news coming out of OpenAI. They have 00:01:08.040 --> 00:01:11.140 just unveiled a new model. It's called O3 Pro. 00:01:11.459 --> 00:01:14.560 Right, O3 Pro. And what's fascinating here, and 00:01:14.560 --> 00:01:17.239 the sources make this super clear, is its core 00:01:17.239 --> 00:01:20.859 identity. This model isn't positioned as another 00:01:20.859 --> 00:01:24.159 lightning -fast chatbot built for casual conversation. 00:01:24.560 --> 00:01:28.180 It's specifically described as a reasoning -focused 00:01:28.180 --> 00:01:31.099 AI model. It's an evolution from their previous 00:01:31.099 --> 00:01:35.040 O3, but it's fundamentally engineered with thinking 00:01:35.040 --> 00:01:38.209 and logic as its primary function. reasoning 00:01:38.209 --> 00:01:40.650 first yeah i kind of love that framing built 00:01:40.650 --> 00:01:43.269 for thinking and the source material really highlights 00:01:43.269 --> 00:01:45.049 where this focus pays off we're not talking about 00:01:45.049 --> 00:01:46.769 like writing a quick email here no definitely 00:01:46.769 --> 00:01:49.810 not the areas it shines in uh are quite demanding 00:01:49.810 --> 00:01:52.709 coding math science complex academic writing 00:01:52.709 --> 00:01:54.709 and detailed business analysis these are places 00:01:54.709 --> 00:01:56.930 where you know you need robust step -by -step 00:01:56.930 --> 00:01:59.170 logic not just speed right and it's not just 00:01:59.170 --> 00:02:02.049 some sort of uh experimental model being tested 00:02:02.049 --> 00:02:04.209 in a lab somewhere it's actually already rolled 00:02:04.209 --> 00:02:07.540 out Oh, really? Yeah, as of Juneteenth, it replaced 00:02:07.540 --> 00:02:11.360 O1 Pro in their ChatGPT Pro and Team plans. So 00:02:11.360 --> 00:02:13.680 if you're using one of those, you now have access 00:02:13.680 --> 00:02:15.740 to this model. It's also available via their 00:02:15.740 --> 00:02:19.969 API, and the sources give specific pricing. Let's 00:02:19.969 --> 00:02:22.669 see, $20 million input tokens and $80 per million 00:02:22.669 --> 00:02:25.370 output tokens. Okay, so the reasoning first approach. 00:02:25.610 --> 00:02:27.770 How does that actually translate into how the 00:02:27.770 --> 00:02:30.110 model works? What's the, like, technical difference 00:02:30.110 --> 00:02:32.330 the sources hint at? Well, this raises an important 00:02:32.330 --> 00:02:35.250 point about model architecture and design philosophy, 00:02:35.370 --> 00:02:37.849 right? Instead of simply predicting the next 00:02:37.849 --> 00:02:40.710 most statistically probable word or phrase in 00:02:40.710 --> 00:02:43.750 a rapid sequence, which is how many models achieve 00:02:43.750 --> 00:02:46.889 speed O3 pro, is designed to process information 00:02:46.889 --> 00:02:50.150 more deliberately. It's built to simulate a step 00:02:50.150 --> 00:02:53.050 by step problem solving process. It kind of thinks 00:02:53.050 --> 00:02:55.030 things through methodically. That's the fundamental 00:02:55.030 --> 00:02:56.949 difference driving its capabilities in those 00:02:56.949 --> 00:02:59.289 complex domains we just mentioned. And the sources 00:02:59.289 --> 00:03:01.330 give us some pretty compelling proof points that 00:03:01.330 --> 00:03:03.710 this isn't just marketing fluff. They show benchmarks 00:03:03.710 --> 00:03:06.770 where it, well, frankly, outsmarts some of the 00:03:06.770 --> 00:03:09.349 top competitors on specific difficult tests. 00:03:09.710 --> 00:03:12.770 Yes, the benchmark results provided are quite 00:03:12.770 --> 00:03:15.310 telling and really give weight to that reasoning 00:03:15.310 --> 00:03:18.389 -focused claim. For instance, it beat Google's 00:03:18.389 --> 00:03:23.500 Gemini 2 .5 Pro on the AIME 2024 math test. Amy, 00:03:23.599 --> 00:03:25.680 and that's like serious math, right? Yeah. Amy 00:03:25.680 --> 00:03:28.020 is the American Invitational Mathematics Examination. 