WEBVTT 00:00:00.000 --> 00:00:02.799 Most people use AI like a fancy search box. We 00:00:02.799 --> 00:00:05.540 type a simple question, get an OK answer, and 00:00:05.540 --> 00:00:08.580 we move on. It's pretty transactional. But what 00:00:08.580 --> 00:00:11.400 separates that casual user from someone who's 00:00:11.400 --> 00:00:14.279 operating in the top 1 %? It's not just asking 00:00:14.279 --> 00:00:16.739 better questions. It's teaching the AI how to 00:00:16.739 --> 00:00:19.699 think and how to build for you. And that's really 00:00:19.699 --> 00:00:22.640 the core difference, isn't it? The AI is built 00:00:22.640 --> 00:00:26.820 for speed, maybe not always for perfection. So 00:00:26.820 --> 00:00:29.440 welcome to the Deep Dive. Today yeah, we're talking 00:00:29.440 --> 00:00:32.119 about transforming chat GPT moving it from just 00:00:32.119 --> 00:00:35.259 a quick answer machine into well a systematic 00:00:35.259 --> 00:00:37.179 personalized thinking partner We've looked at 00:00:37.179 --> 00:00:39.380 a whole stack of power user guides really dug 00:00:39.380 --> 00:00:41.740 in and we're gonna distill it down to three core 00:00:41.740 --> 00:00:43.640 shifts you could make first is making the AI 00:00:43.640 --> 00:00:45.619 self -correct second giving it a kind of long 00:00:45.619 --> 00:00:48.759 -term memory and Finally training it to be your 00:00:48.759 --> 00:00:50.700 specialist coach right our mission here is pretty 00:00:50.700 --> 00:00:52.579 simple We want to move you from just asking for 00:00:52.579 --> 00:00:54.960 those quick answers to actually building customized 00:00:54.960 --> 00:00:57.539 Autonomous systems systems that can save you 00:00:57.539 --> 00:00:59.909 out every single week. Okay, so let's unpack 00:00:59.909 --> 00:01:04.329 that first shift. Systemizing the output. We 00:01:04.329 --> 00:01:06.709 kind of know the first answer Chet GPT gives. 00:01:07.209 --> 00:01:09.150 Well, it's often just average, right? It's the 00:01:09.150 --> 00:01:11.189 fastest thing it can generate. And sometimes 00:01:11.189 --> 00:01:12.769 speed isn't what you need. It can be the enemy 00:01:12.769 --> 00:01:15.530 of quality. Exactly. It defaults to speed. So 00:01:15.530 --> 00:01:17.310 the first big strategy here is something called 00:01:17.310 --> 00:01:20.150 the self -check method. And this basically forces 00:01:20.150 --> 00:01:24.250 the AI to do better work by building critique. 00:01:24.430 --> 00:01:26.709 right into the prompt itself. Yeah, it's a prompt 00:01:26.709 --> 00:01:28.769 trick, basically. You don't need anything fancy. 00:01:28.989 --> 00:01:31.230 You're telling the AI how to think before you 00:01:31.230 --> 00:01:33.109 tell it what to write. It's like we build a little 00:01:33.109 --> 00:01:35.069 internal review process, but it's hidden from 00:01:35.069 --> 00:01:38.010 you, the user. OK, so how does that work in practice? 00:01:38.090 --> 00:01:40.049 How do you build that internal critique into 00:01:40.049 --> 00:01:42.969 just one prompt? What are the steps? You set 00:01:42.969 --> 00:01:45.670 up a four -step sequence. So step one, think 00:01:45.670 --> 00:01:48.390 inside. It outlines what makes a good response. 00:01:48.890 --> 00:01:50.590 But, and this is key, it doesn't show you this 00:01:50.590 --> 00:01:53.069 thinking. Step two is write draft one. which 00:01:53.069 --> 00:01:55.750 also stays hidden. That's usually the fast kind 00:01:55.750 --> 00:01:57.890 of average answer. And the magic happens in step 00:01:57.890 --> 00:02:00.810 three, right? Check and fix. This is where you 00:02:00.810 --> 00:02:03.310 tell it, OK, critique draft one against those 00:02:03.310 --> 00:02:05.250 rules you thought about. Find the weak spots 00:02:05.250 --> 00:02:07.709 and fix them. Exactly. And then finally, the 00:02:07.709 --> 00:02:11.389 AI only shows you the superior draft two. And 00:02:11.389 --> 00:02:14.729 this works because it forces the AI to kind of 00:02:14.729 --> 00:02:18.189 slow down, dedicate resources to self -reflection. 