WEBVTT 00:00:00.000 --> 00:00:03.319 Imagine having two AI experts, you know, dedicated 00:00:03.319 --> 00:00:06.000 just to your private research notes. Right. They 00:00:06.000 --> 00:00:08.439 break it all down. They generate this interactive 00:00:08.439 --> 00:00:11.160 podcast. And you can literally call in. You can 00:00:11.160 --> 00:00:13.240 call in and ask some questions in real time. 00:00:13.439 --> 00:00:16.260 That is the absolute peak of personalized learning. 00:00:16.379 --> 00:00:19.800 It really makes complex stuff stick. Welcome 00:00:19.800 --> 00:00:22.399 back to the Deep Dive. So last time we focused 00:00:22.399 --> 00:00:24.379 on the creation side of things with Google's 00:00:24.379 --> 00:00:27.539 free AI. Today, we're shifting gears. Completely. 00:00:27.760 --> 00:00:32.210 We're all about research, learning, and... Real 00:00:32.210 --> 00:00:34.270 -time assistance. Exactly. We're going to unpack 00:00:34.270 --> 00:00:37.530 these really sophisticated free systems. Think 00:00:37.530 --> 00:00:39.310 of them like your own personal research team. 00:00:39.509 --> 00:00:41.789 We'll get into agentic deep research, a tool 00:00:41.789 --> 00:00:44.130 called Notebook LM. Which is incredible. Oh, 00:00:44.170 --> 00:00:46.350 it's a game changer. And then real -time AI vision 00:00:46.350 --> 00:00:48.670 and even some advanced features that are kind 00:00:48.670 --> 00:00:51.009 of hiding inside tools you already use, like 00:00:51.009 --> 00:00:53.890 docs and sheets. Okay, so let's start there with 00:00:53.890 --> 00:00:56.490 deep research in Gemini Advanced. The source 00:00:56.490 --> 00:00:59.539 material uses this word. agentic what makes a 00:00:59.539 --> 00:01:02.179 model agentic versus you know a standard search 00:01:02.179 --> 00:01:05.420 well it's the difference between Asking a librarian 00:01:05.420 --> 00:01:08.299 for one book versus hiring a research assistant. 00:01:08.519 --> 00:01:11.239 OK. An agentic model doesn't just give you one 00:01:11.239 --> 00:01:13.379 answer and stop. It's like an independent agent. 00:01:13.579 --> 00:01:17.120 It actively starts searching. It cross references 00:01:17.120 --> 00:01:19.659 what it finds and it keeps synthesizing over 00:01:19.659 --> 00:01:21.599 time. So it's making its own decisions about 00:01:21.599 --> 00:01:23.939 where to look next. Precisely. You give it a 00:01:23.939 --> 00:01:27.120 big complex question like the effectiveness of 00:01:27.120 --> 00:01:30.420 different carbon capture tech since 2023. Right. 00:01:30.500 --> 00:01:32.939 And it just goes. It starts pulling from academic 00:01:32.939 --> 00:01:36.560 papers. reports, articles, and it connects the 00:01:36.560 --> 00:01:38.819 dots as it goes. And critically, it's checking 00:01:38.819 --> 00:01:41.459 the claims. It's verifying. Yeah. That part is 00:01:41.459 --> 00:01:43.939 so important because it minimizes the risk of, 00:01:43.959 --> 00:01:46.659 you know, drawing a conclusion from just one 00:01:46.659 --> 00:01:49.719 biased source. The output feels less like a chat 00:01:49.719 --> 00:01:52.299 and more like a real consulting report. And that 00:01:52.299 --> 00:01:54.359 structure is key. You get an executive summary, 00:01:54.579 --> 00:01:57.719 findings, citations for every claim. But the 00:01:57.719 --> 00:01:59.319 part that really stood out to me was the notes 00:01:59.319 --> 00:02:01.799 on conflicting information. That's the gold right 00:02:01.799 --> 00:02:05.799 there. Most AI tools kind of smooth over disagreements. 