WEBVTT 00:00:00.000 --> 00:00:01.840 Have you noticed this? It feels like there's 00:00:01.840 --> 00:00:04.519 this growing idea that you need to be some kind 00:00:04.519 --> 00:00:08.359 of elite prompts engineer now just to talk to 00:00:08.359 --> 00:00:10.160 an AI effectively. Oh, totally. It's like the 00:00:10.160 --> 00:00:11.839 new barrier to entry, isn't it? People think, 00:00:11.880 --> 00:00:14.740 oh, man, I got to write these epic 500 word novels 00:00:14.740 --> 00:00:17.719 or, you know, shell out cash for some course 00:00:17.719 --> 00:00:19.920 to learn the secret handshake. But that's just 00:00:19.920 --> 00:00:21.800 it. Well, it's the big mistake we want to clear 00:00:21.800 --> 00:00:25.399 up today. Turns out. Often the absolute best 00:00:25.399 --> 00:00:28.179 way to get a killer prompt is just letting the 00:00:28.179 --> 00:00:30.660 AI write it for you. It's the ultimate life hack 00:00:30.660 --> 00:00:33.340 for AI, and it doesn't take long to learn, maybe 00:00:33.340 --> 00:00:36.719 15 minutes. So today we're diving deep into these 00:00:36.719 --> 00:00:39.479 non -technical shortcuts, making the AI do the 00:00:39.479 --> 00:00:42.920 hard setup work. No jargon, promise, just stuff 00:00:42.920 --> 00:00:45.420 you can actually use. Okay, so first we'll look 00:00:45.420 --> 00:00:47.299 at why just tossing a simple one -liner at the 00:00:47.299 --> 00:00:49.579 AI often, well, falls flat. And we absolutely 00:00:49.579 --> 00:00:51.939 need to define what a system prompt is. Beat. 00:00:52.119 --> 00:00:55.359 That's like... the AI's job description. Then 00:00:55.359 --> 00:00:57.060 we'll unpack four different methods. There's 00:00:57.060 --> 00:01:00.159 this super simple just ask it method. Then there's 00:01:00.159 --> 00:01:02.619 letting the AI interview you, which is cool for 00:01:02.619 --> 00:01:04.840 complex stuff. We'll touch on optimizer tools, 00:01:05.019 --> 00:01:08.260 some hidden gems there. Finally, how to teach 00:01:08.260 --> 00:01:12.260 it by showing examples. Goal is, better results, 00:01:12.840 --> 00:01:15.959 less headache. Right. Let's start with that common 00:01:15.959 --> 00:01:18.400 failure point. You go online, you see someone 00:01:18.400 --> 00:01:21.400 posts the perfect prompt. And it looks like they 00:01:21.400 --> 00:01:23.700 just, I don't know, channeled it instantly. Yeah, 00:01:23.700 --> 00:01:26.140 that's classic survivorship bias. But for props, 00:01:26.819 --> 00:01:28.640 you're only seeing the shiny finished product. 00:01:28.719 --> 00:01:32.099 You don't see the 20 field attempts, the slow 00:01:32.099 --> 00:01:34.439 tweaking, adding a bit here, taking away a bit 00:01:34.439 --> 00:01:37.540 there, that whole messy process. The real insight, 00:01:37.579 --> 00:01:39.859 then, is starting small. It's kind of like building 00:01:39.859 --> 00:01:42.359 with LEGO blocks, maybe. You add one little instruction, 00:01:42.480 --> 00:01:44.159 you test it out, see if it works, then you add 00:01:44.159 --> 00:01:46.400 the next one. Exactly. Because if you throw, 00:01:46.420 --> 00:01:49.069 like, 10 different rules into one prompt right 00:01:49.069 --> 00:01:51.569 at the start, and the output is garbage. You've 00:01:51.569 --> 00:01:53.590 got no clue which of those 10 things messed it 00:01:53.590 --> 00:01:55.930 up. It's just impossible to fix. Need that simplicity, 00:01:56.049 --> 00:01:58.709 that structure. And that structure really starts 00:01:58.709 --> 00:02:00.329 with that term you mentioned, the system prompt. 00:02:00.609 --> 00:02:03.329 Sounds a bit technical, but it's key. Think of 00:02:03.329 --> 00:02:06.829 it like this. The AI is an actor you just hired. 