WEBVTT 00:00:00.000 --> 00:00:04.660 The difference between a generic AI answer and 00:00:04.660 --> 00:00:07.500 a really excellent one. It's all context, 100%. 00:00:07.500 --> 00:00:10.140 Yeah, it boils down entirely to that. We often 00:00:10.140 --> 00:00:12.900 treat these incredible models, like GPT 5 .2, 00:00:12.939 --> 00:00:14.759 as if they're just... The fancy Google search 00:00:14.759 --> 00:00:17.280 bar. Exactly. We ask these simple questions, 00:00:17.339 --> 00:00:19.320 and then we're surprised when the output is, 00:00:19.320 --> 00:00:22.940 you know, boring. It's flat. It's useless, frankly. 00:00:23.579 --> 00:00:25.239 But they aren't calculators. They're thinking 00:00:25.239 --> 00:00:27.559 machines. And if you want high -level thinking, 00:00:28.039 --> 00:00:29.679 you have to give them a high -level job to do. 00:00:29.739 --> 00:00:33.240 A role, a process, specific rules. And that's 00:00:33.240 --> 00:00:34.939 the foundation of what we're looking at today, 00:00:35.039 --> 00:00:37.799 this mega prompt framework. Welcome to the Deep 00:00:37.799 --> 00:00:40.740 Dive. Today we've synthesized research on 10 00:00:40.740 --> 00:00:44.380 very specific pretty advanced prompting frameworks. 00:00:44.539 --> 00:00:46.439 And these are designed to save you hours of work 00:00:46.439 --> 00:00:48.719 every week. We're talking across research, creative 00:00:48.719 --> 00:00:52.020 stuff, design, even your personal finances. Our 00:00:52.020 --> 00:00:54.299 mission today is to show you how to start treating 00:00:54.299 --> 00:00:59.020 GPT -5 .2 less like a tool and more like a true 00:00:59.020 --> 00:01:01.280 thinking partner. So let's get into it. The most 00:01:01.280 --> 00:01:04.939 challenging area first. Deep research and strategy. 00:01:05.340 --> 00:01:08.299 Okay, let's unpack it. The biggest pain point 00:01:08.299 --> 00:01:10.319 with high stakes research isn't just getting 00:01:10.319 --> 00:01:12.719 an answer, right? It's that thing everyone worries 00:01:12.719 --> 00:01:15.299 about AI hallucination. Where the model just 00:01:15.299 --> 00:01:17.859 confidently makes things up. Yeah, and you can't 00:01:17.859 --> 00:01:22.019 trust the output unless you force the model to... 00:01:22.060 --> 00:01:24.680 Self -correct to check its own work. The fix 00:01:24.680 --> 00:01:27.180 here isn't just asking a better question. It's 00:01:27.180 --> 00:01:30.140 demanding a whole process. Exactly. We introduced 00:01:30.140 --> 00:01:31.939 the chain of thought method. It's actually pretty 00:01:31.939 --> 00:01:33.599 simple in concept. You're just telling the AI 00:01:33.599 --> 00:01:35.719 to show its work. To think step by step before 00:01:35.719 --> 00:01:37.920 it gives you the final answer. And this first 00:01:37.920 --> 00:01:40.939 mega prompt, it turns the AI into a deep dive 00:01:40.939 --> 00:01:43.760 analyst. You give a very specific role. Like 00:01:43.760 --> 00:01:46.599 what? A senior market intelligence officer, 15 00:01:46.599 --> 00:01:49.140 years of experience, specializing in emerging 00:01:49.140 --> 00:01:52.930 tech. Then you tell it how to think. First, it 00:01:52.930 --> 00:01:55.969 has to fact check everything. And this is critical. 00:01:56.390 --> 00:01:59.290 Ignore any data older than 12 months. Keeps the 00:01:59.290 --> 00:02:01.530 analysis fresh. Right. No outdated conclusions? 