WEBVTT 00:00:00.000 --> 00:00:03.399 So you've got access to chat GPT -5. Powerful 00:00:03.399 --> 00:00:06.549 stuff. But maybe the results still feel a bit 00:00:06.549 --> 00:00:08.789 flat sometimes. Yeah, it happens to everyone. 00:00:08.970 --> 00:00:11.730 You've got this amazing tool, but the output 00:00:11.730 --> 00:00:14.890 can sound generic, kind of uninspired. Right. 00:00:15.169 --> 00:00:17.850 Beat. But don't panic. The problem usually isn't 00:00:17.850 --> 00:00:20.329 the AI itself. It's more about how we're asking 00:00:20.329 --> 00:00:22.530 the questions, how they're communicating. Exactly. 00:00:22.589 --> 00:00:24.969 It's like we're giving a jet engine instructions 00:00:24.969 --> 00:00:27.570 meant for, I don't know, a lawnmower. We're not 00:00:27.570 --> 00:00:30.429 tapping into its full capacity. It's that classic 00:00:30.429 --> 00:00:33.450 capacity paradox. advanced tool basic instructions. 00:00:34.170 --> 00:00:36.350 So welcome to the deep dive. Today we're going 00:00:36.350 --> 00:00:38.649 straight past the basics. We've dug into the 00:00:38.649 --> 00:00:42.450 research and pulled out 12 really proven techniques 00:00:42.450 --> 00:00:46.210 to shift your chat GPT -5 results from okay to... 00:00:46.219 --> 00:00:48.759 well, masterpiece. This is going to be your shortcut 00:00:48.759 --> 00:00:50.960 to getting sophisticated results. We've structured 00:00:50.960 --> 00:00:53.000 it around three main areas. What are they? OK, 00:00:53.000 --> 00:00:55.899 first up, the core prompting mindset, like the 00:00:55.899 --> 00:00:58.359 actual words you use to make the AI think harder. 00:00:58.560 --> 00:01:00.340 Then second, we'll get into context and using 00:01:00.340 --> 00:01:03.600 different inputs, files, pictures, the new branches 00:01:03.600 --> 00:01:05.739 feature, building that deep memory. Yeah, the 00:01:05.739 --> 00:01:08.079 multimodal stuff. And finally, we hit what we're 00:01:08.079 --> 00:01:11.260 calling expert mode, looking at automation with 00:01:11.260 --> 00:01:14.579 agent mode, custom personalities, and crucially, 00:01:15.400 --> 00:01:17.980 managing its memory. By the end of this, you 00:01:17.980 --> 00:01:19.819 should feel genuinely well informed on how to 00:01:19.819 --> 00:01:21.799 use this thing at a much higher level. Let's 00:01:21.799 --> 00:01:24.280 dive in. All right, segment one, foundational 00:01:24.280 --> 00:01:27.920 prompt engineering. Let's unpack that core prompting 00:01:27.920 --> 00:01:30.620 mindset first. The research we looked at really 00:01:30.620 --> 00:01:33.700 highlights this idea that your prompt sends a 00:01:33.700 --> 00:01:37.060 signal. It tells the AI how hard it actually 00:01:37.060 --> 00:01:39.590 needs to work on your request. So it's like the 00:01:39.590 --> 00:01:41.790 AI internally assesses the prompt and decides, 00:01:41.849 --> 00:01:44.409 OK, how much energy should I put into this? Pretty 00:01:44.409 --> 00:01:46.969 much. If the prompt seems simple, it defaults 00:01:46.969 --> 00:01:49.250 to like a conservation mode. It doesn't go all 00:01:49.250 --> 00:01:51.290 out. You need to tell it to override that. Ah, 00:01:51.349 --> 00:01:53.750 so the default is kind of lazy mode unless we 00:01:53.750 --> 00:01:56.209 specifically push it. Exactly. And that brings 00:01:56.209 --> 00:01:59.370 us to tip one, the magic phrases. These are specific 00:01:59.370 --> 00:02:01.450 words or sentences you add to your prompt. Things 00:02:01.450 --> 00:02:04.519 like? Things like, think step by step or. be 00:02:04.519 --> 00:02:07.280 extremely thorough, or maybe take your time. 00:02:07.340 --> 00:02:10.259 And even this is critical to get right. So these 00:02:10.259 --> 00:02:12.379 phrases are basically like telling a colleague, 00:02:12.620 --> 00:02:15.199 hey, this one's important. Don't rush it. Put 00:02:15.199 --> 00:02:17.960 some real thought into it. Precisely. It signals 00:02:17.960 --> 00:02:20.659 maximum effort required. And the examples show 00:02:20.659 --> 00:02:23.199 a clear difference, right? Like drafting a simple 00:02:23.199 --> 00:02:25.500 social media post. Yeah, the basic prompt gets 00:02:25.500 --> 00:02:28.639 you basic text. But you add those phrases, specify 00:02:28.639 --> 00:02:31.560 structure, ask for details, maybe even tell it 00:02:31.560 --> 00:02:34.560 to prompt interaction like, tag a friend. And 00:02:34.560 --> 00:02:37.219 suddenly the quality jumps way up. It's much 00:02:37.219 --> 00:02:39.780 more usable immediately. Okay, so we can tell 00:02:39.780 --> 00:02:42.960 it to think harder, but how do we make sure it 00:02:42.960 --> 00:02:45.379 actually meets our specific quality standards, 00:02:45.740 --> 00:02:48.539 especially for complex tasks? That leads right 00:02:48.539 --> 00:02:51.530 into tip two. The self -critique method. This 00:02:51.530 --> 00:02:54.710 is where we make the AI grade its own work before 00:02:54.710 --> 00:02:57.030 it even shows us anything. Ooh, I like that. 00:02:57.050 --> 00:02:59.610 How does it work? You give the AI a quality checklist 00:02:59.610 --> 00:03:02.009 before it starts writing. You define what good 00:03:02.009 --> 00:03:04.310 looks like up front. OK, so say you're writing 00:03:04.310 --> 00:03:07.009 a really important job application email. Right. 00:03:07.319 --> 00:03:10.460 Your checklist might demand tone must be professional 00:03:10.460 --> 00:03:14.020 but also excited, needs perfect clarity, must 00:03:14.020 --> 00:03:16.819 mention specific company details personalization, 00:03:17.439 --> 00:03:19.520 and needs a clear call to action like asking 00:03:19.520 --> 00:03:21.960 for an interview. Got it. And then you tell the 00:03:21.960 --> 00:03:25.580 AI. You tell it to run up to five internal improvement 00:03:25.580 --> 00:03:28.460 cycles using that checklist. It basically edits 00:03:28.460 --> 00:03:31.500 itself, refining the draft against your criteria 00:03:31.500 --> 00:03:33.599 without you seeing the messy in -between steps. 00:03:33.740 --> 00:03:36.020 It acts as its own secret editor. That's clever. 00:03:36.199 --> 00:03:38.340 It is. You know, I still wrestle with prompt 00:03:38.340 --> 00:03:41.120 drift myself sometimes. My intention gets kind 00:03:41.120 --> 00:03:43.240 of fuzzy after a few back and forths. Oh, totally. 00:03:43.500 --> 00:03:46.020 Happens to the best of us. But this self -critique 00:03:46.020 --> 00:03:49.479 idea feels like... Genius, almost, because it 00:03:49.479 --> 00:03:51.639 locks in your definition of success right from 00:03:51.639 --> 00:03:54.479 the very start. It forces the AI to stick to 00:03:54.479 --> 00:03:56.639 the plan. Which connects perfectly to tip three, 00:03:56.659 --> 00:03:58.500 something that trips people up all the time. 00:03:58.759 --> 00:04:00.659 Never give mixed messages. We do it accidentally, 00:04:01.199 --> 00:04:04.500 constantly. The classic contradiction, be brief, 00:04:04.900 --> 00:04:07.240 but provide detailed explanations. Right, the 00:04:07.240 --> 00:04:09.340 AI gets stuck. It's like telling someone to walk 00:04:09.340 --> 00:04:12.060 straight and turn left at the same time. It just 00:04:12.060 --> 00:04:14.699 fuzzes out, can't do both well. So the fix is 00:04:14.699 --> 00:04:18.089 clarity, prioritization. Exactly. separate the 00:04:18.089 --> 00:04:21.350 requests, like first provide brief summaries. 