00:02:07.930 --> 00:02:11.370 The normal prompt is the script for today's scene, 00:02:11.430 --> 00:02:13.770 the specific task you want done right now. Like, 00:02:13.770 --> 00:02:17.280 tell me a joke. The system prompt, though, that's 00:02:17.280 --> 00:02:19.900 the actor's character sheet. It's fixed. It tells 00:02:19.900 --> 00:02:21.919 them who they are, what their overall role is, 00:02:22.240 --> 00:02:25.080 the hard rules they always follow. Beat. It's 00:02:25.080 --> 00:02:26.900 their core identity for the whole play. That 00:02:26.900 --> 00:02:29.159 chef analogy from the sources nails it. You're 00:02:29.159 --> 00:02:32.080 hiring a chef. That's the AI. You set the system 00:02:32.080 --> 00:02:34.159 prompt up front. OK, you're an Italian chef, 00:02:34.460 --> 00:02:37.580 strict vegetarian only. And listen, never use 00:02:37.580 --> 00:02:40.150 garlic. Ever. Right. So now, when you give the 00:02:40.150 --> 00:02:41.770 normal prompt, maybe just, what's for dinner, 00:02:42.550 --> 00:02:44.669 the AI chef has to answer within those rules. 00:02:44.689 --> 00:02:46.969 Yeah. You get a great vegetarian, garlic -free 00:02:46.969 --> 00:02:49.310 Italian suggestion. That fixed guide does all 00:02:49.310 --> 00:02:51.909 the work. OK. So where do you, the listener, 00:02:52.150 --> 00:02:54.750 actually find the setting? Is it buried somewhere 00:02:54.750 --> 00:02:57.189 complicated? No, not usually. It's generally 00:02:57.189 --> 00:03:00.650 in the main settings or setup area. In Chat GPT, 00:03:00.770 --> 00:03:03.310 look for custom instructions. For the bigger 00:03:03.310 --> 00:03:05.669 models, If you use their developer platforms 00:03:05.669 --> 00:03:09.250 like OpenAI Platform or Anthropic Console, it's 00:03:09.250 --> 00:03:11.750 right there waiting for you. So if the system 00:03:11.750 --> 00:03:15.439 prompt is that fixed long -term role... What's 00:03:15.439 --> 00:03:18.280 the biggest risk if someone just skips that step, 00:03:18.580 --> 00:03:21.280 if they don't give the AI its role? Well, the 00:03:21.280 --> 00:03:23.520 AI might just go off the rails, suggest things 00:03:23.520 --> 00:03:25.439 totally outside what you need. Like suggesting 00:03:25.439 --> 00:03:27.919 a steak recipe when you wanted that vegetarian 00:03:27.919 --> 00:03:30.919 pasta. Okay, so we know we need this fixed identity, 00:03:31.240 --> 00:03:33.800 this system prompt. How do we get the AI to write 00:03:33.800 --> 00:03:35.639 a good one for us? Sounds like we need a way 00:03:35.639 --> 00:03:37.620 to choose our approach. Yeah, there's a sort 00:03:37.620 --> 00:03:40.180 of decision matrix. First up, the easiest one. 00:03:40.379 --> 00:03:43.259 The simple method. This is perfect for basic, 00:03:43.419 --> 00:03:45.560 pretty clear tasks. You need a quick email drafted, 00:03:45.800 --> 00:03:47.819 fine, when the stakes aren't super high. And 00:03:47.819 --> 00:03:49.939 the opposite end. That's the complex method, 00:03:50.020 --> 00:03:52.000 what we call the reverse interview. Use this 00:03:52.000 --> 00:03:54.280 when the task is really important, like maybe 00:03:54.280 --> 00:03:57.960 drafting a business plan. Or, crucially, when 00:03:57.960 --> 00:03:59.919 you kind of know what you want, but you're not 00:03:59.919 --> 00:04:02.500 totally sure what details the AI actually needs 00:04:02.500 --> 00:04:04.740 from you to do a good job. And squeezed in the 00:04:04.740 --> 00:04:08.990 middle. Method three. The optimizer tools. These 00:04:08.990 --> 00:04:10.530 are great for taking a prompt you've already 00:04:10.530 --> 00:04:12.590 written, one that maybe sort of works, but is 00:04:12.590 --> 00:04:14.849 messy, and just cleaning it up, making it more 00:04:14.849 --> 00:04:17.790 efficient. Got it. Let's dig into method one 00:04:17.790 --> 00:04:21.689 first. Just