00:02:01.689 --> 00:02:04.290 Here's where it gets really powerful. You force 00:02:04.290 --> 00:02:07.650 a multi -perspective simulation. The AI has to 00:02:07.650 --> 00:02:10.250 run an internal debate. A debate between who? 00:02:10.509 --> 00:02:13.650 Between, say, three different experts. An optimistic 00:02:13.650 --> 00:02:16.110 investor who's all about growth, a skeptical 00:02:16.110 --> 00:02:19.229 regulator focused on risk, and a pragmatic consumer 00:02:19.229 --> 00:02:22.090 who only cares about price and usability. Whoa. 00:02:22.310 --> 00:02:25.629 I mean, just imagine the efficiency there. Simulating 00:02:25.629 --> 00:02:28.490 a debate between three high -level experts instantly 00:02:28.490 --> 00:02:32.250 just to find the key risk. That scale is... that's 00:02:32.250 --> 00:02:34.789 powerful. And the final output has to zero in 00:02:34.789 --> 00:02:39.300 on one thing. The contrarian view. the unpopular 00:02:39.300 --> 00:02:41.919 opinion or the risk that everyone else is missing. 00:02:42.180 --> 00:02:44.139 So why is forcing it to find that contrarian 00:02:44.139 --> 00:02:46.860 view the most valuable part? It reveals risks 00:02:46.860 --> 00:02:48.979 your competitors haven't seen yet. Okay, so if 00:02:48.979 --> 00:02:51.620 deep research is about verification, our next 00:02:51.620 --> 00:02:55.219 area, writing, is all about trust. And attention. 00:02:55.599 --> 00:02:57.860 Prompt two is for structuring technical papers 00:02:57.860 --> 00:02:59.659 that people, you know, decision -makers will 00:02:59.659 --> 00:03:01.639 actually read. The big challenge there is that 00:03:01.639 --> 00:03:04.099 technical writing is often so dry, especially 00:03:04.099 --> 00:03:06.819 for a busy executive. So the solution here is 00:03:06.819 --> 00:03:09.000 to use a psychological flow. You're structuring 00:03:09.000 --> 00:03:11.080 the document to guide the reader on an emotional 00:03:11.080 --> 00:03:13.800 journey, really. You start with the hook, a specific 00:03:13.800 --> 00:03:16.400 costly problem, the pain. Then you show them 00:03:16.400 --> 00:03:19.020 the gap. which is why current solutions are failing. 00:03:19.379 --> 00:03:22.840 You introduce your idea as the logical next step. 00:03:23.300 --> 00:03:25.439 Then you provide the proof, the technical stuff, 00:03:25.639 --> 00:03:29.900 and you end with the payoff. A clear ROI. It's 00:03:29.900 --> 00:03:32.240 a full story. And there's a style rule here that's 00:03:32.240 --> 00:03:35.319 really important. Grade 10 readability. Yeah. 00:03:35.400 --> 00:03:37.439 We know busy executives prefer things that are 00:03:37.439 --> 00:03:39.300 easy to understand. They don't want jargon. It's 00:03:39.300 --> 00:03:41.539 not about sounding smart. It's about being clear. 00:03:41.639 --> 00:03:44.479 And you have to ban certain words. Critically. 00:03:44.860 --> 00:03:47.810 You have to ban the AI filler words. Things like 00:03:47.810 --> 00:03:51.409 unleash, game changer, revolutionary. How does 00:03:51.409 --> 00:03:54.289 banning those typical AI filler words make the 00:03:54.289 --> 00:03:56.550 writing stronger? It forces the model to find 00:03:56.550 --> 00:03:58.689 better, more specific terms. Top three moves 00:03:58.689 --> 00:04:02.310 us from analysis into, I guess, active strategy, 00:04:02.550 --> 00:04:05.409 using the AI to basically play a war game against 00:04:05.409 --> 00:04:07.650 your own product. You assign it the role of a 00:04:07.650 --> 00:04:10.009 ruthlessly efficient competitive strategy consultant, 00:04:10.650 --> 00:04:13.270 someone who is paid to find your flaws. And the 00:04:13.270 --> 00:04:16.470 task is a pre -mortem SWAT analysis. A pre -mortem. 00:04:16.509 --> 00:04:19.310 Exactly. You tell the deep dive into things like 00:04:19.310 --> 00:04:22.250 your pricing psychology, your feature gaps, with 00:04:22.250 --> 00:04:24.149 brutal honesty, and what customers really think. 00:04:24.509 --> 00:04:26.569 The stuff they hate but don't tell you. And the 00:04:26.569 --> 00:04:29.329 prompt ends with this, this power question. If 00:04:29.329 --> 00:04:32.490 you were their CEO, exactly how would you attack 00:04:32.490 --> 00:04:36.730 us to steal our market share? It forces the AI 00:04:36.730 --> 00:04:39.620 to find your weakest spots. to give you the kill 00:04:39.620 --> 00:04:42.000 strategy against yourself. So why is knowing 00:04:42.000 --> 00:04:44.459 your weakness before you launch so vital? It 00:04:44.459 --> 00:04:47.160 lets you fix fatal flaws before they become fatal. 