00:02:18.449 --> 00:02:21.210 It catches its own mistakes before you do. So 00:02:21.210 --> 00:02:23.289 the final output is just dramatically better. 00:02:23.550 --> 00:02:25.310 It's about demanding that internal accountability. 00:02:25.650 --> 00:02:28.289 That self -correction makes the output cleaner, 00:02:28.569 --> 00:02:31.199 for sure. But the builder mindset also means 00:02:31.199 --> 00:02:34.419 connecting the AI outside, right? How do we take 00:02:34.419 --> 00:02:38.099 that better text output and turn it into finished 00:02:38.099 --> 00:02:40.219 products? Ah, yeah. This is where tool integration 00:02:40.219 --> 00:02:42.120 comes in. We're moving beyond just text. You 00:02:42.120 --> 00:02:44.000 can connect third -party apps now, things like 00:02:44.000 --> 00:02:46.400 Canva or Figma, right there in the ChatGPT settings. 00:02:46.520 --> 00:02:48.439 OK, so think about presentations. Let's say you 00:02:48.439 --> 00:02:51.500 upload a five -page document, maybe a quarterly 00:02:51.500 --> 00:02:53.539 marketing report. Pretty dense. You could tell 00:02:53.539 --> 00:02:56.199 ChatGPT, I want a 10 -slide draft, maybe five 00:02:56.199 --> 00:02:59.189 slides summarizing findings, three on key takeaways, 00:02:59.509 --> 00:03:01.569 and I don't know, two on next steps. Right, so 00:03:01.569 --> 00:03:04.949 chat GPT reads the content, figures out the summaries, 00:03:04.990 --> 00:03:07.729 and then tells the connected Canva tool, hey, 00:03:08.050 --> 00:03:09.650 generate some professional designs for this. 00:03:09.810 --> 00:03:12.590 You instantly get like a 90 % finished presentation. 00:03:12.689 --> 00:03:14.569 That's a huge time saver. You're not starting 00:03:14.569 --> 00:03:16.669 from scratch. You're basically editing something 00:03:16.669 --> 00:03:20.009 that's almost done. Or take Figma. If you need 00:03:20.009 --> 00:03:22.830 a workflow diagram, maybe for onboarding a new 00:03:22.830 --> 00:03:25.389 employee, which can get complicated, you just 00:03:25.389 --> 00:03:27.490 give it simple text instructions, define your 00:03:27.490 --> 00:03:30.110 colors, say blue boxes for actions, orange diamonds 00:03:30.110 --> 00:03:32.770 for decisions, and boom, it generates a perfect, 00:03:33.030 --> 00:03:35.310 shareable flowchart. The real power here is that 00:03:35.310 --> 00:03:37.830 it removes the technical skill barrier. You don't 00:03:37.830 --> 00:03:40.250 need to be a Canva wizard or know how to manually 00:03:40.250 --> 00:03:43.750 draw stuff in Figma. You just describe the result 00:03:43.750 --> 00:03:47.340 you want, and the AI builds it. So if the core 00:03:47.340 --> 00:03:49.759 difference is forcing that self -reflection and 00:03:49.759 --> 00:03:52.280 integration, what's the single biggest takeaway 00:03:52.280 --> 00:03:54.879 for everyday prompt writing? Always instruct 00:03:54.879 --> 00:03:57.099 the AI how to think before you tell it what to 00:03:57.099 --> 00:04:00.599 write. Okay, the next big shift then. This involves 00:04:00.599 --> 00:04:04.199 giving the AI a persistent brand persona. Basically 00:04:04.199 --> 00:04:06.379 getting rid of that repetitive work of explaining 00:04:06.379 --> 00:04:08.719 who you are and what your style is every single 00:04:08.719 --> 00:04:11.020 time you start a new chat. Yeah, that constant 00:04:11.020 --> 00:04:13.979 explaining leads to what we call prompt fatigue. 00:04:14.539 --> 00:04:16.980 It's tiring. And that's exactly why the custom 00:04:16.980 --> 00:04:19.860 instructions feature is, well, it's critical. 