WEBVTT 00:00:00.000 --> 00:00:02.180 You type a simple question into the chat window. 00:00:02.560 --> 00:00:05.820 It just spits back a generic robotic wall of 00:00:05.820 --> 00:00:08.019 text. Yeah, and it's practically unusable for 00:00:08.019 --> 00:00:11.099 your actual daily tasks. So you tweak the prompt, 00:00:11.160 --> 00:00:14.160 you try it again. But the result is exactly the 00:00:14.160 --> 00:00:16.780 same frustrating garbage. Welcome to the deep 00:00:16.780 --> 00:00:20.339 dive. That is a deeply relatable frustration 00:00:20.339 --> 00:00:23.500 for all of us. It really is. Today, our mission 00:00:23.500 --> 00:00:27.800 is unpacking a very popular mindset. People constantly 00:00:27.800 --> 00:00:29.879 tell you to learn prompt engineering. Right. 00:00:30.000 --> 00:00:32.859 They say to study frameworks and memorize these 00:00:32.859 --> 00:00:35.000 arbitrary rules. We're going to explore why that 00:00:35.000 --> 00:00:37.280 is pointless. It is honestly a massive waste 00:00:37.280 --> 00:00:39.700 of your time. A total waste. Instead, we're outlining 00:00:39.700 --> 00:00:41.979 a totally different approach today. We will build 00:00:41.979 --> 00:00:45.090 a custom prompt optimizer from scratch. You can 00:00:45.090 --> 00:00:47.310 build it using totally free Google tools. And 00:00:47.310 --> 00:00:49.409 it really only takes about 10 minutes to set 00:00:49.409 --> 00:00:51.609 up. I have a vulnerable admission to make right 00:00:51.609 --> 00:00:53.990 here. I still wrestle with getting generic robotic 00:00:53.990 --> 00:00:56.109 answers myself. Oh, yeah. Yeah, I just throw 00:00:56.109 --> 00:00:59.009 my messy thoughts at the screen. I expect the 00:00:59.009 --> 00:01:01.990 machine to magically understand my intent. But 00:01:01.990 --> 00:01:03.969 it almost never actually does what you want. 00:01:04.170 --> 00:01:06.849 Well, you're definitely not alone in that specific 00:01:06.849 --> 00:01:09.170 struggle. It was a very common friction point 00:01:09.170 --> 00:01:11.510 for everyone. So our roadmap for the discussion 00:01:11.510 --> 00:01:14.090 today is clear. We're going to explore why AI 00:01:14.090 --> 00:01:17.030 constantly guesses. Then we will build this optimizer 00:01:17.030 --> 00:01:19.209 in two steps. Finally, we'll plug it into your 00:01:19.209 --> 00:01:21.530 daily workflows. Let's start with that guessing 00:01:21.530 --> 00:01:24.450 game you just mentioned. Before we can fix the 00:01:24.450 --> 00:01:27.799 AI, we must understand it. We need to know why 00:01:27.799 --> 00:01:30.260 it gives us terrible answers. Especially when 00:01:30.260 --> 00:01:33.180 we just type naturally, you know. Right. The 00:01:33.180 --> 00:01:36.599 core issue is that AI isn't your friend. Think 00:01:36.599 --> 00:01:38.780 about talking to a very close personal friend. 00:01:38.959 --> 00:01:41.340 Okay. You might say you want to lose some weight, 00:01:42.019 --> 00:01:44.319 but you also admit you absolutely hate the gym. 00:01:44.680 --> 00:01:47.180 Right. A good friend knows your personality very 00:01:47.180 --> 00:01:50.299 well. They might suggest swimming or... taking 00:01:50.299 --> 00:01:53.200 a long walk. Exactly. They tailor the advice 00:01:53.200 --> 00:01:56.939 to fit your specific lifestyle. But the AI does 00:01:56.939 --> 00:01:59.379 not know your daily habits. It doesn't know what 00:01:59.379 --> 00:02:01.879 you actually like doing, so it just tries to 00:02:01.879 --> 00:02:04.159 guess what you want. You throw a messy idea at 00:02:04.159 --> 00:02:07.099 the chat window. And it gives you a boring 10 00:02:07.099 --> 00:02:09.560 -page workout plan. It's a rigid plan you're 00:02:09.560 --> 00:02:12.159 never going to use. It is incredibly frustrating 