00:04:47.379 --> 00:04:50.079 Okay, let's shift gears, part two. Creative content 00:04:50.079 --> 00:04:54.160 creation. The focus here moves from data to pure 00:04:54.160 --> 00:04:57.220 attention. For viral social media posts, it is 00:04:57.220 --> 00:04:59.939 all about the hook. The first line is like 80 00:04:59.939 --> 00:05:02.399 % of its success. So the role you assign is a 00:05:02.399 --> 00:05:04.680 viral social media strategist. And the formula 00:05:04.680 --> 00:05:07.350 for viral... requires that the hook is a pattern 00:05:07.350 --> 00:05:10.110 interrupt, something that stops the scroll, a 00:05:10.110 --> 00:05:12.329 controversial statement, a surprising fact, a 00:05:12.329 --> 00:05:14.449 direct question. And the body has to be skimmable, 00:05:14.670 --> 00:05:17.129 short sentences, bullet points, and a twist in 00:05:17.129 --> 00:05:19.430 the middle. Counterintuitive insight, yeah. Yeah. 00:05:19.569 --> 00:05:22.149 And the constraint you add is asking for three 00:05:22.149 --> 00:05:24.649 different hook options for A -B testing, plus 00:05:24.649 --> 00:05:27.550 a visual idea for each, a meme, a chart, whatever, 00:05:27.970 --> 00:05:30.709 saves so much time. So beyond that hook, what 00:05:30.709 --> 00:05:33.709 does asking for the engagement bait do? It forces 00:05:33.709 --> 00:05:36.350 people to comment by ending with a specific question. 00:05:36.709 --> 00:05:39.370 Prompt five tackles what I think is the biggest 00:05:39.370 --> 00:05:43.610 tell for AI writing. That robot smell. The flatness, 00:05:43.810 --> 00:05:45.829 yeah. It sounds clinical because the sentence 00:05:45.829 --> 00:05:47.990 lengths are all the same. The solution is to 00:05:47.990 --> 00:05:51.230 explicitly fix something called burstiness. Which 00:05:51.230 --> 00:05:53.149 is just the variation in sentence length. You 00:05:53.149 --> 00:05:55.889 have to tell the model to mix very short, punchy 00:05:55.889 --> 00:05:58.129 sentences with longer, more complex ones. It 00:05:58.129 --> 00:05:59.970 creates a human rhythm. And again, we're banning 00:05:59.970 --> 00:06:03.019 words. The AI -ism. Oh, this is a big one. You 00:06:03.019 --> 00:06:05.680 have to get rid of words like realm, landscape, 00:06:06.240 --> 00:06:09.819 delve, tapestry, fostering. There did giveaways 00:06:09.819 --> 00:06:12.160 of a generic answer. I have to admit, I still 00:06:12.160 --> 00:06:14.720 wrestle with prompt drift myself, especially 00:06:14.720 --> 00:06:17.160 when the AI defaults to those words like delve 00:06:17.160 --> 00:06:20.000 or tapestry. Banning them is brilliant. And you 00:06:20.000 --> 00:06:22.240 double down on show, don't tell. Don't let it 00:06:22.240 --> 00:06:25.300 say it was efficient. Force it to say it saved 00:06:25.300 --> 00:06:28.819 us four hours. So how does requesting that high 00:06:28.819 --> 00:06:31.600 burstiness help the content pass AI detection 00:06:31.600 --> 00:06:34.120 tools? It breaks the model's uniform safety rules, 00:06:34.300 --> 00:06:36.779 making the rhythm more human. And for the sixth 00:06:36.779 --> 00:06:39.699 prompt, structuring a persuasive presentation, 00:06:40.259 --> 00:06:42.579 the goal is to move beyond bullet points and 00:06:42.579 --> 00:06:45.540 create an actual story. So you assign the AI 00:06:45.540 --> 00:06:48.279 the role of a presentation coach. For TED speakers 00:06:48.279 --> 00:06:50.839 and VCs, you tell it to use the hero's journey 00:06:50.839 --> 00:06:53.740 narrative. Status quo, challenge, resolution. 