WEBVTT 00:00:00.000 --> 00:00:01.720 I want to start with this statistic that feels 00:00:01.720 --> 00:00:05.379 almost counterintuitive today. Yeah. Roughly 00:00:05.379 --> 00:00:08.839 90 % of people using Notebook LM to generate 00:00:08.839 --> 00:00:11.859 slides are doing it completely wrong. They ignore 00:00:11.859 --> 00:00:15.060 one tiny UI element, and because of that, they 00:00:15.060 --> 00:00:18.359 just get generic corporate slop. It is wild. 00:00:18.440 --> 00:00:21.339 They ignore the pencil icon. That one button 00:00:21.339 --> 00:00:23.460 is the difference between a bland summary and 00:00:23.460 --> 00:00:25.940 having an AI act like your own senior creative 00:00:25.940 --> 00:00:28.399 director. Welcome back to the Deep Dive. It is 00:00:28.399 --> 00:00:32.579 Tuesday, February 24th, 2026. Today, we are looking 00:00:32.579 --> 00:00:36.179 at a new guide by Max Ann from AI Fire. It explores 00:00:36.179 --> 00:00:39.359 seven blueprints for moving beyond basic AI summarization 00:00:39.359 --> 00:00:41.780 into something called creative synthesis. Right, 00:00:41.920 --> 00:00:43.780 which is the big shift right now. We are going 00:00:43.780 --> 00:00:46.420 to cover how to transform literally a messy napkin 00:00:46.420 --> 00:00:48.579 into a boardroom deck. We'll look at creating 00:00:48.579 --> 00:00:51.500 manga -style storybooks and the exact 2026 instructions 00:00:51.500 --> 00:00:53.899 you need to make it happen for you. We have a 00:00:53.899 --> 00:00:55.600 lot of ground to cover. We're looking at sources 00:00:55.600 --> 00:00:57.939 that range from handwritten notes to dense technical 00:00:57.939 --> 00:01:00.219 papers. The mission today is to figure out how 00:01:00.219 --> 00:01:03.259 to stop using AI as just a fast reader and start 00:01:03.259 --> 00:01:05.469 using it as a true design part. Let's unpack 00:01:05.469 --> 00:01:08.409 the core problem first. People dump their sources 00:01:08.409 --> 00:01:11.109 into the tool. They click generate. Why is that 00:01:11.109 --> 00:01:13.989 default button such a trap? Because it is the 00:01:13.989 --> 00:01:15.849 path of least resistance. You know, you upload 00:01:15.849 --> 00:01:18.769 your PDFs or audio files into Notebook LM. You 00:01:18.769 --> 00:01:21.349 see that big generate button. You click it. And 00:01:21.349 --> 00:01:24.969 the AI just defaults to the safest, most average 00:01:24.969 --> 00:01:27.310 output. It just gives you text on a screen. Right. 00:01:27.390 --> 00:01:29.650 It is factually accurate, but it has absolutely 00:01:29.650 --> 00:01:32.640 no soul. For those who might be new to this specific 00:01:32.640 --> 00:01:35.780 tool, Notebook LM is Google's free multimodal 00:01:35.780 --> 00:01:38.180 tool. Multimodal just means it understands text, 00:01:38.299 --> 00:01:41.200 images, and audio all at once. And the defining 00:01:41.200 --> 00:01:45.180 feature here is grounded creativity. Unlike other 00:01:45.180 --> 00:01:47.739 AIs that pull random data from the web, this 00:01:47.739 --> 00:01:50.239 one only uses the sources you upload. Which is 00:01:50.239 --> 00:01:53.000 great for accuracy. But that safety rail makes 00:01:53.000 --> 00:01:55.980 the default slides incredibly boring. It is afraid 00:01:55.980 --> 00:01:58.560 to improvise. That is why you need the custom 00:01:58.560 --> 00:02:00.780 instructions menu. The pencil icon. Exactly. 