00:02:05.920 --> 00:02:08.560 Deep research actually highlights them. That's 00:02:08.560 --> 00:02:11.180 essential for any kind of serious strategic planning 00:02:11.180 --> 00:02:14.580 or, say, market research. So this sounds pretty 00:02:14.580 --> 00:02:17.060 exhaustive. I mean, should I just skip regular 00:02:17.060 --> 00:02:19.659 web searches for anything complex now? I'd say 00:02:19.659 --> 00:02:22.659 use deep research when you need that depth. You 00:02:22.659 --> 00:02:25.300 know, where synthesis and conflicting views really 00:02:25.300 --> 00:02:27.360 matter. Okay, so we've gathered all this amazing 00:02:27.360 --> 00:02:29.659 data with deep research. Now, how do we actually 00:02:29.659 --> 00:02:31.460 make it stick? Let's talk about Notebook LM. 00:02:31.900 --> 00:02:34.379 If deep research is the gatherer, Notebook LM 00:02:34.379 --> 00:02:37.960 is the refinery. It's at notebooklm .google .com. 00:02:38.080 --> 00:02:41.199 And it's a completely free private workspace 00:02:41.199 --> 00:02:43.680 for your research. And it can handle a ton of 00:02:43.680 --> 00:02:46.340 material. The source says up to 50 sources, something 00:02:46.340 --> 00:02:48.259 like 25 million words. It's built for scale. 00:02:48.300 --> 00:02:51.400 Yeah. And it takes PDFs, docs, website links. 00:02:51.580 --> 00:02:54.580 And this is huge for learners today. It automatically 00:02:54.580 --> 00:02:57.759 transcribes and analyzes YouTube videos you upload. 00:02:57.960 --> 00:03:00.810 The core concept here. For trust, seems to be 00:03:00.810 --> 00:03:02.849 this thing called ROG. Can you break that down 00:03:02.849 --> 00:03:05.909 for us? Yeah. Our ROG stands for Retrieval Augmented 00:03:05.909 --> 00:03:08.289 Generation. But all that really means is the 00:03:08.289 --> 00:03:10.990 AI answers your questions only using the sources 00:03:10.990 --> 00:03:13.449 you uploaded. So it's a closed system. Exactly. 00:03:13.509 --> 00:03:16.449 It grounds the AI in your facts, which basically 00:03:16.449 --> 00:03:18.990 minimizes it from making stuff up. It makes it 00:03:18.990 --> 00:03:21.789 trustworthy enough for, you know, real academic 00:03:21.789 --> 00:03:24.150 work. You know, I still wrestle with prompt drift 00:03:24.150 --> 00:03:26.969 myself sometimes, where the AI just starts wandering 00:03:26.969 --> 00:03:30.199 off. We all do. So the fact that Notebook LM 00:03:30.199 --> 00:03:32.759 forces it back to the source material every time 00:03:32.759 --> 00:03:36.159 feels like a huge step for accuracy. It is. That 00:03:36.159 --> 00:03:39.000 verification loop is fundamental. Now what about 00:03:39.000 --> 00:03:41.620 retention? Let's talk about the transformation 00:03:41.620 --> 00:03:44.819 features. How does it turn all my documents into 00:03:44.819 --> 00:03:48.139 study materials? This is the magic part. It instantly 00:03:48.139 --> 00:03:51.979 generates quizzes, flashcards, mind maps, even 00:03:51.979 --> 00:03:54.979 full reports, all from your content. But the 00:03:54.979 --> 00:03:56.639 feature that connects back to our intro, the 00:03:56.639 --> 00:03:59.740 really wild one, is podcast mode. Right. It creates 00:03:59.740 --> 00:04:03.259 an interactive podcast where two AI hosts, two 00:04:03.259 --> 00:04:06.000 experts, discuss your research. And you can call 00:04:06.000 --> 00:04:08.120 in. You can literally interrupt the podcast and 00:04:08.120 --> 00:04:10.099 ask a question. Yeah. So if they're talking about 00:04:10.099 --> 00:04:12.639 RAG and vector databases, you can just jump in 00:04:12.639 --> 00:04:14.280 and say, hold on, can you explain what a vector 00:04:14.280 --> 00:04:17.259 database is? And the AI hosts will just answer. 00:04:17.360 --> 00:04:19.959 They'll answer you naturally, define it for you, 00:04:20.040 --> 00:04:22.199 and then they'll just seamlessly go back to their 00:04:22.199 --> 00:04:24.800 high -level discussion. Whoa. I mean, imagine 00:04:24.800 --> 00:04:27.920 scaling that. Personalized learning for a billion 00:04:27.920 --> 00:04:30.399 research queries a day with that kind of interaction. 