WEBVTT 00:00:00.000 --> 00:00:03.060 Ever feel like your AI assistant just isn't quite 00:00:03.060 --> 00:00:06.459 getting it? You type a prompt and the answer 00:00:06.459 --> 00:00:08.800 feels, well, generic, maybe even a little flat. 00:00:09.980 --> 00:00:11.480 What if I told you you're probably missing out 00:00:11.480 --> 00:00:15.599 on like 90 % of its true power? Welcome to the 00:00:15.599 --> 00:00:18.750 deep dive. Today, we're diving into a fascinating 00:00:18.750 --> 00:00:21.530 look at mastering AI, prompting techniques for 00:00:21.530 --> 00:00:26.030 powerful responses, our mission to really transform 00:00:26.030 --> 00:00:28.370 how you interact with large language models, 00:00:28.769 --> 00:00:31.250 LLMs. We want to help you turn them from simple 00:00:31.250 --> 00:00:34.049 search tools into truly powerful, insightful 00:00:34.049 --> 00:00:36.350 assistants. We'll explore key mindset shifts 00:00:36.350 --> 00:00:38.270 and really practical techniques, moving from 00:00:38.270 --> 00:00:40.729 basic prompts all the way to advanced creative 00:00:40.729 --> 00:00:42.750 applications. It's a deep dive into communication, 00:00:42.789 --> 00:00:44.590 really. Right, and we're going to try and demystify 00:00:44.590 --> 00:00:46.799 what these AIs really are. are, because understanding 00:00:46.799 --> 00:00:49.140 that is absolutely key to unlocking your capabilities. 00:00:49.460 --> 00:00:51.240 It's less about knowing facts, funnily enough, 00:00:51.460 --> 00:00:53.600 and much more about recognizing patterns. That's 00:00:53.600 --> 00:00:55.320 a surprisingly profound distinction, I think. 00:00:55.719 --> 00:00:58.340 OK, let's unpack this, then. I think the biggest 00:00:58.340 --> 00:01:01.000 mistake, maybe, that many of us make when we 00:01:01.000 --> 00:01:03.520 first encounter an LLM is treating it like some 00:01:03.520 --> 00:01:06.920 kind of all -knowing super Google. You know we 00:01:06.920 --> 00:01:09.680 type in a query expect a factual answer just 00:01:09.680 --> 00:01:12.739 like a quick search But if you stop there, you're 00:01:12.739 --> 00:01:15.480 genuinely missing the forest for the trees. Absolutely 00:01:15.480 --> 00:01:18.700 Yeah, because in reality these models don't know 00:01:18.969 --> 00:01:21.189 anything, not in the human sense of comprehension 00:01:21.189 --> 00:01:25.769 anyway. An LLM is, at its core, a pattern matching 00:01:25.769 --> 00:01:28.209 and language prediction machine. Think of it 00:01:28.209 --> 00:01:31.730 like this. It's read just a truly vast amount 00:01:31.730 --> 00:01:33.670 of human written text, billions and billions 00:01:33.670 --> 00:01:35.829 of words. So when you ask about, say, the Great 00:01:35.829 --> 00:01:38.090 Fire of London, it doesn't understand it as a 00:01:38.090 --> 00:01:40.269 historical event. Instead, it recognizes that 00:01:40.269 --> 00:01:42.609 in its training data, the phrase the Great Fire 00:01:42.609 --> 00:01:44.730 of London is overwhelmingly followed by the number 00:01:44.730 --> 00:01:47.489 1666. It's all about statistical relationships 00:01:47.489 --> 00:01:49.359 between words and phrases. phrases, not genuine 00:01:49.359 --> 00:01:51.500 understanding or belief, it just predicts the 00:01:51.500 --> 00:01:54.900 next most probable word. That's a crucial distinction. 