00:02:00.920 --> 00:02:02.620 It sits right next to the generator. That is 00:02:02.620 --> 00:02:04.459 your control panel. You can force the AI into 00:02:04.459 --> 00:02:06.799 a specific role. You can tell it you are a Japanese 00:02:06.799 --> 00:02:09.620 manga illustrator or you are a senior workshop 00:02:09.620 --> 00:02:12.780 lead. So it's creativity, but restricted strictly 00:02:12.780 --> 00:02:15.300 to the truth you provide. Exactly. It creates 00:02:15.300 --> 00:02:18.400 style without inventing facts. Let's move to 00:02:18.400 --> 00:02:21.960 the first blueprint. The guide calls this handwritten 00:02:21.960 --> 00:02:25.560 notes to slides. I think this resonates with 00:02:25.560 --> 00:02:29.300 anyone who still prefers analog thinking. The 00:02:29.300 --> 00:02:32.080 messy page of ideas. Oh, the conspiracy wall 00:02:32.080 --> 00:02:34.939 phase. Yeah, exactly. Circles, arrows, half thoughts. 00:02:35.180 --> 00:02:37.120 We all do it. You have this great brainstorming 00:02:37.120 --> 00:02:40.219 session. The notebook looks like chaos. And then 00:02:40.219 --> 00:02:42.039 you realize you have to type it all out. It kills 00:02:42.039 --> 00:02:44.439 the momentum. So how does the pencil icon fix 00:02:44.439 --> 00:02:46.840 that? It's actually really fluid. Step one, you 00:02:46.840 --> 00:02:48.699 just snap a photo of the napkin. Step two, you 00:02:48.699 --> 00:02:51.259 upload it. The AI reads handwriting incredibly 00:02:51.259 --> 00:02:53.680 well now. Even my terrible handwriting. Probably 00:02:53.680 --> 00:02:55.520 better than you can read it. But step three is 00:02:55.520 --> 00:02:57.979 the key. You use the pencil icon to request a 00:02:57.979 --> 00:03:01.159 style. The guide uses a space exploration example. 00:03:01.520 --> 00:03:03.479 Someone uploaded rough notes about milestones 00:03:03.479 --> 00:03:05.400 and future missions. And what did they ask for? 00:03:05.520 --> 00:03:08.099 They asked for a modern, inspiring aesthetic. 00:03:08.479 --> 00:03:11.659 And the AI didn't just transcribe the text. It 00:03:11.659 --> 00:03:14.879 built a stunning deck. And it even kept the original 00:03:14.879 --> 00:03:17.539 handwritten image as a very first slide. Oh, 00:03:17.620 --> 00:03:20.099 to keep the authenticity. Right. It shows exactly 00:03:20.099 --> 00:03:22.199 where the idea started before moving into the 00:03:22.199 --> 00:03:24.379 polished slides. Does it actually understand 00:03:24.379 --> 00:03:26.960 the logic of the arrows and circles? Surprisingly, 00:03:26.960 --> 00:03:29.520 yes. It follows your visual logic perfectly. 00:03:29.800 --> 00:03:32.539 That is fascinating. It translates the visual 00:03:32.539 --> 00:03:36.120 relationship, not just the words. Beat. Let's 00:03:36.120 --> 00:03:38.159 look at the second use case. This addresses the 00:03:38.159 --> 00:03:41.020 opposite problem. The source material isn't messy, 00:03:41.180 --> 00:03:44.259 it's just incredibly dense. The glazed eyes phenomenon. 00:03:44.419 --> 00:03:48.680 Yes. Reading a 40 -page research paper. The information 00:03:48.680 --> 00:03:51.639 density is high, but retention is basically zero. 00:03:51.840 --> 00:03:54.520 Because it is a wall of text, the solution here 00:03:54.520 --> 00:03:57.699 is narrative. You have to treat the AI as a storyteller. 