00:04:21.709 --> 00:04:24.430 Then maybe I'll ask for detailed explanations 00:04:24.430 --> 00:04:26.870 on specific points if I need them. That way it 00:04:26.870 --> 00:04:29.850 can focus all its processing power on one clear 00:04:29.850 --> 00:04:33.310 task at a time. Makes sense. So tying this together, 00:04:34.490 --> 00:04:36.649 since getting that initial quality right seems 00:04:36.649 --> 00:04:39.730 so dependent on having clear intent, how much 00:04:39.730 --> 00:04:41.730 time should people really invest in refining 00:04:41.730 --> 00:04:44.620 that very first instruction? Well, based on this, 00:04:44.800 --> 00:04:46.839 defining success clearly upfront using these 00:04:46.839 --> 00:04:49.240 methods. It'd probably save you ten times that 00:04:49.240 --> 00:04:51.860 amount later, not having to fix vague or confused 00:04:51.860 --> 00:04:54.339 outputs. Good point. Clarity upfront prevents 00:04:54.339 --> 00:04:56.399 messy cleanup. Okay, moving on from the words 00:04:56.399 --> 00:04:59.439 themselves. Segment two gets into building context 00:04:59.439 --> 00:05:02.000 and using different kinds of input. Tip four 00:05:02.000 --> 00:05:05.420 highlights this big shift toward true multimodal 00:05:05.420 --> 00:05:07.480 input. Yeah, this is huge. We're not just limited 00:05:07.480 --> 00:05:10.079 to text anymore. We can use pictures, upload 00:05:10.079 --> 00:05:12.899 files, even use our voice to give the AI context. 00:05:13.160 --> 00:05:15.339 It really changes things, especially for stuff 00:05:15.339 --> 00:05:17.699 that has structure or visual elements. Definitely. 00:05:17.980 --> 00:05:20.139 Take the CV analysis example from the source 00:05:20.139 --> 00:05:23.250 material. You upload your CV, maybe as a PDF, 00:05:23.490 --> 00:05:25.850 maybe even an image. OK. And then you prompt 00:05:25.850 --> 00:05:28.629 ChatGPT, telling it something like, act as an 00:05:28.629 --> 00:05:31.129 expert HR manager with 15 years of experience 00:05:31.129 --> 00:05:33.529 and critique the CV. And because it can actually 00:05:33.529 --> 00:05:36.129 see the layout, the design, the white space, 00:05:36.569 --> 00:05:39.550 it's not just reading the words. Exactly. It 00:05:39.550 --> 00:05:41.589 critiques the visual structure alongside the 00:05:41.589 --> 00:05:44.269 wording. It gives you actionable feedback on 00:05:44.269 --> 00:05:46.029 things you'd normally need a human consultant 00:05:46.029 --> 00:05:48.420 for. That's pretty powerful, like getting expert 00:05:48.420 --> 00:05:50.939 design feedback instantly. And a pro tip they 00:05:50.939 --> 00:05:53.420 mentioned. Upload your version, maybe the bad 00:05:53.420 --> 00:05:56.600 one, and also a good example you like. Then ask 00:05:56.600 --> 00:05:59.259 it for visual comparison advice. Oh, smart. Side 00:05:59.259 --> 00:06:01.720 -by -side critique. Also, quick note on voice. 00:06:02.279 --> 00:06:04.540 The little microphone icon. That's usually just 00:06:04.540 --> 00:06:07.500 for transcription. But the sound wave icon. That's 00:06:07.500 --> 00:06:09.399 the more advanced back and forth conversational 00:06:09.399 --> 00:06:11.540 mode, like a real assistant. Good distinction. 00:06:11.959 --> 00:06:15.540 Okay, tip five takes us from analysis to creation. 00:06:16.160 --> 00:06:19.000 Building simple apps using something called Canvas 00:06:19.000 --> 00:06:21.959 mode. Yeah, this is cool. It's basically designing 00:06:21.959 --> 00:06:24.579 useful little custom tools without needing to 00:06:24.579 --> 00:06:27.500 write any code. Like the example they gave, a 00:06:27.500 --> 00:06:29.240 weekly meal planner and shopping list generator. 