WEBVTT 00:00:00.000 --> 00:00:04.519 What if an AI decided to stop chasing pure speed 00:00:04.519 --> 00:00:07.879 and instead started focusing on deep, slow, and 00:00:07.879 --> 00:00:10.830 just careful reasoning? That fundamental shift 00:00:10.830 --> 00:00:13.210 is here. I mean, it's officially here. It's what 00:00:13.210 --> 00:00:16.149 Google is calling their new thinking model. And 00:00:16.149 --> 00:00:19.550 crucially, it's paired with this massive infrastructural 00:00:19.550 --> 00:00:23.050 upgrade. We're talking a 1 million token context 00:00:23.050 --> 00:00:26.289 window. This is not just another small step. 00:00:26.410 --> 00:00:29.989 It really changes the core architecture of how 00:00:29.989 --> 00:00:33.130 we can handle huge amounts of data. Welcome to 00:00:33.130 --> 00:00:35.880 the Deep Dive. Today, we're doing a really critical 00:00:35.880 --> 00:00:40.200 analysis of Google's new Gemini 3 Pro. And based 00:00:40.200 --> 00:00:43.060 on some pretty extensive real -life testing we've 00:00:43.060 --> 00:00:45.759 done across a lot of different uses from finance 00:00:45.759 --> 00:00:49.100 to visual puzzles, the hype seems, well, it seems 00:00:49.100 --> 00:00:51.240 genuinely warranted. It does. It feels like a 00:00:51.240 --> 00:00:53.640 very serious contender in the high -end AI space 00:00:53.640 --> 00:00:56.439 now, maybe an unavoidable one. So our mission 00:00:56.439 --> 00:00:58.420 for this deep dive is to really distill what 00:00:58.420 --> 00:01:00.579 makes this model unique. We're going to unpack 00:01:00.579 --> 00:01:02.840 the philosophy behind that thinking model and 00:01:02.840 --> 00:01:05.200 explain what that super memory means in a practical 00:01:05.200 --> 00:01:07.560 sense. And then get into the test results. Exactly. 00:01:07.620 --> 00:01:09.560 We're talking about building interactive dashboards, 00:01:09.659 --> 00:01:11.900 solving some tricky visual puzzles, and even 00:01:11.900 --> 00:01:16.079 its surprising ability to analyze silent video 00:01:16.079 --> 00:01:18.500 footage. OK, so let's unpack that core concept 00:01:18.500 --> 00:01:22.480 first. Gemini 3 Pro. It's the first in this new 00:01:22.480 --> 00:01:25.579 Gemini 3 series, and Google is deliberately calling 00:01:25.579 --> 00:01:29.239 it a thinking model. Right. That sounds great. 00:01:29.400 --> 00:01:31.439 But what does that actually mean for someone 00:01:31.439 --> 00:01:34.819 using it day to day? It's a huge shift away from 00:01:34.819 --> 00:01:36.920 latency minimization, you know, that obsession 00:01:36.920 --> 00:01:39.409 with giving you an answer instantly. OK. This 00:01:39.409 --> 00:01:42.469 model is built to pause. When you ask it something 00:01:42.469 --> 00:01:45.170 complex, it doesn't just spit back the first 00:01:45.170 --> 00:01:48.370 thing it finds. It engages in what you could 00:01:48.370 --> 00:01:51.129 call an internal reasoning chain. So less like 00:01:51.129 --> 00:01:53.090 a search engine and more like someone actually 00:01:53.090 --> 00:01:55.290 thinking a problem through. Exactly. Think of 00:01:55.290 --> 00:01:57.549 it like someone carefully working through a multi 00:01:57.549 --> 00:02:00.010 -step math problem instead of just recalling 00:02:00.010 --> 00:02:02.590 an answer. So those extra, what, 10 to 20 seconds 00:02:02.590 --> 00:02:04.870 it takes to respond? That's deliberate. That's 00:02:04.870 --> 00:02:07.959 not wasted time. That's the key. During that 00:02:07.959 --> 00:02:10.120 delay, it's essentially running premortems on 00:02:10.120 --> 00:02:12.539 its own logic. It's checking for consistency, 00:02:13.039 --> 00:02:15.240 making sure step five still aligns with the rules 00:02:15.240 --> 00:02:18.080 you set back in step one. And that rigor is what 00:02:18.080 --> 00:02:20.020 leads to the higher quality. Precisely. It's 00:02:20.020 --> 00:02:22.879 why the final outputs are just so much more accurate. 