00:04:30.680 --> 00:04:32.839 It's a different paradigm. Okay. So with all 00:04:32.839 --> 00:04:34.899 that context for my documents, does Notebook 00:04:34.899 --> 00:04:38.639 LM prevent the AI from making up facts? Yes, 00:04:38.660 --> 00:04:41.420 that's what R does. It grounds answers in your 00:04:41.420 --> 00:04:44.639 documents and minimizes any fabrication. That 00:04:44.639 --> 00:04:46.500 feels like a good transition from deep personal 00:04:46.500 --> 00:04:49.720 research to more immediate real -world help. 00:04:50.040 --> 00:04:52.399 Let's talk about Gemini's voice mode with vision. 00:04:52.730 --> 00:04:56.209 Yeah, this is where the AI becomes truly multimodal. 00:04:56.230 --> 00:04:58.569 It sees what you see through your camera and 00:04:58.569 --> 00:05:00.209 talks to you about it. It's like a collaborative 00:05:00.209 --> 00:05:02.389 partner. It really is. The smartwatch example 00:05:02.389 --> 00:05:04.649 from the source is perfect. You point your phone 00:05:04.649 --> 00:05:07.189 at a watch, ask about it, and the AI doesn't 00:05:07.189 --> 00:05:09.769 just say, that's a watch. Right. It identifies 00:05:09.769 --> 00:05:12.310 the exact model and pulls up detailed research 00:05:12.310 --> 00:05:15.170 like, hey, that's a Galaxy Watch 4 and it might 00:05:15.170 --> 00:05:17.750 not get software updates in 2025. That's real 00:05:17.750 --> 00:05:20.509 context instantly. That is incredibly useful. 00:05:20.990 --> 00:05:24.230 But are there limits? Like, does it fall apart 00:05:24.230 --> 00:05:26.750 if the lighty is bad or if it's not some well 00:05:26.750 --> 00:05:29.529 -known product? That's a fair question. For something 00:05:29.529 --> 00:05:32.569 with a barcode or a known model, it's very accurate. 00:05:32.649 --> 00:05:35.170 If you're pointing it at, I don't know, a weird 00:05:35.170 --> 00:05:38.110 antique in a dark room, your results might vary. 00:05:38.470 --> 00:05:42.129 But the practical uses are still massive. Show 00:05:42.129 --> 00:05:44.149 it the ingredients in your fridge. Ask for a 00:05:44.149 --> 00:05:47.370 recipe. Or point it at two products on a shelf 00:05:47.370 --> 00:05:50.110 in a store for a quick comparison. Or for fixing 00:05:50.110 --> 00:05:52.509 things. You can show it a broken part and ask 00:05:52.509 --> 00:05:55.449 for guidance. It's moving past just recognizing 00:05:55.449 --> 00:05:58.470 an image to helping you solve a problem. Exactly. 00:05:58.470 --> 00:06:01.250 It's a genuinely collaborative partner processing 00:06:01.250 --> 00:06:03.730 what it sees in real time. So this makes the 00:06:03.730 --> 00:06:06.709 AI assistant feel less like a search box. And 00:06:06.709 --> 00:06:08.329 more like a real person. And more like someone 00:06:08.329 --> 00:06:10.370 standing right there with you. Correct. It acts 00:06:10.370 --> 00:06:12.790 as a genuinely collaborative partner processing 00:06:12.790 --> 00:06:15.529 visuals and context in real time. Okay. Let's 00:06:15.529 --> 00:06:17.550 take a quick break. We'll be right back. Sponsor. 00:06:18.009 --> 00:06:21.550 And we are back. So if Notebook LM is the trust 00:06:21.550 --> 00:06:24.310 but verify research space, AI Studio is where 00:06:24.310 --> 00:06:27.769 you get to get under the hood. Yeah. This is 00:06:27.769 --> 00:06:29.689 for the power users who want to pull the levers 00:06:29.689 --> 00:06:32.069 themselves. And there are two features here that 00:06:32.069 --> 00:06:34.509 are just essential for serious users. What are 00:06:34.509 --> 00:06:37.240 they? First is side -by -side model comparison. 00:06:37.560 --> 00:06:40.240 So you can run the exact same prompt across different 00:06:40.240 --> 00:06:43.879 Gemini models, like say Pro versus Flash, and 00:06:43.879 --> 00:06:46.180 see the difference. For someone new to this, 00:06:46.300 --> 00:06:48.379 what's the quick difference between Pro and Flash? 