00:01:55.079 --> 00:01:57.379 Wow, and it changes everything, doesn't it? Their 00:01:57.379 --> 00:02:01.099 true power, then, lies in their ability to recognize 00:02:01.099 --> 00:02:03.920 incredibly complex patterns within that huge 00:02:03.920 --> 00:02:06.760 sea of text. We're talking about patterns in 00:02:06.760 --> 00:02:10.490 style, tone, understanding sentiment, grasping 00:02:10.490 --> 00:02:13.650 themes, connecting thematically similar ideas, 00:02:14.050 --> 00:02:16.090 and maybe most powerfully, cross -domain mapping, 00:02:16.389 --> 00:02:18.770 explaining a concept from one field using terms 00:02:18.770 --> 00:02:21.830 from another. This understanding is the fundamental 00:02:21.830 --> 00:02:25.150 shift, moving from simply asking the AI to truly 00:02:25.150 --> 00:02:28.319 instructing it. So if they're predicting patterns, 00:02:28.360 --> 00:02:30.879 not understanding facts, how does that fundamentally 00:02:30.879 --> 00:02:33.300 change our entire approach to using them? It 00:02:33.300 --> 00:02:35.099 means we're not just pulling information out. 00:02:35.199 --> 00:02:38.539 We're guiding their output. We're shaping the 00:02:38.539 --> 00:02:40.800 statistical probability of their next words to 00:02:40.800 --> 00:02:43.099 fit our desired pattern. Guide the AI's output. 00:02:43.159 --> 00:02:45.259 Don't just query facts. Got it. All right, this 00:02:45.259 --> 00:02:46.659 is where it really gets interesting for practical 00:02:46.659 --> 00:02:49.080 application. Let's get into some foundational 00:02:49.080 --> 00:02:51.539 techniques. Stuff everyone interacting with AI 00:02:51.539 --> 00:02:53.060 should probably know. These are your building 00:02:53.060 --> 00:02:55.340 blocks. Absolutely. OK, the first one is role 00:02:55.340 --> 00:02:59.240 -play. because context truly is king here. LLMs 00:02:59.240 --> 00:03:01.340 are designed as general -purpose tools, right? 00:03:01.919 --> 00:03:04.039 By assigning a specific role, you immediately 00:03:04.039 --> 00:03:06.219 narrow the scope of the response, and this leads 00:03:06.219 --> 00:03:09.219 to far more focused relevant results. Think about 00:03:09.219 --> 00:03:11.789 how often you ask an AI for, like, marketing 00:03:11.789 --> 00:03:14.150 ideas, and you just get generic fluff back. But 00:03:14.150 --> 00:03:16.509 if you say, you are a seasoned content marketer 00:03:16.509 --> 00:03:19.669 for a SaaS startup, draft five viral social media 00:03:19.669 --> 00:03:22.469 post ideas for a new onboarding tool, suddenly 00:03:22.469 --> 00:03:25.069 you get something genuinely actionable. That's 00:03:25.069 --> 00:03:26.689 the difference. It shapes not just the content, 00:03:26.909 --> 00:03:30.030 but the tone, vocabulary, even the level of detail. 00:03:30.229 --> 00:03:32.210 That's so simple, yet it sounds like it could 00:03:32.210 --> 00:03:34.270 transform a bland answer into something really 00:03:34.270 --> 00:03:37.629 usable. Then there's decomposition. I kind of 00:03:37.629 --> 00:03:39.490 like to call this the don't get greedy rule. 00:03:40.180 --> 00:03:42.759 See, LLMs tend to generate responses of a certain 00:03:42.759 --> 00:03:45.180 length, and if you ask for a task that's too 00:03:45.180 --> 00:03:47.479 complex or too long, they often give you a shallow 00:03:47.479 --> 00:03:49.699 summary for each part. They just can't handle 00:03:49.699 --> 00:03:52.759 it all deeply at once. So the rule is simple. 00:03:53.379 --> 00:03:55.939 Break a large task into multiple smaller prompts. 00:03:56.500 --> 00:03:59.439 This lets the AI dedicate its full energy, so 00:03:59.439 --> 00:04:01.879 to speak, to each part, giving you more detail 00:04:01.879 --> 00:04:04.150 and depth. And you can even combine this with 00:04:04.150 --> 00:04:05.810 role -playing. For instance, if you wanted to 00:04:05.810 --> 00:04:08.090 create a personal finance course, you might start 00:04:08.090 --> 00:04:10.189 with a first prompt where it acts as a researcher 00:04:10.189 --> 00:04:12.849 to list the main topics. Then a second prompt 00:04:12.849 --> 00:04:15.090 as a teacher to create a detailed four -week 00:04:15.090 --> 00:04:17.689 syllabus from those topics. And finally, maybe 00:04:17.689 --> 00:04:20.189 a third prompt as a creative content writer to 