00:04:20.259 --> 00:04:22.579 It acts like ChatGPT's long -term memory about 00:04:22.579 --> 00:04:25.060 you and what you like. We saw this great example 00:04:25.060 --> 00:04:27.959 using a small local business. Let's call it... 00:04:27.720 --> 00:04:30.620 Happy Bakery. You set up the instructions to 00:04:30.620 --> 00:04:33.360 define the brand as warm, friendly, and you enforce 00:04:33.360 --> 00:04:35.939 specific rules like use short, simple sentences, 00:04:36.100 --> 00:04:38.100 maybe use certain emojis like a cupcake carry, 00:04:38.300 --> 00:04:40.519 and always ask customers to visit the actual 00:04:40.519 --> 00:04:43.060 bakery. And now every single new chat, whether 00:04:43.060 --> 00:04:44.819 you ask for an email, an Instagram post, whatever, 00:04:45.259 --> 00:04:47.759 it automatically adopts that voice. Those rules, 00:04:47.959 --> 00:04:50.259 the systemic benefit is just huge. You save so 00:04:50.259 --> 00:04:52.279 much mental energy because you're not constantly 00:04:52.279 --> 00:04:55.220 reminding it, hey, be friendly or don't use complicated 00:04:55.220 --> 00:04:57.360 words. But sometimes you don't actually know 00:04:57.360 --> 00:04:59.439 your own brand voice that clearly right. You 00:04:59.439 --> 00:05:01.879 just kind of write what feels right at the time. 00:05:02.319 --> 00:05:04.120 Which leads to this really interesting strategy. 00:05:04.959 --> 00:05:07.800 Let the AI figure out your brand guide for you 00:05:07.800 --> 00:05:11.279 first. Exactly. You upload, say, five to ten 00:05:11.279 --> 00:05:13.360 pieces of your own content that you really like. 00:05:13.639 --> 00:05:15.879 Maybe blog posts, newsletters, emails you've 00:05:15.879 --> 00:05:18.699 written. Then you ask ChatGPT to analyze them. 00:05:19.040 --> 00:05:21.439 Act like a brand consultant for me. And it creates 00:05:21.439 --> 00:05:24.019 this detailed brand voice guide. It looks at 00:05:24.019 --> 00:05:27.269 your existing tone. Is it motivating? Gentle, 00:05:27.870 --> 00:05:30.970 your sentence style short and punchy, long and 00:05:30.970 --> 00:05:33.810 detailed, even your brand personality like. Do 00:05:33.810 --> 00:05:35.790 you sound more like a tough coach or a helpful 00:05:35.790 --> 00:05:38.170 friend? You know, I still wrestle with prompt 00:05:38.170 --> 00:05:40.370 drift myself sometimes. Honestly, I'll be like 00:05:40.370 --> 00:05:42.230 15 minutes into writing something with it. And 00:05:42.230 --> 00:05:44.329 I realized the tone has gone completely off track, 00:05:44.470 --> 00:05:46.110 maybe too corporate or something. And then I 00:05:46.110 --> 00:05:49.180 have to stop. backtrack, reset the context. Having 00:05:49.180 --> 00:05:52.420 the AI define my preferred style upfront saves 00:05:52.420 --> 00:05:54.740 me from that constant resetting. It's a major 00:05:54.740 --> 00:05:57.079 time drain otherwise. And once you've got that 00:05:57.079 --> 00:05:59.759 memory institutionalized, you can actually organize 00:05:59.759 --> 00:06:02.839 multiple specific personalities. The sources 00:06:02.839 --> 00:06:05.779 talk about using the memory feature to save these 00:06:05.779 --> 00:06:08.639 personas linked to a trigger word. Yeah, this 00:06:08.639 --> 00:06:11.100 is pretty cool. You could create, say, the tough 00:06:11.100 --> 00:06:13.660 editor persona, train it to be super direct, 00:06:13.819 --> 00:06:16.879 really strict, just cut out every unnecessary 00:06:16.879 --> 00:06:19.500 word, and you trigger it just by typing edit 00:06:19.500 --> 00:06:22.420 now. Or maybe you need a brainstorm buddy sometimes, 00:06:22.779 --> 00:06:25.519 right? Triggered by crazy ideas. It just instantly 00:06:25.519 --> 00:06:27.779 shifts its whole role and tone based on