WEBVTT 00:00:00.000 --> 00:00:03.279 So I spent, I think, an entire evening just recently 00:00:03.279 --> 00:00:05.759 trying to cross -reference these three different 00:00:05.759 --> 00:00:08.320 internal reports, just looking for one specific 00:00:08.320 --> 00:00:10.400 number. We're all just pile up information, aren't 00:00:10.400 --> 00:00:13.339 we? PDFs, articles we save, transcripts from 00:00:13.339 --> 00:00:17.440 videos. Sometimes it feels less like a library 00:00:17.440 --> 00:00:19.719 and more like, I don't know, just drowning in 00:00:19.719 --> 00:00:22.140 data. Oh, yeah, it's the classic research problem, 00:00:22.239 --> 00:00:24.320 right? But it's amplified now with all these 00:00:24.320 --> 00:00:27.359 tools. We need info fast. But then you turn to 00:00:27.359 --> 00:00:30.120 a general AI, and sometimes it just starts. Well, 00:00:30.480 --> 00:00:32.299 making things up because it's looking at the 00:00:32.299 --> 00:00:35.219 whole messy internet. Right. So imagine if you 00:00:35.219 --> 00:00:38.299 had this AI assistant, but it was totally personalized, 00:00:38.659 --> 00:00:42.119 a research guide that only and strictly uses 00:00:42.119 --> 00:00:44.479 your trusted information. It basically stops 00:00:44.479 --> 00:00:46.880 that whole hallucination problem because, well, 00:00:46.939 --> 00:00:48.479 it can't talk about something it hasn't read 00:00:48.479 --> 00:00:50.140 from you. And that's exactly what we're digging 00:00:50.140 --> 00:00:52.859 into today. We're exploring a notebook LM. It's 00:00:52.859 --> 00:00:55.420 this really interesting tool from Google. It 00:00:55.420 --> 00:00:58.060 fundamentally changes how you can organize and, 00:00:58.100 --> 00:01:00.380 more importantly, use all that knowledge you've 00:01:00.380 --> 00:01:02.979 gathered. OK, so today we'll break down the core 00:01:02.979 --> 00:01:04.980 tech behind it. We'll look at the workspace, 00:01:05.060 --> 00:01:07.500 which is actually pretty simple. Then we'll walk 00:01:07.500 --> 00:01:09.359 you through setting up your first notebook focused 00:01:09.359 --> 00:01:11.760 on your stuff. And then we'll get into the cool 00:01:11.760 --> 00:01:14.980 creative tools like making quizzes automatically 00:01:14.980 --> 00:01:18.079 or even presentation videos. This is really about 00:01:18.079 --> 00:01:21.250 giving you a way to master large chunks of information, 00:01:21.609 --> 00:01:23.510 but specifically targeted to the data you've 00:01:23.510 --> 00:01:27.030 already collected. Okay, let's get into it. So 00:01:27.030 --> 00:01:30.049 fundamentally, Notebook LM is built to be your 00:01:30.049 --> 00:01:32.670 personal research assistant, super focused. But 00:01:32.670 --> 00:01:35.670 what's the single biggest tech difference between 00:01:35.670 --> 00:01:38.620 this and just a standard chatbot? It really boils 00:01:38.620 --> 00:01:40.799 down to the constraint. It's forced limitation. 00:01:41.359 --> 00:01:43.260 Standard AI looks everywhere, the whole web. 00:01:43.480 --> 00:01:45.819 Notebook LM is strictly limited. Only the documents 00:01:45.819 --> 00:01:48.659 you upload, your PDFs, your notes, your transcripts, 00:01:48.760 --> 00:01:51.019 nothing else. And that's what makes it genuinely 00:01:51.019 --> 00:01:53.879 personal and maybe most importantly, trustworthy. 00:01:54.260 --> 00:01:57.359 So if I upload, say, 15 really specific files 00:01:57.359 --> 00:02:01.359 about internal project metrics, the AI can only 00:02:01.359 --> 00:02:03.040 quote from those 15 files. It won't bring in 00:02:03.040 --> 00:02:05.730 anything else. Exactly. That's it. And the tech 00:02:05.730 --> 00:02:07.750 that makes this happen, that guarantees it, is 00:02:07.750 --> 00:02:10.949 called RAG -G, Retrieval Augmented Generation. 