ask the AI to write its own system 00:04:21.689 --> 00:04:23.889 prompt. Sounds like the quickest win if your 00:04:23.889 --> 00:04:26.769 goal is clear. It is, yeah. But you got to ask 00:04:26.769 --> 00:04:28.310 the right way. There's a template that works 00:04:28.310 --> 00:04:31.250 really well. You basically say, hey, AI, I want 00:04:31.250 --> 00:04:33.129 you to create a system prompt for model name, 00:04:33.310 --> 00:04:36.279 like GPT -4. My goal is to describe your task 00:04:36.279 --> 00:04:38.779 clearly. Can you research the best ways to prompt 00:04:38.779 --> 00:04:41.519 for this as a put today's date? Then write me 00:04:41.519 --> 00:04:44.220 that system prompt. Okay, hold on. Why specify 00:04:44.220 --> 00:04:46.899 the model like GPT -4 versus CLOD -3? Does it 00:04:46.899 --> 00:04:48.699 make that much difference? It actually does. 00:04:49.220 --> 00:04:51.000 Different models have their own little quirks, 00:04:51.180 --> 00:04:54.420 their own best practices for how they like constructions. 00:04:54.620 --> 00:04:57.399 Telling it the model lets it tap into its specific 00:04:57.399 --> 00:04:59.879 internal knowledge base for that engine. You 00:04:59.879 --> 00:05:02.620 get better advice. And the date. Why at today's 00:05:02.620 --> 00:05:05.279 date? Ah, that's super important because prompting 00:05:05.279 --> 00:05:07.860 techniques change fast. What was best practice 00:05:07.860 --> 00:05:11.420 six months ago might be outdated now. Putting 00:05:11.420 --> 00:05:14.639 the date forces it to use the very latest information 00:05:14.639 --> 00:05:17.459 it has. Right, keeps it fresh. OK, the coffee 00:05:17.459 --> 00:05:19.279 shop example in the sources showed this well. 00:05:19.500 --> 00:05:21.839 Asking simply resulted in a prompt that defined 00:05:21.839 --> 00:05:25.230 the AI's role. Okay, you're a social media copywriter. 00:05:25.750 --> 00:05:29.149 It set the brand voice be warm, authentic, and 00:05:29.149 --> 00:05:31.970 added rules. Keep captions short, use hashtags, 00:05:32.350 --> 00:05:34.949 no jargon. Yeah, it builds that structure automatically. 00:05:35.250 --> 00:05:37.649 It even included a line like, avoid cliches such 00:05:37.649 --> 00:05:40.579 as best coffee in town. I wonder, though, why 00:05:40.579 --> 00:05:42.639 get that specific? Isn't the AI supposed to be 00:05:42.639 --> 00:05:44.920 creative? Aren't we boxing it in too much? It's 00:05:44.920 --> 00:05:47.620 about guiding the creativity. Specific constraints 00:05:47.620 --> 00:05:49.540 like that actually help it sound more authentic, 00:05:49.740 --> 00:05:51.560 less like generic marketing fluff that screams, 00:05:51.939 --> 00:05:54.600 AI wrote this. It avoids the cliches everyone 00:05:54.600 --> 00:05:57.180 else uses. OK, that makes sense. Let's shift 00:05:57.180 --> 00:05:59.500 gears to the high -stakes stuff. Method two. 00:05:59.660 --> 00:06:02.879 The reverse AI interview. This is for when it's 00:06:02.879 --> 00:06:05.399 critical, or maybe the idea is still a bit fuzzy 00:06:05.399 --> 00:06:08.079 in your own head. Exactly. Here, you tell the 00:06:08.079 --> 00:06:11.319 AI to interview you, to ask questions until it 00:06:11.319 --> 00:06:13.860 fully gets what you need. But here's the absolute 00:06:13.860 --> 00:06:16.639 key instruction you must give it. Tell the AI 00:06:16.639 --> 00:06:19.800 to ask only one question at a time. Ah, okay. 00:06:20.180 --> 00:06:22.079 Because otherwise... Otherwise you get hit with 00:06:22.079 --> 00:06:24.540 a wall of text, like five questions bundled together, 00:06:24.819 --> 00:06:27.199 and it just kills the conversation. It's overwhelming. 