00:02:23.159 --> 00:02:25.560 And the performance data seems to back this up. 00:02:25.699 --> 00:02:27.919 I mean, we're seeing benchmarks where Gemini 00:02:27.919 --> 00:02:31.439 3 Pro beat its competitors. Like Claude and the 00:02:31.439 --> 00:02:33.919 latest GPT models. By the biggest performance 00:02:33.919 --> 00:02:35.740 gap ever recorded in these kinds of head -to 00:02:35.740 --> 00:02:38.120 -head tests. Right, but reasoning is only half 00:02:38.120 --> 00:02:40.259 the picture. You can have the best reasoning 00:02:40.259 --> 00:02:42.740 in the world, but if you forget what the initial 00:02:42.740 --> 00:02:45.879 request was, it's useless. Which brings us to 00:02:45.879 --> 00:02:49.629 the memory. The context window. The other massive 00:02:49.629 --> 00:02:51.849 technical leap here. Okay, so let's talk numbers. 00:02:52.009 --> 00:02:54.110 Remind us what a context window is in a practical 00:02:54.110 --> 00:02:57.090 sense. It's basically the AI's short -term memory 00:02:57.090 --> 00:02:58.909 during your conversation. It's all the info it 00:02:58.909 --> 00:03:01.129 can hold in its head at once. And most models 00:03:01.129 --> 00:03:04.550 today are around, what, 256 ,000 tokens? Yeah, 00:03:04.710 --> 00:03:06.509 around there, which is already pretty impressive. 00:03:06.969 --> 00:03:10.080 And a token is... you know, roughly a word or 00:03:10.080 --> 00:03:13.300 part of a word. But Gemini 3 Pro is quadrupling 00:03:13.300 --> 00:03:16.819 that. It's quadrupling it up to a full one million 00:03:16.819 --> 00:03:19.099 tokens. So you can think of the super memory 00:03:19.099 --> 00:03:22.360 like upgrading the AIs RAM by four times. So 00:03:22.360 --> 00:03:25.300 the practical benefit for you listening is what? 00:03:25.360 --> 00:03:28.960 It means the AI can digest truly huge documents. 00:03:29.020 --> 00:03:31.580 We're talking thick legal contracts, 300 page 00:03:31.580 --> 00:03:34.439 reports, even entire books without forgetting 00:03:34.439 --> 00:03:36.780 the details from page one. So it just reduces 00:03:36.780 --> 00:03:39.400 that that management overhead. You don't have 00:03:39.400 --> 00:03:41.520 to keep reminding it of things. That's it. It 00:03:41.520 --> 00:03:43.560 just manages the complexity better. It gets rid 00:03:43.560 --> 00:03:45.939 of that frustrating moment where an AI gets three 00:03:45.939 --> 00:03:47.680 quarters through a project and starts making 00:03:47.680 --> 00:03:50.240 things up because its memory just ran out. The 00:03:50.240 --> 00:03:52.939 best news for getting started, it's free. It's 00:03:52.939 --> 00:03:54.819 accessible right now. Yep. Just go to Gemini 00:03:54.819 --> 00:03:57.699 .Google .com and it's the default model. But 00:03:57.699 --> 00:03:59.900 you have an important tip here. A very important 00:03:59.900 --> 00:04:02.240 tip. To make sure you're actually using this 00:04:02.240 --> 00:04:04.840 high -quality