00:02:11.430 --> 00:02:14.729 Basically, RAG just means in simple terms. The 00:02:14.729 --> 00:02:17.729 AI first reads and kind of indexes your documents. 00:02:18.210 --> 00:02:21.050 Then it uses only that reading to generate answers. 00:02:21.330 --> 00:02:23.110 So the answers are accurate. You can check them. 00:02:23.169 --> 00:02:26.289 It's like a closed loop. Huh. So it's almost 00:02:26.289 --> 00:02:29.870 like a lawyer arguing a case. They can only use 00:02:29.870 --> 00:02:32.169 the specific evidence you handed them. They can't 00:02:32.169 --> 00:02:34.530 just speculate or pull something from across 00:02:34.530 --> 00:02:36.569 the street. That's a great analogy, yeah. Perfect. 00:02:36.750 --> 00:02:38.569 The output is always anchored in the evidence 00:02:38.569 --> 00:02:41.870 you provide. OK, so then if the AI is strictly 00:02:41.870 --> 00:02:45.590 limited like that, only using my sources, how 00:02:45.590 --> 00:02:48.030 does that really change the quality, the reliability 00:02:48.030 --> 00:02:50.060 of the answers I get? Well, the answers become 00:02:50.060 --> 00:02:52.340 really accurate. You can trust them. And crucially, 00:02:52.460 --> 00:02:54.520 they're always tied directly back to your original 00:02:54.520 --> 00:02:57.039 source material. So when you first open it up, 00:02:57.159 --> 00:02:58.900 the interface, it's really built for function. 00:02:59.000 --> 00:03:01.780 It avoids confusion. It's basically split into 00:03:01.780 --> 00:03:04.319 three main parts, which makes handling complex 00:03:04.319 --> 00:03:07.000 research, well, simpler. OK, so on the left, 00:03:07.080 --> 00:03:09.379 you've got your library. That's like your personal 00:03:09.379 --> 00:03:12.460 bookshelf, right? For all sorts of files, Google 00:03:12.460 --> 00:03:16.770 Docs, PDFs. website links, even YouTube transcripts, 00:03:17.030 --> 00:03:19.810 that's where your raw stuff lives. Then right 00:03:19.810 --> 00:03:21.669 in the middle, that's the conversation space, 00:03:21.849 --> 00:03:23.569 that's your main workspace, that's where you 00:03:23.569 --> 00:03:26.449 actually chat with the AI, ask it to summarize 00:03:26.449 --> 00:03:29.590 that long technical manual, or maybe compare 00:03:29.590 --> 00:03:31.729 ideas from two different files you uploaded. 00:03:31.969 --> 00:03:34.090 And the really key thing there, the quality check, 00:03:34.490 --> 00:03:37.169 is the citation. Every single thing it says every 00:03:37.169 --> 00:03:39.509 fact has this little number next to it, right? 00:03:39.870 --> 00:03:42.150 And it links directly back to the exact spot 00:03:42.150 --> 00:03:44.069 in your original document So you can check it 00:03:44.069 --> 00:03:46.370 instantly super important. Yep, and then over 00:03:46.370 --> 00:03:49.270 on the right. There's the creative toolbox They 00:03:49.270 --> 00:03:51.270 call it the studio panel. Think of that as the 00:03:51.270 --> 00:03:53.490 workshop That's where the AI takes your simple 00:03:53.490 --> 00:03:56.090 data and turns it into useful things quizzes 00:03:56.090 --> 00:03:58.930 mind maps flashcards that kind of stuff The flexibility 00:03:58.930 --> 00:04:01.710 seems like the big win here. Students could use 00:04:01.710 --> 00:04:04.770 it for prepping for complex exams, easily grabbing 00:04:04.770 --> 00:04:07.530 citations for essays. Writers could use it to 00:04:07.530 --> 00:04:10.469 organize tons of research from different articles 00:04:10.469 --> 00:04:12.689 for a book or song. Absolutely. And for professionals. 00:04:12.830 --> 00:04:15.530 Imagine you get that 50 -page business report 00:04:15.530 --> 00:04:19.439 Monday morning. Dense stuff. Instead of spending 00:04:19.439 --> 00:04:22.379 hours reading it, you just feed it into the conversation 00:04:22.379 --> 00:04:24.819 space and ask, okay, what are the three main 00:04:24.819 --> 00:04:27.399 risks here and how do they suggest mitigating 00:04:27.399 --> 00:04:29.720 them? Boom, you're ready for the meeting. Right. 