00:03:28.020 --> 00:03:29.840 It's a very challenging competition math test, 00:03:29.979 --> 00:03:33.379 way beyond simple arithmetic. Beating a top competitor 00:03:33.379 --> 00:03:35.639 there is significant. Oh, wow. So not just high 00:03:35.639 --> 00:03:38.379 school math, but like competitive math. Exactly. 00:03:38.580 --> 00:03:42.060 And it also beat Claude For Opus on the GPQA 00:03:42.060 --> 00:03:44.919 diamond test. GPQA. What's that? GPQA stands 00:03:44.919 --> 00:03:46.919 for Graduate Level General Knowledge Question 00:03:46.919 --> 00:03:49.900 Answering. The Diamond subset is specifically 00:03:49.900 --> 00:03:52.419 curated for extremely difficult questions that 00:03:52.419 --> 00:03:54.580 often require deep, nuanced understanding across 00:03:54.580 --> 00:03:57.139 various scientific fields. It's essentially testing 00:03:57.139 --> 00:03:59.479 PhD -level science knowledge and reasoning. PhD 00:03:59.479 --> 00:04:01.900 -level science. OK. Yeah. And expert reviewers 00:04:01.900 --> 00:04:05.439 who tested O3 Pro alongside O1 Pro and O3 also 00:04:05.439 --> 00:04:07.419 consistently ranked it higher across various 00:04:07.419 --> 00:04:09.800 tested areas, confirming the perceived improvement 00:04:09.800 --> 00:04:13.520 in logical processing. Beating top models on 00:04:13.520 --> 00:04:16.639 both challenging math and complex multidisciplinary 00:04:16.639 --> 00:04:19.519 science, that really does back up the idea that 00:04:19.519 --> 00:04:21.399 it's designed for deeper thinking. It seems so. 00:04:21.759 --> 00:04:24.180 And it still has all the modern AI capabilities, 00:04:24.439 --> 00:04:27.300 right? Like browsing the web, using Python code 00:04:27.300 --> 00:04:29.740 interpreters, analyzing documents, processing 00:04:29.740 --> 00:04:33.019 visual inputs, even using memory for personalized 00:04:33.019 --> 00:04:35.259 interaction. Correct. It's fully featured in 00:04:35.259 --> 00:04:38.040 terms of tool access and multimodal capabilities. 00:04:38.139 --> 00:04:40.519 You get that enhanced reasoning, plus the ability 00:04:40.519 --> 00:04:42.779 to interact with various data types and tools. 00:04:42.939 --> 00:04:45.639 Okay, so you get all that power. But the sources 00:04:45.639 --> 00:04:47.620 also point out a significant trade -off, right? 00:04:47.660 --> 00:04:49.730 There's got to be a catch. They do. And this 00:04:49.730 --> 00:04:52.209 is crucial. The sources emphasize that responses 00:04:52.209 --> 00:04:55.470 from O3 Pro are notably slower than those from 00:04:55.470 --> 00:04:58.730 its predecessor, O1 Pro, and certainly slower 00:04:58.730 --> 00:05:01.670 than models optimized purely for speed. And this 00:05:01.670 --> 00:05:03.910 isn't a bug. It's a direct consequence of its 00:05:03.910 --> 00:05:07.040 design. That deeper step -by -step reasoning 00:05:07.040 --> 00:05:09.740 process simply takes more computational time 00:05:09.740 --> 00:05:11.579 than quick prediction. It just takes longer to 00:05:11.579 --> 00:05:14.180 think. Okay, so what does this all mean? Why 00:05:14.180 --> 00:05:17.480 would open AI in a market obsessed with speed 00:05:17.480 --> 00:05:21.560 release a model that is explicitly slower? What's 00:05:21.560 --> 00:05:23.709 the play here? This connects directly to the 00:05:23.709 --> 00:05:26.470 bigger picture of OpenAI's strategic positioning, 00:05:26.589 --> 00:05:29.250 according to the sources anyway. They frame O3 00:05:29.250 --> 00:05:31.629 Pro as their response to what they see as a growing 00:05:31.629 --> 00:05:34.449 issue with some speedy AI models in the market. 00:05:34.629 --> 00:05:37.050 Right. Which they imply can hallucinate too much 00:05:37.050 --> 00:05:39.430 or produce illogical outputs when under pressure 00:05:39.430 --> 00:05:42.310 to respond instantly. This new model is explicitly 00:05:42.310 --> 00:05:44.889 designed for reliability and trust in complex 00:05:44.889 --> 00:05:48.170 tasks over sheer speed. Ah, gotcha. So they're 00:05:48.170 --> 00:05:50.480 splitting their offerings. kind of specializing. 