thinking model, you have to go 00:04:04.840 --> 00:04:07.639 into the settings and find the specific option 00:04:07.639 --> 00:04:10.280 to turn it on. Otherwise, it might default to 00:04:10.280 --> 00:04:13.060 a faster, less thorough mode. Exactly. And for 00:04:13.060 --> 00:04:15.560 developers listening, it's already integrated 00:04:15.560 --> 00:04:19.920 into Google AI Studio and Vertex AI. So, if the 00:04:19.920 --> 00:04:23.379 model is designed to pause... Is that extra quality 00:04:23.379 --> 00:04:27.459 really worth the annoyance of waiting? For complex 00:04:27.459 --> 00:04:31.259 tasks, the quality leap is undeniable. That waiting 00:04:31.259 --> 00:04:34.160 period is where the magic happens. Speaking of 00:04:34.160 --> 00:04:35.920 complexity, let's look at the first real -life 00:04:35.920 --> 00:04:38.779 test, building interactive dashboards. We gave 00:04:38.779 --> 00:04:41.579 it a deliberately practical challenge. We asked 00:04:41.579 --> 00:04:44.120 it to create a full financial calculator for 00:04:44.120 --> 00:04:47.000 a multi -unit rental property. Not just math. 00:04:47.310 --> 00:04:50.009 but a structured, interactive tool. Right. So 00:04:50.009 --> 00:04:52.410 the prompt had a lot of variables. Purchase price, 00:04:52.589 --> 00:04:55.290 loan rate, down payment. Income, costs, all the 00:04:55.290 --> 00:04:57.529 usual stuff. But the real challenge we threw 00:04:57.529 --> 00:04:59.790 in was an interactive slider for the vacancy 00:04:59.790 --> 00:05:01.870 rate. OK, so the user could drag it from, what, 00:05:01.930 --> 00:05:04.629 0 % to 30%. Exactly. And the result was genuinely 00:05:04.629 --> 00:05:06.930 impressive. It built a full functional dashboard 00:05:06.930 --> 00:05:08.970 right there in the chat. So as you move the slider. 00:05:09.089 --> 00:05:11.610 The profit numbers instantly update. You can 00:05:11.610 --> 00:05:14.009 see your profit shrink or even watch the property 00:05:14.009 --> 00:05:17.290 start losing money in real time. And it also 00:05:17.290 --> 00:05:20.730 very smartly included a Generate Report button 00:05:20.730 --> 00:05:23.470 that gave you a written summary based on wherever 00:05:23.470 --> 00:05:26.089 the slider was set. That is a huge time saver 00:05:26.089 --> 00:05:28.310 for just checking if an idea is feasible. It 00:05:28.310 --> 00:05:30.750 is. And a big pro tip here is you can share these 00:05:30.750 --> 00:05:33.709 tools just by sending the Gemini link so you 00:05:33.709 --> 00:05:35.709 can collaborate on the model with someone else. 00:05:35.790 --> 00:05:38.629 OK, so from complex math to pure writing, the 00:05:38.629 --> 00:05:41.740 next test was about nuance, right? Yeah, passing 00:05:41.740 --> 00:05:44.519 subtle, human -imposed constraints. The challenge 00:05:44.519 --> 00:05:49.860 was to write a 700 -word SEO article on AI in 00:05:49.860 --> 00:05:52.240 marketing. For a non -technical audience, using 00:05:52.240 --> 00:05:55.000 web search for current info. And here's the tricky 00:05:55.000 --> 00:05:58.180 part. It was forbidden from using M dashes. A 00:05:58.180 --> 00:06:00.920 style rule that even human writers mess up all 00:06:00.920 --> 00:06:02.939 the time. And how did it do? The writing was 00:06:02.939 --> 00:06:05.899 excellent, very natural, very helpful, and most 00:06:05.899 --> 00:06:09.029 importantly, it worked. No M dashes. It showed 00:06:09.029 --> 00:06:11.569 a really high level of obedience to a subtle 00:06:11.569 --> 00:06:13.769 style guide. But there was a small issue. There 00:06:13.769 --> 00:06:16.949 was. And here's a vulnerable admission for me. 00:06:17.370 --> 00:06:19.990 I still wrestle with prompt drift myself, and 00:06:19.990 --> 00:06:22.209 even this top tier model had a little bit of 00:06:22.209 --> 00:06:24.129 it. What happened? It missed the word count. 