00:06:29.279 --> 00:06:31.709 Right. You tell it what you want, specify the 00:06:31.709 --> 00:06:33.750 design, use light colors, make two sections, 00:06:34.110 --> 00:06:36.430 and the functionality, I need seven boxes for 00:06:36.430 --> 00:06:40.209 meals, a shopping list I can check off, and crucially, 00:06:40.670 --> 00:06:43.810 a button labeled generate my week that automatically 00:06:43.810 --> 00:06:45.810 fills in the ingredients based on the meals. 00:06:46.050 --> 00:06:48.569 And the AI just builds it, handles the layout, 00:06:48.709 --> 00:06:51.009 the buttons, the logic. In minutes, yeah. You 00:06:51.009 --> 00:06:53.389 get a working visual tool that solves a real 00:06:53.389 --> 00:06:55.709 problem, no coding required. Pretty amazing. 00:06:55.889 --> 00:06:59.740 Wow. Okay, next up. Tip six, using branches. 00:07:00.399 --> 00:07:02.779 What problem does this solve? This tackles that 00:07:02.779 --> 00:07:04.699 really common issue where you're exploring one 00:07:04.699 --> 00:07:07.000 idea in a chat, but then you think of an alternative 00:07:07.000 --> 00:07:09.420 you want to explore, but you don't want to derail 00:07:09.420 --> 00:07:11.959 the main conversation or lose the context. Yeah, 00:07:12.379 --> 00:07:14.620 conversation drift. Exactly. Branches let you 00:07:14.620 --> 00:07:17.660 create separate parallel conversations that all 00:07:17.660 --> 00:07:19.959 stem from the same starting point, the same original 00:07:19.959 --> 00:07:22.699 query. So, the trip planning example. You start 00:07:22.699 --> 00:07:25.660 by asking about two places, say, Daulat and Fuquak. 00:07:25.839 --> 00:07:28.600 Mm -hmm. Then you can create a branch for Daulat 00:07:28.600 --> 00:07:30.779 to plan the budget and activities just for there, 00:07:31.160 --> 00:07:33.480 and another branch for Fuquak to do the same. 00:07:33.720 --> 00:07:36.339 But both branches remember the original context, 00:07:36.439 --> 00:07:38.660 like the total trip length or overall budget. 00:07:39.139 --> 00:07:42.819 Precisely. The core info remains, but the detailed 00:07:42.819 --> 00:07:45.220 exploration happens in separate threads without 00:07:45.220 --> 00:07:48.100 confusing each other. Great for comparing options 00:07:48.100 --> 00:07:51.040 or exploring what -ifs. Very useful. And that 00:07:51.040 --> 00:07:54.100 brings us to tip 7 for this segment. Use projects. 00:07:54.600 --> 00:07:56.860 What are these? Think of these as more than just 00:07:56.860 --> 00:07:59.949 folders. They're sophisticated, continuous memory 00:07:59.949 --> 00:08:02.490 workspaces. OK, so you'd set up a project, maybe 00:08:02.490 --> 00:08:05.689 call it Q4 marketing project. Yep. And then you 00:08:05.689 --> 00:08:08.389 upload relevant files directly into that project. 00:08:08.490 --> 00:08:11.149 Brand style guides, past campaign reports, competitor 00:08:11.149 --> 00:08:13.430 analysis. And custom instructions, too. Like, 00:08:13.430 --> 00:08:15.529 always write in our friendly but professional 00:08:15.529 --> 00:08:18.350 brand voice. Exactly. And here's the key value. 00:08:19.029 --> 00:08:22.089 Continuous memory. Anything you put in that project. 00:08:22.910 --> 00:08:26.050 Project files, instructions, even past conversations 00:08:26.050 --> 00:08:29.399 within it. chat GPT remembers it for every new 00:08:29.399 --> 00:08:32.179 chat you start inside that same project. So it's 00:08:32.179 --> 00:08:34.860 always building on previous contexts within that 00:08:34.860 --> 00:08:37.139 specific project. No more re -explaining the 00:08:37.139 --> 00:08:39.620 brand voice every single time. That's the power. 