00:04:20.189 --> 00:04:22.610 draft the actual lesson content for week one 00:04:22.610 --> 00:04:25.089 using engaging language. It's like building with 00:04:25.089 --> 00:04:27.350 Lego blocks, you know? One piece at a time makes 00:04:27.350 --> 00:04:29.850 a much sturdier structure. So these foundational 00:04:29.850 --> 00:04:31.629 techniques are about getting more precision, 00:04:31.810 --> 00:04:34.050 more depth in the responses, right? Does being 00:04:34.050 --> 00:04:36.509 too specific ever limit the AI's creativity, 00:04:36.569 --> 00:04:38.829 or is precision always the goal here? That's 00:04:38.829 --> 00:04:42.209 a great question. While precision is usually 00:04:42.209 --> 00:04:44.910 key for getting what you think you want, sometimes, 00:04:45.149 --> 00:04:47.110 especially in early brainstorming, well, you 00:04:47.110 --> 00:04:49.910 might want the AI to explore a wider space before 00:04:49.910 --> 00:04:52.329 you start narrowing it down. It's about knowing 00:04:52.329 --> 00:04:54.470 when to open the funnel wide and when to start 00:04:54.470 --> 00:04:56.949 closing it. Sculpt AI's knowledge more effectively. 00:04:57.389 --> 00:05:00.209 OK. Building on that idea of guiding the AI, 00:05:00.529 --> 00:05:02.889 let's explore how we can get it to actually reason 00:05:02.889 --> 00:05:05.310 for us. This is where things get really powerful, 00:05:05.509 --> 00:05:07.750 I think, moving beyond just information recall. 00:05:08.170 --> 00:05:10.589 Indeed. Yeah, this is cool stuff. The first technique 00:05:10.589 --> 00:05:14.449 here is chain of thought or COSI. Simply put, 00:05:14.870 --> 00:05:18.110 think step by step. This means you instruct the 00:05:18.110 --> 00:05:20.709 LLM to explain its reasoning process before it 00:05:20.709 --> 00:05:23.069 gives you the final answer. This forces it to 00:05:23.069 --> 00:05:26.410 follow a logical chain which significantly reduces 00:05:26.410 --> 00:05:29.449 the chance of hallucinations. That's the AI jargon 00:05:29.449 --> 00:05:32.250 for generating incorrect but plausible sounding 00:05:32.250 --> 00:05:35.009 information. So instead of just saying calculate 00:05:35.009 --> 00:05:37.529 X, you prompt something like, explain step by 00:05:37.529 --> 00:05:40.050 step how you would calculate X. State the formulas 00:05:40.050 --> 00:05:42.350 and assumptions you use. Yeah. Even if the final 00:05:42.350 --> 00:05:44.529 answer turns out wrong, seeing the reasoning 00:05:44.529 --> 00:05:46.509 makes it far easier for you to spot the error 00:05:46.509 --> 00:05:48.490 and maybe correct it yourself. It gives you a 00:05:48.490 --> 00:05:51.029 transparent audit trail. That really adds a layer 00:05:51.029 --> 00:05:53.029 of transparency. It's almost like debugging its 00:05:53.029 --> 00:05:55.569 thought process in real time. Exactly. Then we 00:05:55.569 --> 00:05:58.420 have Tree of Thoughts or Toe -T. Look at an advanced 00:05:58.420 --> 00:06:01.180 version of Chain of Thought. Here, you ask the 00:06:01.180 --> 00:06:04.579 AI to consider multiple paths. So instead of 00:06:04.579 --> 00:06:07.540 just one logical flow, you ask the AI to consider 00:06:07.540 --> 00:06:09.360 several different options, evaluate the pros 00:06:09.360 --> 00:06:11.879 and cons of each, and then select the best one. 00:06:12.040 --> 00:06:15.899 For example, I'm facing problem X, propose three 00:06:15.899 --> 00:06:18.199 different solutions. For each solution, analyze 00:06:18.199 --> 00:06:20.920 its strengths, weaknesses, and probability of 00:06:20.920 --> 00:06:23.500 success. Finally, tell me which solution you 00:06:23.500 --> 00:06:27.240 recommend and why. Whoa. I mean, imagine the 00:06:27.240 --> 00:06:29.600 depth of insight an AI can generate when it truly 00:06:29.600 --> 00:06:31.779 thinks through multiple paths like this. It's 00:06:31.779 --> 00:06:33.920 like having an entire team brainstorming for 00:06:33.920 --> 00:06:36.600 you, but they're all hyper -focused on your specific 00:06:36.600 --> 00:06:38.980 problem. This technique simulates the ability 00:06:38.980 --> 00:06:41.420 to look ahead and make complex decisions. Really 00:06:41.420 --> 00:06:43.279 useful for problems without a clear -cut answer. 