00:03:57.800 --> 00:04:00.280 You upload your dense sources. But instead of 00:04:00.280 --> 00:04:02.360 asking for bullet points, you use the pencil 00:04:02.360 --> 00:04:04.639 icon to ask for a visual storybook. How does 00:04:04.639 --> 00:04:07.120 that change the output? It changes the delivery 00:04:07.120 --> 00:04:09.500 mechanism. You are asking for metaphors. You 00:04:09.500 --> 00:04:12.300 are asking for cause and effect. Whoa, imagine 00:04:12.300 --> 00:04:15.159 taking a dense quantum physics paper and instantly 00:04:15.159 --> 00:04:17.519 turning it into a graphic novel for a 12 year 00:04:17.519 --> 00:04:19.660 old. That completely changes how we approach 00:04:19.660 --> 00:04:22.569 education. It really does. It is perfect for 00:04:22.569 --> 00:04:25.149 visual learners or teaching kids complex topics. 00:04:25.290 --> 00:04:28.350 You aren't dumbing down the science. You are 00:04:28.350 --> 00:04:31.189 just stacking Lego blocks of data into a story 00:04:31.189 --> 00:04:34.129 they can actually digest. Is the retention actually 00:04:34.129 --> 00:04:36.470 higher with this narrative approach? Research 00:04:36.470 --> 00:04:41.170 says yes. Story stick, raw data fades. That brings 00:04:41.170 --> 00:04:43.970 us to use case three, which tackles a massive 00:04:43.970 --> 00:04:47.699 corporate pain point, ugly slides. Inconsistent 00:04:47.699 --> 00:04:50.360 fonts. The wall of text. The Frankenstein deck, 00:04:50.439 --> 00:04:52.180 where you stitch together five different presentations. 00:04:52.579 --> 00:04:54.660 It is the worst. But the guide has this brand 00:04:54.660 --> 00:04:56.819 guide hack. You basically build a content -only 00:04:56.819 --> 00:04:58.980 deck in Google Slides. It can be completely ugly. 00:04:59.139 --> 00:05:02.199 You export it as a PDF and upload it. Okay, but 00:05:02.199 --> 00:05:04.100 how does it know your brand? You upload a second 00:05:04.100 --> 00:05:06.699 PDF, a brand Bible. This has your hex codes, 00:05:06.980 --> 00:05:09.519 your fonts, your logo rules. And then you prompt 00:05:09.519 --> 00:05:12.439 it. You use the pencil icon and say, apply the 00:05:12.439 --> 00:05:15.459 visual rules from the brand Bible to the content 00:05:15.459 --> 00:05:18.439 deck. And it automatically maps the style. It 00:05:18.439 --> 00:05:20.680 aligns the fonts. It creates visual consistency. 00:05:20.959 --> 00:05:23.379 So you never have to manually format a slide 00:05:23.379 --> 00:05:26.259 again. Not if you upload that brand Bible once. 00:05:26.519 --> 00:05:30.620 That is a massive time saver. Sponsor. Placeholder 00:05:30.620 --> 00:05:32.339 for separate script. Do not include content. 00:05:33.339 --> 00:05:36.100 We're back. Let's shift from corporate workflows 00:05:36.100 --> 00:05:40.279 to personal skill mastery. Use case four is the 00:05:40.279 --> 00:05:43.519 step -by -step guide. I have a vulnerable admission 00:05:43.519 --> 00:05:45.639 here. I still wrestle with prompt drift myself. 00:05:46.000 --> 00:05:47.779 Oh, totally. We all do. I'll want to learn a 00:05:47.779 --> 00:05:49.959 new coding language, and I end up with 20 tabs 00:05:49.959 --> 00:05:52.480 open, but I don't know the sequence. I want to 00:05:52.480 --> 00:05:54.220 learn, but I don't know the structure. That is 00:05:54.220 --> 00:05:56.660 tutorial hell. The progression plane use case 00:05:56.660 --> 00:05:59.360 solves this. The guide uses the one -armed pull 00:05:59.360 --> 00:06:01.779 -up as an example. It's a really hard skill. 