WEBVTT 00:00:00.000 --> 00:00:02.940 Welcome to the deep dive. We are going on a very 00:00:02.940 --> 00:00:05.860 specific mission today. But before we get into 00:00:05.860 --> 00:00:08.560 the mechanics of it, I want you to picture something. 00:00:08.580 --> 00:00:11.119 Oh, I know the exact scenario. You finally sit 00:00:11.119 --> 00:00:13.640 down to work on a major project, you open up 00:00:13.640 --> 00:00:16.219 a totally fresh chat window, and immediately 00:00:16.219 --> 00:00:19.260 you just feel this agonizing sense of exhaustion. 00:00:19.719 --> 00:00:22.359 Yeah. Because you suddenly realize you have to 00:00:22.359 --> 00:00:25.300 upload the exact same PDF. You have to explain 00:00:25.300 --> 00:00:28.719 your entire brand voice. Again, every single 00:00:28.719 --> 00:00:31.559 session basically starts from absolute zero. 00:00:31.859 --> 00:00:34.859 It is a profound friction. I think we are constantly 00:00:34.859 --> 00:00:37.179 introducing ourselves to our own tools. We totally 00:00:37.179 --> 00:00:39.520 are. It's exhausting. Today, we are looking at 00:00:39.520 --> 00:00:42.710 a brilliant guide on curing AI amnesia. The goal 00:00:42.710 --> 00:00:44.929 here is simple. We're going to build a long -term 00:00:44.929 --> 00:00:47.030 memory system for your work. And the best part 00:00:47.030 --> 00:00:49.450 is we'll use two completely free tools for this, 00:00:49.909 --> 00:00:52.670 Cloud Projects and Notebook LM. There is no coding 00:00:52.670 --> 00:00:55.229 involved, there are no complicated scripts, and 00:00:55.229 --> 00:00:57.329 it takes under 20 minutes to actually set up. 00:00:57.530 --> 00:00:59.189 Yeah, and once you finally build this system 00:00:59.189 --> 00:01:01.549 out, you'll never have to brief your AI from 00:01:01.549 --> 00:01:04.030 scratch again. It honestly changes everything 00:01:04.030 --> 00:01:06.569 about how you work day to day. Let's explore 00:01:06.569 --> 00:01:09.280 the hidden cost of this amnesia first. Before 00:01:09.280 --> 00:01:11.159 we can actually build the solution, we have to 00:01:11.159 --> 00:01:14.140 understand it. Why is the current way we use 00:01:14.140 --> 00:01:17.939 AI quietly draining our productivity? Well, it's 00:01:17.939 --> 00:01:20.700 sneaky. It feels like a minor annoyance in the 00:01:20.700 --> 00:01:23.760 moment. But it actually scales terribly across 00:01:23.760 --> 00:01:26.319 a work week. Right. Every time you start a new 00:01:26.319 --> 00:01:29.500 chat, you're wasting valuable time. You're wasting 00:01:29.500 --> 00:01:32.379 your own mental energy just doing setup work. 00:01:32.540 --> 00:01:34.819 The source material uses a really great analogy 00:01:34.819 --> 00:01:37.670 here. Claude's memory is basically like a dry 00:01:37.670 --> 00:01:40.469 erase whiteboard. The exact moment you close 00:01:40.469 --> 00:01:43.430 that chat tad, it's wiped completely clean. Exactly 00:01:43.430 --> 00:01:45.810 that. It just vanishes into thin air. Now Claude 00:01:45.810 --> 00:01:47.790 does try to synthesize your conversations every 00:01:47.790 --> 00:01:50.349 24 hours, right? It does, yeah. But that synthesis 00:01:50.349 --> 00:01:53.310 only applies to new standalone chats. It's very 00:01:53.310 --> 00:01:56.150 broad. So if you had this incredibly deep work 00:01:56.150 --> 00:01:59.090 session on Tuesday, By Friday morning, it is 00:01:59.090 --> 00:02:01.049 entirely