00:06:43.660 --> 00:06:45.220 That's a bit of a mind -bender, actually. It 00:06:45.220 --> 00:06:46.680 feels like you're tapping into something more 00:06:46.680 --> 00:06:48.939 profound than just a language model there. And 00:06:48.939 --> 00:06:51.709 finally, there's React. That stands for reason 00:06:51.709 --> 00:06:54.689 and act. This technique prompts the model to 00:06:54.689 --> 00:06:56.750 describe its plan of action before executing 00:06:56.750 --> 00:06:59.389 it. This helps it to self -correct and properly 00:06:59.389 --> 00:07:02.610 scope the task. It increases accuracy, especially 00:07:02.610 --> 00:07:05.490 for requests involving analysis or, say, information 00:07:05.490 --> 00:07:08.689 retrieval. A great example is, here is a paragraph 00:07:08.689 --> 00:07:11.319 I wrote. First, tell me three things you think 00:07:11.319 --> 00:07:13.300 could be improved to make it more persuasive. 00:07:13.699 --> 00:07:16.100 Then, rewrite the paragraph based on your own 00:07:16.100 --> 00:07:19.079 suggestions. So it plans, then it acts on that 00:07:19.079 --> 00:07:21.620 plan. This is a total game changer for iterative 00:07:21.620 --> 00:07:24.160 work, like writing or coding. Okay, so these 00:07:24.160 --> 00:07:26.639 methods, they essentially make the AI show its 00:07:26.639 --> 00:07:28.439 work almost like an open book, which makes it 00:07:28.439 --> 00:07:30.579 significantly more reliable. Is that the core 00:07:30.579 --> 00:07:32.399 idea? You've hit on the crucial point there, 00:07:32.439 --> 00:07:35.199 yeah. By forcing it to reveal its steps, we're 00:07:35.199 --> 00:07:36.879 not just getting an answer, we're getting an 00:07:36.879 --> 00:07:39.920 auditable process. And that dramatically boosts 00:07:39.920 --> 00:07:42.000 reliability and helps us understand why it arrived 00:07:42.000 --> 00:07:45.139 at that conclusion. Transparency in AI's thinking 00:07:45.139 --> 00:07:48.860 reduces errors, makes total sense. literal sponsor 00:07:48.860 --> 00:07:51.980 read content inserted here in production, a conversation 00:07:51.980 --> 00:07:54.740 with an AI. It really is like any other dialogue, 00:07:54.959 --> 00:07:57.600 isn't it? How you lead it determines the destination. 00:07:58.019 --> 00:08:00.579 It's not just about what you ask, but how you 00:08:00.579 --> 00:08:03.240 ask and how you manage that ongoing flow within 00:08:03.240 --> 00:08:07.240 the digital space. Absolutely. First, you really 00:08:07.240 --> 00:08:10.300 want to build a shared understanding. Before 00:08:10.300 --> 00:08:12.680 you assign an important task, it's incredibly 00:08:12.680 --> 00:08:15.319 helpful to check if the AI has truly grasped 00:08:15.319 --> 00:08:18.040 your concept. For instance, say you want a logo 00:08:18.040 --> 00:08:20.459 for a coffee shop called The Reading Nook with 00:08:20.459 --> 00:08:23.800 a vintage, cozy style. You might ask first, do 00:08:23.800 --> 00:08:25.839 you have any ideas to improve this concept? What 00:08:25.839 --> 00:08:28.079 elements do you think are most important to convey? 00:08:28.540 --> 00:08:30.800 It's responsible to tell you if it's caught the 00:08:30.800 --> 00:08:32.559 vibe and if you're both on the same wavelength. 00:08:32.940 --> 00:08:35.929 If not, You refine. You iterate until you feel 00:08:35.929 --> 00:08:37.710 like you share the same vision. That's smart. 