00:08:40.039 --> 00:08:42.529 Super efficient. Though, note... there might 00:08:42.529 --> 00:08:44.470 be limits on file uploads, depending on your 00:08:44.470 --> 00:08:46.889 plan, like maybe five files for free users. Still 00:08:46.889 --> 00:08:49.210 very useful. OK, lots of tools there. So for 00:08:49.210 --> 00:08:51.409 someone just starting to explore these more advanced 00:08:51.409 --> 00:08:54.190 features, which one of these files, canvas, branches, 00:08:54.529 --> 00:08:57.110 projects, offers the quickest, most immediate 00:08:57.110 --> 00:09:00.190 payoff, like fastest return on investment? Good 00:09:00.190 --> 00:09:03.129 question. I'd probably argue using files, especially 00:09:03.129 --> 00:09:05.750 that CV critique example. Uploading a document 00:09:05.750 --> 00:09:07.830 you already have and getting instant expert level 00:09:07.830 --> 00:09:09.929 feedback on its structure and design. That's 00:09:09.929 --> 00:09:11.730 immediate value you previously had to pay quite 00:09:11.730 --> 00:09:14.490 a bit for. Yeah, that makes sense. Instant expert 00:09:14.490 --> 00:09:18.070 feedback is hard to beat. All right, let's move 00:09:18.070 --> 00:09:21.549 into segment three. Expert mode. This is about 00:09:21.549 --> 00:09:24.789 really leveraging the system for learning, honing 00:09:24.789 --> 00:09:28.389 voice, and getting ultimate control. Tip 8 introduces 00:09:28.389 --> 00:09:30.610 study mode. Yeah, this isn't just about getting 00:09:30.610 --> 00:09:32.509 information from the AI. It's about learning 00:09:32.509 --> 00:09:35.029 through a conversation that actively challenges 00:09:35.029 --> 00:09:37.129 your thinking. How does that work in practice? 00:09:37.570 --> 00:09:40.409 The example prompt is great. Create an AI board 00:09:40.409 --> 00:09:42.970 of experts to help evaluate my business idea, 00:09:43.190 --> 00:09:45.840 and then you describe your idea. So the example 00:09:45.840 --> 00:09:48.659 was healthy meal kits for busy professionals 00:09:48.659 --> 00:09:52.100 in Ho Chi Minh City HCMC. Right. And the AI doesn't 00:09:52.100 --> 00:09:54.899 just give you facts. It responds by adopting 00:09:54.899 --> 00:09:57.240 specific roles. You might say, OK, I'm the operations 00:09:57.240 --> 00:09:59.820 expert. Question, how will you ensure delivery 00:09:59.820 --> 00:10:03.450 freshness during peak HCMC traffic? Ah, so it's 00:10:03.450 --> 00:10:05.289 asking challenging questions from different perspectives. 00:10:05.649 --> 00:10:07.490 Exactly. Then maybe it switches. Now I'm the 00:10:07.490 --> 00:10:09.889 finance expert. Your proposed price point seems 00:10:09.889 --> 00:10:12.769 low. How will you ensure profitability considering 00:10:12.769 --> 00:10:15.330 ingredient cost fluctuations? So you're forced 00:10:15.330 --> 00:10:18.350 to actively problem solve and defend your idea 00:10:18.350 --> 00:10:21.149 against expert level scrutiny rather than just 00:10:21.149 --> 00:10:22.929 passively reading info. That's it. It really 00:10:22.929 --> 00:10:25.289 pushes your assumptions. Invaluable for stress 00:10:25.289 --> 00:10:29.230 testing any idea. OK, next. Combating that robotic 00:10:29.230 --> 00:10:34.200 AI tone. Tip nine. Write like a real person. 