WEBVTT 00:00:00.000 --> 00:00:03.520 Imagine this. You describe a simple video game, 00:00:03.919 --> 00:00:06.580 maybe a classic brick breaker, just using plain 00:00:06.580 --> 00:00:09.439 English. No code. Not a single line of Python 00:00:09.439 --> 00:00:12.800 or C++ fun. Yeah. You just watch the code write 00:00:12.800 --> 00:00:16.160 itself. And then, maybe 15 seconds later, you're 00:00:16.160 --> 00:00:19.280 playing that game. It's fully functional, right 00:00:19.280 --> 00:00:21.579 there on your screen. And that's not science 00:00:21.579 --> 00:00:24.039 fiction anymore. That is the immediate reality 00:00:24.039 --> 00:00:27.629 of Google's Gemini 3 model. Okay. Look, we know 00:00:27.629 --> 00:00:30.350 the flood of new AI tools is constant, but this 00:00:30.350 --> 00:00:34.630 update is a genuine architectural shift. It feels 00:00:34.630 --> 00:00:37.469 less like a typical chatbot. Yeah. And much more 00:00:37.469 --> 00:00:39.689 like an intelligent coworker who can actually 00:00:39.689 --> 00:00:41.950 do stuff for you. So for this deep dive, we're 00:00:41.950 --> 00:00:45.159 focusing on seven immediate practical ways you 00:00:45.159 --> 00:00:47.920 can use this upgrade. We are moving way past 00:00:47.920 --> 00:00:51.020 just generating text. The core feature that makes 00:00:51.020 --> 00:00:52.960 all this possible is something they call the 00:00:52.960 --> 00:00:55.399 Canvas. You have to get this term from the start. 00:00:55.579 --> 00:00:57.000 The Canvas, essentially, it's that interactive 00:00:57.000 --> 00:00:59.179 window. Exactly. It's the space where all the 00:00:59.179 --> 00:01:01.140 generated stuff, the games, the diagrams, the 00:01:01.140 --> 00:01:04.680 websites, actually appears. You can use it directly. 00:01:04.719 --> 00:01:07.140 You can click on it. It's the difference between 00:01:07.140 --> 00:01:11.579 getting a static blueprint of a house and getting 00:01:11.579 --> 00:01:14.060 a fully built interactive Lego model you can 00:01:14.060 --> 00:01:17.359 play with. So today our mission is to cover complex 00:01:17.359 --> 00:01:21.420 logic solving, zero code game creation, visual 00:01:21.420 --> 00:01:24.659 learning, and even high stakes job interview 00:01:24.659 --> 00:01:26.939 prep. And here's the critical insight we want 00:01:26.939 --> 00:01:29.530 you to hold on to. The real advantage isn't going 00:01:29.530 --> 00:01:32.010 to come from being a deep technical AI expert. 00:01:32.090 --> 00:01:34.590 Yeah. It's going to come from automating the 00:01:34.590 --> 00:01:37.530 small daily tasks that save you massive amounts 00:01:37.530 --> 00:01:39.829 of time. This is your shortcut. Exactly. So let's 00:01:39.829 --> 00:01:41.790 start with the architecture. OK, let's unpack 00:01:41.790 --> 00:01:44.750 this. For the longest time, AI models were basically 00:01:44.750 --> 00:01:47.290 just predicting the next word, right? Yeah. Like 00:01:47.290 --> 00:01:49.890 a super advanced autocorrect. Yeah. What's the 00:01:49.890 --> 00:01:51.950 fundamental change here that allows for such 00:01:51.950 --> 00:01:55.530 accurate functional output? Especially with complex 00:01:55.530 --> 00:01:57.969 logic. So this is the whole crux of the upgrade. 00:01:58.170 --> 00:02:01.010 It's built on three key improvements, but the 00:02:01.010 --> 00:02:02.790 mechanism behind it all is what we're calling 00:02:02.790 --> 00:02:05.069 pre -thought logic. Pre -thought logic. Instead 00:02:05.069 --> 00:02:07.530 of just guessing the next word, the model actually 00:02:07.530 --> 00:02:10.870 pauses and it uses what you could think of as 00:02:10.870 --> 00:02:13.949 like an internal scratch pad. It outlines the 00:02:13.949 --> 00:02:16.110 steps to the solution for itself before it even 00:02:16.110 --> 00:02:17.969 starts writing the final answer. So it's like 00:02:17.969 --> 00:02:20.050 a mathematician working out the problem on scrap 00:02:20.050 --> 00:02:23.169 paper first. Precisely. Before showing the final 00:02:23.169 --> 00:02:26.139 clean