00:04:30.680 --> 00:04:33.240 So beyond just studying your reading summaries, 00:04:34.100 --> 00:04:36.899 what's maybe the fastest way a professional could 00:04:36.899 --> 00:04:39.699 use that citation feature? especially when they're 00:04:39.699 --> 00:04:41.980 on a tight deadline. Those quick citation checks 00:04:41.980 --> 00:04:44.560 just build credibility instantly. You know, I'll 00:04:44.560 --> 00:04:46.560 admit, I still wrestle with prompt drift myself 00:04:46.560 --> 00:04:49.459 sometimes. So being able to quickly verify the 00:04:49.459 --> 00:04:51.680 AI's claim against the source, that's essential. 00:04:52.079 --> 00:04:54.040 Okay, let's make this practical. Let's walk through 00:04:54.040 --> 00:04:56.220 setting up a notebook. Say we're researching. 00:04:56.660 --> 00:04:59.439 Sustainable living. Pretty common topic. Sure. 00:05:00.000 --> 00:05:02.959 So step one is easy. Sign in, click PUP plus 00:05:02.959 --> 00:05:05.620 new notebook. Step two is all about the sourcing. 00:05:05.660 --> 00:05:07.519 This is where you feed the AI your knowledge 00:05:07.519 --> 00:05:10.339 base. You can pull files from Google Drive, upload 00:05:10.339 --> 00:05:12.920 that important PDF guide you found, paste in 00:05:12.920 --> 00:05:15.379 a Wikipedia link. And this is cool. Paste a YouTube 00:05:15.379 --> 00:05:17.339 video link. It'll grab the whole transcript. 00:05:17.579 --> 00:05:19.899 Okay, so everything's loaded in. Now you start 00:05:19.899 --> 00:05:21.779 the conversation, right, in that chat window. 00:05:22.439 --> 00:05:25.300 But the key is getting specific with your prompts, 00:05:25.459 --> 00:05:27.480 isn't it? Not just general questions. Absolutely. 00:05:27.540 --> 00:05:29.399 You've got to move beyond what is sustainable 00:05:29.399 --> 00:05:32.800 living. Ask for something specific like, are 00:05:32.800 --> 00:05:34.779 there any tips about saving water mentioned in 00:05:34.779 --> 00:05:37.560 the PDF and the transcript? Or maybe a comparison 00:05:37.560 --> 00:05:40.259 prompt. Compare how the Wikipedia article defines 00:05:40.259 --> 00:05:42.500 sustainable living versus how the YouTube video 00:05:42.500 --> 00:05:45.910 explains it. Get specific. And again, every answer 00:05:45.910 --> 00:05:48.910 it gives is grounded by those little clickable 00:05:48.910 --> 00:05:51.449 citation numbers. It's like built -in fact checking. 00:05:51.709 --> 00:05:53.889 Makes the whole thing reliable. Yeah, it's like 00:05:53.889 --> 00:05:56.129 having a super efficient librarian who points 00:05:56.129 --> 00:05:58.550 you to the exact sentence on the right page every 00:05:58.550 --> 00:06:00.750 single time. Cuts down massively on that time 00:06:00.750 --> 00:06:02.910 you'd normally spend checking footnotes. That 00:06:02.910 --> 00:06:05.870 sounds really useful. But what about, what if 00:06:05.870 --> 00:06:08.819 I feed it two sources that totally clash? Like, 00:06:09.060 --> 00:06:11.459 one article says plastic recycling is great and 00:06:11.459 --> 00:06:13.680 another says it's basically useless. Does the 00:06:13.680 --> 00:06:16.300 AI try to pick a side or smooth it over? Good 00:06:16.300 --> 00:06:19.019 question. No, it's designed to actually highlight 00:06:19.019 --> 00:06:21.740 the conflict. It'll usually say something like, 00:06:21.860 --> 00:06:25.720 source one says X, but source two says Y. It 00:06:25.720 --> 00:06:28.779 presents the differing views with the citations 00:06:28.779 --> 00:06:31.420 so you can see the conflict and decide based 00:06:31.420 --> 00:06:33.740 on the evidence. It won't create some kind of 00:06:33.740 --> 00:06:36.620 false consensus. Okay, that makes sense. So, 00:06:36.620 --> 00:06:38.360 say I've just started looking into this topic. 