00:06:24.689 --> 00:06:27.990 The goal was exactly 700 words, and it produced 00:06:27.990 --> 00:06:31.670 785. And for anyone who hasn't run into it, what 00:06:31.670 --> 00:06:34.339 is prompt drift? It's when the model deep in 00:06:34.339 --> 00:06:36.800 a conversation starts to prioritize the quality 00:06:36.800 --> 00:06:39.220 of what it's creating over your original structural 00:06:39.220 --> 00:06:41.519 rules. So it wanted to finish the thought properly, 00:06:41.720 --> 00:06:43.720 even if it meant going over the word count. That's 00:06:43.720 --> 00:06:45.759 my guess. It's also just a reminder that these 00:06:45.759 --> 00:06:48.660 AIs count in tokens, not exact words. So there's 00:06:48.660 --> 00:06:51.220 always a bit of ambiguity there. These dashboards 00:06:51.220 --> 00:06:54.639 seem great for initial modeling, but how reliable 00:06:54.639 --> 00:06:57.899 are they for serious financial work? They're 00:06:57.899 --> 00:07:01.209 outstanding for that initial analysis. For precision, 00:07:01.610 --> 00:07:03.250 always double check the logic, but the structure 00:07:03.250 --> 00:07:06.329 is very reliable. Okay, moving beyond text and 00:07:06.329 --> 00:07:09.290 numbers into the area where this model is really 00:07:09.290 --> 00:07:12.990 expected to shine. Multimodal power. Right, analyzing 00:07:12.990 --> 00:07:16.389 images and video. So test three was about visual 00:07:16.389 --> 00:07:19.470 reasoning. We gave it two classic spatial puzzles. 00:07:20.110 --> 00:07:22.949 The first was a photo of stacked colored cubes 00:07:22.949 --> 00:07:25.430 and we asked it to count the total, including 00:07:25.430 --> 00:07:27.899 the ones it couldn't see. A puzzle that famously 00:07:27.899 --> 00:07:31.040 trips up older AIs. They just can't infer what 00:07:31.040 --> 00:07:33.560 isn't directly visible. But Gemini 3 Pro got 00:07:33.560 --> 00:07:36.319 it. Yep. It took about 10 seconds, using that 00:07:36.319 --> 00:07:38.300 thinking model, and then it showed its work. 00:07:38.779 --> 00:07:40.639 It explained how it counted the hidden blocks 00:07:40.639 --> 00:07:42.939 needed for support, and it gave the correct total. 00:07:43.220 --> 00:07:45.720 And the second puzzle. was identifying the top 00:07:45.720 --> 00:07:48.660 -down view of a pyramid based on subtle color 00:07:48.660 --> 00:07:51.079 patterns from different photos. It got that one 00:07:51.079 --> 00:07:53.839 right too, explaining its reasoning based on 00:07:53.839 --> 00:07:56.060 how the colors had to line up. That level of 00:07:56.060 --> 00:07:59.259 spatial analysis is a huge deal. It's not just 00:07:59.259 --> 00:08:01.740 recognizing an object. It's not. And this is 00:08:01.740 --> 00:08:04.519 critical for fields like architecture, engineering, 00:08:04.920 --> 00:08:07.220 or even medical image analysis, where you need 00:08:07.220 --> 00:08:09.459 to understand what lies beneath the surface. 00:08:09.639 --> 00:08:11.399 The next test was maybe even more impressive. 