00:06:48.740 --> 00:06:52.300 Think of Pro as the heavy lifter. It's for deep, 00:06:52.339 --> 00:06:55.439 complex reasoning. Flash is built for speed, 00:06:55.579 --> 00:06:58.319 for quick chat applications. You compare them 00:06:58.319 --> 00:07:00.439 to find the right balance for what you're trying 00:07:00.439 --> 00:07:03.000 to do. Got it. And the second feature is temperature 00:07:03.000 --> 00:07:05.720 controls. This is basically the creativity slider. 00:07:06.189 --> 00:07:08.370 Yeah, that's a perfect way to put it. A low temperature, 00:07:08.550 --> 00:07:11.970 like 0 .1, gives you very predictable, factual 00:07:11.970 --> 00:07:14.149 answers. You'd use that for summarizing meeting 00:07:14.149 --> 00:07:16.430 notes or writing code. Where you can't have mistakes. 00:07:16.850 --> 00:07:19.449 Zero mistakes. Then a high temperature, like 00:07:19.449 --> 00:07:22.649 0 .9, is for maximum creativity, brainstorming. 00:07:22.750 --> 00:07:24.509 So if I'm trying to come up with a new company 00:07:24.509 --> 00:07:26.490 mission statement, I'd start with a high temperature 00:07:26.490 --> 00:07:28.569 to get a bunch of ideas. And then I'd drop the 00:07:28.569 --> 00:07:31.170 temperature way down to refine the final wording. 00:07:31.329 --> 00:07:33.560 You start wide. then you narrow for consistency. 00:07:33.899 --> 00:07:37.000 That alone saves hours of back and forth. Then 00:07:37.000 --> 00:07:38.800 there's this feature for learning new software. 00:07:39.279 --> 00:07:42.680 Stream real time, the screen sharing with live 00:07:42.680 --> 00:07:45.800 guidance. It is basically a free tutor watching 00:07:45.800 --> 00:07:48.600 over your shoulder. You share your screen, you 00:07:48.600 --> 00:07:50.660 talk to Gemini, and it walks you through things 00:07:50.660 --> 00:07:53.100 step by step. The Canva example they gave was 00:07:53.100 --> 00:07:56.120 great. A user asks how to remove a background. 00:07:56.699 --> 00:07:58.720 they click on the wrong button and gemini sees 00:07:58.720 --> 00:08:01.060 the mistake on the screen and says nope not that 00:08:01.060 --> 00:08:04.060 one click over here it corrects you based on 00:08:04.060 --> 00:08:07.319 live visual input that's an amazing feedback 00:08:07.319 --> 00:08:10.540 loop for mastering unfamiliar software it's transformative 00:08:10.540 --> 00:08:13.399 yeah for corporate training for just learning 00:08:13.399 --> 00:08:16.379 a new skill on your own it's huge so does this 00:08:16.379 --> 00:08:18.480 live guidance only work with google's own software 00:08:18.860 --> 00:08:21.439 No, not at all. This tool is designed for learning 00:08:21.439 --> 00:08:24.300 things like Canva or, you know, even those obscure 00:08:24.300 --> 00:08:26.379 settings in Excel. Okay, so we've covered the 00:08:26.379 --> 00:08:28.839 big standalone tools. Let's talk about the AI 00:08:28.839 --> 00:08:31.160 that's woven into the things we use every single 00:08:31.160 --> 00:08:34.399 day, the hidden AI. This is Google's biggest 00:08:34.399 --> 00:08:36.600 advantage, I think, the seamless integration. 00:08:36.659 --> 00:08:40.000 We can kind of group them into, say, data and 00:08:40.000 --> 00:08:42.899 document mastery and then communication accelerators. 00:08:43.179 --> 00:08:46.429 Let's start with data and document mastery. Google 00:08:46.429 --> 00:08:49.669 Sheets AI formula generation. This is a massive 00:08:49.669 --> 00:08:51.409 time saver. You used to have to know all this 00:08:51.409 --> 00:08:54.610 complex syntax. Oh, I know it well. Now you just 00:08:54.610 --> 00:08:57.129 describe what you want in plain English. So instead 00:08:57.129 --> 00:08:59.809 of trying to remember some nested average function, 00:09:00.250 --> 00:09:02.950 I can just type, get the average of column C, 00:09:03.009 --> 00:09:05.710 but only where column A says sales. And it just 00:09:05.710 --> 00:09:08.049 builds the