step, It checks its own logic before giving 00:02:26.139 --> 00:02:28.219 you an output that might look confident, but 00:02:28.219 --> 00:02:30.919 is actually, you know, flawed. Right. And that 00:02:30.919 --> 00:02:33.240 leads to the other improvements. It does. That 00:02:33.240 --> 00:02:35.879 leads to number two, speed. Because the whole 00:02:35.879 --> 00:02:38.259 process is more streamlined, it starts writing 00:02:38.259 --> 00:02:40.719 almost instantly. No more of that frustrating 00:02:40.719 --> 00:02:43.560 10 -second lag on complex queries. Right. And 00:02:43.560 --> 00:02:45.840 the third piece is the interactivity, which we 00:02:45.840 --> 00:02:48.680 tied back to that Canvas feature. You get functional, 00:02:48.900 --> 00:02:50.900 clickable results, not just a block of text. 00:02:51.210 --> 00:02:54.349 So since AI is so rooted in pattern recognition, 00:02:54.729 --> 00:02:57.669 why is that moment of structured thinking before 00:02:57.669 --> 00:03:01.009 speaking so essential for complex output? It 00:03:01.009 --> 00:03:03.530 allows the model to correctly sequence logical 00:03:03.530 --> 00:03:06.550 steps rather than just defaulting to the most 00:03:06.550 --> 00:03:09.569 common statistical answer. So once we understood 00:03:09.569 --> 00:03:11.810 that shift, we had to test the reasoning itself. 00:03:12.490 --> 00:03:14.129 And this is where it gets really interesting. 00:03:14.569 --> 00:03:16.469 Let's start with that classic family riddle. 00:03:16.490 --> 00:03:18.770 Oh, this one's great. Mike has three brothers. 00:03:19.069 --> 00:03:21.889 Each of his brothers has two sisters. The question 00:03:21.889 --> 00:03:24.629 is, how many children are in the family? And 00:03:24.629 --> 00:03:27.150 this is a great test because the common mistake 00:03:27.150 --> 00:03:30.229 which older AIs always made is to just rush to 00:03:30.229 --> 00:03:32.250 the math. They'd go three brothers plus Mike 00:03:32.250 --> 00:03:35.490 plus three times two sisters. And you get 10 00:03:35.490 --> 00:03:38.639 children. Which is wrong. Totally wrong. But 00:03:38.639 --> 00:03:41.300 this model didn't do that. No, it actually demonstrated 00:03:41.300 --> 00:03:43.939 an understanding of family relationships. It 00:03:43.939 --> 00:03:46.500 reasoned that if Mike has three brothers, that's 00:03:46.500 --> 00:03:49.719 four boys total. And if each of those boys shares 00:03:49.719 --> 00:03:52.400 the same two sisters, there are only two sisters 00:03:52.400 --> 00:03:55.500 total in the family. So the result, four boys 00:03:55.500 --> 00:03:58.280 plus two girls equals six children. It got it. 00:03:58.360 --> 00:04:00.419 It understood the linguistic trick. So we took 00:04:00.419 --> 00:04:03.060 that and applied it to a real world business 00:04:03.060 --> 00:04:06.129 problem, the coffee shop roster test. Yeah, this 00:04:06.129 --> 00:04:08.050 was a tough one. The prompt had all these conflicting 00:04:08.050 --> 00:04:11.669 rules. The shift is 8 a .m. to 4 p .m. Alex must 00:04:11.669 --> 00:04:14.229 work before noon. And Sarah and Tom cannot work 00:04:14.229 --> 00:04:17.850 together ever. And the key constraint, Tom needs 00:04:17.850 --> 00:04:20.870 exactly four hours of work. And it didn't just 00:04:20.870 --> 00:04:23.649 guess. It produced a perfect schedule instantly. 00:04:24.209 --> 00:04:27.430 Alex from 8 a .m. to 12 p .m. and Tom from 12 00:04:27.430 --> 00:04:29.850 p .m. to 4 p .m. And it explained its logic. 00:04:29.889 --> 00:04:32.430 It said Tom must take the afternoon shift to 00:04:32.430 --> 00:04:35.069 guarantee his four hours, which also, you know, 00:04:35.129 --> 00:04:37.550 conveniently avoids the conflict with Sarah. 