00:06:38.939 --> 00:06:41.639 How can I find more trustworthy info without 00:06:41.639 --> 00:06:43.959 having to leave the notebook interface? Ah, yeah. 00:06:43.959 --> 00:06:46.000 You can use the new Discover Sources feature. 00:06:46.319 --> 00:06:48.660 It suggests related, high -quality sources, often 00:06:48.660 --> 00:06:50.800 from places like universities or research groups 00:06:50.800 --> 00:06:52.819 based on what you've already uploaded. All right. 00:06:52.819 --> 00:06:54.360 So we've talked about pulling information out. 00:06:54.600 --> 00:06:57.120 Now let's shift to creating things with it. That's 00:06:57.120 --> 00:06:59.079 where the Studio Panel, the creative toolkit, 00:06:59.160 --> 00:07:02.160 comes in. This seems like where the AI goes beyond 00:07:02.160 --> 00:07:04.259 just summarizing. It actually starts generating 00:07:04.259 --> 00:07:07.180 useful, finished, knowledge products for you. 00:07:07.230 --> 00:07:09.689 Yeah, and the strategic value here is important. 00:07:09.750 --> 00:07:12.970 It's not just about, say, making flashcards. 00:07:13.050 --> 00:07:15.949 It's about making flashcards based only on the 00:07:15.949 --> 00:07:18.769 specific technical terms from your proprietary 00:07:18.769 --> 00:07:20.990 documents that you need to memorize. It's targeted. 00:07:21.509 --> 00:07:23.750 OK, like that discover sources feature we just 00:07:23.750 --> 00:07:26.029 mentioned. Yeah. That sounds like a real research 00:07:26.029 --> 00:07:28.850 booster. It looks at your stuff and then actively 00:07:28.850 --> 00:07:31.449 suggests related academic sources. And you can 00:07:31.449 --> 00:07:32.990 get really specific with the prompt there, too. 00:07:33.089 --> 00:07:34.550 Oh, yeah, definitely. You could ask something 00:07:34.550 --> 00:07:37.699 like, Based on my documents about recycling and 00:07:37.699 --> 00:07:40.620 composting, find me five recent academic papers 00:07:40.620 --> 00:07:44.420 on the circular economy. And also effective waste 00:07:44.420 --> 00:07:47.040 management policies, specifically in major port 00:07:47.040 --> 00:07:49.980 cities in Southeast Asia. I mean, that's a pretty 00:07:49.980 --> 00:07:52.839 sophisticated targeted research request in just 00:07:52.839 --> 00:07:56.079 one go. Then you got the study tools, like quizzes. 00:07:56.379 --> 00:07:58.079 It just makes them automatically to test your 00:07:58.079 --> 00:08:01.100 knowledge. Exactly. And you can demand really 00:08:01.100 --> 00:08:03.199 specific kinds of quizzes, too. You might tell 00:08:03.199 --> 00:08:06.459 it, create a quiz, make it 15 difficult questions, 00:08:06.660 --> 00:08:09.279 mix of multiple choice and true -false, and focus 00:08:09.279 --> 00:08:11.779 only on the stats and scientific terms from these 00:08:11.779 --> 00:08:14.819 specific science reports I uploaded. Ignore the 00:08:14.819 --> 00:08:17.100 general concepts. It tests exactly what you need 00:08:17.100 --> 00:08:20.759 it to test. And flashcards for just straightforward 00:08:20.759 --> 00:08:23.360 memorization. Like, make me 30 flashcards to 00:08:23.360 --> 00:08:26.319 memorize key definitions about sustainable farming 00:08:26.319 --> 00:08:30.139 techniques. on things like permaculture or hydroponics 00:08:30.139 --> 00:08:32.220 from my sources. Yeah. And personally, I really 00:08:32.220 --> 00:08:34.539 like the mind maps. For visual learners, seeing 00:08:34.539 --> 00:08:37.720 that big picture is so helpful. It lays it out 00:08:37.720 --> 00:08:39.860 visually. You know, like sustainable living in 00:08:39.860 --> 00:08:42.399 the middle, then main branches like reduce waste, 00:08:42.600 --> 00:08:44.519 save energy, and then smaller branches off those, 00:08:44.659 --> 00:08:46.679 like composting, recycling. It turns a dense 00:08:46.679 --> 00:08:48.759 topic into something you can see. These tools 00:08:48.759 --> 00:08:51.379 really do seem to move way beyond just summarizing. 