00:02:12.159 --> 00:02:14.620 to deal with that output. The source material 00:02:14.620 --> 00:02:17.500 introduces a brilliant core analogy here. Yeah, 00:02:17.580 --> 00:02:20.719 I love this part. It calls the prompt optimizer 00:02:20.719 --> 00:02:24.979 a cleaning machine. It's essentially a cleaning 00:02:24.979 --> 00:02:28.000 machine for your brain. I really love that specific 00:02:28.000 --> 00:02:30.840 framing for this tool. Instead of stressing to 00:02:30.840 --> 00:02:34.360 write a mathematically perfect instruction, you 00:02:34.360 --> 00:02:37.039 just throw in your messy half -formed human thoughts. 00:02:37.280 --> 00:02:39.620 The system automatically cleans them up for you. 00:02:39.740 --> 00:02:42.300 It turns a simple sentence into a highly structured 00:02:42.300 --> 00:02:45.060 prompt. It fills in the missing pieces of your 00:02:45.060 --> 00:02:47.919 logic. Right. And the AI then knows exactly what 00:02:47.919 --> 00:02:50.560 you actually need. It stops making wild guesses 00:02:50.560 --> 00:02:53.650 about your underlying intent. Yeah. Beat. But 00:02:53.650 --> 00:02:55.669 that brings us to the dreaded fluff problem. 00:02:55.849 --> 00:02:57.729 Oh, the fluff problem is absolutely everywhere 00:02:57.729 --> 00:03:00.629 in AI today. The model thinks talking a lot means 00:03:00.629 --> 00:03:03.610 being helpful. It adds these incredibly long, 00:03:03.669 --> 00:03:05.930 useless introductory paragraphs. It gives you 00:03:05.930 --> 00:03:08.229 extra advice you never even asked for. It equates 00:03:08.229 --> 00:03:10.530 word count with actual value and intelligence. 00:03:10.949 --> 00:03:14.370 It really does operate on that exact flawed logic. 00:03:15.310 --> 00:03:18.110 When you give a short prompt, you leave empty 00:03:18.110 --> 00:03:20.960 space. And the AI desperately has to fill those 00:03:20.960 --> 00:03:23.060 empty gaps. It starts guessing what your true 00:03:23.060 --> 00:03:25.520 goals might be. Let me ask you this. Why does 00:03:25.520 --> 00:03:28.539 the AI always generate useless fluff when I ask 00:03:28.539 --> 00:03:31.259 a short question? Well, it tries filling the 00:03:31.259 --> 00:03:33.740 empty space by guessing your goals. It just plays 00:03:33.740 --> 00:03:36.800 it safe. It assumes high word counts mean it's 00:03:36.800 --> 00:03:39.500 being helpful. Right. It talks too much because 00:03:39.500 --> 00:03:42.939 it thinks length equals value. Exactly. The optimizer 00:03:42.939 --> 00:03:44.979 acts like a professional editor sitting beside 00:03:44.979 --> 00:03:47.830 you. OK. It finds the structural holes in your 00:03:47.830 --> 00:03:50.669 messy story. It fills them in automatically before 00:03:50.669 --> 00:03:53.349 the AI answers. You avoid doing any of the heavy 00:03:53.349 --> 00:03:56.830 lifting yourself. Two sec silence. So we know 00:03:56.830 --> 00:03:59.150 the AI desperately needs an editor. Definitely. 00:03:59.289 --> 00:04:02.009 Needs a rigid framework to stop the endless guessing. 00:04:02.349 --> 00:04:04.349 Let's walk through exactly how to build that 00:04:04.349 --> 00:04:06.550 editor. You just need a completely free Google 00:04:06.550 --> 00:04:09.490 account. OK. The setup is way easier than it 00:04:09.490 --> 00:04:11.969 might sound. You don't need to be a complex tech 00:04:11.969 --> 00:04:16.009 wizard. Good to know. You need Notebook LM. Gemini, 00:04:16.230 --> 00:04:19.029 and two simple text files. That is your entire 00:04:19.029 --> 00:04:21.850 toolkit for this entire build. Just to be clear 00:04:21.850 --> 00:04:24.569 on the technical requirements here, we're just 00:04:24.569 --> 00:04:27.329 using basic text files for this setup. That's 00:04:27.329 --> 00:04:29.610 it. Let's walk through the contents