00:05:50.660 --> 00:05:52.680 Precisely. It looks like a deliberate split lineup 00:05:52.680 --> 00:05:54.939 strategy. Yeah. You have GPT -4 -0, which is 00:05:54.939 --> 00:05:56.959 incredibly fast, great for real -time interactions, 00:05:57.300 --> 00:05:59.459 handles multimodal inputs, seamlessly good for 00:05:59.459 --> 00:06:01.759 creative tasks, quick summaries, conversations. 00:06:02.079 --> 00:06:04.920 Yeah, the flashy one. Kind of. And now you have 00:06:04.920 --> 00:06:07.259 O3 -FRO, which is purpose -built for deep logic, 00:06:07.379 --> 00:06:09.939 accuracy, and trust in those highly demanding, 00:06:10.040 --> 00:06:12.319 reasoning -heavy applications. Okay, that makes 00:06:12.319 --> 00:06:14.800 a lot of sense. So for you, the listener, this 00:06:14.800 --> 00:06:16.519 distinction is really important because it means 00:06:16.519 --> 00:06:19.490 choosing the right tool for the specific job 00:06:19.490 --> 00:06:22.990 you need done. If you need rapid creative brainstorming, 00:06:22.990 --> 00:06:25.329 quick information retrieval, or just conversational 00:06:25.329 --> 00:06:28.930 flow, GPC 4 .0 might be your best bet. But if 00:06:28.930 --> 00:06:31.709 you're tackling a complex coding problem, analyzing 00:06:31.709 --> 00:06:35.029 dense financial reports, writing a detailed academic 00:06:35.029 --> 00:06:38.370 paper, or trying to solve a difficult scientific 00:06:38.370 --> 00:06:40.930 query where the accuracy and soundness of the 00:06:40.930 --> 00:06:43.829 logic are paramount. Yeah, where you really needed 00:06:43.829 --> 00:06:47.089 to be right. Then O3 Pro is specifically designed 00:06:47.089 --> 00:06:49.689 for that kind of deep work, even if it takes 00:06:49.689 --> 00:06:51.410 a little longer to give you the answer. It's 00:06:51.410 --> 00:06:53.629 about balancing speed with trustworthiness depending 00:06:53.629 --> 00:06:56.089 on your task. Exactly. It caters to different 00:06:56.089 --> 00:06:58.250 user needs by offering specialized strengths. 00:06:58.569 --> 00:07:00.870 Makes sense. All right. Shifting gears a bit, 00:07:00.930 --> 00:07:03.149 let's look at what Google has been up to. According 00:07:03.149 --> 00:07:05.050 to these sources, they've got some pretty diverse 00:07:05.050 --> 00:07:07.370 things happening, too. Definitely. On the creative 00:07:07.370 --> 00:07:09.689 side, building on their existing capabilities, 00:07:10.050 --> 00:07:13.250 they've unveiled VO3 Fast. This is an update 00:07:13.250 --> 00:07:15.949 to their video generation tool. VO3 fast. Okay. 00:07:16.069 --> 00:07:19.230 The key highlight here is speed vidits. The sources 00:07:19.230 --> 00:07:21.629 say it's generating videos two times faster now. 00:07:21.889 --> 00:07:24.790 They're also mentioning improved serving optimizations, 00:07:24.829 --> 00:07:27.310 which implies it's also getting better at delivering 00:07:27.310 --> 00:07:30.389 those videos efficiently. It maintains a 720p 00:07:30.389 --> 00:07:34.339 resolution as well. Two times faster is... Pretty 00:07:34.339 --> 00:07:36.420 significant for video generation. That can be 00:07:36.420 --> 00:07:38.560 a bottleneck, right? Waiting around for renders. 00:07:38.600 --> 00:07:40.920 Oh, absolutely. Cut sound waiting time. And there's 00:07:40.920 --> 00:07:43.100 this wild user example the source has pointed 00:07:43.100 --> 00:07:47.139 to. Someone apparently used VO3 for these Stormtrooper 00:07:47.139 --> 00:07:49.540 -style vlogs. Yeah, the Stormtrooper -style vlogs. 00:07:50.670 --> 00:07:52.790 Quite a mental image. The report mentioned that 00:07:52.790 --> 00:07:54.629 one such account reportedly garnered something 00:07:54.629 --> 00:07:56.930 like 8 million views on Instagram in a single 00:07:56.930 --> 00:08:00.509 day using videos generated with VO3. Whoa! 8 00:08:00.509 --> 00:08:03.230 million views in one day. That's a crazy viral. 