00:10:34.480 --> 00:10:37.019 This is crucial if you want the output to sound 00:10:37.019 --> 00:10:39.720 genuinely human. It's a two -step process they 00:10:39.720 --> 00:10:42.580 recommend. Step one. Provide two or three samples 00:10:42.580 --> 00:10:45.159 of your own writing. Stuff you think sounds like 00:10:45.159 --> 00:10:47.240 you. Okay, so it learns your style. And step 00:10:47.240 --> 00:10:49.720 two. You give it a framework prompt with some 00:10:49.720 --> 00:10:52.960 unbreakable rules. Things like keep most sentences 00:10:52.960 --> 00:10:56.600 under 15 words. Use simple vocabulary. Strictly 00:10:56.600 --> 00:10:59.220 avoid buzzwords. You know, synergy, optimized 00:10:59.220 --> 00:11:01.919 leverage. Oh, yes. Please avoid those. And connect 00:11:01.919 --> 00:11:03.960 ideas naturally using simple conjunctions like 00:11:03.960 --> 00:11:07.320 and, but, so. So it's a combination. Model your 00:11:07.320 --> 00:11:09.940 specific voice, but also enforce general rules 00:11:09.940 --> 00:11:12.600 of clear, simple human writing. That's the magic 00:11:12.600 --> 00:11:15.240 combo. Makes the output feel much more conversational 00:11:15.240 --> 00:11:17.500 and less like, well, like a machine wrote it. 00:11:17.879 --> 00:11:20.740 Now, for the really advanced stuff. Tip 10, Agent 00:11:20.740 --> 00:11:24.259 Mode. This sounds serious. It is. It requires 00:11:24.259 --> 00:11:27.190 a paid plan. But Agent Mode... essentially puts 00:11:27.190 --> 00:11:31.110 Chet GPT on autopilot for complex multi -step 00:11:31.110 --> 00:11:34.730 tasks. Autopilot? Like what? Think research projects, 00:11:35.250 --> 00:11:37.830 browsing the web for information, even designing 00:11:37.830 --> 00:11:40.350 visuals. It can string together multiple actions 00:11:40.350 --> 00:11:43.509 to achieve a goal you set. Whoa! Okay, hang on. 00:11:43.629 --> 00:11:46.320 Imagine scaling that. Having an agent plan an 00:11:46.320 --> 00:11:48.580 entire trip, research a whole market report, 00:11:49.000 --> 00:11:51.399 and create the presentation slides while you 00:11:51.399 --> 00:11:53.740 just sort of watch. That's the potential. It 00:11:53.740 --> 00:11:56.059 handles the research, the browsing, the integration 00:11:56.059 --> 00:11:58.759 between different tools, like pulling info and 00:11:58.759 --> 00:12:00.539 then feeding it to an image generator. Okay, 00:12:00.600 --> 00:12:02.720 give me the specific example they used. Planning 00:12:02.720 --> 00:12:05.700 a company team building event. The task was... 00:12:05.519 --> 00:12:09.059 Plan a two -day event for 30 employees near HCMC. 00:12:09.220 --> 00:12:12.700 Budget is 1 .5 million VND per person, about 00:12:12.700 --> 00:12:15.259 60 US dollars. That's a complex task. Lots of 00:12:15.259 --> 00:12:17.360 variables. And Agent Mode apparently handled 00:12:17.360 --> 00:12:19.740 it. It autonomously researched three suitable 00:12:19.740 --> 00:12:22.720 resorts nearby, compared them, created a detailed 00:12:22.720 --> 00:12:24.379 minute -by -minute schedule for the two -day. 00:12:24.440 --> 00:12:27.580 Okay, impressive. And it designed a fun announcement 00:12:27.580 --> 00:12:29.940 poster for the event using the built -in image 00:12:29.940 --> 00:12:33.039 generator. It delivered the whole package. a 00:12:33.039 --> 00:12:35.120 task that would normally take a person, what, 00:12:35.840 --> 00:12:38.279 days? Then automatically, that's... Yeah. Yeah, 00:12:38.419 --> 00:12:39.940 that's a glimpse of the future right there. It 00:12:39.940 --> 00:12:42.840 really is. Now, finishing up with tips 11 and 00:12:42.840 --> 00:12:45.500 12, these are quicker hits on maximizing control 00:12:45.500 --> 00:12:47.820 and making sure the AI works smoothly for you 00:12:47.820 --> 00:12:51.000 over time. Tip 11 covers extra controls. Right. 