of file one. 00:04:29.790 --> 00:04:32.189 This file contains your core system instructions, 00:04:32.230 --> 00:04:34.189 right? Yeah. This is where you establish the 00:04:34.189 --> 00:04:37.069 permanent ground rules. You open a blank document 00:04:37.069 --> 00:04:39.629 on your computer screen. You tell the system 00:04:39.629 --> 00:04:42.600 exactly how it should behave. You tell the AI 00:04:42.600 --> 00:04:45.600 it's a world -class prompt engineer. That is 00:04:45.600 --> 00:04:47.459 the exact phrase to use right there. You tell 00:04:47.459 --> 00:04:50.319 it to read the user's messy input. Then you ask 00:04:50.319 --> 00:04:53.160 it to create a JSON schema. Which is just a very 00:04:53.160 --> 00:04:55.600 organized list that keeps the AI boxed in. That 00:04:55.600 --> 00:04:57.879 is a perfect way to describe the mechanism. It 00:04:57.879 --> 00:05:01.300 forces the AI to answer specific targeted points. 00:05:01.399 --> 00:05:03.519 It stops the endless wandering through random 00:05:03.519 --> 00:05:06.459 topics completely. It cuts out the useless conversational 00:05:06.459 --> 00:05:09.420 fluff right away. File 1 also asks for a brief 00:05:09.420 --> 00:05:11.579 thinking guide. We'll dig into the mechanics 00:05:11.579 --> 00:05:14.279 of that shortly. You keep the overall tone professional, 00:05:14.680 --> 00:05:17.319 clear, and helpful. Everything outputs neatly 00:05:17.319 --> 00:05:19.899 inside a single formatted code block. You save 00:05:19.899 --> 00:05:22.620 that text file to your local computer. Name it 00:05:22.620 --> 00:05:25.680 prompt -optimizer -system -instructions for clarity. 00:05:26.819 --> 00:05:29.220 Next, we move on to building file number two. 00:05:29.360 --> 00:05:31.939 This is your very important example pairs file. 00:05:32.360 --> 00:05:34.000 Ground rules are great for setting the baseline 00:05:34.000 --> 00:05:37.600 behavior, but the AI also needs to see tangible, 00:05:37.720 --> 00:05:40.459 good examples. This second file might actually 00:05:40.459 --> 00:05:43.720 be even more important. It shows the AI the exact 00:05:43.720 --> 00:05:46.000 style you want. It operates in a very simple 00:05:46.000 --> 00:05:48.240 before and after logic. Let's look at a specific 00:05:48.240 --> 00:05:50.980 example from our source. The before state is 00:05:50.980 --> 00:05:53.860 a messy, raw user thought. Like you might just 00:05:53.860 --> 00:05:56.199 type, I want to learn to code. Right. That is 00:05:56.199 --> 00:05:59.420 a very broad, messy, and unstructured goal. The 00:05:59.420 --> 00:06:01.939 after state is what the optimizer builds. It 00:06:01.939 --> 00:06:04.920 becomes a highly structured, detailed JSON plan. 00:06:05.180 --> 00:06:07.839 It specifies learning Python specifically for 00:06:07.839 --> 00:06:10.420 an absolute beginner. It outlines a manageable 00:06:10.420 --> 00:06:13.180 weekly plan for the user. It also adds that crucial 00:06:13.180 --> 00:06:15.399 thinking guide we mentioned. It tells the AI 00:06:15.399 --> 00:06:18.379 to focus on underlying logic first. It explicitly 00:06:18.379 --> 00:06:20.839 says to avoid using complex technical jargon. 00:06:21.180 --> 00:06:24.240 It provides small daily wins for ongoing user 00:06:24.240 --> 00:06:26.920 motivation. You save that second file to your 00:06:26.920 --> 00:06:31.579 computer too. Name it prompt. Optimizer -example 00:06:31.579 --> 00:06:34.399 pairs to stay organized. Now you have your two 00:06:34.399 --> 00:06:36.459 core structural files ready. You open a brand 00:06:36.459 --> 00:06:39.839 new notebook in Google's Notebook LM. You upload 00:06:39.839 --> 00:06:42.519 both of these text files as sources. Then you 00:06:42.519 --> 00:06:44.319 do something very important for the system. You 00:06:44.319 --> 00:06:47.160 click the Convoke to Source button for both files. 