forgotten. You are completely back to 00:02:01.049 --> 00:02:03.109 square one. Yep. You're re -explaining all the 00:02:03.109 --> 00:02:05.129 core concepts before you can do any real work. 00:02:05.230 --> 00:02:07.969 And then there is the token tax. Oh yes. This 00:02:07.969 --> 00:02:09.930 is the part almost nobody talks about. Let's 00:02:09.930 --> 00:02:12.050 define that quickly for everyone. Tokens are 00:02:12.050 --> 00:02:15.750 tiny pieces of words AI uses to charge you. Perfect 00:02:15.750 --> 00:02:18.870 definition. So, I mean, think about pasting a 00:02:18.870 --> 00:02:21.389 5 ,000 -word brand guide. Right. If you drop 00:02:21.389 --> 00:02:23.389 that into every single new session you open, 00:02:23.870 --> 00:02:26.349 you are repeatedly paying for that exact same 00:02:26.349 --> 00:02:29.990 data. Over and over again. Exactly. Across a 00:02:29.990 --> 00:02:33.189 busy month, that financial cost really adds up. 00:02:33.689 --> 00:02:36.770 It is deeply inefficient, but I think there is 00:02:36.770 --> 00:02:40.530 also an intangible loss here. The text describes 00:02:40.530 --> 00:02:43.449 it as losing the feel of your project. Oh, this 00:02:43.449 --> 00:02:46.340 is so incredibly real. When you work with an 00:02:46.340 --> 00:02:49.360 AI for a solid hour, it actually starts to learn 00:02:49.360 --> 00:02:51.780 your unique rhythm. It picks up on your tone. 00:02:51.900 --> 00:02:53.680 It learns exactly what you're going to push back 00:02:53.680 --> 00:02:55.960 on. It starts to feel like an actual creative 00:02:55.960 --> 00:02:57.819 partner. And then you just close the tab. It's 00:02:57.819 --> 00:03:00.860 gone. That shared feeling completely resets. 00:03:00.939 --> 00:03:03.280 The very next morning, it is a total stranger 00:03:03.280 --> 00:03:05.840 again. You lose that incredibly valuable shared 00:03:05.840 --> 00:03:08.460 context. It's tragic, honestly. I actually have 00:03:08.460 --> 00:03:10.939 to admit something here. I still wrestle with 00:03:10.939 --> 00:03:14.639 this terrible habit myself. I catch myself pasting 00:03:14.639 --> 00:03:17.879 the exact same context brief three times a week, 00:03:18.180 --> 00:03:21.219 just out of sheer laziness. It's such a ridiculously 00:03:21.219 --> 00:03:25.259 common trap. We just default to the easiest immediate 00:03:25.259 --> 00:03:27.680 action in the moment. Even if we know it is going 00:03:27.680 --> 00:03:31.219 to cost us more time later. Exactly. So why is 00:03:31.219 --> 00:03:33.800 this specific cycle so incredibly hard for us 00:03:33.800 --> 00:03:36.680 to break? Well, because the AI sounds so confident. 00:03:36.919 --> 00:03:39.740 It's conversational every single time. So we 00:03:39.740 --> 00:03:43.939 naturally project human traits onto it. We subconsciously 00:03:43.939 --> 00:03:46.599 assume it remembers us, because the human colleague 00:03:46.599 --> 00:03:49.580 would. And when it obviously doesn't, we just 00:03:49.580 --> 00:03:51.520 panic -paste our documents to catch it up. We 00:03:51.520 --> 00:03:53.620 project human memory onto a machine that sounds 00:03:53.620 --> 00:03:55.900 confident. Yeah, that's it entirely. Which leads 00:03:55.900 --> 00:03:58.520 us directly to the actual fix. To stop starting 00:03:58.520 --> 00:04:01.120 from zero, we need to combine two very different 00:04:01.120 --> 00:04:02.840 tools. Wait, I want to push back on that a