00:06:27.680 --> 00:06:29.660 Forcing it to go one question, then you answer 00:06:29.660 --> 00:06:32.000 the next question. that lets the understanding 00:06:32.000 --> 00:06:34.180 build step by step. That feels really useful. 00:06:34.399 --> 00:06:36.720 It's got two big benefits. First, obviously the 00:06:36.720 --> 00:06:39.240 AI gets super specific details about your situation. 00:06:39.680 --> 00:06:41.740 Second, it actually helps you get clearer on 00:06:41.740 --> 00:06:44.279 what you want. Sometimes you don't fully know 00:06:44.279 --> 00:06:46.720 until the AI asks that one perfect question you 00:06:46.720 --> 00:06:49.339 hadn't considered. You know, I'll admit, I still 00:06:49.339 --> 00:06:51.540 wrestle with palm drift myself, especially when 00:06:51.540 --> 00:06:53.899 starting a big new project. Just nailing down 00:06:53.899 --> 00:06:56.319 that initial scope, getting it really defined 00:06:56.319 --> 00:06:58.740 before the AI starts generating, that can be 00:06:58.740 --> 00:07:01.360 genuinely tricky. Oh, for sure. The study tutor 00:07:01.360 --> 00:07:04.060 example illustrates this perfectly. The AI asks 00:07:04.060 --> 00:07:06.560 something broad first, like, okay, what grade 00:07:06.560 --> 00:07:09.199 level and subject? You answer. Then it narrows 00:07:09.199 --> 00:07:12.000 down. Got it. What's the preferred learning style? 00:07:12.759 --> 00:07:15.600 Visuals, reading, hands -on practice. It keeps 00:07:15.600 --> 00:07:17.639 digging until it knows the real goal. Are we 00:07:17.639 --> 00:07:19.860 just trying to pass the test, or do we want, 00:07:19.860 --> 00:07:22.379 like, deep understanding? OK, let's jump ahead 00:07:22.379 --> 00:07:24.060 slightly to method four for a sec, because it 00:07:24.060 --> 00:07:26.000 feels related to getting the feel right. Method 00:07:26.000 --> 00:07:28.740 four is improve with examples. Yeah, this is 00:07:28.740 --> 00:07:30.899 technically called few shot prompting. Fancy 00:07:30.899 --> 00:07:34.220 name, simple idea. You just give the AI a couple 00:07:34.220 --> 00:07:36.279 of good examples of what you want it to do. You 00:07:36.279 --> 00:07:38.860 show it the pattern. Like the new employee analogy 00:07:38.860 --> 00:07:40.720 again. You wouldn't just say, hey, write professional 00:07:40.720 --> 00:07:42.699 emails. You'd probably show them two or three 00:07:42.699 --> 00:07:44.180 emails that are perfect and say, make it sound 00:07:44.180 --> 00:07:46.939 like this. Exactly that. So if you want the AI 00:07:46.939 --> 00:07:49.980 to say, turn messy notes into polished sentences. 00:07:50.220 --> 00:07:52.240 You give it a few pairs. Here's a messy note, 00:07:52.399 --> 00:07:54.240 and here's the professional sentence I want. 00:07:54.779 --> 00:07:57.019 It learns the style, the tone, the formatting, 00:07:57.339 --> 00:07:59.680 just from those examples. So follow -up question 00:07:59.680 --> 00:08:03.300 then. If the AI's tone is just off, maybe too 00:08:03.300 --> 00:08:06.220 formal or too robotic, is using method four, 00:08:06.439 --> 00:08:08.319 giving it example is usually the fastest way 00:08:08.319 --> 00:08:10.279 to fix that. Faster than trying to write more 00:08:10.279 --> 00:08:12.779 rules in the system prompt. Oh, absolutely. Examples 00:08:12.779 --> 00:08:15.790 teach tone and style. way faster and more effectively 00:08:15.790 --> 00:08:18.509 than trying to describe it abstractly with rules 00:08:18.509 --> 00:08:21.629 alone. Show. Don't just tell. OK. Looping back 00:08:21.629 --> 00:08:24.269 to method three, then. Using those free prompt 00:08:24.269 --> 00:08:26.149 optimizer tools. This is for when you've got 00:08:26.149 --> 00:08:28.110 a prompt. Maybe it kind of works, but it's just 00:08:28.110 --> 00:08:30.290 messy. Right. And these tools are often built 00:08:30.290 --> 00:08:33.230 right into the AI platforms