WEBVTT 00:00:00.000 --> 00:00:03.279 What if a single passing thought on your commute 00:00:03.279 --> 00:00:06.379 could automatically build a fully researched, 00:00:06.639 --> 00:00:09.800 organized project plan before you even sit at 00:00:09.800 --> 00:00:11.880 your desk? I mean, that sounds total like science 00:00:11.880 --> 00:00:15.640 fiction. Right. Beat. It really does. But you 00:00:15.640 --> 00:00:17.940 can actually build this today. You just need 00:00:17.940 --> 00:00:20.699 to connect your personal archives to an AI action 00:00:20.699 --> 00:00:23.039 layer. Yeah, it's a massive shift in how we approach 00:00:23.039 --> 00:00:25.379 our daily work, honestly. Welcome to the Deep 00:00:25.379 --> 00:00:28.760 Dive. Today, our mission is unpacking a fascinating 00:00:28.760 --> 00:00:32.280 workflow. We're looking at how two specific tools, 00:00:32.579 --> 00:00:36.100 Hermes and Notebook LM, create a 247 personal 00:00:36.100 --> 00:00:38.659 research assistant. And we're moving from a world 00:00:38.659 --> 00:00:41.560 where we just, you know, passively query information 00:00:41.560 --> 00:00:44.600 to one where our systems actively execute tasks 00:00:44.600 --> 00:00:46.560 for us. We're going to explore what these tools 00:00:46.560 --> 00:00:49.079 do. We'll look at why combining them is such 00:00:49.079 --> 00:00:51.359 a breakthrough and dive into the standout features. 00:00:51.859 --> 00:00:54.719 Plus the real world applications. Exactly. And 00:00:54.719 --> 00:00:56.939 crucially, we'll cover the severe security risks 00:00:56.939 --> 00:00:59.840 you need to navigate. But before we build the 00:00:59.840 --> 00:01:02.119 machine... We need to understand the parts. Right. 00:01:02.200 --> 00:01:04.280 So let's start with the research brain of the 00:01:04.280 --> 00:01:07.359 operation. That is Notebook LM. Most people are 00:01:07.359 --> 00:01:09.319 probably familiar with it by now. Yeah, it's 00:01:09.319 --> 00:01:12.540 this incredibly powerful platform. It's designed 00:01:12.540 --> 00:01:16.379 specifically to store and synthesize your personal 00:01:16.379 --> 00:01:18.840 sources. So you upload your own stuff. Exactly. 00:01:19.319 --> 00:01:23.079 Dense PDFs, long strategy documents, or like 00:01:23.079 --> 00:01:25.840 web links. And then it interacts with you based 00:01:25.840 --> 00:01:28.260 entirely on those specific materials. It essentially 00:01:28.260 --> 00:01:30.359 walls itself off from the broader internet. It 00:01:30.359 --> 00:01:33.060 does. It turns your private data into highly 00:01:33.060 --> 00:01:36.359 accurate summaries. It builds intricate mind 00:01:36.359 --> 00:01:40.260 maps. It even generates those those incredibly 00:01:40.260 --> 00:01:44.239 popular podcast style audio overviews. Which 00:01:44.239 --> 00:01:46.379 are wild to listen to. They really are. But the 00:01:46.379 --> 00:01:48.819 key is that it grounds every single response 00:01:48.819 --> 00:01:52.040 in your provided data. Right. And then we have 00:01:52.040 --> 00:01:54.680 the second player in this workflow, Hermes. Yeah. 00:01:54.780 --> 00:01:57.319 So if Notebook LM is the brain, Hermes is, well, 00:01:57.379 --> 00:01:59.840 it's the ham. The action layer. Exactly. Hermes 00:01:59.840 --> 00:02:02.200 operates as the control center for this entire 00:02:02.200 --> 00:02:04.219 workflow. It doesn't just passively retrieve 00:02:04.219 --> 00:02:06.299 information. It actually triggers actions. Right. 