00:12:51.230 --> 00:12:53.750 Little things, but important. Like the edit message 00:12:53.750 --> 00:12:55.850 feature lets you tweak your prompt after you 00:12:55.850 --> 00:12:57.370 send it without losing the whole conversation 00:12:57.370 --> 00:13:00.769 context. Super useful for fixing typos or refining 00:13:00.769 --> 00:13:02.929 your ask. Oh, that's handy. No more starting 00:13:02.929 --> 00:13:05.889 over for one mistake. And connectors, these are 00:13:05.889 --> 00:13:08.789 integrations, usually paid, linking chat GPT 00:13:08.789 --> 00:13:11.230 to other tools like your Google Drive or Calendar. 00:13:11.730 --> 00:13:14.549 So you could ask it to, say, summarize my unread 00:13:14.549 --> 00:13:17.090 emails from the last 24 hours. Integration is 00:13:17.090 --> 00:13:19.809 key. What else? Personalities. You can choose 00:13:19.809 --> 00:13:22.149 different default interaction styles, like robot 00:13:22.149 --> 00:13:26.070 for purely factual, lookner for supportive, cynic 00:13:26.070 --> 00:13:29.350 for critical feedback, or nerd for deep dives. 00:13:29.590 --> 00:13:31.330 So you can kind of set the default mood for the 00:13:31.330 --> 00:13:34.029 conversation. Yeah. And also, the ability to 00:13:34.029 --> 00:13:37.429 access older models, like GPT -4, usually tucked 00:13:37.429 --> 00:13:39.169 away in the settings if you need them for some 00:13:39.169 --> 00:13:42.470 reason. OK. And finally, tip 12. This sounds 00:13:42.470 --> 00:13:46.620 important. Manage memory. hugely important, and 00:13:46.620 --> 00:13:49.299 probably the most underrated control. ChatGPT 00:13:49.299 --> 00:13:51.120 remembers things about you and your preferences 00:13:51.120 --> 00:13:54.379 across different chats to personalize responses 00:13:54.379 --> 00:13:57.059 that's stored in its memory. Which sounds good, 00:13:57.100 --> 00:14:00.379 but... But that memory can get outdated. If you 00:14:00.379 --> 00:14:02.399 told it six months ago you were working on Project 00:14:02.399 --> 00:14:05.200 X, but now you're on Project Y, it might still 00:14:05.200 --> 00:14:08.059 be tailoring answers based on old, irrelevant 00:14:08.059 --> 00:14:11.500 context from Project X. Ah, so its memory needs 00:14:11.500 --> 00:14:13.700 tidying up. Exactly. You need to regularly go 00:14:13.700 --> 00:14:16.559 into settings, personalization and review, edit, 00:14:17.000 --> 00:14:18.840 or clear out old memories that are no longer 00:14:18.840 --> 00:14:21.080 relevant to your current goals. So keeping the 00:14:21.080 --> 00:14:23.399 memory clean ensures the answers stay accurate 00:14:23.399 --> 00:14:25.679 and focused on what you need now, not what you 00:14:25.679 --> 00:14:28.340 needed last year. Precisely. It prevents context 00:14:28.340 --> 00:14:30.759 drift and keeps the AI aligned with your current 00:14:30.759 --> 00:14:33.200 reality. It's essential maintenance for long 00:14:33.200 --> 00:14:35.720 -term reliable use. OK, we've covered a ton of 00:14:35.720 --> 00:14:38.059 features there, from basic mindset shifts to 00:14:38.059 --> 00:14:41.379 complex automation and control. Based on everything 00:14:41.379 --> 00:14:44.299 we've synthesized, what's the single most underrated 00:14:44.299 --> 00:14:47.059 control people skip that really costs them accuracy 00:14:47.059 --> 00:14:49.539 over time? I think it has to be that memory management. 