00:08:11.629 --> 00:08:14.689 video analysis, but with no sound. This is a 00:08:14.689 --> 00:08:16.689 breakthrough capability I think people are sleeping 00:08:16.689 --> 00:08:20.009 on. I uploaded a silent five -minute screen recording 00:08:20.009 --> 00:08:22.490 of me just working. And you just asked it, what 00:08:22.490 --> 00:08:25.430 am I doing? Pretty much. What am I doing and 00:08:25.430 --> 00:08:28.589 what info can you see? And the result was? Startlingly 00:08:28.589 --> 00:08:31.449 detailed. It spotted a tiny pop -up notification 00:08:31.449 --> 00:08:34.610 that was only on screen for a second. It read 00:08:34.610 --> 00:08:38.250 my name from the user interface. And it accurately 00:08:38.250 --> 00:08:40.529 described what I was doing, saying I was changing 00:08:40.529 --> 00:08:43.450 parameters in a financial calculator tool. It 00:08:43.450 --> 00:08:45.769 understood the purpose of my actions from visuals 00:08:45.769 --> 00:08:49.009 alone. Whoa. Yeah. Just imagine scaling that. 00:08:49.269 --> 00:08:51.470 Imagine scaling that kind of video understanding 00:08:51.470 --> 00:08:55.730 to a billion security camera queries. Oh. Or 00:08:55.730 --> 00:08:58.149 analyzing sports footage. Exactly. We did another 00:08:58.149 --> 00:09:00.389 test with a pickleball clip, and it worked perfectly 00:09:00.389 --> 00:09:02.809 there, too. It opens up entirely new workflows. 00:09:02.990 --> 00:09:05.210 And the final test in this segment was building 00:09:05.210 --> 00:09:08.940 a simple app. a game. It did. It created a playable 00:09:08.940 --> 00:09:11.460 game called Neon Swarm, kind of like Galaga, 00:09:11.659 --> 00:09:13.980 using the system's Canvas tool. It had keyboard 00:09:13.980 --> 00:09:17.120 controls, score tracking, the works. So crucial 00:09:17.120 --> 00:09:19.279 tip for anyone listening who wants to try making 00:09:19.279 --> 00:09:21.320 something visual like that. Always, always make 00:09:21.320 --> 00:09:23.220 sure you've activated the Canvas tool. That's 00:09:23.220 --> 00:09:24.940 the environment where it can actually execute 00:09:24.940 --> 00:09:27.399 and build these interactive things. Okay. And 00:09:27.399 --> 00:09:29.399 you had a quick note on debugging these projects. 00:09:29.639 --> 00:09:31.860 You do. If you go back and forth with the AI, 00:09:32.250 --> 00:09:35.870 more than say 10 times, the canvas can sometimes 00:09:35.870 --> 00:09:38.629 get confused and things start to break. So what's 00:09:38.629 --> 00:09:41.750 the fix? Honestly, it's usually faster to just 00:09:41.750 --> 00:09:44.190 start a new chat with a clean prompt than to 00:09:44.190 --> 00:09:46.990 try and fix the old one. So if the model is so 00:09:46.990 --> 00:09:50.009 capable, what's the biggest limiting factor right 00:09:50.009 --> 00:09:52.529 now when you're trying to create these complex 00:09:52.529 --> 00:09:55.909 visual tools? It's that the memory of the tool's 00:09:55.909 --> 00:09:58.269 visual state can sometimes break after too many 00:09:58.269 --> 00:10:01.250 back and forth refinements. A fresh chat usually 00:10:01.250 --> 00:10:04.049 solves it. Let's talk about daily workflow. We've 00:10:04.049 --> 00:10:05.870 confirmed the power of the thinking model, but 00:10:05.870 --> 00:10:08.929 it takes 10 to 20 seconds. In a fast -paced job, 00:10:09.210 --> 00:10:11.990 isn't that delay a deal -breaker? It's a fair 00:10:11.990 --> 00:10:14.509 question. That