that 00:06:27.779 --> 00:06:30.920 one word. It's powerful. So by institutionalizing 00:06:30.920 --> 00:06:34.139 our style, these personalities... Are we maybe 00:06:34.139 --> 00:06:36.819 risking creative stagnation or is it just about 00:06:36.819 --> 00:06:39.339 increasing speed? Institutionalizing style frees 00:06:39.339 --> 00:06:41.160 up your mental energy for the higher level creative 00:06:41.160 --> 00:06:42.779 thinking. All right, let's shift gears a bit. 00:06:42.800 --> 00:06:45.160 Let's talk about complexity. The next set of 00:06:45.160 --> 00:06:47.639 strategies really focuses on using AI to find 00:06:47.639 --> 00:06:50.839 patterns, whether that's in like raw data or 00:06:50.839 --> 00:06:52.920 generating creative stuff without needing technical 00:06:52.920 --> 00:06:55.259 skills. Yeah, let's start with data storytelling. 00:06:56.040 --> 00:06:57.699 The problem is pretty common. You get these big 00:06:57.699 --> 00:07:00.430 Excel files full of numbers, right? sales figures, 00:07:00.449 --> 00:07:03.230 whatever. And it's hard to see the trends hidden 00:07:03.230 --> 00:07:05.730 in all those rows and columns. So the solution 00:07:05.730 --> 00:07:10.410 is you upload the raw file like a .xlsx or .csv 00:07:10.410 --> 00:07:13.850 - and you ask the AI to find the story in the 00:07:13.850 --> 00:07:15.449 numbers. You give it a specific prompt. Tell 00:07:15.449 --> 00:07:18.170 me the best and worst months. What's the general 00:07:18.170 --> 00:07:21.250 trend? Summarize the key takeaways. Maybe four 00:07:21.250 --> 00:07:24.410 bullet points. And crucially, ask for one suggested 00:07:24.410 --> 00:07:27.339 action based on that data. It turns confusing 00:07:27.339 --> 00:07:29.420 numbers, like sales reports or even big bank 00:07:29.420 --> 00:07:31.720 statements, into simple, actionable insights 00:07:31.720 --> 00:07:33.180 you can use right away. It's kind of like having 00:07:33.180 --> 00:07:35.680 an analyst on call instantly finding patterns 00:07:35.680 --> 00:07:38.120 your own eyes might just skim over. Okay, now 00:07:38.120 --> 00:07:39.939 shifting to creation. This is where things get 00:07:39.939 --> 00:07:42.060 really interesting, I think. Professional video. 00:07:42.160 --> 00:07:44.000 We know AI video tools are coming. They're getting 00:07:44.000 --> 00:07:46.240 better fast. But the bottleneck is often the 00:07:46.240 --> 00:07:49.139 prompt quality. So Chet, GPT's role here is writing 00:07:49.139 --> 00:07:51.959 that perfect, super detailed cinematic prompt. 00:07:52.259 --> 00:07:54.620 Right. Take a simple idea like a happy dog running 00:07:54.620 --> 00:07:56.439 on a beach. This is not going to give you a great 00:07:56.439 --> 00:07:59.699 video, it's too basic. So you ask ChatGPT to 00:07:59.699 --> 00:08:02.449 rewrite it cinematically. Force it to add details. 00:08:02.910 --> 00:08:05.870 Instruct it. What breed of dog? What specific 00:08:05.870 --> 00:08:08.269 time of day? What camera angle should we use? 00:08:08.529 --> 00:08:10.910 What precise feeling should the scene evoke? 00:08:11.029 --> 00:08:13.129 And the result is a prompt that specifies, like, 00:08:13.550 --> 00:08:15.769 a golden retriever joyfully running along the 00:08:15.769 --> 00:08:19.129 wet sand precisely at sunset, camera flying low 00:08:19.129 --> 00:08:21.629 beside the dog, capturing the glint of the golden 00:08:21.629 --> 00:08:24.810 light on its fur. The overall feeling must be 00:08:24.810 --> 00:08:27.370 warmth and freedom. That's a prompt that an AI 00:08:27.370 --> 00:08:29.850 video tool can actually work with. Whoa, yeah, 00:08:29.990 --> 00:08:32.909 imagine scaling that. A million unique high -quality 00:08:32.909 --> 00:08:35.470 video scenes generated every single day, all 00:08:35.470 --> 