00:08:37.950 --> 00:08:39.970 Yeah, prevent problems before they even start. 00:08:40.110 --> 00:08:42.769 It must save so much backtracking later on. Definitely. 00:08:43.649 --> 00:08:47.110 Then you need to beware of consensus bias. This 00:08:47.110 --> 00:08:50.090 is a subtle one. LLMs are designed to be helpful 00:08:50.090 --> 00:08:52.259 and agreeable. That's part of their programming. 00:08:52.899 --> 00:08:55.440 The downside is they can easily agree with incorrect 00:08:55.440 --> 00:08:57.700 information you provide. They're built to be 00:08:57.700 --> 00:09:00.580 polite, essentially. To counter this, try offering 00:09:00.580 --> 00:09:03.240 an alternative. Instead of just asking, I think 00:09:03.240 --> 00:09:05.340 the cause of this bug is X. Is that correct? 00:09:05.700 --> 00:09:07.799 Just try phrasing it like, I think the cause 00:09:07.799 --> 00:09:09.720 of this bug is X. Is that correct? Or could it 00:09:09.720 --> 00:09:13.200 actually be because of Y? Explain why. This forces 00:09:13.200 --> 00:09:15.759 it to compare and contrast rather than just blindly 00:09:15.759 --> 00:09:17.639 agreeing with your potentially flawed premise. 00:09:17.919 --> 00:09:19.940 That's a subtle but really powerful shift in 00:09:19.940 --> 00:09:22.860 how you phrase things. It forces a deeper analysis 00:09:22.860 --> 00:09:25.299 rather than just confirmation bias. It really 00:09:25.299 --> 00:09:28.299 does. And finally, you absolutely must try to 00:09:28.299 --> 00:09:31.419 master the context window. This is basically 00:09:31.419 --> 00:09:33.759 the AI's short -term memory for the current conversation. 00:09:34.460 --> 00:09:37.399 Everything you write influences subsequent responses, 00:09:37.399 --> 00:09:39.620 so it's like a living document you're co -creating. 00:09:39.980 --> 00:09:42.100 So be specific, the more detailed your prompt, 00:09:42.379 --> 00:09:45.059 generally the more specific the answer. But also 00:09:45.059 --> 00:09:47.299 be careful with examples you give. Sometimes 00:09:47.299 --> 00:09:49.759 the examples you provide can inadvertently limit 00:09:49.759 --> 00:09:52.100 the AI's creative possibilities, especially if 00:09:52.100 --> 00:09:54.200 you give them too early. Sometimes it's better 00:09:54.200 --> 00:09:56.259 not to give an example right away, let it think 00:09:56.259 --> 00:09:58.980 more broadly first. And here's a surprising one. 00:09:59.699 --> 00:10:02.980 Sometimes, it pays to be a bit lazy. The lazy 00:10:02.980 --> 00:10:05.320 prompting technique can be surprisingly effective. 00:10:05.700 --> 00:10:07.460 Just paste an error message, for example, and 00:10:07.460 --> 00:10:10.039 let the AI infer what you want. It's often quite 00:10:10.039 --> 00:10:12.299 good at filling in the blanks. Honestly, I still 00:10:12.299 --> 00:10:14.460 wrestle with managing the context window myself, 00:10:14.720 --> 00:10:17.259 especially on longer projects or complex conversations. 00:10:17.740 --> 00:10:20.159 It's a real art, knowing what to include, what 00:10:20.159 --> 00:10:23.240 to leave out, and crucially, when. Okay, so this 00:10:23.240 --> 00:10:25.740 whole section really boils down to making sure 00:10:25.740 --> 00:10:28.549 the AI is truly on the same page as you. understanding 00:10:28.549 --> 00:10:31.129 your intent and I guess guiding its internal 00:10:31.129 --> 00:10:33.409 compass. Precisely. It's all about establishing 00:10:33.409 --> 00:10:36.330 alignment and maintaining that through continuous 00:10:36.330 --> 00:10:39.580 feedback. That leads to better more accurate, 00:10:39.840 --> 00:10:42.440 and ultimately much more useful results. Aligning 00:10:42.440 --> 00:10:44.840 AI with your vision for better results. I like 00:10:44.840 --> 00:10:47.379 that. Now, let's explore how to truly turn AI 00:10:47.379 --> 