00:08:03.449 --> 00:08:06.189 Just for stormtrooper blogs. It really underscores 00:08:06.189 --> 00:08:08.730 the power of combining a novel creative idea 00:08:08.730 --> 00:08:12.389 with accessible, fast -generation tools. You 00:08:12.389 --> 00:08:14.589 can produce content at a volume and speed previously 00:08:14.589 --> 00:08:17.050 impossible, and apparently turning everyone into 00:08:17.050 --> 00:08:19.569 a stormtrooper resonates with 8 million people 00:08:19.569 --> 00:08:22.779 very quickly. Who knew? Unbelievable. OK, so 00:08:22.779 --> 00:08:26.860 from viral stormtrooper vlogs to government bureaucracy, 00:08:27.180 --> 00:08:29.819 the sources also mention Google partnering with 00:08:29.819 --> 00:08:32.039 the UK government. That seems like a jump. Yes. 00:08:32.100 --> 00:08:34.779 And in my view, this particular application is 00:08:34.779 --> 00:08:37.179 one of the most practical and immediately impactful 00:08:37.179 --> 00:08:40.840 deployments highlighted in the sources. Google's 00:08:40.840 --> 00:08:43.120 Gemini extract is being leveraged to tackle a 00:08:43.120 --> 00:08:46.879 massive real world bottleneck in the UK public 00:08:46.879 --> 00:08:49.980 sector. The incredibly slow infrastructure planning 00:08:49.980 --> 00:08:52.480 process. Right. Think about everything involved 00:08:52.480 --> 00:08:54.759 in getting approval to build houses, roads or 00:08:54.759 --> 00:08:57.539 other essential infrastructure. Just mountains 00:08:57.539 --> 00:09:00.700 of paperwork. Government paperwork. And planning 00:09:00.700 --> 00:09:03.600 documents specifically can be a notoriously complex, 00:09:03.860 --> 00:09:06.200 messy thing. Absolutely. And the problem is often 00:09:06.200 --> 00:09:08.289 the format. These aren't always neat digital 00:09:08.289 --> 00:09:11.490 files. Extract is designed to scan and process 00:09:11.490 --> 00:09:14.769 incredibly messy, handwritten or scanned planning 00:09:14.769 --> 00:09:17.250 documents. OK. And the source is specific here. 00:09:17.309 --> 00:09:19.590 It can handle things like blurry maps and even 00:09:19.590 --> 00:09:21.669 handwritten notes scrolled in the margins. Stuff 00:09:21.669 --> 00:09:23.710 that is usually really hard for computers. Oh, 00:09:23.730 --> 00:09:26.649 wow. So it's not just OCR on a clean typed page. 00:09:26.769 --> 00:09:30.789 It's dealing with like. The real messy analog 00:09:30.789 --> 00:09:34.190 world, coffee stains and all. Exactly. It's built 00:09:34.190 --> 00:09:36.850 to interpret and understand unstructured data 00:09:36.850 --> 00:09:40.409 that isn't in a standard digital format. Its 00:09:40.409 --> 00:09:43.470 job is to convert that physical, often chaotic 00:09:43.470 --> 00:09:47.090 information, whether it's a faded stamp, a drawing 00:09:47.090 --> 00:09:50.149 on a map, or handwritten comments into searchable, 00:09:50.190 --> 00:09:53.129 structured digital data. Ah, searchable and structured. 00:09:53.269 --> 00:09:55.470 That's key. Data that planners and decision makers 00:09:55.470 --> 00:09:57.750 can actually work with efficiently in a database 00:09:57.750 --> 00:10:00.220 or system. OK, so it doesn't just digitize an 00:10:00.220 --> 00:10:02.960 image. It makes the information within that image 00:10:02.960 --> 00:10:05.539 usable and searchable. That's a big step. And 00:10:05.539 --> 00:10:07.779 the statistic quoted in the trials for this is 00:10:07.779 --> 00:10:10.720 pretty jaw dropping. It really is. According 00:10:10.720 --> 00:10:12.500 to the early trials mentioned in the sources, 00:10:12.639 --> 00:10:14.779 a process that previously took a human planner 00:10:14.779 --> 00:10:17.620 two hours of manual work to extract key information 00:10:17.620 --> 00:10:21.000 is cut down to just 40 seconds using Gemini extract. 