00:06:47.279 --> 00:06:50.279 This tells the system these files are the absolute 00:06:50.279 --> 00:06:52.500 truth. It must follow them without any deviation 00:06:52.500 --> 00:06:57.000 at all. Whoa! Imagine turning a blank text file 00:06:57.000 --> 00:06:59.600 into a professional engineering brain in seconds. 00:07:00.000 --> 00:07:02.459 It really feels like unlocking a weird new superpower. 00:07:02.860 --> 00:07:04.819 Totally. The instructions mention a chain of 00:07:04.819 --> 00:07:07.180 thought process. What exactly is that doing behind 00:07:07.180 --> 00:07:09.319 the scenes? It acts as a mandatory guide for 00:07:09.319 --> 00:07:12.540 the model. It forces the AI to explicitly plan 00:07:12.540 --> 00:07:14.920 its logic. Oh, okay. It makes it think through 00:07:14.920 --> 00:07:16.860 the problem step by step. It does this before 00:07:16.860 --> 00:07:19.019 it starts writing the final answer. Got it. It 00:07:19.019 --> 00:07:21.879 gives the AI a clear thinking path to follow. 00:07:22.199 --> 00:07:24.300 Right. And that eliminates the guessing game 00:07:24.300 --> 00:07:27.480 completely. The AI has a concrete plan before 00:07:27.480 --> 00:07:31.920 it ever speaks. Two -sec silence. We have built 00:07:31.920 --> 00:07:35.579 the foundational brain in Notebook LM. Before 00:07:35.579 --> 00:07:37.579 we connect it to our chat window to see the magic 00:07:37.579 --> 00:07:40.740 happen, let's take a quick break. Sponsor. And 00:07:40.740 --> 00:07:43.240 we are back. We just built our custom prompt 00:07:43.240 --> 00:07:46.540 optimizer in Notebook LM. Now we need to connect 00:07:46.540 --> 00:07:49.009 it to our daily workflow. This is where the real 00:07:49.009 --> 00:07:51.810 practical magic kicks in. Linking the two distinct 00:07:51.810 --> 00:07:54.089 tools is actually quite simple. It really only 00:07:54.089 --> 00:07:57.689 takes a few very quick clicks. First, you open 00:07:57.689 --> 00:08:00.149 the standard Gemini website interface. You start 00:08:00.149 --> 00:08:02.829 a brand new chat session right there. Look closely 00:08:02.829 --> 00:08:04.829 at the prompt box at the bottom. You will see 00:08:04.829 --> 00:08:06.750 a little plus button sitting there. You click 00:08:06.750 --> 00:08:09.269 that plus button to open the menu. A list of 00:08:09.269 --> 00:08:11.490 integration options appears on your screen. You 00:08:11.490 --> 00:08:14.170 select Notebook LM as your primary data source. 00:08:14.350 --> 00:08:16.509 Then you choose the specific notebook you just 00:08:16.509 --> 00:08:18.430 built. You confirm the source is active in your 00:08:18.430 --> 00:08:21.389 chat window. From that exact second on, Geminite 00:08:21.389 --> 00:08:23.750 checks your instruction files. It checks them 00:08:23.750 --> 00:08:26.250 every single time you send a message. You do 00:08:26.250 --> 00:08:29.079 not have to keep uploading the text files. You 00:08:29.079 --> 00:08:31.819 never have to re -explain the basic ground rules. 00:08:32.480 --> 00:08:34.700 The notebook stays permanently connected to the 00:08:34.700 --> 00:08:37.039 chat session. By attaching that notebook as the 00:08:37.039 --> 00:08:39.480 primary source document. You're giving Gemini 00:08:39.480 --> 00:08:43.679 a rigid set of boundary conditions. It cannot 00:08:43.679 --> 00:08:46.440 just pull from its vast, messy training data. 00:08:46.519 --> 00:08:48.379 It has to filter everything through the brain 00:08:48.379 --> 00:08:50.940 you built. That brings us directly to the two 00:08:50.940 --> 00:08:53.639 -step workflow rule. This is the core workflow 00:08:53.639 --> 00:08:57.039 for your daily tasks. Step one. Type a messy 00:08:57.039 --> 00:08:59.879 idea into your new notebook LM. This quickly 00:08:59.879 --> 00:09:02.120 gives you the highly structured smart prompt. 