wait time does feel a little annoying 00:10:14.509 --> 00:10:16.789 at first. We're all conditioned for instant AI 00:10:16.789 --> 00:10:18.950 replies. But the trade -off is that the quality 00:10:18.950 --> 00:10:21.090 is so much higher that you spend less time revising 00:10:21.090 --> 00:10:23.669 and fact -checking. For high -stakes work like 00:10:23.669 --> 00:10:26.429 drafting a contract summary, that depth is absolutely 00:10:26.429 --> 00:10:29.190 worth the wait. You use fast tools for fast work. 00:10:29.320 --> 00:10:31.639 and this for deliberate work. For those ready 00:10:31.639 --> 00:10:34.179 to adopt it, you mentioned two essential settings 00:10:34.179 --> 00:10:37.500 to turn on right away. Yes. First is personal 00:10:37.500 --> 00:10:41.399 context. Turn this on. It lets Gemini learn your 00:10:41.399 --> 00:10:45.220 style, your tone, your jargon from past chats. 00:10:45.879 --> 00:10:48.340 So it becomes more personalized over time. Exactly. 00:10:48.539 --> 00:10:50.799 It'll start writing sales emails in your formal 00:10:50.799 --> 00:10:53.019 tone automatically, for example. And the second 00:10:53.019 --> 00:10:56.259 must -have setting. Custom instructions. This 00:10:56.259 --> 00:10:58.259 is why you set permanent rules for every single 00:10:58.259 --> 00:11:01.190 chat. Things like always avoid complex jargon 00:11:01.190 --> 00:11:04.470 or frame all advice around website speed. And 00:11:04.470 --> 00:11:06.350 you're saying these stick much better in this 00:11:06.350 --> 00:11:08.309 new version. Far more reliably than they did 00:11:08.309 --> 00:11:11.450 before. Yes. OK, what about the AI mode in Google 00:11:11.450 --> 00:11:14.080 search? How did that perform against? Traditional 00:11:14.080 --> 00:11:16.419 search. I ran a comparison looking for a hotel 00:11:16.419 --> 00:11:19.559 in a specific price range. Regular Google search 00:11:19.559 --> 00:11:23.220 was perfect. Fast, precise, clean links. The 00:11:23.220 --> 00:11:25.600 AI mode was inconsistent. It would sometimes 00:11:25.600 --> 00:11:28.360 show conflicting information. How so? The little 00:11:28.360 --> 00:11:30.940 AI generated summary might quote one price, but 00:11:30.940 --> 00:11:32.820 the actual link it provided showed a price that 00:11:32.820 --> 00:11:35.679 was 50 % higher. It was hallucinating details 00:11:35.679 --> 00:11:37.720 to make the summary sound good. So the problem 00:11:37.720 --> 00:11:40.659 isn't speed, it's reliability for real -time 00:11:40.659 --> 00:11:44.399 facts. Exactly. For simple factual queries, regular 00:11:44.399 --> 00:11:46.700 Google search is still much more reliable right 00:11:46.700 --> 00:11:48.500 now. And we have to talk about the future here. 00:11:49.159 --> 00:11:51.500 Agent mode. This is the big one. This is the 00:11:51.500 --> 00:11:54.139 idea that signals the shift from a chat tool 00:11:54.139 --> 00:11:57.149 to a digital colleague. So what does it do? It 00:11:57.149 --> 00:12:00.169 can autonomously do things on the web for you. 00:12:00.269 --> 00:12:02.549 It's not just giving you information, it's acting 00:12:02.549 --> 00:12:04.669 on it. Like booking a reservation or filling 00:12:04.669 --> 00:12:08.409 out forms. Exactly. It moves the AI from being 00:12:08.409 --> 00:12:11.070 a passive answer generator to an active doer 00:12:11.070 --> 00:12:13.909 in your workflow. Given those search inconsistencies, 00:12:14.110 --> 