00:04:37.689 --> 00:04:39.189 Right. She's just not on the schedule. So for 00:04:39.189 --> 00:04:41.529 a busy professional, what does that automated 00:04:41.529 --> 00:04:44.649 roster creation really signify for broader business 00:04:44.649 --> 00:04:47.910 applications? It provides instant, viable and 00:04:47.910 --> 00:04:50.430 explained solutions for painfully complex resource 00:04:50.430 --> 00:04:53.329 allocation and scheduling tasks. Right. But quick 00:04:53.329 --> 00:04:55.329 note of caution here. It's still just a logic 00:04:55.329 --> 00:04:57.509 solver. What do you mean? It can't handle interpersonal 00:04:57.509 --> 00:05:00.569 office politics. You know, if Tom is mad at Alex 00:05:00.569 --> 00:05:02.930 and refuses to work the fryer, it can't solve 00:05:02.930 --> 00:05:05.990 that yet. That's a fair point. It solves the 00:05:05.990 --> 00:05:09.310 math, not the drama. So once the AI has mastered 00:05:09.310 --> 00:05:12.649 reasoning like that, the next question is, can 00:05:12.649 --> 00:05:15.370 it translate that logic into something that actually 00:05:15.370 --> 00:05:18.930 works? And this is where we get to the real wow 00:05:18.930 --> 00:05:21.769 factor building actual software without knowing 00:05:21.769 --> 00:05:24.829 Python or JavaScript. Right. You tested this 00:05:24.829 --> 00:05:27.310 by asking for a simple brick breaker game. I 00:05:27.310 --> 00:05:29.870 did. I just dictated the rules. Mouse control, 00:05:30.110 --> 00:05:33.089 colorful bricks, score counter, dark blue background. 00:05:33.439 --> 00:05:35.939 I hit enter. And what happened? The canvas opened 00:05:35.939 --> 00:05:38.540 up and I literally watched the HTML and JavaScript 00:05:38.540 --> 00:05:40.920 write themselves. I went from prompt to playing 00:05:40.920 --> 00:05:43.519 the game in like, I don't know, 20 seconds. That 00:05:43.519 --> 00:05:45.339 the real game changer wasn't just creating it, 00:05:45.379 --> 00:05:47.920 it was iterating on it. Yes. You didn't open 00:05:47.920 --> 00:05:50.379 a code editor. You just said, conversationally, 00:05:50.540 --> 00:05:53.279 the ball's too slow, make it 2x faster, and it 00:05:53.279 --> 00:05:55.459 just did it. I have to admit, I still wrestle 00:05:55.459 --> 00:05:58.149 with prompt drift myself. You know, wondering 00:05:58.149 --> 00:06:00.970 if the AI will lose context. Yeah. But this ability 00:06:00.970 --> 00:06:03.970 to update code just by talking to it is incredibly 00:06:03.970 --> 00:06:06.230 helpful. And that power translates seamlessly 00:06:06.230 --> 00:06:08.970 from games to design. You tested a link in Bio 00:06:08.970 --> 00:06:12.689 website. Yep. Minimalist, circular photo, four 00:06:12.689 --> 00:06:15.500 buttons, black and white. Looked great. And then 00:06:15.500 --> 00:06:17.519 you changed the whole theme with one command 00:06:17.519 --> 00:06:21.139 from serious black and white to what was it a 00:06:21.139 --> 00:06:24.699 cozy coffee shop? Beige, soft brown, cream, and 00:06:24.699 --> 00:06:26.860 it redrew the entire thing instantly with the 00:06:26.860 --> 00:06:29.040 new palette. It was amazing. You could also do 00:06:29.040 --> 00:06:32.100 app mockups like a fitness app UI. Which lets 00:06:32.100 --> 00:06:35.660 a manager show a developer a clickable vision 00:06:35.660 --> 00:06:38.560 before they even start coding. So why is that 00:06:38.560 --> 00:06:41.560 conversational iteration editing code just by 00:06:41.560 --> 00:06:44.420 talking such a game changer for speed. It bypasses 00:06:44.420 --> 00:06:47.500 the need to manually hunt for and edit a specific 00:06:47.500 --> 00:06:50.660 line of code. The AI just handles the scaffolding 00:06:50.660 --> 00:06:53.379 for you. OK, let's shift to education. The teaching 00:06:53.379 --> 00:06:55.620 aspect here is, I think, next level because it 00:06:55.620 --> 00:06:57.399 builds lessons just for you. Oh, absolutely. 