bit. 00:04:02.919 --> 00:04:05.199 Isn't the whole point of these massive AI models 00:04:05.199 --> 00:04:06.879 that they do everything? That's the marketing 00:04:06.879 --> 00:04:09.539 pitch, sure. So why would I complicate my life 00:04:09.539 --> 00:04:12.039 by using two different tools? Because people 00:04:12.039 --> 00:04:14.919 fail. when they expect one tool to do absolutely 00:04:14.919 --> 00:04:17.439 everything. You can't use a hammer to turn a 00:04:17.439 --> 00:04:20.339 screw. Ah, I see. Claude Projects and Notebook 00:04:20.339 --> 00:04:22.939 LM are built for entirely different jobs. When 00:04:22.939 --> 00:04:25.100 you finally put them together, the entire system 00:04:25.100 --> 00:04:27.160 just clicks. Okay, let's look at that first tool 00:04:27.160 --> 00:04:30.339 then. Claude Projects. This basically acts as 00:04:30.339 --> 00:04:33.399 your persistent, everyday workspace. Right. You 00:04:33.399 --> 00:04:35.660 upload a file once and it stays there forever. 00:04:35.949 --> 00:04:39.170 It uses AI to constantly remember your unique 00:04:39.170 --> 00:04:42.410 context. And just to clarify, Rag, for a second, 00:04:42.769 --> 00:04:46.029 it is a way for AI to pull your exact files instantly. 00:04:46.290 --> 00:04:49.569 Yeah. Great way to put it. So inside Claude projects, 00:04:49.910 --> 00:04:52.029 the AI can constantly read those files. Right. 00:04:52.410 --> 00:04:55.230 But it can also still search the broader open 00:04:55.230 --> 00:04:57.449 web. It still has its vast general knowledge. 00:04:57.550 --> 00:05:00.129 Yeah. Your specific files just add an extra layer 00:05:00.129 --> 00:05:02.750 of personal context. But Notebook LM works in 00:05:02.750 --> 00:05:04.870 the exact opposite way, doesn't it? It acts strictly 00:05:04.870 --> 00:05:07.089 as the archive. This is the massive difference. 00:05:07.490 --> 00:05:10.649 Notebook LM only pulls from the specific documents 00:05:10.649 --> 00:05:13.100 you provide it. Interesting. It has incredibly 00:05:13.100 --> 00:05:15.779 strict data adherence. There is absolutely no 00:05:15.779 --> 00:05:17.800 external web knowledge allowed in there. So it 00:05:17.800 --> 00:05:20.779 is kind of like having a brilliant creative director 00:05:20.779 --> 00:05:24.600 in Claude and then an incredibly strict literal 00:05:24.600 --> 00:05:27.860 archivist in Notebook LM. They're doing totally 00:05:27.860 --> 00:05:29.600 different jobs. That's a really great way to 00:05:29.600 --> 00:05:31.279 visualize it. They're a team. Now here's why 00:05:31.279 --> 00:05:33.379 we actually do this workflow manually instead 00:05:33.379 --> 00:05:35.600 of using a script. You mean advanced things like 00:05:35.600 --> 00:05:38.500 MCP. Right. Which is basically code that connects 00:05:38.500 --> 00:05:40.500 apps together automatically in the background. 00:05:40.639 --> 00:05:44.339 Yeah, exactly. Advanced users absolutely love 00:05:44.339 --> 00:05:47.980 setting up those automated complex scripts. But 00:05:47.980 --> 00:05:50.319 they require you to do complicated terminal work. 00:05:50.759 --> 00:05:53.079 You constantly have to authenticate tokens through 00:05:53.079 --> 00:05:56.139 Google. That sounds fragile. It is. For most 00:05:56.139 --> 00:05:58.339 normal people, those setups just break without 00:05:58.339 --> 00:06:00.889 any warning. a random library updates overnight, 00:06:01.209 --> 00:06:03.769 