00:09:02.480 --> 00:09:05.220 Step two is just as simple as the first. You 00:09:05.220 --> 00:09:07.220 copy and paste that smart prompt into Gemini. 00:09:07.480 --> 00:09:10.059 And Gemini then gets the actual real work done 00:09:10.059 --> 00:09:12.519 perfectly. Let's look at some real -world examples 00:09:12.519 --> 00:09:14.740 from the source material. These show exactly 00:09:14.740 --> 00:09:17.399 how powerful this two -step workflow is. Let's 00:09:17.399 --> 00:09:19.759 start with a very relatable daily fitness example. 00:09:19.899 --> 00:09:21.679 I think everyone can relate to this specific 00:09:21.679 --> 00:09:24.259 scenario. You are completely exhausted after 00:09:24.259 --> 00:09:26.960 a long day of work. Your daily schedule is an 00:09:26.960 --> 00:09:29.840 absolute total disaster right now. You do not 00:09:29.840 --> 00:09:32.019 have the energy to write perfectly. You don't 00:09:32.019 --> 00:09:34.159 need a mathematically perfect instruction here. 00:09:34.600 --> 00:09:38.019 You just type out your messy, chaotic daily reality. 00:09:38.340 --> 00:09:41.379 You type, I am tired and want to be fit, but 00:09:41.379 --> 00:09:43.860 I have zero time and I absolutely hate running. 00:09:44.080 --> 00:09:47.360 Without the optimizer, a standard AI just says, 00:09:47.360 --> 00:09:50.720 try harder. It gives you a generic 5K marathon 00:09:50.720 --> 00:09:53.919 training plan. But with the optimizer, it builds 00:09:53.919 --> 00:09:56.659 a tailored structure. It creates a detailed JSON 00:09:56.659 --> 00:09:59.120 list just for you. It understands the actual 00:09:59.120 --> 00:10:01.840 underlying problem you are facing. It specifies 00:10:01.840 --> 00:10:05.059 15 minutes of gentle, low -impact exercise. The 00:10:05.059 --> 00:10:07.539 thinking guide tells the AI to focus on consistency. 00:10:07.769 --> 00:10:11.350 It strictly deprioritizes intense exhausting 00:10:11.350 --> 00:10:14.110 daily workouts completely. You just copy that 00:10:14.110 --> 00:10:17.090 professional clean code block output. You paste 00:10:17.090 --> 00:10:19.149 it directly into your Gemini chat window. You 00:10:19.149 --> 00:10:22.250 get a totally realistic 15 minute daily workout. 00:10:22.610 --> 00:10:25.610 It fits your messy chaotic life absolutely perfectly. 00:10:25.750 --> 00:10:27.210 Let's look at a completely different technical 00:10:27.210 --> 00:10:29.470 coding example. Let's say you want to learn Tailwind 00:10:29.470 --> 00:10:32.230 CSS programming. If you just ask AI to teach 00:10:32.230 --> 00:10:35.490 you, it fails. The topic is way too broad for 00:10:35.490 --> 00:10:38.350 a simple prompt. The standard AI answer will 00:10:38.350 --> 00:10:41.090 be incredibly confusing. It will overwhelm you 00:10:41.090 --> 00:10:43.409 with advanced technical details immediately. 00:10:43.610 --> 00:10:46.730 But you put that messy idea into Notebook LM 00:10:46.730 --> 00:10:49.830 first. The system realizes you need a clear step 00:10:49.830 --> 00:10:52.389 -by -step path. It actively stops you from feeling 00:10:52.389 --> 00:10:55.330 totally overwhelmed right away. It systematically 00:10:55.330 --> 00:10:58.299 organizes the topic into core foundation. concepts. 00:10:58.320 --> 00:11:01.279 It creates small, manageable daily coding challenges 00:11:01.279 --> 00:11:04.179 for you. The thinking guide forces the AI to 00:11:04.179 --> 00:11:06.919 explain the underlying why. Like why use Tailwind 00:11:06.919 --> 00:11:10.159 CSS instead of older CSS files. You take that 00:11:10.159 --> 00:11:12.139 beautifully optimized prompt from the screen. 