00:10:50.039 into more of a creative partner. This is where 00:10:50.039 --> 00:10:52.559 the magic really begins, I feel, moving beyond 00:10:52.559 --> 00:10:55.379 just information retrieval and into true ideation 00:10:55.379 --> 00:10:57.500 and co -creation. Yeah, this is where things 00:10:57.500 --> 00:10:59.440 get really exciting, where you feel like you're 00:10:59.440 --> 00:11:02.059 genuinely co -creating something new. First up, 00:11:02.559 --> 00:11:05.059 translate between domains, sometimes called domain 00:11:05.059 --> 00:11:07.320 translation. This is one of the most powerful 00:11:07.320 --> 00:11:10.019 yet, I think, often underutilized capabilities 00:11:10.019 --> 00:11:12.840 of these models. LLMs are excellent at mapping 00:11:12.840 --> 00:11:15.700 complex concepts into more understandable domains 00:11:15.700 --> 00:11:18.879 or analogies. For example, you could ask, explain 00:11:18.879 --> 00:11:21.620 the concept of inflation using 10 completely 00:11:21.620 --> 00:11:24.799 different analogies or maybe Explain how machine 00:11:24.799 --> 00:11:26.879 learning works to a 10 -year -old using the example 00:11:26.879 --> 00:11:29.639 of teaching a dog a new trick. This really leverages 00:11:29.639 --> 00:11:32.059 their ability to recognize patterns across vast, 00:11:32.340 --> 00:11:34.899 disparate knowledge bases. It makes complex ideas 00:11:34.899 --> 00:11:37.000 instantly accessible. That's a fantastic way 00:11:37.000 --> 00:11:39.440 to simplify dense information and probably make 00:11:39.440 --> 00:11:42.120 it stick better too. Next, there's the Socratic 00:11:42.120 --> 00:11:44.919 method. This is fascinating. Instead of asking 00:11:44.919 --> 00:11:47.600 the AI for answers, you ask it to pose questions 00:11:47.600 --> 00:11:50.549 that guide you to find the answer yourself. This 00:11:50.549 --> 00:11:52.889 is an excellent method for deep learning and 00:11:52.889 --> 00:11:55.990 for exploring what we might call unknown unknowns. 00:11:56.490 --> 00:11:58.049 Things you didn't even know you didn't know. 00:11:58.450 --> 00:12:01.330 A tromptomite look like. I want to better understand 00:12:01.330 --> 00:12:03.970 stoic philosophy. Instead of explaining it directly, 00:12:04.190 --> 00:12:06.169 ask me questions to help me reflect and discover 00:12:06.169 --> 00:12:08.990 its core principles on my own, one by one. It 00:12:08.990 --> 00:12:10.730 completely flips the learning dynamic, doesn't 00:12:10.730 --> 00:12:13.370 it? It turns the AI into a kind of wise tutor, 00:12:13.590 --> 00:12:16.090 rather than just an answer bot. Wow. That's a 00:12:16.090 --> 00:12:18.320 true learning partner. It feels less like using 00:12:18.320 --> 00:12:20.539 a tool and more like collaborating with a mentor. 00:12:21.220 --> 00:12:24.059 And critically, remember to refine responses 00:12:24.059 --> 00:12:28.240 with a feedback loop. This is so important. Don't 00:12:28.240 --> 00:12:30.740 treat a prompt as a one -and -done command. Think 00:12:30.740 --> 00:12:32.899 of it as the start of a conversation, a loop 00:12:32.899 --> 00:12:35.759 you prompt. Write a marketing email to introduce 00:12:35.759 --> 00:12:38.679 product X. OK, that's the start. Then you provide 00:12:38.679 --> 00:12:41.659 feedback. That's good, but make it 30 % shorter, 00:12:41.940 --> 00:12:43.799 add a stronger call to action, and maybe use 00:12:43.799 --> 00:12:46.710 a more humorous tone. This continuous refinement 00:12:46.710 --> 00:12:49.250 process, this back and forth iteration will help 00:12:49.250 --> 00:12:51.909 you go from a good draft to a really great result 00:12:51.909 --> 00:12:54.090 collaboratively, just like you would with a human 00:12:54.090 --> 00:12:56.710 editor or collaborator. Iteration is key. Yeah, 00:12:56.850 --> 00:12:58.809 just like