and boom, your entire workflow is totally dead 00:06:03.769 --> 00:06:06.189 right before a major deadline. So this manual 00:06:06.189 --> 00:06:08.670 method is just much more resilient. It is completely 00:06:08.670 --> 00:06:11.529 free. There is zero coding required. Not at all. 00:06:11.689 --> 00:06:13.829 You can even do it on your phone. It stays perfectly 00:06:13.829 --> 00:06:16.110 stable. It actually stays working. That's the 00:06:16.110 --> 00:06:18.389 entire point of building it this way. I want 00:06:18.389 --> 00:06:20.629 to linger on Notebook LM for just a second, though. 00:06:21.000 --> 00:06:23.600 Why are strict boundaries sometimes better than 00:06:23.600 --> 00:06:26.420 having endless web knowledge? Because the open 00:06:26.420 --> 00:06:28.839 web is totally full of contradictions. It's full 00:06:28.839 --> 00:06:31.120 of generic advice. That is very true. When you're 00:06:31.120 --> 00:06:33.259 making critical business decisions based on your 00:06:33.259 --> 00:06:36.459 own specific client data, you want the AI locked 00:06:36.459 --> 00:06:39.620 in a room with only your proven facts. Endless 00:06:39.620 --> 00:06:42.279 internet knowledge dilutes the truth when you 00:06:42.279 --> 00:06:45.680 need factual boundaries. Yeah, exactly. So let's 00:06:45.680 --> 00:06:47.819 get hands -on and actually build out this creative 00:06:47.819 --> 00:06:50.519 director's workspace first. Let's do it. Setting 00:06:50.519 --> 00:06:54.100 up Claude projects. The guide is incredibly clear 00:06:54.100 --> 00:06:56.839 about the pacing here. You do not need to do 00:06:56.839 --> 00:06:58.779 absolutely everything in one single sitting. 00:06:58.959 --> 00:07:01.699 You start simple. People always overcomplicate 00:07:01.699 --> 00:07:03.839 this step. You just go into Claude. You click 00:07:03.839 --> 00:07:06.459 on new project. Right. You give it a highly specific 00:07:06.459 --> 00:07:09.660 name, something like newsletter Q2 or client 00:07:09.660 --> 00:07:11.759 strategy redesign. Inside that project, you will 00:07:11.759 --> 00:07:14.779 see a file section. This acts as your new long 00:07:14.779 --> 00:07:18.139 term memory. And the golden rule here is. Exactly 00:07:18.139 --> 00:07:20.899 three files to start. Right. Do not upload 20 00:07:20.899 --> 00:07:23.519 chaotic files on day one. Just three. First, 00:07:23.839 --> 00:07:27.519 you need your context. .md file. This explains 00:07:27.519 --> 00:07:30.660 who you are, your core goals, your specific tone. 00:07:30.800 --> 00:07:33.699 Keep it super lean, under 500 words. Second, 00:07:33.779 --> 00:07:37.000 you upload your formal brand guidelines. Third, 00:07:37.040 --> 00:07:40.459 and most importantly, your decisions log. The 00:07:40.459 --> 00:07:43.199 decisions log. This is basically a running document 00:07:43.199 --> 00:07:45.720 of choices you have made and the reasons why. 00:07:45.860 --> 00:07:48.410 Yeah. Once those three are finally uploaded, 00:07:48.930 --> 00:07:51.350 you set up the project instructions. These are 00:07:51.350 --> 00:07:53.509 basically standing rules. Claude has to follow 00:07:53.509 --> 00:07:55.970 them before you even type a single word. You 00:07:55.970 --> 00:07:58.870 literally tell it, check my decisions log before 00:07:58.870 --> 00:08:01.730 answering. always match my brand guidelines. 