00:08:37.169 from just text instructions. Yeah. I mean, that 00:08:37.169 --> 00:08:39.029 really is the future of digital content, isn't 00:08:39.029 --> 00:08:41.649 it? And finally, for maybe folks with websites 00:08:41.649 --> 00:08:44.309 that feel a bit static or boring, our sources 00:08:44.309 --> 00:08:46.669 point into building simple website widgets using 00:08:46.669 --> 00:08:48.929 ChatGPT. You don't need to hire a programmer 00:08:48.929 --> 00:08:51.029 for everything. You can ask ChatGPT to write 00:08:51.029 --> 00:08:54.090 the full code, HTML, CSS, JavaScript, for something 00:08:54.090 --> 00:08:56.330 like a simple interest calculator. or maybe a 00:08:56.330 --> 00:08:58.669 calorie counter. And it generates the code. Usually 00:08:58.669 --> 00:09:00.990 pretty well. You can just copy that code block 00:09:00.990 --> 00:09:02.990 and paste it directly into your website builder 00:09:02.990 --> 00:09:05.789 like WordPress or Wix or whatever you use. It 00:09:05.789 --> 00:09:08.110 adds useful interactive features to your site 00:09:08.110 --> 00:09:10.889 in minutes. That's a massive return on your time 00:09:10.889 --> 00:09:13.610 investment. So which do you think is the bigger 00:09:13.610 --> 00:09:17.769 time saver overall? using AI to analyze existing 00:09:17.769 --> 00:09:20.529 data or using it to generate totally new creative 00:09:20.529 --> 00:09:23.690 assets. Analysis reveals where you need to act. 00:09:24.029 --> 00:09:26.889 Creation implements how to act quickly. Okay, 00:09:26.970 --> 00:09:30.129 our final core shift. This one feels maybe the 00:09:30.129 --> 00:09:32.909 most personal. It's about transforming the AI 00:09:32.909 --> 00:09:35.049 into more of a thinking partner, even a decision 00:09:35.049 --> 00:09:36.950 maker alongside you. Yeah, and this can start 00:09:36.950 --> 00:09:39.500 with active learning. Say you upload a complex 00:09:39.500 --> 00:09:42.879 document, maybe a report on Vietnam's economic 00:09:42.879 --> 00:09:45.279 history. Don't just ask for a summary. That's 00:09:45.279 --> 00:09:48.299 passive. Instead, ask ChatGPT, act like a demanding 00:09:48.299 --> 00:09:51.320 teacher, tell it. Quiz me one question at a time, 00:09:51.740 --> 00:09:53.820 hold me accountable, explain why my answers are 00:09:53.820 --> 00:09:56.080 right or wrong, and keep going until you hit 00:09:56.080 --> 00:09:58.279 a certain score. That's active learning. It really 00:09:58.279 --> 00:10:00.480 helps with retaining the information much more 00:10:00.480 --> 00:10:03.980 than just reading a summary. Okay, next up, organizing 00:10:03.980 --> 00:10:06.500 big decisions. And when you're planning something 00:10:06.500 --> 00:10:09.360 complex, maybe comparing Joe A versus Joe B, 00:10:09.679 --> 00:10:12.019 or deciding between Da Nang and Sapa for a vacation, 00:10:13.299 --> 00:10:15.500 that single chat thread can get really messy, 00:10:15.720 --> 00:10:17.960 right? Pros and cons all mixed up. The power 00:10:17.960 --> 00:10:21.120 move here is using the branch chat feature, which 00:10:21.120 --> 00:10:23.019 basically just means duplicating the current 00:10:23.019 --> 00:10:26.100 chat thread. So you explore option A completely 00:10:26.100 --> 00:10:30.480 separately from option B to distinct, clean conversations. 