00:06:54.060 --> 00:06:56.319 But the detail here is about the emotional goal 00:06:56.319 --> 00:06:58.680 for every single slide. Yeah, this is the key. 00:06:58.800 --> 00:07:00.980 You say, for slide five, I want the audience 00:07:00.980 --> 00:07:04.779 to feel fear. For slide eight, hope. For slide 00:07:04.779 --> 00:07:08.360 12, trust. Why is it so vital to define the emotional 00:07:08.360 --> 00:07:11.399 goal for every single slide? Because people buy 00:07:11.399 --> 00:07:14.360 and invest based on feeling, not just cold, hard 00:07:14.360 --> 00:07:16.680 numbers. Welcome back to The Deep Dive. We're 00:07:16.680 --> 00:07:18.759 moving into our final section now, looking at 00:07:18.759 --> 00:07:20.879 design, learning, and some critical financial 00:07:20.879 --> 00:07:23.879 analysis. Prompt 7 is all about designing beautiful 00:07:23.879 --> 00:07:27.379 and, more importantly, functional UI UX. You're 00:07:27.379 --> 00:07:29.579 not just asking for a layout. You're telling 00:07:29.579 --> 00:07:32.579 the AI to act as a senior product designer and 00:07:32.579 --> 00:07:35.319 to focus on the whole user journey and the edge 00:07:35.319 --> 00:07:38.120 cases. Right, the edge cases. The critical deliverable 00:07:38.120 --> 00:07:40.860 you ask for is the interaction stays. What happens 00:07:40.860 --> 00:07:44.079 on hover, active disabled, and especially on 00:07:44.079 --> 00:07:47.420 error? This prevents so much bad design. It forces 00:07:47.420 --> 00:07:50.420 the AI to plan for messy situations like slow 00:07:50.420 --> 00:07:52.939 internet or when a user makes a mistake. And 00:07:52.939 --> 00:07:55.319 that gives you a complete design framework, usually 00:07:55.319 --> 00:07:58.160 with the code ready to go. So why is planning 00:07:58.160 --> 00:08:00.579 for those edge cases the real difference maker? 00:08:00.879 --> 00:08:03.040 A bad design forgets what the button looks like 00:08:03.040 --> 00:08:05.740 when the user messes up. Promptate is about personalized 00:08:05.740 --> 00:08:07.759 learning. I think we all know online courses 00:08:07.759 --> 00:08:10.660 can be way too long. Packed with theory you don't 00:08:10.660 --> 00:08:13.160 need. So here we apply the Pareto principle, 00:08:13.519 --> 00:08:16.600 the 80 -20 rule. The AI becomes an expert curriculum 00:08:16.600 --> 00:08:19.540 designer. The goal is to find the critical 20 00:08:19.540 --> 00:08:22.079 % of concepts that get you 80 % of the results. 00:08:22.399 --> 00:08:24.279 So if you're learning Python for finance, it 00:08:24.279 --> 00:08:26.579 just cuts out all the abstract computer science 00:08:26.579 --> 00:08:28.980 and focuses right on the libraries you'll actually 00:08:28.980 --> 00:08:31.699 use. And the output has to include specific topics, 00:08:32.080 --> 00:08:34.419 links to free resources, and one micro project 00:08:34.419 --> 00:08:37.259 each week. and a list of common pitfalls, the 00:08:37.259 --> 00:08:40.080 beginner mistakes. How does including those pitfalls 00:08:40.080 --> 00:08:42.620 help the learner save time? It helps you avoid 00:08:42.620 --> 00:08:45.679 time -wasting routes and useless theory. OK, 00:08:45.759 --> 00:08:48.519 prompt nine, analyzing stocks and investments. 00:08:48.799 --> 00:08:50.440 And we have to start in the disclaimer here. 00:08:50.740 --> 00:08:53.580 A big one. AI is a research assistant, not a 00:08:53.580 --> 00:08:55.700 financial advisor. Always check the facts yourself. 00:08:55.940 --> 00:08:58.500 So we give it the role of a financial analyst, 00:08:58.720 --> 00:09:01.340 but one with a very conservative, risk -averse 00:09:01.340 --> 00:09:04.360 mindset. And the framework focuses on two things. 00:09:04.940 --> 00:09:07.539 The business model, is it recurring revenue or 00:09:07.539 --> 00:09:10.460 one -off sales? And the moat, what protects them 00:09:10.460 --> 00:09:13.559 from competitors. But the crucial part, the mandatory 00:09:13.559 --> 00:09:16.659 part, is the bear case. Absolutely. The AI has 00:09:16.659 --> 00:09:18.940 to list five specific reasons the investment 00:09:18.940 --> 00:09:22.379 could fail. It forces critical, defensive thinking. 