00:10:21.200 --> 00:10:24.360 40 seconds from two hours. I mean, that's a monumental 00:10:24.360 --> 00:10:27.259 efficiency gain for that specific task. It's 00:10:27.259 --> 00:10:30.299 a concrete. quantifiable benefit. And the goal 00:10:30.299 --> 00:10:33.519 here is clear and directly addresses a major 00:10:33.519 --> 00:10:36.700 public sector issue. By accelerating the data 00:10:36.700 --> 00:10:39.080 extraction from these planning documents, they 00:10:39.080 --> 00:10:42.419 can speed up notoriously slow decisions on infrastructure 00:10:42.419 --> 00:10:45.299 and housing projects in the UK. Makes sense. 00:10:45.460 --> 00:10:48.620 It removes that incredibly tedious, mind -numbing 00:10:48.620 --> 00:10:51.600 manual data entry, freeing up trained planners 00:10:51.600 --> 00:10:54.519 to focus on their actual expertise making informed 00:10:54.519 --> 00:10:57.120 planning decisions. It's designed to cut through 00:10:57.120 --> 00:11:00.179 the massive backlogs that have reportedly stalled 00:11:00.179 --> 00:11:02.129 development for years. That feels like such a 00:11:02.129 --> 00:11:04.870 powerful practical use case, not some, you know, 00:11:04.870 --> 00:11:08.250 far off futuristic concept, but using AI to fix 00:11:08.250 --> 00:11:10.809 a deeply rooted systemic problem that has real 00:11:10.809 --> 00:11:12.669 consequences for things like housing shortages. 00:11:12.909 --> 00:11:15.049 It absolutely is. The source specifically highlights 00:11:15.049 --> 00:11:17.230 this, calling it one of the most practical AI 00:11:17.230 --> 00:11:19.590 deployments yet in the public sector. High praise. 00:11:19.789 --> 00:11:22.950 And this move doesn't just solve a UK problem. 00:11:23.090 --> 00:11:26.990 It also significantly strengthens Google's position 00:11:26.990 --> 00:11:31.000 in the enterprise and government AI market. By 00:11:31.000 --> 00:11:34.539 directly unblocking these slow, data -intensive 00:11:34.539 --> 00:11:37.659 processes, they're supporting a major national 00:11:37.659 --> 00:11:41.279 target, like the UK's goal of building 1 .5 million 00:11:41.279 --> 00:11:45.500 homes. It's AI solving a fundamental real world 00:11:45.500 --> 00:11:48.879 bottleneck using existing documents. That's incredibly 00:11:48.879 --> 00:11:50.899 insightful. OK, let's zoom out a bit and hit 00:11:50.899 --> 00:11:52.580 some of the other quick takes from the sources, 00:11:52.639 --> 00:11:54.179 because there are quite a few other interesting 00:11:54.179 --> 00:11:56.120 nuggets that give us a broader picture of the 00:11:56.120 --> 00:11:58.379 AI landscape right now. Sure. Yeah. There are 00:11:58.379 --> 00:12:00.279 a number of points that paint a wider picture. 00:12:00.519 --> 00:12:04.039 For instance, the recent chat GPT worldwide outage. 00:12:04.039 --> 00:12:05.940 Oh, yeah. I remember seeing reports on that down 00:12:05.940 --> 00:12:08.720 detector lit up with nearly 2000 reports of it 00:12:08.720 --> 00:12:11.500 just being down globally for a bit. Right. While 00:12:11.500 --> 00:12:13.799 it was a temporary inconvenience for many, it 00:12:13.799 --> 00:12:15.899 really just served as a stark reminder of something 00:12:15.899 --> 00:12:18.799 important, a rapidly growing reliance on these 00:12:18.799 --> 00:12:21.100 AI systems. Totally. When they're integral to 00:12:21.100 --> 00:12:23.700 so many workflows, even a relatively short outage 00:12:23.700 --> 00:12:26.080 highlights how dependent we're becoming. Things 00:12:26.080 --> 00:12:28.059 just stop. Yeah, it makes you think about the 00:12:28.059 --> 00:12:30.559 infrastructure supporting all this. And then 00:12:30.559 --> 00:12:33.259 there was this little kerfuffle the sources mentioned 00:12:33.259 --> 00:12:37.139 about X users kind of dragging Apple. What was 00:12:37.139 --> 00:12:39.779 that about? Ah, yes. That was a bit of digital 00:12:39.779 --> 00:12:42.240 commentary. Apparently, Apple published some 00:12:42.240 --> 00:12:45.480 research or analysis pointing out perceived flaws 00:12:45.480 --> 00:12:48.799 or limitations in current AI reasoning models. 