00:02:06.379 --> 00:02:08.879 It triggers external workflows, like it can draft 00:02:08.879 --> 00:02:10.879 your scripts, save your notes into your system, 00:02:10.919 --> 00:02:13.520 or even schedule follow -up reminders. So to 00:02:13.520 --> 00:02:17.180 synthesize this, it's like having a genius researcher 00:02:17.180 --> 00:02:20.530 locked in the archives. That's Notebook LM. And 00:02:20.530 --> 00:02:22.789 at the front desk, you have a highly capable 00:02:22.789 --> 00:02:26.810 executive assistant. That's Hermes. And finally, 00:02:26.909 --> 00:02:28.969 they have a walkie -talkie to talk to each other. 00:02:29.090 --> 00:02:32.150 That is a perfect analogy. Notebook LM does the 00:02:32.150 --> 00:02:35.150 heavy lifting of reading dense texts, and Hermes 00:02:35.150 --> 00:02:37.150 parses your intent and delivers the results. 00:02:37.509 --> 00:02:39.770 I do have to ask, though, about a common problem. 00:02:40.370 --> 00:02:43.069 How do we know the AI won't just invent fake 00:02:43.069 --> 00:02:46.469 facts? Ah, hallucinations. Yeah, that's a massive 00:02:46.469 --> 00:02:49.310 issue when automating research. Right. Because 00:02:49.310 --> 00:02:51.909 you aren't watching it work. But the answer lies 00:02:51.909 --> 00:02:55.389 in Notebook LM's core design. It has a highly 00:02:55.389 --> 00:02:58.550 constrained retrieval system. It forces the model 00:02:58.550 --> 00:03:01.650 to restrict its answers exclusively to the documents 00:03:01.650 --> 00:03:03.909 you uploaded. So if it's not in the text. It 00:03:03.909 --> 00:03:06.129 just refuses to answer. It won't guess. So it 00:03:06.129 --> 00:03:07.969 only reads your uploaded documents, not the whole 00:03:07.969 --> 00:03:10.949 internet. Exactly. And that hard limitation is 00:03:10.949 --> 00:03:13.780 what makes it trustworthy. Now, because both 00:03:13.780 --> 00:03:16.680 of these tools are so powerful alone, we have 00:03:16.680 --> 00:03:18.680 to ask why we need them to talk to each other. 00:03:18.780 --> 00:03:20.780 Right. Why bother connecting them? Yeah. Beat. 00:03:21.060 --> 00:03:24.680 And the answer is friction. Friction is the enemy 00:03:24.680 --> 00:03:27.479 of execution. Just think about the standard notebook 00:03:27.479 --> 00:03:29.740 LM workflow right now. You have an idea. You 00:03:29.740 --> 00:03:32.030 have to break your focus. Totally. You manually 00:03:32.030 --> 00:03:34.870 open a browser, navigate to the site, log in, 00:03:34.990 --> 00:03:37.289 scroll to find the notebook, type your query. 00:03:37.409 --> 00:03:39.810 And then copy and paste the output. Right. Back 00:03:39.810 --> 00:03:41.789 into whatever app you were originally using. 00:03:41.889 --> 00:03:44.409 It's exhausting. I have to admit, I still wrestle 00:03:44.409 --> 00:03:46.990 with having 20 tabs open just to research a new 00:03:46.990 --> 00:03:50.189 microphone. Huh. We all do. The context switching 00:03:50.189 --> 00:03:52.990 is brutal on your working memory. Yeah. But Hermes 00:03:52.990 --> 00:03:55.530 eliminates this. Walk me through the actual sequence. 00:03:55.789 --> 00:03:58.389 Sure. So you just message Hermes in your chat. 00:03:58.550 --> 00:04:02.389 Okay. Beat. Hermes analyzes it and decides if 00:04:02.389 --> 00:04:04.810 Notebook LM is needed. It makes that choice itself. 00:04:05.150 --> 00:04:08.169 Yep. Beat, beat. Then Notebook LM processes the 00:04:08.169 --> 00:04:11.430 sources. Beat. And finally, Hermes returns the 00:04:11.430 --> 00:04:13.370 answer and helps you act on it. That's amazing. 