00:06:57.519 --> 00:06:59.779 I started with the solar system. I asked for 00:06:59.779 --> 00:07:02.120 an interactive 3D visualization. And what did 00:07:02.120 --> 00:07:04.579 it build? The result in the canvas showed all 00:07:04.579 --> 00:07:06.899 the planets orbiting the sun, but, and this is 00:07:06.899 --> 00:07:08.339 the cool part, they were moving at the correct 00:07:08.339 --> 00:07:11.379 relative speeds. Mercury was zipping by and Neptune 00:07:11.379 --> 00:07:13.959 was slowly crawling along. And you could interact 00:07:13.959 --> 00:07:16.500 with it. Yeah, you could hover over any planet 00:07:16.500 --> 00:07:19.019 and get a fun fact like Mars having the tallest 00:07:19.019 --> 00:07:21.620 mountain in the solar system. It built the exact 00:07:21.620 --> 00:07:23.399 lesson you asked for. Then we took it inside 00:07:23.399 --> 00:07:25.939 the human body, the heart. The request was very 00:07:25.939 --> 00:07:30.370 specific. Show blood flow. Use red for oxygenated 00:07:30.370 --> 00:07:33.350 blood, blue for deoxygenated. The result was, 00:07:33.810 --> 00:07:37.509 it was stunning. A dynamic diagram appeared where 00:07:37.509 --> 00:07:39.069 you could see these little particles moving. 00:07:39.360 --> 00:07:41.720 blue dots going into the right side, red dots 00:07:41.720 --> 00:07:43.360 coming out of the left. And you could click on 00:07:43.360 --> 00:07:45.040 the aorta or the ventricles and get a little 00:07:45.040 --> 00:07:48.000 description. Whoa! I mean, imagine scaling that 00:07:48.000 --> 00:07:51.060 kind of immediate custom -built lesson to every 00:07:51.060 --> 00:07:53.939 student in the world who needs a specific visual 00:07:53.939 --> 00:07:56.420 explanation for something. The personalization 00:07:56.420 --> 00:07:58.939 is just immense. So what advantage does building 00:07:58.939 --> 00:08:01.540 your own custom diagram have over just watching 00:08:01.540 --> 00:08:04.300 a standard YouTube video? You can build the exact 00:08:04.300 --> 00:08:07.339 lesson you need. customizing the data visualization 00:08:07.339 --> 00:08:10.620 precisely to your own knowledge gap. All right. 00:08:10.740 --> 00:08:13.459 Let's talk about high stakes preparation. Job 00:08:13.459 --> 00:08:16.040 interviews. People get so nervous and practices 00:08:16.040 --> 00:08:18.879 everything. It is. So I tested role playing a 00:08:18.879 --> 00:08:21.199 project manager interview using the voice mode. 00:08:21.459 --> 00:08:24.319 And you asked it to be tough. I did. I told it 00:08:24.319 --> 00:08:26.699 to act as the hiring manager and give me detailed 00:08:26.699 --> 00:08:28.939 feedback after every answer. And when I gave 00:08:28.939 --> 00:08:31.019 a simple answer, like I tell everyone to work 00:08:31.019 --> 00:08:34.440 harder, the AI pushed back hard. What did it 00:08:34.440 --> 00:08:36.870 say? It coached me on the spot. It demanded a 00:08:36.870 --> 00:08:39.950 more strategic answer like prioritizing tasks, 00:08:40.429 --> 00:08:43.309 cutting features, or reassigning resources. It 00:08:43.309 --> 00:08:45.769 wasn't gentle at all. Which is what a real hiring 00:08:45.769 --> 00:08:48.289 manager would do. Exactly. And if you use the 00:08:48.289 --> 00:08:51.149 webcam, it's optional, of course. It can even 00:08:51.149 --> 00:08:54.470 analyze things like your posture or eye contact. 00:08:54.929 --> 00:08:56.610 It might say, you know, you're looking down a 00:08:56.610 --> 00:08:59.509 lot, which can make you seem unsure. That's incredible 00:08:59.509 --> 00:09:02.769 feedback. But we have to ask, how accurate is 00:09:02.769 --> 00:09:04.850 that kind of analysis really? Well, it's still 00:09:04.850 --> 00:09:07.289 early, and it depends on your lighting and camera 00:09:07.289 --> 00:09:10.230 angle. But the main benefit isn't perfect accuracy. 00:09:10.669 --> 00:09:13.070 It's the ability to practice 100 times without 00:09:13.070 --> 00:09:15.889 bothering a human coach. You get rid of the nerves 00:09:15.889 --> 00:09:17.990 and refine your answers without any pressure. 