formula for you. It democratizes data 00:09:08.049 --> 00:09:11.820 analysis. And in Google Docs, The drive file 00:09:11.820 --> 00:09:15.360 references are huge. You type the A at symbol 00:09:15.360 --> 00:09:18.399 and you can instantly pull quotes or figures 00:09:18.399 --> 00:09:21.059 from another document in your drive right into 00:09:21.059 --> 00:09:23.100 the paragraph you're writing. So no more having 00:09:23.100 --> 00:09:25.779 three different windows open trying to copy and 00:09:25.779 --> 00:09:27.620 paste between a report and your source note. 00:09:27.679 --> 00:09:29.740 Exactly. It keeps everything consistent. Okay. 00:09:29.779 --> 00:09:31.759 Now what about those communication accelerators? 00:09:32.120 --> 00:09:35.159 Gmail and Meet are the big ones here. In Gmail, 00:09:35.320 --> 00:09:37.759 you've got thread summarization, which turns 00:09:37.759 --> 00:09:40.879 a 20 email chain into a few bullet points. That's 00:09:40.879 --> 00:09:43.779 a lifesaver. It really is. And contextual search, 00:09:43.940 --> 00:09:46.519 where you can search by description, like, find 00:09:46.519 --> 00:09:48.759 that email about the Chicago hotel from last 00:09:48.759 --> 00:09:51.899 June. Not just keywords. And Google Meets features. 00:09:52.639 --> 00:09:54.639 They're basically replacing a lot of paid assistant 00:09:54.639 --> 00:09:57.759 tools now. Totally. Auto -generated summaries 00:09:57.759 --> 00:10:00.700 and action items are now standard. It frees up 00:10:00.700 --> 00:10:02.600 the person who was stuck taking notes to actually 00:10:02.600 --> 00:10:06.139 participate. And don't forget AI mode in Google 00:10:06.139 --> 00:10:08.279 Search itself, which lets you have these long 00:10:08.279 --> 00:10:10.659 conversational searches without opening 50 tabs. 00:10:10.879 --> 00:10:13.299 So out of all of those, which integrations do 00:10:13.299 --> 00:10:15.700 you think save the most time for the average 00:10:15.700 --> 00:10:18.539 person doing complex work? I'd say the Sheets 00:10:18.539 --> 00:10:21.519 formula generation and the Docs references, those 00:10:21.519 --> 00:10:24.799 are massive accelerators for complex tasks. Stepping 00:10:24.799 --> 00:10:27.279 back from all of this, the overall strategy seems 00:10:27.279 --> 00:10:30.139 pretty clear. There are really two big takeaways 00:10:30.139 --> 00:10:33.639 from our deep dive today. First, Google is offering 00:10:33.639 --> 00:10:36.779 these top -tier free AI models across every category 00:10:36.779 --> 00:10:39.799 that matters. Text, research, vision. All of 00:10:39.799 --> 00:10:43.039 it. And second, that AI is so deeply woven into 00:10:43.039 --> 00:10:45.840 the ecosystem you already use, Docs, Gmail, Search, 00:10:46.019 --> 00:10:48.220 that you might not even notice it's there. And 00:10:48.220 --> 00:10:50.220 those capabilities, the ones we just talked about, 00:10:50.279 --> 00:10:53.980 they replace dozens of expensive paid tools. 00:10:54.600 --> 00:10:57.700 And they're just available to everyone right 00:10:57.700 --> 00:11:00.220 now. The limitation isn't cost or access anymore. 00:11:00.440 --> 00:11:03.240 The tools are free. They're powerful. The only 00:11:03.240 --> 00:11:05.340 real limitation is just your knowledge of what's 00:11:05.340 --> 00:11:07.600 out there and your willingness to explore a bit. 00:11:07.820 --> 00:11:10.600 Absolutely. So our challenge to you is to pick 00:11:10.600 --> 00:11:12.940 one of these tools. Maybe it's trying Notebook 00:11:12.940 --> 00:11:15.659 LM's podcast mode or just generating one formula 00:11:15.659 --> 00:11:18.639 in Sheets. Try one today. See how it can accelerate 00:11:18.639 --> 00:11:21.360 what you do. Yeah. Thanks for joining us as we 00:11:21.360 --> 00:11:23.360 dove deep into the source material on this one. 00:11:23.460 --> 00:11:24.220 We'll see you next time.