00:04:13.610 --> 00:04:16.949 But why is a simple chat interface actually superior 00:04:16.949 --> 00:04:20.250 to a dedicated visual dashboard? Dashboards give 00:04:20.250 --> 00:04:22.689 you control. Because chat removes the barrier 00:04:22.689 --> 00:04:26.110 between having an idea and executing it. It captures 00:04:26.110 --> 00:04:28.439 your raw intent immediately. You don't have to 00:04:28.439 --> 00:04:31.079 navigate drop -down menus. Exactly. You just 00:04:31.079 --> 00:04:33.899 state your goal in plain English, and Hermes 00:04:33.899 --> 00:04:37.259 acts as the translation layer. Chat makes complex, 00:04:37.480 --> 00:04:40.740 multi -step research as easy as texting a smart 00:04:40.740 --> 00:04:43.439 friend. Yeah, intent -driven execution. It's 00:04:43.439 --> 00:04:45.720 a total paradigm shift. Now that the friction 00:04:45.720 --> 00:04:48.420 is gone, let's look at the wild features this 00:04:48.420 --> 00:04:50.699 actually unlocks. This is the fun part. Let's 00:04:50.699 --> 00:04:53.060 talk about voice notes. Okay. Imagine you're 00:04:53.060 --> 00:04:56.300 walking down a busy street. No screen. You just 00:04:56.300 --> 00:04:58.519 hold a button on your phone and ask for research 00:04:58.519 --> 00:05:01.060 on, like, Ralph Lauren jumpers. You're just rambling 00:05:01.060 --> 00:05:04.620 naturally. Rambling naturally. Hermes transcribes 00:05:04.620 --> 00:05:07.300 it, extracts the intent, and turns it into a 00:05:07.300 --> 00:05:10.040 notebook LM task instantly. You don't even touch 00:05:10.040 --> 00:05:12.579 a keyboard. Not at all. And it gets better. You 00:05:12.579 --> 00:05:14.899 know those podcast -style audio overviews and 00:05:14.899 --> 00:05:16.600 mind maps? Yeah, they're great. You can retrieve 00:05:16.600 --> 00:05:18.800 those directly in the chat. Hermes pulls those 00:05:18.800 --> 00:05:21.019 heavy files right into your thread. You just 00:05:21.019 --> 00:05:24.750 ask? and the image renders right there but here's 00:05:24.750 --> 00:05:27.410 where it gets crazy you can introduce external 00:05:27.410 --> 00:05:31.029 connective tools like nan okay let's clarify 00:05:31.029 --> 00:05:34.670 that n8n a tool that visually connects different 00:05:34.670 --> 00:05:37.009 apps together perfect it's the digital plumbing 00:05:37.009 --> 00:05:40.629 and you can also use mcp mcp a universal plug 00:05:40.629 --> 00:05:43.470 that lets ai control other software right so 00:05:43.470 --> 00:05:46.350 you use mcp to connect hermes to your nnn workflows 00:05:46.350 --> 00:05:49.490 now your assistant is actively moving data across 00:05:49.490 --> 00:05:53.490 your digital life whoa Two sec silence. Imagine 00:05:53.490 --> 00:05:55.910 turning a passing thought into an automated daily 00:05:55.910 --> 00:05:58.769 morning briefing. Just waking up to a fully formatted 00:05:58.769 --> 00:06:01.810 intelligence report. That's incredible. But I 00:06:01.810 --> 00:06:04.529 have to push back here, Beat. Is it truly safe 00:06:04.529 --> 00:06:08.089 to let an AI agent use MCP to control other apps? 00:06:08.370 --> 00:06:11.329 That is a very valid concern. The text issues 00:06:11.329 --> 00:06:14.050 a strong warning about this exact thing. You 00:06:14.050 --> 00:06:16.529 must be incredibly careful. Especially with authentication, 00:06:16.829 --> 00:06:19.680 right? Yes. Do not ever paste private credentials, 00:06:19.939 --> 00:06:23.379 API keys, or cookies into insecure web -based 00:06:23.379 --> 00:06:25.639 environments. Because if it's compromised, they 00:06:25.639 --> 00:06:27.759 have your digital identity. Precisely. Never 00:06:27.759 --> 00:06:30.040 paste them into random tools. Keep your digital 00:06:30.040 --> 00:06:33.100 keys safe. Never paste them into random or unverified 00:06:33.100 --> 00:06:35.560 tools. Exactly. Security has to be the top priority. 00:06:35.899 --> 00:06:38.040 Let's take a quick break here. Insert mid -roll 00:06:38.040 --> 00:06:42.259 sponsor read here. Welcome back. So knowing these 00:06:42.259 --> 00:06:45.060 capabilities is really exciting, but we need 00:06:45.060 --> 00:06:47.889 to ground this in reality. Yeah, we have to talk 00:06:47.889 --> 00:06:51.529 about the setup. Exactly. How hard is the actual 00:06:51.529 --> 00:06:53.709 plumbing to put this together? I won't sugarcoat 00:06:53.709 --> 00:06:56.389 it. It is unglamorous. There are some highly 00:06:56.389 --> 00:06:59.350 necessary, tedious steps. Walk us through it. 00:07:00.009 --> 00:07:03.089 First, you have to download the notebook LM skill. 