00:06:01.920 --> 00:06:03.759 So you upload all the scattered advice. Yeah, 00:06:03.779 --> 00:06:05.639 you upload YouTube transcripts, Reddit threads, 00:06:05.860 --> 00:06:08.199 fitness blogs. You intentionally feed it the 00:06:08.199 --> 00:06:10.579 chaos. Yeah. Then you use the pencil icon to 00:06:10.579 --> 00:06:12.800 ask for a structured progression plan. And it 00:06:12.800 --> 00:06:15.079 outputs a roadmap. Right. It gives you baseline 00:06:15.079 --> 00:06:18.839 strength requirements. Progressive drills. Stage 00:06:18.839 --> 00:06:20.879 -by -stage advancement. It acts like your own 00:06:20.879 --> 00:06:23.819 personal coach. It synthesizes the conflicting 00:06:23.819 --> 00:06:26.519 advice from Reddit and YouTube. It finds the 00:06:26.519 --> 00:06:29.199 consensus and orders it logically. That is incredibly 00:06:29.199 --> 00:06:32.860 practical. Let's move to use cases 5 and 6. These 00:06:32.860 --> 00:06:35.060 are for high -pressure settings. The classic 00:06:35.060 --> 00:06:37.740 Monday morning panic for a Friday workshop. Right, 00:06:37.800 --> 00:06:39.740 where you have no plan and you just end up making 00:06:39.740 --> 00:06:42.240 a boring lecture. But use case 5 changes that. 00:06:42.480 --> 00:06:45.279 You upload your case studies. Then you ask for 00:06:45.279 --> 00:06:47.339 a workshop structure. And it doesn't just give 00:06:47.339 --> 00:06:49.379 you slides. It generates discussion prompts. 00:06:49.579 --> 00:06:52.040 It creates live activities and reflection questions. 00:06:52.319 --> 00:06:54.740 It essentially writes a full lesson plan for 00:06:54.740 --> 00:06:57.850 you. It builds the actual experience. Two sec 00:06:57.850 --> 00:07:00.569 silence. Use case six applies a similar logic 00:07:00.569 --> 00:07:03.290 to debates. We usually only research things that 00:07:03.290 --> 00:07:06.389 confirm our own biases. Which is dangerous. The 00:07:06.389 --> 00:07:08.810 debate use case forces critical thinking. You 00:07:08.810 --> 00:07:11.509 upload conflicting sources. For example, articles 00:07:11.509 --> 00:07:13.889 saying AI is great for jobs and articles saying 00:07:13.889 --> 00:07:16.529 AI will destroy the economy. And you ask for 00:07:16.529 --> 00:07:19.019 a head -to -head comparison deck. Exactly. You 00:07:19.019 --> 00:07:21.199 get a slide showing the productivity gains right 00:07:21.199 --> 00:07:23.379 next to a slide showing the inequality risks. 00:07:23.579 --> 00:07:25.959 It forces you to see the trade -offs. Does it 00:07:25.959 --> 00:07:29.180 pick a winner in the debate? No. It steelmans 00:07:29.180 --> 00:07:32.000 both sides for clarity. It builds the strongest 00:07:32.000 --> 00:07:36.000 possible argument for both perspectives. I love 00:07:36.000 --> 00:07:40.560 that. Now, use case 7. The guide calls this the 00:07:40.560 --> 00:07:43.509 recursive speaker notes trick. This is the absolute 00:07:43.509 --> 00:07:46.129 smartest workflow in the entire guide. It sounds 00:07:46.129 --> 00:07:48.230 a bit meta. Walk us through it. Okay, so you 00:07:48.230 --> 00:07:50.110 generate your slide deck using everything we 00:07:50.110 --> 00:07:51.730 just talked about. You download that finished 00:07:51.730 --> 00:07:54.399 deck. Then you re -upload that finished deck 00:07:54.399 --> 00:07:57.379 back into Notebook LM as a new source. You feed 00:07:57.379 --> 00:07:59.420 the final product back into the machine. Yes. 00:07:59.699 --> 00:08:02.040 Then you go to the standard chat, not the slide 00:08:02.040 --> 00:08:04.620 generator, and you ask for structured speaker 00:08:04.620 --> 00:08:06.639 notes for the file you just uploaded. Because 00:08:06.639 --> 00:08:08.759 it can see the slides it just made. Exactly. 