00:12:49.000 --> 00:12:52.879 Okay. And some users on X were quick to retort, 00:12:52.899 --> 00:12:56.100 essentially pointing out that while Apple is 00:12:56.100 --> 00:12:58.419 critiquing reasoning models, they haven't actually 00:12:58.419 --> 00:13:00.500 launched their own foundational large language 00:13:00.500 --> 00:13:03.000 model yet. A bit of a glass houses situation, 00:13:03.340 --> 00:13:06.519 maybe. Pot calling the kettle black. Slight chuckle. 00:13:06.750 --> 00:13:08.669 Something like that. It sparked a bit of debate 00:13:08.669 --> 00:13:11.129 online, though, about the state of LLMs and who 00:13:11.129 --> 00:13:13.370 has the right to critique whom. It really shows 00:13:13.370 --> 00:13:14.909 the competitive heat in the space right now. 00:13:14.970 --> 00:13:17.250 Yeah, definitely. But OK, completely different 00:13:17.250 --> 00:13:18.909 note. Here's one that really surprised me and 00:13:18.909 --> 00:13:21.110 shows the unexpected places AI is going on medical 00:13:21.110 --> 00:13:24.169 application. Doctors at Columbia University reportedly 00:13:24.169 --> 00:13:27.149 used AI to help a couple with 19 years of infertility 00:13:27.149 --> 00:13:30.230 finally achieve pregnancy. That story was really 00:13:30.230 --> 00:13:33.190 quite moving, wasn't it? The source specifically 00:13:33.190 --> 00:13:36.169 states it's the first known case of pregnancy 00:13:36.169 --> 00:13:39.110 made possible through AI. Now, it wasn't like 00:13:39.110 --> 00:13:41.370 the AI performed the medical procedure itself. 00:13:41.529 --> 00:13:44.090 Right, right. But rather, it likely analyzed 00:13:44.090 --> 00:13:46.929 vast amounts of patient data, treatment histories, 00:13:47.029 --> 00:13:50.090 maybe genetic information, to identify factors 00:13:50.090 --> 00:13:52.509 or potential pathways that human doctors might 00:13:52.509 --> 00:13:55.269 have missed over those nearly two decades. Wow. 00:13:55.919 --> 00:13:58.320 Beyond the productivity tools and the big models, 00:13:58.539 --> 00:14:01.379 AI directly impacting lives in such a profound 00:14:01.379 --> 00:14:04.620 human way. That's pretty incredible. Really shows 00:14:04.620 --> 00:14:06.960 the potential breadth. And speaking of unexpected 00:14:06.960 --> 00:14:09.240 connections, there was that intriguing business 00:14:09.240 --> 00:14:12.639 detail, OpenAI quietly signing a cloud deal. 00:14:13.389 --> 00:14:15.409 With Google, aren't they like direct competitors? 00:14:15.529 --> 00:14:16.710 Yeah, this is definitely one of those behind 00:14:16.710 --> 00:14:18.429 the scenes moves that caught attention in the 00:14:18.429 --> 00:14:21.090 sources. OpenAI is famously funded and heavily 00:14:21.090 --> 00:14:23.929 supported by Microsoft, one of Google's fiercest 00:14:23.929 --> 00:14:26.850 competitors in the cloud and AI space. For OpenAI 00:14:26.850 --> 00:14:30.590 to quietly sign a cloud deal with Google Cloud 00:14:30.590 --> 00:14:34.330 Platform. The sources frame this as a kind of 00:14:34.330 --> 00:14:37.330 arms dealer. Well, by Google, basically selling 00:14:37.330 --> 00:14:39.350 their infrastructure capabilities even to rivals. 00:14:39.730 --> 00:14:43.210 If you need compute. We've got it. And the sources 00:14:43.210 --> 00:14:45.610 suggested it might indicate something about OpenAI's 00:14:45.610 --> 00:14:47.889 relationship with Microsoft, like things are 00:14:47.889 --> 00:14:50.529 shifting. Potentially. It could be interpreted 00:14:50.529 --> 00:14:53.409 in a few ways. One, as the source mentions, it 00:14:53.409 --> 00:14:56.009 could suggest that Microsoft's tight grip on 00:14:56.009 --> 00:14:59.070 OpenAI might be loosening slightly, or at least 00:14:59.070 --> 00:15:01.210 that OpenAI is asserting more independence in 00:15:01.210 --> 00:15:03.870 its infrastructure choices, making its own decisions. 