00:07:03.490 --> 00:07:06.209 Because there's no official API yet, this is 00:07:06.209 --> 00:07:08.509 a workaround. You're manually installing it into 00:07:08.509 --> 00:07:10.730 your environment. Right. Then comes the browser 00:07:10.730 --> 00:07:13.180 authentication. This is the tricky part. Because 00:07:13.180 --> 00:07:15.160 you're dealing with live sessions. Yeah, you 00:07:15.160 --> 00:07:16.980 have to connect it to your Google account by 00:07:16.980 --> 00:07:19.839 extracting live access tokens and cookies directly 00:07:19.839 --> 00:07:22.379 from your browser's developer tools. Which goes 00:07:22.379 --> 00:07:24.379 back to our critical safety warning. Exactly. 00:07:24.420 --> 00:07:26.160 You have to handle those cookies very carefully. 00:07:26.339 --> 00:07:28.800 Once you feed them into Hermes, you run a test 00:07:28.800 --> 00:07:31.879 prompt. Like what? You just say, find one of 00:07:31.879 --> 00:07:34.240 my last three notebooks. If it works, the bridge 00:07:34.240 --> 00:07:37.500 is up. But relying on an unofficial integration 00:07:37.500 --> 00:07:40.800 skill, isn't that system incredibly fragile? 00:07:41.290 --> 00:07:44.470 It is. Because it's unofficial, any future updates 00:07:44.470 --> 00:07:46.730 to the Notebook LM interface could temporarily 00:07:46.730 --> 00:07:49.470 break your entire workflow. Unofficial tools 00:07:49.470 --> 00:07:52.870 can break, so expect to tweak the plumbing occasionally. 00:07:53.269 --> 00:07:55.589 Yeah, that's the price of being an early adopter 00:07:55.589 --> 00:07:57.490 right now. Assuming you're willing to manage 00:07:57.490 --> 00:08:00.209 that setup, who is this actually built for in 00:08:00.209 --> 00:08:01.930 the real world? The applications are actually 00:08:01.930 --> 00:08:05.439 vast. Let's look at creators first. A creator 00:08:05.439 --> 00:08:08.000 can dump historical scripts and analytics into 00:08:08.000 --> 00:08:11.500 a notebook and have Hermes generate 10 new video 00:08:11.500 --> 00:08:14.560 hook ideas that match their specific voice. That 00:08:14.560 --> 00:08:17.040 cures blank page syndrome instantly. Totally. 00:08:17.180 --> 00:08:20.220 Or founders. They can upload 50 dense competitor 00:08:20.220 --> 00:08:23.779 PDFs. Nobody wants to read 50 PDFs. Exactly. 00:08:23.959 --> 00:08:26.939 So they ask Hermes to find positioning gaps in 00:08:26.939 --> 00:08:28.920 the market based on that research. What about 00:08:28.920 --> 00:08:32.210 students? Oh, it's huge for them. They turn dense 00:08:32.210 --> 00:08:34.490 academic papers into beginner -friendly study 00:08:34.490 --> 00:08:37.870 guides, pulling out only the key terms. And everyday 00:08:37.870 --> 00:08:41.009 shoppers. Yeah, say you're comparing $500 standing 00:08:41.009 --> 00:08:43.750 desks. You just feed the specs into the notebook, 00:08:43.850 --> 00:08:46.190 and Hermes cross -references the warranties and 00:08:46.190 --> 00:08:48.850 materials without you opening 20 tabs. It saves 00:08:48.850 --> 00:08:52.320 so much manual effort. And travelers. You're 00:08:52.320 --> 00:08:55.519 going to Rome. You put historical articles and 00:08:55.519 --> 00:08:58.379 maps into a notebook, and Hermes generates a 00:08:58.379 --> 00:09:01.519 day -by -day guide to the Colosseum. The underlying 00:09:01.519 --> 00:09:04.960 theme here is really interesting. Beat. It stops 00:09:04.960 --> 00:09:07.320 your second brain from becoming a graveyard of 00:09:07.320 --> 00:09:09.940 untouched notes. Oh, absolutely. We all have 00:09:09.940 --> 00:09:13.039 those folders of saved PDFs we never open. This 00:09:13.039 --> 00:09:15.080 actually makes them useful. But I have to ask 00:09:15.080 --> 00:09:17.299 thoughtfully here, is there a danger in letting 00:09:17.299 --> 00:09:21.179 AI synthesize everything? Do we lose our own 00:09:21.179 --> 00:09:23.710 judgment? That's a fair question. The cognitive 00:09:23.710 --> 00:09:26.450 friction of reading is important, but the AI 00:09:26.450 --> 00:09:29.049 just organizes and retrieves the data. So it 00:09:29.049 --> 00:09:31.090 preps the landscape. Right. It's meant to prep 00:09:31.090 --> 00:09:33.450 the information, not make the final human decision. 00:09:33.809 --> 00:09:36.049 It serves up the menu, but you still have to 00:09:36.049 --> 00:09:37.649 pick the meal. That's a great way to put it. 00:09:37.669 --> 00:09:40.269 You still make the call. So having seen the menu 00:09:40.269 --> 00:09:42.929 of possibilities, it's time for the ultimate 00:09:42.929 --> 00:09:46.009 cost -benefit analysis. Let's weigh it out. Before 00:09:46.009 --> 00:09:48.289 the listener decides to build this, what are 00:09:48.289 --> 00:09:50.480 the ultimate benefits? Well, the gains are pure 00:09:50.480 --> 00:09:53.340 speed, the reuse of forgotten knowledge, and 00:09:53.340 --> 00:09:55.980 drastically less context switching between apps. 00:09:56.259 --> 00:09:59.080 But the risks are real. You have that technical 00:09:59.080 --> 00:10:02.100 setup curve for beginners, the danger of leaked 00:10:02.100 --> 00:10:05.139 credentials if you mishandle cookies, and the 00:10:05.139 --> 00:10:08.740 inherent variability of AI output quality. It 00:10:08.740 --> 00:10:10.919 can still hallucinate if the grounding fails. 00:10:11.159 --> 00:10:14.399 What is the single biggest behavioral trap once 00:10:14.399 --> 00:10:18.000 someone gets this working? Over -reliance. Without 00:10:18.000 --> 00:10:20.759 a doubt. Specifically, letting automations run 00:10:20.759 --> 00:10:23.500 without human review. Like sending emails. Yes. 00:10:23.539 --> 00:10:26.240 Do not set up a workflow that auto sends an email 00:10:26.240 --> 00:10:28.600 or a client report without you reading it first. 00:10:28.779 --> 00:10:31.860 Never let the AI auto send an email or report 00:10:31.860 --> 00:10:33.940 without reading it first. Absolutely never. You 00:10:33.940 --> 00:10:36.539 are still responsible for the output. As we wrap 00:10:36.539 --> 00:10:38.279 up, let's look at the broader paradigm shift 00:10:38.279 --> 00:10:41.440 here. Beat. We are moving away from simple chatbots 00:10:41.440 --> 00:10:43.940 that just answer questions. Right. The old model 00:10:43.940 --> 00:10:46.480 was just a fancy search box. Exactly. And now 00:10:46.480 --> 00:10:49.480 we're moving toward integrated AI systems that 00:10:49.480 --> 00:10:52.340 research, organize, and execute physical digital 00:10:52.340 --> 00:10:55.480 tasks from a single interface. It's an active 00:10:55.480 --> 00:10:57.779 assistant, not a passive tool. It really is. 00:10:58.159 --> 00:10:59.980 So I want to leave you with a final thought to 00:10:59.980 --> 00:11:04.019 mull over. Beat. Think about the biggest, messiest 00:11:04.019 --> 00:11:06.700 folder of saved articles or PDFs you currently 00:11:06.700 --> 00:11:08.600 have sitting on your desktop. We all have one. 00:11:08.740 --> 00:11:11.460 Ask yourself, what would happen if that dead 00:11:11.460 --> 00:11:13.759 information could suddenly talk back to you and 00:11:13.759 --> 00:11:16.090 draft your next project? The possibilities are 00:11:16.090 --> 00:11:18.110 incredible once you remove the friction. Thank 00:11:18.110 --> 00:11:20.149 you for joining us on this deep dive. Stay curious. 00:11:20.269 --> 00:11:20.950 See you next time.