00:09:23.019 --> 00:09:25.259 And you make it compare valuation metrics against 00:09:25.259 --> 00:09:28.059 the top three competitors. So why is forcing 00:09:28.059 --> 00:09:31.600 the AI to act like a critic, the bear case, so 00:09:31.600 --> 00:09:33.940 good for decision making? It combats our bias 00:09:33.940 --> 00:09:36.340 to only see the good side of a stock we like. 00:09:36.759 --> 00:09:40.159 And finally, prompt 10. Managing taxes and deductions. 00:09:40.440 --> 00:09:44.000 This has huge immediate value. Especially for 00:09:44.000 --> 00:09:46.779 freelancers or small business owners, yeah. The 00:09:46.779 --> 00:09:49.799 AI becomes a tax strategist for your specific 00:09:49.799 --> 00:09:52.139 country and state. The high value output is called 00:09:52.139 --> 00:09:54.559 the deduction hunter. Which lists every possible 00:09:54.559 --> 00:09:56.539 write -off you might be eligible for. Home office, 00:09:57.179 --> 00:09:59.320 electricity, software subscriptions, everything. 00:09:59.519 --> 00:10:02.100 It also gives you a list of red flags, what might 00:10:02.100 --> 00:10:04.620 trigger an audit in your industry, and a checklist 00:10:04.620 --> 00:10:06.940 of all the documents you'll need. So which specific 00:10:06.940 --> 00:10:09.080 output from this prompt gives the most immediate 00:10:09.080 --> 00:10:11.320 value for a freelancer? The deduction hunter. 00:10:11.639 --> 00:10:13.779 It's a list of write -offs to discuss with their 00:10:13.779 --> 00:10:16.200 accountant. So if we zoom out here and look at 00:10:16.200 --> 00:10:18.639 the big idea connecting all 10 of these, what 00:10:18.639 --> 00:10:22.379 is it? It's that GPT 5 .2 is, well, it's pretty 00:10:22.379 --> 00:10:24.759 useless if you don't control it. The power is 00:10:24.759 --> 00:10:27.340 in the frameworks. It's not about just copy pasting. 00:10:27.360 --> 00:10:29.419 It's about understanding the thinking process 00:10:29.419 --> 00:10:32.740 you're creating. Exactly. The difference is moving 00:10:32.740 --> 00:10:36.019 from a simple question and answer to providing 00:10:36.019 --> 00:10:39.360 context, the role, and forcing self -correction 00:10:39.360 --> 00:10:42.279 with things like chain of thought or the bear 00:10:42.279 --> 00:10:45.340 case. The path to becoming an expert user isn't 00:10:45.340 --> 00:10:48.059 about some secret command. It's learning how 00:10:48.059 --> 00:10:50.820 to tell the AI how to think, not just what to 00:10:50.820 --> 00:10:53.220 say. The source material we looked at recommends 00:10:53.220 --> 00:10:55.179 choosing just one big problem you have right 00:10:55.179 --> 00:10:57.379 now and just trying one of these mega prompts 00:10:57.379 --> 00:11:00.279 on it. And here's a final thought to Chuan. We've 00:11:00.279 --> 00:11:03.059 seen that every really effective prompt forces 00:11:03.059 --> 00:11:05.559 the AI to find the opposite view, the contrarian, 00:11:05.600 --> 00:11:09.039 the bear case, the kill strategy. What if we 00:11:09.039 --> 00:11:11.220 applied that same chain of thought framework 00:11:11.220 --> 00:11:14.360 not to an AI, but to our own most stubborn personal 00:11:14.360 --> 00:11:17.440 biases? Could forcing ourselves to simulate the 00:11:17.440 --> 00:11:19.659 skeptical regulator or the bear case against 00:11:19.659 --> 00:11:22.379 our own deeply held beliefs, could that be the 00:11:22.379 --> 00:11:24.940 ultimate prompt for personal growth? Some to 00:11:24.940 --> 00:11:26.820 chew on. Go try one of these frameworks, see 00:11:26.820 --> 00:11:28.299 what a difference it makes. And thank you for 00:11:28.299 --> 00:11:29.299 joining us for this deep dive.