00:14:49.899 --> 00:14:52.860 It's not flashy, but forgetting to prune that 00:14:52.860 --> 00:14:55.980 old context means the AI is working with bad 00:14:55.980 --> 00:14:59.289 data about you. Cleaning the memory ensures consistency 00:14:59.289 --> 00:15:02.190 and accuracy. It's the foundation for reliable 00:15:02.190 --> 00:15:04.269 long -term results. Yeah, that makes perfect 00:15:04.269 --> 00:15:07.470 sense. Ongoing maintenance is key. So, wrapping 00:15:07.470 --> 00:15:09.970 this up, the big idea, the core takeaway from 00:15:09.970 --> 00:15:12.110 this deep dive, really seems to be that the difference 00:15:12.110 --> 00:15:15.549 between getting amazing AI results and just OK 00:15:15.549 --> 00:15:18.669 results. It isn't really about the AI's raw capability 00:15:18.669 --> 00:15:20.730 anymore. No, it's about us. It's about how well 00:15:20.730 --> 00:15:23.409 we communicate with it, how clearly we give instructions, 00:15:23.490 --> 00:15:25.850 how we provide context, how we guide its thinking. 00:15:26.090 --> 00:15:28.409 These 12 techniques we covered aren't just clever 00:15:28.409 --> 00:15:30.330 tricks. They feel more like fundamental shifts 00:15:30.330 --> 00:15:32.190 in how we need to interact with these powerful 00:15:32.190 --> 00:15:34.990 systems. Better communication leads to better 00:15:34.990 --> 00:15:37.909 outcomes. Absolutely. So, if you're listening 00:15:37.909 --> 00:15:39.710 and want to try just a few things this week, 00:15:40.210 --> 00:15:42.690 what would you prioritize? Based on our discussion, 00:15:42.690 --> 00:15:46.370 I'd say. One, try using those magic phrases like, 00:15:46.610 --> 00:15:48.950 think step -by -step on your next complex prompt. 00:15:49.049 --> 00:15:51.110 See if you notice a difference. Good one. Two. 00:15:51.429 --> 00:15:54.620 Two, implement that self -critique method. Create 00:15:54.620 --> 00:15:56.879 a quality checklist for an important email or 00:15:56.879 --> 00:16:00.360 document before you ask the AI to write it. Define 00:16:00.360 --> 00:16:02.980 success upfront. And three. Set up your first 00:16:02.980 --> 00:16:05.899 project. Upload a key document, like a style 00:16:05.899 --> 00:16:08.100 guide or project debrief, and experience that 00:16:08.100 --> 00:16:11.059 continuous memory benefit. Solid starting points. 00:16:11.700 --> 00:16:13.759 Now, for a final thought to lead people with. 00:16:14.610 --> 00:16:16.450 Considering that agent mode's power, we talked 00:16:16.450 --> 00:16:19.009 about autonomously researching, planning, designing 00:16:19.009 --> 00:16:21.809 complex things. Yeah. And the source material 00:16:21.809 --> 00:16:23.929 even hinted at people using these techniques 00:16:23.929 --> 00:16:26.169 to build multiple businesses rapidly. Yeah. It 00:16:26.169 --> 00:16:28.269 makes you wonder, doesn't it? If you truly master 00:16:28.269 --> 00:16:30.529 this kind of AI communication, what parts of 00:16:30.529 --> 00:16:32.450 your own work, your own projects, maybe even 00:16:32.450 --> 00:16:34.830 parts of running a small business, could you 00:16:34.830 --> 00:16:37.529 realistically automate this year? That's a provocative 00:16:37.529 --> 00:16:41.840 question. Moving beyond just assistance to actual 00:16:41.840 --> 00:16:45.299 autonomous execution of complex tasks, what becomes 00:16:45.299 --> 00:16:48.240 possible then? Something to think about. We really 00:16:48.240 --> 00:16:50.100 hope this deep dive into the source material 00:16:50.100 --> 00:16:52.259 saved you some serious learning time and gave 00:16:52.259 --> 00:16:54.600 you practical tools. If you found these insights 00:16:54.600 --> 00:16:56.879 useful, maybe share this with a friend or colleague 00:16:56.879 --> 00:16:59.259 who's also trying to level up their AI game. 00:16:59.460 --> 00:17:00.940 Thanks for joining us for this deep dive.