00:10:31.279 --> 00:10:33.240 It keeps your thinking organized and stops that 00:10:33.240 --> 00:10:35.600 feeling of cognitive overload. And we shouldn't 00:10:35.600 --> 00:10:38.480 overlook practicing soft skills. Using the mobile 00:10:38.480 --> 00:10:40.279 app's voice mode lets you practice difficult 00:10:40.279 --> 00:10:42.370 conversations out loud. Ask it to role play, 00:10:42.450 --> 00:10:45.090 like be a professional, serious HR manager and 00:10:45.090 --> 00:10:47.090 ask me five tough interview questions. Yeah, 00:10:47.169 --> 00:10:49.269 saying the words out loud actually builds muscle 00:10:49.269 --> 00:10:51.649 memory. You get to practice your phrasing, work 00:10:51.649 --> 00:10:54.149 through the awkward pauses, fix the stammers 00:10:54.149 --> 00:10:56.929 in private before you face the real high stakes 00:10:56.929 --> 00:10:59.409 conversation. It really helps. And all this kind 00:10:59.409 --> 00:11:02.429 of culminates in the ability to build a custom 00:11:02.429 --> 00:11:05.110 GPT advisor. This is like a specialized version 00:11:05.110 --> 00:11:08.889 of the AI that you train to think and talk like 00:11:08.889 --> 00:11:11.879 a specific persona or maybe a certain philosophy. 00:11:12.059 --> 00:11:14.539 Right, like the example of building a stoic advisor 00:11:14.539 --> 00:11:17.279 GPT. You train it specifically on stoic principles, 00:11:17.620 --> 00:11:19.600 instructing it to focus on what's controllable, 00:11:19.639 --> 00:11:22.779 on wisdom, on calm advice. So when you ask it 00:11:22.779 --> 00:11:25.440 about, say, a conflict at work, it won't just 00:11:25.440 --> 00:11:27.840 give you standard HR advice. It might say something 00:11:27.840 --> 00:11:30.019 like, you cannot control your co -workers' actions, 00:11:30.200 --> 00:11:32.899 only your reaction. Focus on what is within your 00:11:32.899 --> 00:11:35.600 sphere of influence. It gives you this entirely 00:11:35.600 --> 00:11:38.340 different specialized lens to look at your problems. 00:11:38.659 --> 00:11:40.620 It really becomes a targeted thinking partner. 00:11:40.730 --> 00:11:43.210 So does building a specialized GPT like that 00:11:43.210 --> 00:11:46.190 for advice fundamentally change how we relate 00:11:46.190 --> 00:11:48.710 to AI? Does it move from just a tool to more 00:11:48.710 --> 00:11:51.169 of a collaborator? Yes, I think it does. It evolves 00:11:51.169 --> 00:11:54.090 the tool into a specialized, highly targeted 00:11:54.090 --> 00:11:56.649 thinking partner. So the common thread here... 00:11:56.649 --> 00:11:58.850 Across all of these methods, it seems pretty 00:11:58.850 --> 00:12:01.909 simple, really. Top users don't just use ChatGBT 00:12:01.909 --> 00:12:04.529 for one -off answers. They build systems. They 00:12:04.529 --> 00:12:06.889 teach the AI about their brand, their voice, 00:12:06.929 --> 00:12:09.970 their processes, their goals. It's about persistence. 00:12:10.309 --> 00:12:12.809 Exactly. The real shift is moving from that sort 00:12:12.809 --> 00:12:15.830 of transactional input output to a more persistent 00:12:15.830 --> 00:12:18.309 contextual collaboration. The difference between 00:12:18.309 --> 00:12:21.950 a regular user and that top 1%, it isn't really 00:12:21.950 --> 00:12:24.269 about knowing more hidden features. It's about 00:12:24.269 --> 00:12:27.620 using the AI as a partner. a partner that holds 00:12:27.620 --> 00:12:30.960 long -term context and critically self -corrects 00:12:30.960 --> 00:12:33.759 its own output before you even see it. So maybe 00:12:33.759 --> 00:12:36.139 start small. Pick just one method that feels 00:12:36.139 --> 00:12:38.600 like it fits your workflow right now. Maybe setting 00:12:38.600 --> 00:12:40.840 up a simple custom instruction for your brand 00:12:40.840 --> 00:12:42.559 voice. That could be a good first step. The more 00:12:42.559 --> 00:12:44.440 you teach it about what matters to you, the more 00:12:44.440 --> 00:12:46.460 valuable it's going to become. Now here's something 00:12:46.460 --> 00:12:48.779 to think about. If you used this system building 00:12:48.779 --> 00:12:51.840 approach to automate one really major, painful 00:12:51.840 --> 00:12:53.960 task in your week, what system would you build 00:12:53.960 --> 00:12:54.299 first?