00:08:02.069 --> 00:08:04.410 You essentially hand it a custom instruction 00:08:04.410 --> 00:08:06.850 manual for your own brain. But I know people 00:08:06.850 --> 00:08:09.610 assume AI just naturally learns over time. Oh, 00:08:09.610 --> 00:08:12.410 they do. Without this next manual step, all that 00:08:12.410 --> 00:08:14.769 brilliant brainstorming just vanishes into the 00:08:14.769 --> 00:08:17.329 ether. It vanishes completely. You absolutely 00:08:17.329 --> 00:08:19.649 cannot just close the tab when you're done working. 00:08:19.750 --> 00:08:21.810 Right. At the end of every single session, you 00:08:21.810 --> 00:08:24.029 have to run the wrap -up habit. The prompt they 00:08:24.029 --> 00:08:26.209 suggest for this is actually really simple. Summarize 00:08:26.209 --> 00:08:28.430 what we worked on, decisions made, and next steps. 00:08:28.509 --> 00:08:31.430 Keep it under 200 words. You just copy that short 00:08:31.430 --> 00:08:34.370 summary, you paste it directly into your decisions 00:08:34.370 --> 00:08:37.429 log document, and you re -upload it to the project. 00:08:37.590 --> 00:08:40.090 The whole thing takes two minutes, Tops. Literally 00:08:40.090 --> 00:08:42.769 two minutes. So why is this specific decisions 00:08:42.769 --> 00:08:46.049 log the Absort secret weapon here? Well, human 00:08:46.049 --> 00:08:48.289 memory is honestly terrible. I mean, we forget 00:08:48.289 --> 00:08:50.549 what we had for breakfast, let alone why we discarded 00:08:50.549 --> 00:08:53.440 a marketing angle three weeks ago. Yeah. If the 00:08:53.440 --> 00:08:55.899 AI doesn't have a log of that specific choice, 00:08:56.220 --> 00:08:58.220 it's just going to suggest the exact same bad 00:08:58.220 --> 00:09:01.059 angle next month. It records the logic of your 00:09:01.059 --> 00:09:03.419 choices to prevent recurring arguments. Yeah, 00:09:03.440 --> 00:09:06.039 it creates a perfectly linear history of your 00:09:06.039 --> 00:09:08.879 own logic. We are going to pause for a brief 00:09:08.879 --> 00:09:11.620 moment. When we come back, we will explore adding 00:09:11.620 --> 00:09:14.840 the heavy research layer, mid -roll sponsor read 00:09:14.840 --> 00:09:20.759 placeholder. Welcome back to the deep dive. We 00:09:20.759 --> 00:09:22.980 have successfully built the active workspace 00:09:22.980 --> 00:09:25.919 using Claude, but Claude naturally hits upload 00:09:25.919 --> 00:09:28.659 limits very fast. It really does, even if you're 00:09:28.659 --> 00:09:31.379 on the expensive paid plans. You upload a few 00:09:31.379 --> 00:09:33.980 massive research PDFs and you immediately run 00:09:33.980 --> 00:09:36.980 out of contextual space. So where do we actually 00:09:36.980 --> 00:09:40.100 put those massive industry reports? Where do 00:09:40.100 --> 00:09:43.139 we store our giant competitor analyses? This 00:09:43.139 --> 00:09:46.389 brings us to the second tool. Notebook LM. This 00:09:46.389 --> 00:09:49.549 serves as your heavy research layer. It comfortably 00:09:49.549 --> 00:09:52.389 handles up to 50 different sources per notebook. 00:09:52.529 --> 00:09:55.429 Right. You can upload dense PDFs, long Google 00:09:55.429 --> 00:09:57.809 Docs, or endless website links. You can even 00:09:57.809 --> 00:10:00.710 drop in YouTube transcripts and huge audio files. 00:10:00.769 --> 00:10:02.750 And you just organize them by topic. Put your 00:10:02.750 --> 00:10:04.789 audience research in one notebook, competitor 00:10:04.789 --> 00:10:07.029 landscape goes in another, industry trends in 00:10:07.029 --> 00:10:09.820 a third. The actual workflow here is fascinating. 