00:11:12.259 --> 00:11:14.740 You paste it into your preferred AI chat tool. 00:11:15.059 --> 00:11:17.580 A clear, professional step -by -step lesson plan 00:11:17.580 --> 00:11:20.419 appears immediately. The final practical example 00:11:20.419 --> 00:11:22.419 is about growing a business. Growing a weekly 00:11:22.419 --> 00:11:24.440 newsletter or getting more Twitter followers? 00:11:24.500 --> 00:11:27.299 Do not just ask the AI for a generic growth plan. 00:11:27.519 --> 00:11:30.259 You type your raw, unpolished goal into Notebook 00:11:30.259 --> 00:11:33.179 LM directly. The optimizer acts exactly like 00:11:33.179 --> 00:11:35.700 a professional marketing consultant. It generates 00:11:35.700 --> 00:11:38.460 a strategy tailored for your specific brand niche. 00:11:38.840 --> 00:11:41.679 It builds a rigid content structure for your 00:11:41.679 --> 00:11:44.200 specific niche. It clearly defines your core 00:11:44.200 --> 00:11:47.639 daily content pillars. It maps out a realistic 00:11:47.639 --> 00:11:50.379 daily social engagement strategy. The underlying 00:11:50.379 --> 00:11:53.610 thinking guide is absolutely crucial right here. 00:11:53.990 --> 00:11:56.830 It tells the AI to focus on providing real value. 00:11:56.970 --> 00:11:59.450 It strictly avoids annoying your followers with 00:11:59.450 --> 00:12:02.070 constant ads. You copy the resulting code block 00:12:02.070 --> 00:12:04.669 from the optimizer. You paste it back into your 00:12:04.669 --> 00:12:07.389 main Gemini window. You get a brilliant 30 -day 00:12:07.389 --> 00:12:10.070 content calendar with great headlines. The real 00:12:10.070 --> 00:12:13.039 secret is knowing which tool to use when. Notebook 00:12:13.039 --> 00:12:15.120 LM is just the filter that cleans your messy 00:12:15.120 --> 00:12:17.320 thoughts. Gemini is the engine that does the 00:12:17.320 --> 00:12:19.480 actual heavy lifting. It is exactly like having 00:12:19.480 --> 00:12:22.200 a dedicated professional coach. It stops embarrassing 00:12:22.200 --> 00:12:25.179 mistakes before you even start typing. So I generate 00:12:25.179 --> 00:12:28.100 this smart prompt using Notebook LM. Am I locked 00:12:28.100 --> 00:12:30.539 into only using Gemini for the final result? 00:12:30.679 --> 00:12:32.840 Not at all. You can copy that optimized structure. 00:12:32.919 --> 00:12:35.519 You can paste it right into ChatGPT, Claude, 00:12:35.620 --> 00:12:37.700 or any other model you prefer. So the cleaned 00:12:37.700 --> 00:12:40.320 up prompt works perfectly in ChatGPT or Claude 00:12:40.320 --> 00:12:43.299 too. Absolutely. The prompt structure is already 00:12:43.299 --> 00:12:45.240 thoroughly cleaned up. It will work beautifully 00:12:45.240 --> 00:12:48.899 on literally any modern AI model. Two -sec silence. 00:12:50.580 --> 00:12:53.659 This basic workflow clearly saves hours of daily 00:12:53.659 --> 00:12:57.320 frustration. Now let's explore leveling up this 00:12:57.320 --> 00:12:59.889 entire system. We can tweak it to fit perfectly 00:12:59.889 --> 00:13:03.230 with highly unique tasks. Small incremental changes 00:13:03.230 --> 00:13:05.669 make a truly massive difference here. You do 00:13:05.669 --> 00:13:07.490 not have to rebuild everything from scratch, 00:13:07.730 --> 00:13:09.669 ever. You just push a few buttons to alter the 00:13:09.669 --> 00:13:12.529 overall vibe. Let's discuss changing the AI's 00:13:12.529 --> 00:13:15.110 underlying conversational tone. You can easily 00:13:15.110 --> 00:13:17.370 change the personality of your custom optimizer. 00:13:17.570 --> 00:13:20.009 You just swap out one single line in the core 00:13:20.009 --> 00:13:21.690 instructions. You might want a patient and clear 00:13:21.690 --> 00:13:24.250 teacher for learning. That tone is great for 00:13:24.250 --> 00:13:27.529 grasping new, complex topics. But for daily business 00:13:27.529 --> 00:13:30.169 tasks, that tone might feel too soft. Right. 