00:12:16.990 should users rely on any AI, including this one, 00:12:17.409 --> 00:12:20.759 for precise real -time data. Not yet for simple 00:12:20.759 --> 00:12:23.000 facts. Regular search is still king for that. 00:12:23.200 --> 00:12:25.419 But for complex analysis of data you provide, 00:12:25.700 --> 00:12:28.120 the thinking model is far superior. So let's 00:12:28.120 --> 00:12:30.220 bring it all together. Where does Gemini 3 Pro 00:12:30.220 --> 00:12:32.860 really shine in a professional workflow? I'd 00:12:32.860 --> 00:12:35.580 say three clear areas. First, creating structured 00:12:35.580 --> 00:12:38.399 things like interactive tools. Second, deep reasoning 00:12:38.399 --> 00:12:41.399 for complex problems. And third, that advanced 00:12:41.399 --> 00:12:43.679 multimodal analysis. And the context window. 00:12:43.899 --> 00:12:46.820 Honestly, the 1 million token context window 00:12:46.820 --> 00:12:49.419 alone is a compelling reason to try it, especially 00:12:49.419 --> 00:12:51.700 if you work with long documents. And compared 00:12:51.700 --> 00:12:54.549 to the competition, where does it fit? Well, 00:12:54.750 --> 00:12:57.409 chat GPT still has the edge on third -party integrations. 00:12:57.610 --> 00:13:00.970 The app ecosystem is huge. Claude is still fantastic, 00:13:01.149 --> 00:13:03.610 especially for high -level coding. But Gemini 00:13:03.610 --> 00:13:06.990 3 Pro is a serious contender. A very serious 00:13:06.990 --> 00:13:09.110 one. And if you're already deep in the Google 00:13:09.110 --> 00:13:13.110 ecosystem, Gmail, Docs, Workspace, the integration 00:13:13.110 --> 00:13:15.850 is so seamless, it makes it a really powerful 00:13:15.850 --> 00:13:18.029 and easy choice. So what's your final advice 00:13:18.029 --> 00:13:21.169 for our listeners today? If you're totally happy 00:13:21.169 --> 00:13:23.250 with your current tools for simple things, there's 00:13:23.250 --> 00:13:25.789 no need to rush. But you should absolutely try 00:13:25.789 --> 00:13:28.769 Gemini 3 Pro for three specific tasks. Which 00:13:28.769 --> 00:13:31.730 are? Interactive tool creation, any kind of complex 00:13:31.730 --> 00:13:34.769 analysis that requires deep reasoning, and processing 00:13:34.769 --> 00:13:37.789 any long document that's over, say, 100 pages. 00:13:38.009 --> 00:13:40.389 So the core features, the reasoning, the quality, 00:13:41.190 --> 00:13:43.889 they're top -notch. They are, despite some minor 00:13:43.889 --> 00:13:46.649 issues like the word count thing. We really encourage 00:13:46.649 --> 00:13:49.049 you to try them all yourself. It's at Gemini 00:13:49.049 --> 00:13:52.269 .Google .com. And remember to go into those settings 00:13:52.269 --> 00:13:55.509 and turn on the thinking model option to really 00:13:55.509 --> 00:13:58.169 test it on your own complex work. And building 00:13:58.169 --> 00:14:01.769 on that idea of agent mode, just imagine an AI 00:14:01.769 --> 00:14:04.009 not just answering your questions, but automatically 00:14:04.009 --> 00:14:06.389 completing the next three steps of your workflow. 00:14:06.649 --> 00:14:09.330 Scheduling, drafting, data entry. All of it, 00:14:09.509 --> 00:14:11.590 autonomously. What kind of fundamental shift 00:14:11.590 --> 00:14:14.409 happens to our idea of work when the chat tool 00:14:14.409 --> 00:14:17.129 becomes a true digital colleague? Thank you for 00:14:17.129 --> 00:14:18.350 joining us for the deep dive.