00:10:09.960 --> 00:10:12.019 You do not ask Claude first. Right. Never ask 00:10:12.019 --> 00:10:14.320 Claude first for hard research. You go directly 00:10:14.320 --> 00:10:17.740 to Notebook LM. You ask a highly specific, targeted 00:10:17.740 --> 00:10:20.500 question. Something like, what are the top three 00:10:20.500 --> 00:10:22.919 audience patterns found in these sources? And 00:10:22.919 --> 00:10:25.539 Notebook LM gives you a grounded, perfectly cited 00:10:25.539 --> 00:10:27.720 answer. It is only using your uploaded files. 00:10:27.879 --> 00:10:30.679 Exactly. You then take that clean answer, copy 00:10:30.679 --> 00:10:32.820 it, and paste it back into your Claude project. 00:10:33.120 --> 00:10:35.860 You basically tell Claude, here is my verified 00:10:35.860 --> 00:10:39.169 research. Based on this exact data, write out 00:10:39.169 --> 00:10:42.289 the new strategy. Whoa. Imagine having your entire 00:10:42.289 --> 00:10:45.210 project history perfectly recalled in an instant 00:10:45.210 --> 00:10:47.730 without a single hallucination. It is a superpower. 00:10:48.059 --> 00:10:50.740 You get the strict factual recall of Notebook 00:10:50.740 --> 00:10:53.600 LM seamlessly combined with the creative reasoning 00:10:53.600 --> 00:10:56.059 of Claude. It's the best of both worlds. A quick 00:10:56.059 --> 00:10:58.340 note on data safety here. People always ask if 00:10:58.340 --> 00:11:00.879 their private data is actually safe. Notebook 00:11:00.879 --> 00:11:03.720 LM is officially a Google product. It follows 00:11:03.720 --> 00:11:06.860 standard Google workspace privacy terms. So it's 00:11:06.860 --> 00:11:09.000 perfectly fine for general market research and 00:11:09.000 --> 00:11:11.860 creative drafts. But you still need to be careful. 00:11:12.000 --> 00:11:14.870 Right. Don't go uploading highly sensitive legal 00:11:14.870 --> 00:11:17.549 or financial data unless you explicitly clear 00:11:17.549 --> 00:11:20.149 it with your IT team first. Let's talk about 00:11:20.149 --> 00:11:23.389 prompt drift for a second. How exactly does this 00:11:23.389 --> 00:11:26.549 dual step process prevent the AI from slowly 00:11:26.549 --> 00:11:28.649 drifting completely off topic? Well, when you 00:11:28.649 --> 00:11:31.450 just chat loosely with an AI for hours, it naturally 00:11:31.450 --> 00:11:33.970 starts losing the plot. It gets overly creative, 00:11:34.110 --> 00:11:36.870 just starts filling in the blanks. By forcing 00:11:36.870 --> 00:11:40.110 the AI to start from notebook elements hard facts 00:11:40.110 --> 00:11:43.289 every single session, you constantly reset its 00:11:43.289 --> 00:11:45.809 compass back to reality. Anchoring Claude to 00:11:45.809 --> 00:11:47.970 factual sources stops it from making up creative 00:11:47.970 --> 00:11:51.450 data. Precisely. It keeps the final output highly 00:11:51.450 --> 00:11:54.370 rigorous. Let's look at how this actually applies 00:11:54.370 --> 00:11:57.870 in the real world. Real use cases. Because this 00:11:57.870 --> 00:12:00.009 definitely is not just a system for writers. 