00:13:30.389 --> 00:13:32.629 For serious business, you want something much 00:13:32.629 --> 00:13:34.970 sharper. You chained it to a direct and data 00:13:34.970 --> 00:13:38.210 -driven consultant. You want crisp, highly actionable 00:13:38.210 --> 00:13:40.809 insights without the fluff. For purely creative 00:13:40.809 --> 00:13:43.629 projects, you try an inspiring and imaginative 00:13:43.629 --> 00:13:47.110 coach. That one single line completely shifts 00:13:47.110 --> 00:13:50.570 the semantic space. It changes how the AI responds 00:13:50.570 --> 00:13:53.440 to your messy thoughts. The source material also 00:13:53.440 --> 00:13:56.820 introduces gems inside of Gemini. If you use 00:13:56.820 --> 00:13:59.159 Gemini constantly, attaching the notebook gets 00:13:59.159 --> 00:14:03.039 tedious. Doing it every single time causes unnecessary 00:14:03.039 --> 00:14:06.559 daily friction. The gems feature fixes that annoying 00:14:06.559 --> 00:14:09.139 repetitive hassle completely. Think of a gem 00:14:09.139 --> 00:14:11.759 as a little dedicated shortcut app. It lives 00:14:11.759 --> 00:14:13.960 permanently right there in your left sidebar. 00:14:14.120 --> 00:14:16.580 It saves you so many unnecessary manual clicks 00:14:16.580 --> 00:14:18.440 daily. You don't have to attach the notebook 00:14:18.440 --> 00:14:20.779 every single time. You just click your custom 00:14:20.779 --> 00:14:23.120 optimizer gem and start working. You start typing 00:14:23.120 --> 00:14:25.779 your messy idea into the box immediately. It 00:14:25.779 --> 00:14:27.759 keeps everything you need perfectly organized 00:14:27.759 --> 00:14:30.740 in one place. It is a massive time saver for 00:14:30.740 --> 00:14:33.320 writing emails or planning dinner. You can also 00:14:33.320 --> 00:14:35.480 make the AI significantly smarter over time. 00:14:35.610 --> 00:14:37.950 You do this by adding more example pairs to your 00:14:37.950 --> 00:14:40.490 logic file. This is where the real personalization 00:14:40.490 --> 00:14:42.590 actually starts happening. The more before and 00:14:42.590 --> 00:14:44.669 after stories you show the system. The smarter 00:14:44.669 --> 00:14:46.950 and more refined its outputs eventually become. 00:14:47.230 --> 00:14:50.269 It gradually learns your highly specific personal 00:14:50.269 --> 00:14:53.149 preferences and quirks. It is exactly like teaching 00:14:53.149 --> 00:14:55.350 a friend how you like your coffee. Once they 00:14:55.350 --> 00:14:57.490 see you make it a few times, they stop asking. 00:14:58.110 --> 00:15:00.669 They just intuitively know exactly what you want. 00:15:00.830 --> 00:15:03.570 If the AI consistently gives you way too much 00:15:03.570 --> 00:15:06.149 text, you just add very short punchy examples 00:15:06.149 --> 00:15:09.009 to the file. If it misses crucial details, show 00:15:09.009 --> 00:15:11.590 it a JSON with many fields. Every single new 00:15:11.590 --> 00:15:14.309 pair makes the system feel more like you. This 00:15:14.309 --> 00:15:16.429 system works incredibly beautifully for technical 00:15:16.429 --> 00:15:19.070 coding as well. It makes absolutely sure you 00:15:19.070 --> 00:15:21.570 include important technical details, things like 00:15:21.570 --> 00:15:24.750 web accessibility, proper meta tags, and clean 00:15:24.750 --> 00:15:27.129 structure. It actively prevents you from getting 00:15:27.129 --> 00:15:29.509 sloppy buggy code back. Even if you think your 00:15:29.509 --> 00:15:31.330 prompt is already pretty good, you should still 00:15:31.330 --> 00:15:33.669 run it through the optimizer system anyway. The 