00:15:04.070 --> 00:15:07.450 Two, it could simply be a pragmatic move by OpenAI 00:15:07.450 --> 00:15:10.370 to diversify its infrastructure providers, which 00:15:10.370 --> 00:15:12.009 is a standard practice for ensuring resilience 00:15:12.009 --> 00:15:14.090 and redundancy, and maybe leveraging competitive 00:15:14.090 --> 00:15:16.279 pricing. hedging their bets. Right. Don't put 00:15:16.279 --> 00:15:18.019 all your eggs in one basket. Either way, it's 00:15:18.019 --> 00:15:20.340 a fascinating dynamic in the competitive AI ecosystem. 00:15:20.899 --> 00:15:23.139 Lots of maneuvering. Complex corporate stuff 00:15:23.139 --> 00:15:25.679 going on there. Also saw some funding news that 00:15:25.679 --> 00:15:27.899 highlights where investment is heading. Right. 00:15:27.960 --> 00:15:30.139 Plug and Play, the global accelerator, secured 00:15:30.139 --> 00:15:33.919 a substantial $50 million fintech and AI fund. 00:15:34.120 --> 00:15:36.100 $50 million. Okay. The focus is specifically 00:15:36.100 --> 00:15:38.700 on AI applications within financial services, 00:15:38.860 --> 00:15:42.120 which is a massive industry ripe for AI disruption 00:15:42.120 --> 00:15:44.980 and efficiency gains. Shows investor confidence 00:15:44.980 --> 00:15:46.820 in that particular vertical. Money's flowing 00:15:46.820 --> 00:15:50.159 there. And just a few quick hits to sort of round 00:15:50.159 --> 00:15:52.100 things out and give a sense of the pace of development. 00:15:52.220 --> 00:15:55.360 Saw mentions of Apple's expected AI announcements 00:15:55.360 --> 00:15:59.210 at WWDC 2025. Hinching perhaps at things like 00:15:59.210 --> 00:16:02.070 visual AI capabilities integrated into their 00:16:02.070 --> 00:16:04.409 ecosystem. Yeah, always anticipation around Apple 00:16:04.409 --> 00:16:07.450 events. And OpenAI hitting that pretty staggering 00:16:07.450 --> 00:16:09.990 $10 billion annual recurring revenue mark. Wow, 00:16:10.090 --> 00:16:13.250 $10 billion ARR. That's approximately $833 million 00:16:13.250 --> 00:16:15.929 a month. It shows the commercial scale they've 00:16:15.929 --> 00:16:18.870 reached surprisingly quickly. Also, news that 00:16:18.870 --> 00:16:21.090 their first planned open model in years has been 00:16:21.090 --> 00:16:23.409 delayed until later this summer. Yeah, that delay 00:16:23.409 --> 00:16:25.789 is interesting, especially after their big closed 00:16:25.789 --> 00:16:29.610 model announcements like O3 Pro. Also saw AI 00:16:29.610 --> 00:16:32.309 companies prominently featured on the CNBC Disruptor 00:16:32.309 --> 00:16:34.629 50 list, which again just underscores how much 00:16:34.629 --> 00:16:37.370 AI is seen as reshaping industries right now. 00:16:37.509 --> 00:16:40.139 No surprise there. And Microsoft backed Mistral 00:16:40.139 --> 00:16:42.279 launching its own reasoning model specifically 00:16:42.279 --> 00:16:46.700 positioned to rival OpenAI. Exactly. That Mistral 00:16:46.700 --> 00:16:48.779 point is key because it shows that the competition 00:16:48.779 --> 00:16:51.840 isn't just in speed or size, but also in that 00:16:51.840 --> 00:16:54.940 specific crucial capability of logical reasoning 00:16:54.940 --> 00:16:58.580 that OpenAI is emphasizing with O3 Pro. So everyone's 00:16:58.580 --> 00:17:00.559 jumping into the reasoning game now. The race 00:17:00.559 --> 00:17:02.360 is definitely happening on multiple fronts. It's 00:17:02.360 --> 00:17:04.440 not just one dimension. And just a couple examples 00:17:04.440 --> 00:17:07.420 of like practical tools being able, things like. 00:17:07.769 --> 00:17:10.250 Hunter, an AI tool that can provide a resume 00:17:10.250 --> 00:17:12.730 review in under five minutes. Or Bubble, which 00:17:12.730 --> 00:17:14.589 lets you build no -code applications powered 00:17:14.589 --> 00:17:17.430 by AI. So tools for everyone. Right. Those highlight 00:17:17.430 --> 00:17:19.390 that it's not just about the foundational models, 00:17:19.470 --> 00:17:22.029 but the proliferation of applications and tools 00:17:22.029 --> 00:17:24.430 built on top of them that are directly impacting 00:17:24.430 --> 00:17:27.269 workflows and creating new possibilities for 