00:12:00.129 --> 00:12:02.330 Not at all. Think about digital marketers for 00:12:02.330 --> 00:12:05.360 a second. You create a totally separate cloud 00:12:05.360 --> 00:12:08.029 project for each individual client. Okay. You 00:12:08.029 --> 00:12:10.429 upload their specific brand voice. You upload 00:12:10.429 --> 00:12:13.029 all their past campaign results. So every single 00:12:13.029 --> 00:12:15.149 time you open up that client's project, Claude 00:12:15.149 --> 00:12:17.129 already knows them intimately. Right. There's 00:12:17.129 --> 00:12:19.070 no tedious re -briefing. You just jump in and 00:12:19.070 --> 00:12:22.190 ask, what specific content format worked best 00:12:22.190 --> 00:12:24.269 for them last quarter? Educators are using this 00:12:24.269 --> 00:12:26.750 heavily too. You upload your massive core site 00:12:26.750 --> 00:12:29.070 line. You upload anonymous student feedback. 00:12:29.210 --> 00:12:32.110 You can easily ask Claude to rewrite a confusing 00:12:32.110 --> 00:12:35.149 lesson based entirely on your specific teaching 00:12:35.149 --> 00:12:37.919 philosophy. And for solo founders, this setup 00:12:37.919 --> 00:12:40.399 is just brilliant. They call it creating a business 00:12:40.399 --> 00:12:43.220 brain project. This is honestly my favorite application. 00:12:44.080 --> 00:12:48.360 You upload your big yearly goals, your external 00:12:48.360 --> 00:12:51.460 communication guidelines, your internal pricing 00:12:51.460 --> 00:12:53.899 rationale. Yeah. You can actively use it to fight 00:12:53.899 --> 00:12:56.200 against scope creep. A client wants a brand new 00:12:56.200 --> 00:12:59.799 service, right? Right. You ask, based on my past 00:12:59.799 --> 00:13:01.620 decisions about scope, what should I actually 00:13:01.620 --> 00:13:04.159 propose here? It essentially acts as a perfectly 00:13:04.159 --> 00:13:06.399 objective business partner. It forcefully holds 00:13:06.399 --> 00:13:08.620 you to your own established rules. It totally 00:13:08.620 --> 00:13:11.500 does. But obviously, any system has its traps. 00:13:12.000 --> 00:13:14.720 The guide highlights four major mistakes that 00:13:14.720 --> 00:13:17.080 will completely break this setup if you aren't 00:13:17.080 --> 00:13:19.340 careful. We definitely have to cover these. Mistake 00:13:19.340 --> 00:13:21.700 number one is uploading way too much data all 00:13:21.700 --> 00:13:24.440 at once. People always want to just dump 30 chaotic 00:13:24.440 --> 00:13:26.259 documents in on day one. Yeah, you have to start 00:13:26.259 --> 00:13:28.919 with three to five. Mistake number two is never 00:13:28.919 --> 00:13:31.480 bothering to update your core files. Your business 00:13:31.480 --> 00:13:34.620 context changes. Your goals shift. You really 00:13:34.620 --> 00:13:36.799 need to review those core foundation files every 00:13:36.799 --> 00:13:39.539 four weeks. It's a really small habit, but it 00:13:39.539 --> 00:13:42.220 makes a massive difference. Mistake number three 00:13:42.220 --> 00:13:44.240 is skipping the wrap up. We said it before, but 00:13:44.240 --> 00:13:46.620 it absolutely bears repeating here. Without that 00:13:46.620 --> 00:13:48.940 two minute summary step at the end of a session, 00:13:49.039 --> 00:13:51.559 you are not building a memory. You are just having 00:13:51.559 --> 00:13:55.539 totally isolated, forgettable chats. Yep. And 00:13:55.539 --> 00:13:58.039 mistake four is using Claude projects to hoard 00:13:58.039 --> 00:14:00.679 all your heavy research. Right. Do not treat 00:14:00.679 --> 00:14:04.100 Claude like a giant filing cabinet. That is exactly 00:14:04.100 --> 00:14:07.399 what Notebook LM's job is. You have to keep the 00:14:07.399 --> 00:14:10.039 tools in their respective lanes. Keep them separated. 