00:08:08.860 --> 00:08:11.279 It looks at the visual cues, it pulls the underlying 00:08:11.279 --> 00:08:14.420 data, and it writes a slide -by -slide script 00:08:14.420 --> 00:08:16.740 for you. You're feeding the output back in as 00:08:16.740 --> 00:08:19.459 input. Exactly. It analyzes its own creation 00:08:19.459 --> 00:08:22.110 to coach you. Let's pause there. We've talked 00:08:22.110 --> 00:08:24.490 a lot about the magic, but we need to talk about 00:08:24.490 --> 00:08:27.310 reality. What breaks? What is the catch here? 00:08:27.490 --> 00:08:29.410 There are three main limitations you have to 00:08:29.410 --> 00:08:32.730 know about. First is the watermarks. There is 00:08:32.730 --> 00:08:35.470 a notebook LM watermark in the bottom right corner 00:08:35.470 --> 00:08:38.029 of every slide. You can't remove it inside the 00:08:38.029 --> 00:08:39.950 tool. You have to export it and crop it manually. 00:08:39.990 --> 00:08:42.350 Yeah, which is annoying. The second limitation 00:08:42.350 --> 00:08:46.450 is no editing. You cannot change the text inside 00:08:46.450 --> 00:08:48.710 Notebook LM. If it makes a typo, you can't just 00:08:48.710 --> 00:08:51.289 click and fix it. That seems like a massive friction 00:08:51.289 --> 00:08:53.950 point. It is. You have to export to Google Slides 00:08:53.950 --> 00:08:56.850 or PowerPoint to fix any text. And the third 00:08:56.850 --> 00:08:58.889 limitation is what they call the clean trap. 00:08:59.090 --> 00:09:01.840 Meaning the aesthetic. Right. Unless you use 00:09:01.840 --> 00:09:05.220 a really radical style prompt like a 1960s NASA 00:09:05.220 --> 00:09:08.379 poster or a chalkboard style, the visuals tend 00:09:08.379 --> 00:09:10.919 to look very similar. They lean toward a clean, 00:09:11.039 --> 00:09:13.539 generic corporate look. So this is for the first 00:09:13.539 --> 00:09:16.080 80 % of the work? Yes. It gets you to the polish 00:09:16.080 --> 00:09:18.899 phase instantly. You still have to do the final 00:09:18.899 --> 00:09:22.240 20 % yourself. That seems fair for how much time 00:09:22.240 --> 00:09:25.240 it saves. Let's do a quick recap. The big idea 00:09:25.240 --> 00:09:27.679 here is a fundamental shift in how we view the 00:09:27.679 --> 00:09:30.960 software. Right. In 2026, you cannot treat this 00:09:30.960 --> 00:09:33.179 like a search engine or a photocopier. It is 00:09:33.179 --> 00:09:35.600 a creative synthesizer. You have to stop clicking 00:09:35.600 --> 00:09:37.799 generate and start clicking the pencil. Treat 00:09:37.799 --> 00:09:40.559 the AI as a designer. Give it a role. Give it 00:09:40.559 --> 00:09:43.440 constraints. That is where the magic happens. 00:09:43.740 --> 00:09:45.960 If you want to try this out today, the guide 00:09:45.960 --> 00:09:48.379 suggests starting with the handwritten note trick. 00:09:48.639 --> 00:09:51.159 Take a picture of your messy notepad, upload 00:09:51.159 --> 00:09:54.019 it. and use the pencil icon. It is the ultimate 00:09:54.019 --> 00:09:56.799 gateway drug to this entire workflow. Once you 00:09:56.799 --> 00:09:59.100 see it work, you will never look at your notes 00:09:59.100 --> 00:10:01.480 the same way. Before we sign off, I want to leave 00:10:01.480 --> 00:10:04.360 you with a thought. If this tool allows anyone 00:10:04.360 --> 00:10:06.860 to generate a perfectly structured, beautifully 00:10:06.860 --> 00:10:09.980 narrated presentation in 10 seconds, what happens 00:10:09.980 --> 00:10:12.360 to the baseline? When a perfect slide deck is 00:10:12.360 --> 00:10:14.779 no longer impressive, what will make your actual 00:10:14.779 --> 00:10:18.139 ideas stand out? Beat? Thanks for diving in. 00:10:18.200 --> 00:10:18.659 See you next time.