00:17:27.269 --> 00:17:29.609 individuals and businesses. The ecosystem growing 00:17:29.609 --> 00:17:32.269 around the big models. Okay. So wrapping up this 00:17:32.269 --> 00:17:34.759 deep dive. What are the big picture takeaways 00:17:34.759 --> 00:17:36.779 that surface from this collection of sources 00:17:36.779 --> 00:17:39.480 today? What should we leave people with? I think 00:17:39.480 --> 00:17:41.740 the key insights. Pulling it all together are 00:17:41.740 --> 00:17:44.400 quite clear. First, we're seeing the AI frontier 00:17:44.400 --> 00:17:47.299 push towards not just bigger and faster, but 00:17:47.299 --> 00:17:50.500 smarter and more reliable models like OpenAI's 00:17:50.500 --> 00:17:53.380 O3 Pro. The reasoning focus. Specifically engineered 00:17:53.380 --> 00:17:56.740 for complex logic and trustworthiness, even if 00:17:56.740 --> 00:17:59.380 it means accepting a tradeoff in speed. Second, 00:17:59.559 --> 00:18:02.720 AI is clearly enabling faster, more accessible 00:18:02.720 --> 00:18:05.460 creative workflows like Google's VO3, potentially 00:18:05.460 --> 00:18:07.920 changing how content is generated and consumed, 00:18:08.079 --> 00:18:10.890 you know, like those. Stormtrooper vlogs. And 00:18:10.890 --> 00:18:14.029 perhaps most significantly, AI is moving beyond 00:18:14.029 --> 00:18:17.609 just high tech or creative domains and is increasingly 00:18:17.609 --> 00:18:22.109 tackling fundamental, often mundane, but absolutely 00:18:22.109 --> 00:18:25.509 critical real world problems like that government 00:18:25.509 --> 00:18:28.329 paperwork bottleneck with Gemini Extract or even 00:18:28.329 --> 00:18:30.730 enabling breakthroughs in deeply human areas 00:18:30.730 --> 00:18:34.039 like health care. Yeah, it's not just about the 00:18:34.039 --> 00:18:36.359 raw advancement of the models themselves anymore, 00:18:36.480 --> 00:18:39.059 is it? It's this parallel trend of specialization 00:18:39.059 --> 00:18:42.980 speed versus logic, for instance, and this really 00:18:42.980 --> 00:18:45.119 accelerating integration into fundamental parts 00:18:45.119 --> 00:18:48.140 of society, business, and even our personal lives. 00:18:48.319 --> 00:18:50.319 The integration point is vital. It's showing 00:18:50.319 --> 00:18:52.859 up not just in our chatbots, but in the infrastructure 00:18:52.859 --> 00:18:55.799 of government, in healthcare, in finance. It's 00:18:55.799 --> 00:18:58.079 becoming woven into the fabric of how things 00:18:58.079 --> 00:19:00.700 work, sometimes invisibly. Which brings us to 00:19:00.700 --> 00:19:02.740 a final provocative thought for you to consider 00:19:02.740 --> 00:19:04.599 based on everything we've just explored in these 00:19:04.599 --> 00:19:06.839 sources. Given the clear tradeoffs we're seeing 00:19:06.839 --> 00:19:09.039 in these new AI models, things like balancing 00:19:09.039 --> 00:19:12.000 speed versus reliability and deep reasoning and 00:19:12.000 --> 00:19:14.299 recognizing our increasing dependence on these 00:19:14.299 --> 00:19:16.740 systems, starkly highlighted by things like that 00:19:16.740 --> 00:19:19.079 recent widespread outage. Yeah, that dependence 00:19:19.079 --> 00:19:22.789 is growing fast. What aspects of AI do you find 00:19:22.789 --> 00:19:25.549 yourself prioritizing as these tools become more 00:19:25.549 --> 00:19:27.910 powerful and integrated into your own life or 00:19:27.910 --> 00:19:30.089 work? How do we collectively and individually 00:19:30.089 --> 00:19:32.789 think about balancing this incredible push for 00:19:32.789 --> 00:19:35.569 innovation and speed with the absolute critical 00:19:35.569 --> 00:19:38.869 need for robustness, accuracy, and trust in systems 00:19:38.869 --> 00:19:42.069 we're starting to rely on so heavily? That's 00:19:42.069 --> 00:19:43.630 the big question, isn't it? It's something worth 00:19:43.630 --> 00:19:46.450 mulling over as AI continues its deep dive into 00:19:46.450 --> 00:19:47.009 our world.