00:14:10.279 --> 00:14:12.639 I am actually curious about that very first mistake. 00:14:13.320 --> 00:14:16.360 Why does giving an AI more files actually make 00:14:16.360 --> 00:14:19.019 it less smart? Shouldn't more data be better? 00:14:19.340 --> 00:14:21.669 You'd think so, but no. It's this phenomenon 00:14:21.669 --> 00:14:24.169 called context dilution. OK. When you give it 00:14:24.169 --> 00:14:27.230 50 vaguely related files, the AI's attention 00:14:27.230 --> 00:14:29.710 mechanism really struggles to weigh what is actually 00:14:29.710 --> 00:14:32.350 important. It might pull a totally mediocre point 00:14:32.350 --> 00:14:34.649 from an outdated file instead of the critical, 00:14:34.649 --> 00:14:37.169 nuanced point from yesterday's file. Too much 00:14:37.169 --> 00:14:39.710 background noise hides the critical signals the 00:14:39.710 --> 00:14:42.509 AI needs. Yeah. Keep the active workspace lean. 00:14:43.110 --> 00:14:46.129 Keep the archive layer deep. Let's recap the 00:14:46.129 --> 00:14:48.799 big idea we were exploring here. We have all 00:14:48.799 --> 00:14:51.879 been paying for this incredibly powerful evolving 00:14:51.879 --> 00:14:55.440 partner, but we have been treating it like a 00:14:55.440 --> 00:14:58.919 basic, completely amnesiac question answering 00:14:58.919 --> 00:15:01.000 machine. We've seriously been limiting its true 00:15:01.000 --> 00:15:03.519 potential just because of our own lazy habits. 00:15:03.700 --> 00:15:06.200 The roadmap to fix this is genuinely simple. 00:15:06.820 --> 00:15:09.639 Week one, build the Claude Foundation. Right. 00:15:09.940 --> 00:15:12.899 Right. Your short context file. Upload it. Start 00:15:12.899 --> 00:15:15.240 practicing that two minute wrap up habit. Week 00:15:15.240 --> 00:15:18.379 two, you add the heavy research layer. Set up 00:15:18.379 --> 00:15:20.960 your notebook LM, add your massive reference 00:15:20.960 --> 00:15:24.200 sources, but strictly keep the two tools separated. 00:15:24.399 --> 00:15:26.779 Absolutely crucial. This fundamentally transforms 00:15:26.779 --> 00:15:29.840 the tool into a true creative partner. It finally 00:15:29.840 --> 00:15:31.879 stops being a complete stranger every morning. 00:15:31.960 --> 00:15:34.039 Because it already knows where you are. It knows 00:15:34.039 --> 00:15:37.019 exactly what you are building toward. So here's 00:15:37.019 --> 00:15:39.100 the immediate challenge for you. Okay. In your 00:15:39.100 --> 00:15:41.500 very next work session, don't just close the 00:15:41.500 --> 00:15:44.639 window. Force yourself to ask for that 200 -word 00:15:44.639 --> 00:15:47.659 summary. Start your decisions log today. Stop 00:15:47.659 --> 00:15:50.159 starting from zero. It is a relatively small 00:15:50.159 --> 00:15:52.519 shift with incredibly profound leverage, but 00:15:52.519 --> 00:15:54.879 it really does make you think. If you successfully 00:15:54.879 --> 00:15:57.220 build an AI partner that perfectly remembers 00:15:57.220 --> 00:16:00.440 every decision, every pivot, and every preference 00:16:00.440 --> 00:16:03.100 you have ever had, five years from now, where 00:16:03.100 --> 00:16:05.179 exactly does your original thinking end and the 00:16:05.179 --> 00:16:06.139 AI's memory begin?