00:15:33.669 --> 00:15:36.269 source notes it improves good prompts by 20 to 00:15:36.269 --> 00:15:38.789 30 percent. It consistently finds a reliable 00:15:38.789 --> 00:15:42.110 way to add a clearer thinking path. It actively 00:15:42.110 --> 00:15:44.730 catches the small technical details we naturally 00:15:44.730 --> 00:15:47.929 forget. What if I want to type my messy ideas 00:15:47.929 --> 00:15:51.250 in a language other than English? Will this system 00:15:51.250 --> 00:15:55.259 still work? Yes. Gemini is excellent at translating, 00:15:55.460 --> 00:15:57.679 so you can type in your native language and the 00:15:57.679 --> 00:15:59.779 system will still automatically optimize the 00:15:59.779 --> 00:16:02.320 logic. It takes your native language and still 00:16:02.320 --> 00:16:05.039 optimizes the prompt perfectly. It seamlessly 00:16:05.039 --> 00:16:07.419 handles the complex translation and the logic 00:16:07.419 --> 00:16:10.460 cleanup simultaneously. It is an incredibly robust 00:16:10.460 --> 00:16:14.700 system for international users. Beat! We have 00:16:14.700 --> 00:16:17.059 covered a massive amount of ground today. Let's 00:16:17.059 --> 00:16:20.120 zoom out and recap the really big overarching 00:16:20.120 --> 00:16:23.259 idea. The biggest takeaway is shifting your fundamental 00:16:23.259 --> 00:16:26.299 mental framework. You do not need to become a 00:16:26.299 --> 00:16:28.360 professional prompt engineer. You do not need 00:16:28.360 --> 00:16:31.519 to memorize complex, rigid conversational frameworks. 00:16:31.700 --> 00:16:34.100 Your instruction text files are the absolute 00:16:34.100 --> 00:16:37.120 undeniable truth. Notebook LM is simply the filter 00:16:37.120 --> 00:16:39.440 for your messy human thoughts. Gemini simply 00:16:39.440 --> 00:16:42.100 takes the required action. based on that clean 00:16:42.100 --> 00:16:44.960 logic. You just need 10 uninterrupted minutes 00:16:44.960 --> 00:16:47.360 to set up the brain. Then you just let the automated 00:16:47.360 --> 00:16:49.700 filter do its daily job. It completely changes 00:16:49.700 --> 00:16:52.039 the dynamic of how you work. You stop fighting 00:16:52.039 --> 00:16:54.259 the machine to get what you want. You start actually 00:16:54.259 --> 00:16:56.220 collaborating with it on a structural level. 00:16:56.600 --> 00:16:58.899 I strongly encourage everyone listening to try 00:16:58.899 --> 00:17:02.460 this out. Do not wait for a massive, overwhelming 00:17:02.460 --> 00:17:05.839 project to begin. Start with something very small 00:17:05.839 --> 00:17:08.559 and incredibly low stakes today. Plan a simple 00:17:08.559 --> 00:17:11.220 dinner menu for the upcoming week. Plan a very 00:17:11.220 --> 00:17:14.099 simple weekend trip to a nearby city. Type in 00:17:14.099 --> 00:17:16.599 your messiest, least organized thoughts about 00:17:16.599 --> 00:17:19.160 the topic. Just sit back and watch how the system 00:17:19.160 --> 00:17:21.420 cleans it up. See the massive difference in the 00:17:21.420 --> 00:17:23.640 final structured output. Once you see the power, 00:17:23.779 --> 00:17:25.299 you're never going to go back to guessing. It 00:17:25.299 --> 00:17:27.680 really is a fundamental paradigm shift in daily 00:17:27.680 --> 00:17:31.440 computing. I want to leave you with one final 00:17:31.440 --> 00:17:34.559 thought to mull over today. If two free tools 00:17:34.559 --> 00:17:37.000 can take our messiest half -formed thoughts and 00:17:37.000 --> 00:17:39.119 translate them perfectly into structured machine 00:17:39.119 --> 00:17:42.259 logic, how long until we stop typing instructions 00:17:42.259 --> 00:17:44.740 altogether and just let AI listen to our messy 00:17:44.740 --> 00:17:46.500 conversations to run our lives for us?