00:08:51.700 --> 00:08:53.679 So how quickly can they actually take us from 00:08:53.679 --> 00:08:57.080 just raw data, I mean, messy notes, to a finished 00:08:57.230 --> 00:09:00.029 maybe even professional -level output. Whoa. 00:09:00.710 --> 00:09:03.309 I mean, imagine creating a whole structured presentation 00:09:03.309 --> 00:09:06.409 video, fully sourced from Documency Trust, in 00:09:06.409 --> 00:09:09.730 just minutes. It drastically speeds up both the 00:09:09.730 --> 00:09:11.750 learning part and the actual production part. 00:09:12.370 --> 00:09:14.389 And that speed in production brings us to the 00:09:14.389 --> 00:09:17.549 really polished, media -ready outputs. First 00:09:17.549 --> 00:09:20.450 up, there's the audio overview. This thing turns 00:09:20.450 --> 00:09:22.870 your documents into a short, kind of flexible 00:09:22.870 --> 00:09:25.909 audio summary, like a custom mini podcast just 00:09:25.909 --> 00:09:28.549 for you, based on your stuff. That sounds great 00:09:28.549 --> 00:09:31.029 for absorbing complex info while you're, like, 00:09:31.129 --> 00:09:32.929 commuting or exercising. You could prompt it 00:09:32.929 --> 00:09:35.129 something like, make a seven -minute audio overview, 00:09:35.549 --> 00:09:37.730 use the style of a thoughtful documentary narrator, 00:09:38.269 --> 00:09:40.230 tell a story about plastic pollution's impact, 00:09:40.509 --> 00:09:42.110 and present the solutions found in my notes. 00:09:42.519 --> 00:09:45.259 Pretty cool and then the video overview. Honestly, 00:09:45.399 --> 00:09:47.779 this one feels pretty groundbreaking. It automatically 00:09:47.779 --> 00:09:49.659 generates a three to four minute presentation 00:09:49.659 --> 00:09:51.840 video We're talking dynamic slides a voiceover 00:09:51.840 --> 00:09:54.399 all created automatically and based only on your 00:09:54.399 --> 00:09:57.559 source documents Wow, so you could build a professional 00:09:57.559 --> 00:10:01.379 looking deck on say the impact of fast fashion 00:10:01.529 --> 00:10:04.590 You'd tell it to structure the video into maybe 00:10:04.590 --> 00:10:07.470 environmental costs, social impacts, and sustainable 00:10:07.470 --> 00:10:09.929 alternatives, feed it the source articles and 00:10:09.929 --> 00:10:12.350 reports, and you've got a presentation ready 00:10:12.350 --> 00:10:14.370 in minutes. Pretty much. And it's not just for 00:10:14.370 --> 00:10:16.570 English speakers. You can generate the script 00:10:16.570 --> 00:10:18.649 and slides in other languages too, like Vietnamese, 00:10:18.730 --> 00:10:21.090 for example. Makes your research accessible much 00:10:21.090 --> 00:10:23.629 more broadly. That speed is definitely incredible. 00:10:23.850 --> 00:10:27.070 But it does make me wonder, if you generate a 00:10:27.070 --> 00:10:30.100 video presentation, in, say, three minutes, instead 00:10:30.100 --> 00:10:33.460 of spending 20 hours crafting it, does that maybe 00:10:33.460 --> 00:10:36.860 bypass some of the critical thinking needed to 00:10:36.860 --> 00:10:39.320 really internalize the material? That's a really 00:10:39.320 --> 00:10:41.360 thoughtful point. I think the value shifts, you 00:10:41.360 --> 00:10:43.740 know? You save a ton of time on the design aspect, 00:10:44.200 --> 00:10:46.259 but you probably need to spend more focused time 00:10:46.259 --> 00:10:48.480 on the prompting, making sure the structure is 00:10:48.480 --> 00:10:50.980 exactly right. and then critically verifying 00:10:50.980 --> 00:10:53.779 the content using those citations. So the critical 00:10:53.779 --> 00:10:56.100 thought moves maybe from creation to validation 00:10:56.100 --> 00:10:58.080 and synthesis. It's a different kind of engagement. 