00:09:18.730 --> 00:09:21.710 So why is being able to practice 100 times so 00:09:21.710 --> 00:09:24.149 essential before a big interview? It refines 00:09:24.149 --> 00:09:26.690 your responses and eliminates nervousness without 00:09:26.690 --> 00:09:30.029 the pressure of a real human coach. Okay, finally, 00:09:30.129 --> 00:09:33.370 let's tackle real life chaos. For most of us, 00:09:33.389 --> 00:09:35.389 that means food and travel. Let's start with 00:09:35.389 --> 00:09:37.970 the fridge test. Right. The constraints were 00:09:37.970 --> 00:09:40.970 simple, but very common. I have three eggs, some 00:09:40.970 --> 00:09:43.970 salsa, cheese, spinach, and frozen rice. What 00:09:43.970 --> 00:09:46.389 can I make, and what do I need? And the result 00:09:46.389 --> 00:09:48.710 was just incredibly practical. It immediately 00:09:48.710 --> 00:09:51.750 gave you two recipes, an omelet and a fried rice 00:09:51.750 --> 00:09:53.990 bowl, using exactly what you had. But it didn't 00:09:53.990 --> 00:09:57.049 stop there. It then created this perfectly formatted 00:09:57.049 --> 00:10:00.830 table with a grocery list, chicken, beans, tortillas, 00:10:01.149 --> 00:10:03.769 to expand those ingredients into five full dinners 00:10:03.769 --> 00:10:06.470 for the week. It turns chaos into organization. 00:10:06.750 --> 00:10:08.490 It really does. And then there's travel planning. 00:10:08.669 --> 00:10:10.909 Yes, for a tough one. A two -day California trip 00:10:10.909 --> 00:10:13.309 focused on non -crowded spots, great coffee, 00:10:13.549 --> 00:10:16.360 and quiet beaches. Yeah, and the itinerary it 00:10:16.360 --> 00:10:18.639 gave back was so detailed. It felt like it was 00:10:18.639 --> 00:10:21.480 from a local friend. It had specific non -chain 00:10:21.480 --> 00:10:24.179 coffee shops. Not Starbucks. No, definitely not 00:10:24.179 --> 00:10:27.000 Starbucks. It found quiet, less -visited beaches, 00:10:27.259 --> 00:10:29.580 specific museums, and even gave estimated costs 00:10:29.580 --> 00:10:32.220 in U .S. dollars for everything. So beyond just 00:10:32.220 --> 00:10:34.679 the convenience of that instant planning, what's 00:10:34.679 --> 00:10:38.049 the key mindset shift required to really... use 00:10:38.049 --> 00:10:40.549 AI effectively for these tasks. Don't accept 00:10:40.549 --> 00:10:43.169 the first draft as final. Constant refinement 00:10:43.169 --> 00:10:46.409 in conversation leads to perfect, personalized 00:10:46.409 --> 00:10:49.169 results. So to summarize this whole deep dive, 00:10:49.450 --> 00:10:52.990 Gemini 3 really is functioning like a smart coworker 00:10:52.990 --> 00:10:56.610 now. It builds, it visualizes, and because of 00:10:56.610 --> 00:10:58.649 that pre -thought logic, it actually reasons 00:10:58.649 --> 00:11:01.870 through complex problems. Yeah, and the key takeaway 00:11:01.870 --> 00:11:03.990 is simple. If you're looking for an edge, focus 00:11:03.990 --> 00:11:06.470 on automating the small daily stuff. Writing 00:11:06.470 --> 00:11:08.909 emails, fixing spreadsheets, meal planning. All 00:11:08.909 --> 00:11:11.090 that compounding time you save. All that time. 00:11:11.129 --> 00:11:13.429 And my advice is that the barrier to entry is 00:11:13.429 --> 00:11:15.750 just so low now. Just go try it. Talk to it like 00:11:15.750 --> 00:11:17.590 a colleague. You don't need to be a prompt engineer. 00:11:17.710 --> 00:11:20.149 Right. And remember, the big mistake to avoid. 00:11:20.570 --> 00:11:22.730 Don't just accept the first output. Even if it's 00:11:22.730 --> 00:11:25.649 good, always ask for refinements. Demand more 00:11:25.649 --> 00:11:28.370 from it. And that brings us to our final thought. 00:11:28.409 --> 00:11:30.330 Yeah. The people who are truly going to get ahead 00:11:30.330 --> 00:11:32.710 in the next few years won't be the AI experts 00:11:32.710 --> 00:11:36.029 focused on all the technical details. Nope. It's 00:11:36.029 --> 00:11:37.730 going to be the regular people who just figure 00:11:37.730 --> 00:11:40.529 out how to leverage these tools to save 30 minutes 00:11:40.529 --> 00:11:44.269 every single day. That accumulated time, it changes 00:11:44.269 --> 00:11:46.830 everything about what you can achieve. So think 00:11:46.830 --> 00:11:49.929 of that one thing you hate doing, that one frustrating 00:11:49.929 --> 00:11:53.529 daily task. And just ask the AI for help. What 00:11:53.529 --> 00:11:54.330 are you going to build first?