themselves. OpenAI 00:08:33.230 --> 00:08:35.210 has its playground, which you find at platform 00:08:35.210 --> 00:08:37.909 .openai .com. And the trick here is you can actually 00:08:37.909 --> 00:08:40.809 use method one inside the playground. You feed 00:08:40.809 --> 00:08:42.929 it your messy prompt and ask it to optimize it. 00:08:42.990 --> 00:08:45.289 Interesting. So I take my kind of clunky prompt, 00:08:45.330 --> 00:08:47.429 something like plan healthy meals for the week, 00:08:47.490 --> 00:08:49.909 no fish. I put that into the playground and then 00:08:49.909 --> 00:08:52.450 I asked GPT -4 to rewrite that prompt using its 00:08:52.450 --> 00:08:54.649 own best practices. What kind of changes would 00:08:54.649 --> 00:08:57.470 I typically see? You'll see much better structure. 00:08:57.600 --> 00:09:00.980 The AI will likely add clear sections using things 00:09:00.980 --> 00:09:03.340 called delimiters, often just a hash symbol, 00:09:03.759 --> 00:09:05.659 hashtag to separate different parts of the instruction. 00:09:06.279 --> 00:09:08.679 It'll make vague terms specific like changing 00:09:08.679 --> 00:09:12.279 healthy to maybe focus on lean protein and whole 00:09:12.279 --> 00:09:14.379 grains. And I noticed in the source material 00:09:14.379 --> 00:09:17.159 it often adds instructions about the output format 00:09:17.159 --> 00:09:20.299 too, like present the answer as a table. Yes, 00:09:20.860 --> 00:09:23.679 that's huge for consistency. It tells the AI 00:09:23.679 --> 00:09:26.440 exactly how you want the information back. Whoa. 00:09:27.039 --> 00:09:29.039 Okay, imagine the kind of precision you can get 00:09:29.039 --> 00:09:30.899 when you start telling it how to structure the 00:09:30.899 --> 00:09:33.299 output, like use a table or use these specific 00:09:33.299 --> 00:09:35.899 hashtag section markers. That feels like a big 00:09:35.899 --> 00:09:37.820 step up in control. It really is. It's about 00:09:37.820 --> 00:09:40.299 making communication with the AI super clear, 00:09:40.519 --> 00:09:42.779 which leads to way better, more reliable results, 00:09:42.960 --> 00:09:45.139 especially if you're doing it often. Now, Anthropix 00:09:45.139 --> 00:09:47.399 platform, the workbench for Claude, it does something 00:09:47.399 --> 00:09:50.759 similar, but it uses XML tags, things like text 00:09:50.759 --> 00:09:53.690 and instruction instead of hashtags. Why the 00:09:53.690 --> 00:09:55.830 difference? It's basically just a different formatting 00:09:55.830 --> 00:09:59.110 convention they prefer. Those tags, like text, 00:09:59.409 --> 00:10:02.149 serve the same purpose. They help Claude clearly 00:10:02.149 --> 00:10:04.830 separate your instructions from the actual content 00:10:04.830 --> 00:10:07.350 or context you provide. It just helps the model 00:10:07.350 --> 00:10:10.350 organize the input better and avoids confusion. 00:10:11.549 --> 00:10:14.350 Think of them like labeled folders for the AI's 00:10:14.350 --> 00:10:17.669 brain. Mid -roll sponsor, read placeholder. All 00:10:17.669 --> 00:10:19.269 right, so we've got these methods for creating 00:10:19.269 --> 00:10:21.610 great system prompts, but you mentioned testing. 00:10:21.950 --> 00:10:24.309 A prompt isn't really done once the AI writes 00:10:24.309 --> 00:10:27.250 it, is it? No way. You absolutely have to test 00:10:27.250 --> 00:10:30.389 its durability, its robustness. A truly good 00:10:30.389 --> 00:10:32.690 prompt needs to work across a few different but 00:10:32.690 --> 00:10:35.149 related tasks. So take that coffee shop prompt 00:10:35.149 --> 00:10:37.470 we talked about. Okay. Don't just test it with 00:10:37.470 --> 00:10:40.009 one type of post. Try it for a cheerful Happy 00:10:40.009 --> 00:10:42.019 Monday Latte