00:10:58.940 --> 00:11:01.100 OK, so to make sure that synthesis is high quality, 00:11:01.539 --> 00:11:03.879 we have those three best practices. First one, 00:11:04.620 --> 00:11:07.720 quality over quantity. Garbage in, garbage out. 00:11:08.059 --> 00:11:10.480 Always holds true, right? Absolutely. Second, 00:11:10.879 --> 00:11:14.000 be specific. Don't just ask vaguely, Tell me 00:11:14.000 --> 00:11:17.100 about recycling. Ask. List the step -by -step 00:11:17.100 --> 00:11:19.500 process for recycling plastic bottles correctly, 00:11:19.940 --> 00:11:23.320 based only on the guidelines in Document X. Specific 00:11:23.320 --> 00:11:25.919 prompts get accurate results. Makes sense. And 00:11:25.919 --> 00:11:28.279 third, break down big topics. If you're tackling 00:11:28.279 --> 00:11:30.000 something huge like, I don't know, the entire 00:11:30.000 --> 00:11:31.840 history of the space shuttle program, maybe split 00:11:31.840 --> 00:11:34.500 it up. Make smaller, focused notebooks, one for 00:11:34.500 --> 00:11:36.279 design, one for missions, one for the end of 00:11:36.279 --> 00:11:38.580 the program. Keep the AI focused on manageable 00:11:38.580 --> 00:11:41.659 chunks. Right. So given how fast these AI tools 00:11:41.659 --> 00:11:43.919 are changing all the time, what should listeners 00:11:43.919 --> 00:11:46.139 maybe keep an eye out for? What should they prioritize 00:11:46.139 --> 00:11:48.779 checking for next week with Notebook LM? Yeah, 00:11:48.860 --> 00:11:51.200 good question. I'd say just check back often. 00:11:51.340 --> 00:11:54.159 See what new creative tools pop up or what new 00:11:54.159 --> 00:11:56.740 kinds of sources you can upload. The pace of 00:11:56.740 --> 00:11:59.039 updates is pretty fast. And it's all aimed at 00:11:59.039 --> 00:12:01.100 helping you learn faster and get more inventive 00:12:01.100 --> 00:12:04.669 with your own knowledge. So when you step back, 00:12:04.889 --> 00:12:06.970 Notebook LM really feels like it changes how 00:12:06.970 --> 00:12:08.850 we interact with our own knowledge. It's not 00:12:08.850 --> 00:12:11.330 just another general search tool. It becomes 00:12:11.330 --> 00:12:14.269 this personalized, creative, and importantly, 00:12:14.809 --> 00:12:17.210 highly accurate assistant for managing knowledge. 00:12:17.470 --> 00:12:19.590 Yeah, the goal shifts, doesn't it? It's not just 00:12:19.590 --> 00:12:22.049 about collecting facts anymore. It's about actively 00:12:22.049 --> 00:12:24.830 using them, synthesizing them, creating new things 00:12:24.830 --> 00:12:27.750 based on them. Outputs you can actually trust 00:12:27.750 --> 00:12:29.970 because they're tied directly back to your sources. 00:12:30.200 --> 00:12:33.539 So the question for you listening is, what complex 00:12:33.539 --> 00:12:36.639 50 page report or maybe what huge online course 00:12:36.639 --> 00:12:38.519 or lecture series are you going to finally master 00:12:38.519 --> 00:12:40.960 using these kinds of features? It's time to stop 00:12:40.960 --> 00:12:43.600 feeling buried by all that data and actually 00:12:43.600 --> 00:12:45.559 start organizing that knowledge effectively. 00:12:45.779 --> 00:12:47.960 Yeah, just start building that first notebook. 00:12:48.720 --> 00:12:50.940 You might genuinely be surprised how fast you 00:12:50.940 --> 00:12:54.460 can take maybe years of saved articles and notes 00:12:54.460 --> 00:12:56.919 and turn them into a clear, verifiable summary. 00:12:57.360 --> 00:12:58.700 We'll catch you on the next Deep Dive.