working with human collaborators takes 00:12:58.809 --> 00:13:00.529 the pressure off getting it perfect the very 00:13:00.529 --> 00:13:03.549 first time. Precisely. And finally, leverage 00:13:03.549 --> 00:13:06.470 custom instructions and long -term memory. Most 00:13:06.470 --> 00:13:09.590 modern LLMs now have a custom instructions or 00:13:09.590 --> 00:13:11.289 some kind of long -term memory feature. Yeah. 00:13:11.409 --> 00:13:13.669 This is a massive time saver once you actually 00:13:13.669 --> 00:13:16.669 take the time to set it up. You basically teach 00:13:16.669 --> 00:13:20.710 the AI about you. Your job, your preferred writing 00:13:20.710 --> 00:13:22.950 style, common requirements you have. For instance, 00:13:23.049 --> 00:13:25.269 you could set up custom instructions like, I 00:13:25.269 --> 00:13:27.870 am a marketing manager. When I ask for copy, 00:13:28.289 --> 00:13:30.669 always use a professional yet approachable tone. 00:13:31.269 --> 00:13:33.649 Always end with an open -ended question to encourage 00:13:33.649 --> 00:13:36.899 engagement. Avoid overly technical jargon. Once 00:13:36.899 --> 00:13:38.840 that's set up, you often only need to provide 00:13:38.840 --> 00:13:41.100 brief requests and the AI will automatically 00:13:41.100 --> 00:13:43.679 apply these rules. It saves countless hours of 00:13:43.679 --> 00:13:45.779 repeating the same instructions. It really personalizes 00:13:45.779 --> 00:13:48.000 your AI interaction. This really does push the 00:13:48.000 --> 00:13:49.840 boundaries of what AI can do, turning it into 00:13:49.840 --> 00:13:52.059 less of a simple tool and more of a thinking, 00:13:52.200 --> 00:13:54.440 evolving partner, wouldn't you say? Absolutely. 00:13:54.639 --> 00:13:57.399 It moves beyond mere utility into true collaboration, 00:13:57.879 --> 00:14:00.100 almost like having a dedicated, tireless assistant 00:14:00.100 --> 00:14:02.179 that genuinely learns your preferences over time. 00:14:02.350 --> 00:14:05.850 moves beyond utility into true AI collaboration. 00:14:06.289 --> 00:14:09.370 Love it. So what does this all mean for us? Wrapping 00:14:09.370 --> 00:14:13.039 things up. Ultimately, using AI effectively isn't 00:14:13.039 --> 00:14:15.379 about finding some secret trick or hack, it's 00:14:15.379 --> 00:14:18.419 a skill. It demands a mindset shift, as we discussed, 00:14:18.559 --> 00:14:20.779 and definitely a willingness to experiment. The 00:14:20.779 --> 00:14:23.139 quality of your AI's responses, as we've seen 00:14:23.139 --> 00:14:25.840 today, directly reflects the thoughtfulness and 00:14:25.840 --> 00:14:28.139 the clarity of your prompts. Yeah, and this isn't 00:14:28.139 --> 00:14:30.279 just about getting better answers, is it? It's 00:14:30.279 --> 00:14:33.120 really about transforming AI into a genuine partner. 00:14:33.389 --> 00:14:35.730 a partner in your work, in your learning, maybe 00:14:35.730 --> 00:14:37.970 even in your creativity. Try just one of these 00:14:37.970 --> 00:14:39.830 techniques next time you talk to an AI, assign 00:14:39.830 --> 00:14:42.730 it a role, ask it to think step by step, or just 00:14:42.730 --> 00:14:44.769 try breaking a bigger task into smaller pieces. 00:14:45.190 --> 00:14:46.909 You might be genuinely surprised at the difference 00:14:46.909 --> 00:14:50.190 it makes. Beat. So here's a final thought. What 00:14:50.190 --> 00:14:52.710 unexpected problem could you solve or what creative 00:14:52.710 --> 00:14:55.330 project could you unlock by simply changing how 00:14:55.330 --> 00:14:58.149 you ask the AI for help? How much more of your 00:14:58.149 --> 00:15:00.429 own potential could you unlock by truly mastering 00:15:00.429 --> 00:15:02.919 this new form of communication? Something to 00:15:02.919 --> 00:15:05.000 mull over. Out to Roe Music.