post. Then try it for something 00:10:42.019 --> 00:10:44.000 more formal, like announcing an early closing 00:10:44.000 --> 00:10:46.179 time. And maybe try something a bit quirky or 00:10:46.179 --> 00:10:48.139 funny, like a post about a customer who dramatically 00:10:48.139 --> 00:10:50.259 dropped their coffee. See if it handles all three 00:10:50.259 --> 00:10:52.460 scenarios well. And if it fails one of those 00:10:52.460 --> 00:10:55.159 tests, say the early closing post sounds way 00:10:55.159 --> 00:10:58.480 too casual. Then you tweak the rules in the system 00:10:58.480 --> 00:11:01.480 prompt. You might add a line like, adjust tone 00:11:01.480 --> 00:11:04.289 based on the seriousness of the message. Or you 00:11:04.289 --> 00:11:07.110 could even use all -cap PS for really crucial 00:11:07.110 --> 00:11:10.610 rules like important keep all social media posts 00:11:10.610 --> 00:11:14.509 under 50 words Don't be afraid to emphasize and 00:11:14.509 --> 00:11:16.850 then once it passes the stress test then you 00:11:16.850 --> 00:11:19.309 save it Seriously start building your own prompt 00:11:19.309 --> 00:11:22.070 library doesn't need to be fancy a simple text 00:11:22.070 --> 00:11:24.669 file or note is fine Just give each prompt a 00:11:24.669 --> 00:11:27.409 clear name Maybe note its goal and paste in the 00:11:27.409 --> 00:11:29.789 full system prompt text that you know works. 00:11:29.809 --> 00:11:33.059 This will save you so much time later Okay, let's 00:11:33.059 --> 00:11:35.000 tackle the troubleshooting. What are the most 00:11:35.000 --> 00:11:36.899 common headaches people run into when a prompt 00:11:36.899 --> 00:11:39.159 isn't quite working, and what are the quick fixes? 00:11:39.480 --> 00:11:43.120 First big one. The AI just seems to ignore a 00:11:43.120 --> 00:11:46.000 rule you gave it. Ah, the classic. Quick fix. 00:11:46.259 --> 00:11:48.600 Try moving that specific rule to the very end 00:11:48.600 --> 00:11:50.980 of your system prompt. Models often pay extra 00:11:50.980 --> 00:11:52.720 attention to the last instruction they read. 00:11:53.039 --> 00:11:55.980 Worth a shot. Okay. Problem two, the output format 00:11:55.980 --> 00:11:57.759 is all over the place. Sometimes it's a list, 00:11:57.860 --> 00:12:00.039 sometimes it's a paragraph, sometimes it's something 00:12:00.039 --> 00:12:03.139 else. Inconsistency, yeah. The fix here is usually 00:12:03.139 --> 00:12:05.440 just being hyper -specific in your prompt. Don't 00:12:05.440 --> 00:12:07.759 just say, make a list. Provide the answer is 00:12:07.759 --> 00:12:10.100 a bulleted list. Ensure each bullet point starts 00:12:10.100 --> 00:12:12.059 with a capital letter and ends with a period. 00:12:12.320 --> 00:12:15.000 And remember method four, giving examples. That's 00:12:15.000 --> 00:12:17.139 really powerful for teaching specific formatting. 00:12:17.320 --> 00:12:21.039 Got it. And the third common one, the AI just 00:12:21.039 --> 00:12:25.000 sounds like a robot. stiff, unnatural, no personality. 00:12:25.240 --> 00:12:28.559 Right, the robot voice problem. The fix is defining 00:12:28.559 --> 00:12:31.279 the desired tone much more clearly than just 00:12:31.279 --> 00:12:34.320 saying Be friendly. Get really granular. Think 00:12:34.320 --> 00:12:36.820 about how a helpful friend talks. So you might 00:12:36.820 --> 00:12:39.200 add instructions like write in a warm, encouraging 00:12:39.200 --> 00:12:41.720 tone like a helpful friend offering advice. Use 00:12:41.720 --> 00:12:45.580 simple, everyday words. Avoid jargon. Limit emojis 00:12:45.580 --> 00:12:48.360 to maybe one per response. Treat it like giving 00:12:48.360 --> 00:12:50.860 stage directions to an actor. So putting it all 00:12:50.860 --> 00:12:53.240 together in a workflow, let's use that vocabulary 00:12:53.240 --> 00:12:55.299 tutor example again. You'd maybe start with method 00:12:55.299 --> 00:12:57.200 one, the simple ask, right? Make me a prompt 00:12:57.200 --> 00:12:59.539 for vocabulary tutor. Yep. Then you test it. 00:12:59.690 --> 00:13:01.450 And maybe you find, ooh, the words it's giving 00:13:01.450 --> 00:13:03.470 me are way too simple. So then you'd maybe use 00:13:03.470 --> 00:13:05.690 method four, you'd add a few examples into the 00:13:05.690 --> 00:13:07.490 system prompt, showing the kind of words you 00:13:07.490 --> 00:13:10.649 want, say, B2 level vocabulary words. Exactly. 00:13:10.769 --> 00:13:13.929 Add those examples, test it again. Oh, perfect. 00:13:14.350 --> 00:13:16.690 Now it's giving challenging but appropriate words. 00:13:17.190 --> 00:13:20.269 Then you save that finished tested prompt to 00:13:20.269 --> 00:13:24.190 your library. It's that loop. Ask, create, test, 00:13:24.370 --> 00:13:28.269 mad, adjust, bash, test, Dave. OK. So wrapping 00:13:28.269 --> 00:13:30.919 things up. The big idea here, the main thing 00:13:30.919 --> 00:13:33.879 we hope you take away today, is this. You really 00:13:33.879 --> 00:13:36.639 don't need to become some kind of coding wizard 00:13:36.639 --> 00:13:39.259 or prompt engineer in the complex sense. That's 00:13:39.259 --> 00:13:41.019 kind of the old way of thinking. Yeah, the real 00:13:41.019 --> 00:13:43.240 shortcut, the smarter way, is learning how to 00:13:43.240 --> 00:13:46.259 guide the AI to define its own best identity. 00:13:46.659 --> 00:13:48.779 How to get it to write the most effective instructions 00:13:48.779 --> 00:13:50.919 for itself based on your goal. And we've walked 00:13:50.919 --> 00:13:53.279 through four pretty clear methods to do that. 00:13:53.340 --> 00:13:55.500 You can simply ask the AI. You can let the AI 00:13:55.500 --> 00:13:57.259 interview you for complex things. You can use 00:13:57.259 --> 00:13:59.679 those built -in optimizer tools. Or you can teach 00:13:59.679 --> 00:14:02.179 it by showing examples. The underlying principle 00:14:02.179 --> 00:14:05.279 for all of them, though, is start simple. Test 00:14:05.279 --> 00:14:08.519 it. Iterate. And remember, that structure how 00:14:08.519 --> 00:14:11.039 you organize the prompt is often just as important 00:14:11.039 --> 00:14:14.179 as the content itself. Beat. Which makes you 00:14:14.179 --> 00:14:16.659 wonder, since these AI models are getting so 00:14:16.659 --> 00:14:18.659 much better at understanding how we prompt them, 00:14:19.159 --> 00:14:21.519 What happens when the models start proactively 00:14:21.519 --> 00:14:24.500 suggesting which of these four methods you should 00:14:24.500 --> 00:14:26.940 use for the specific thing you're trying to achieve? 00:14:27.460 --> 00:14:30.279 Like, the AI itself becomes your prompt strategy 00:14:30.279 --> 00:14:33.159 coach. That's a really interesting thought for 00:14:33.159 --> 00:14:35.440 the future. Definitely something to explore another 00:14:35.440 --> 00:14:37.659 time. But for now, the challenge to you, the 00:14:37.659 --> 00:14:40.139 listener, is this. Just pick one method we talked 00:14:40.139 --> 00:14:42.399 about today. Maybe the simple ask, if you have 00:14:42.399 --> 00:14:44.559 a clear goal. Or try the reverse interview, if 00:14:44.559 --> 00:14:46.600 you're feeling adventurous. And just try creating 00:14:46.600 --> 00:14:49.519 one really solid, durable system prompt for something 00:14:49.519 --> 00:14:51.759 you do regularly. Yeah, give it a go. See what 00:14:51.759 --> 00:14:53.759 a difference having that clear AI -generated 00:14:53.759 --> 00:14:56.000 structure makes to your results. Bet you'll be 00:14:56.000 --> 00:14:57.779 surprised. Alturo music.