WEBVTT 00:00:00.000 --> 00:00:02.379 learning how to master artificial intelligence. 00:00:03.680 --> 00:00:06.120 It can feel a bit like being totally lost, right? 00:00:06.259 --> 00:00:10.199 Like inside this giant chaotic library that just 00:00:10.199 --> 00:00:13.039 rearranges its shelves every single week. Exactly. 00:00:13.179 --> 00:00:16.899 Yeah. New models, new tools, new books appearing 00:00:16.899 --> 00:00:19.660 almost daily, it seems. You're constantly chasing 00:00:19.660 --> 00:00:21.980 them, you know, collecting information and courses, 00:00:21.980 --> 00:00:24.079 but you never actually settle down to finish 00:00:24.079 --> 00:00:26.480 building anything useful. We feel that pain. 00:00:27.079 --> 00:00:30.160 So today. We are trying to provide the clear 00:00:30.160 --> 00:00:33.060 strategic map. This deep dive is your shortcut, 00:00:33.079 --> 00:00:36.100 hopefully, to stop wasting months just going 00:00:36.100 --> 00:00:37.920 in circles, trying to figure out where to even 00:00:37.920 --> 00:00:40.679 begin. We're extracting the clearest path we 00:00:40.679 --> 00:00:44.380 could find to real AI proficiency from all the 00:00:44.380 --> 00:00:46.500 source material you shared. And our core mission 00:00:46.500 --> 00:00:48.780 here is really about structure and mindset. It's 00:00:48.780 --> 00:00:51.500 not about chasing the specific tool of the week. 00:00:51.740 --> 00:00:54.460 We want you to stop kind of consuming and really 00:00:54.460 --> 00:00:57.729 start creating. OK, let's unpack this then. We're 00:00:57.729 --> 00:00:59.630 going to cover the four biggest, most common 00:00:59.630 --> 00:01:02.450 mistakes we see keeping people stuck in that 00:01:02.450 --> 00:01:05.590 cycle, that consumption cycle. Then we'll tackle 00:01:05.590 --> 00:01:07.930 those essential tactical questions, you know, 00:01:08.150 --> 00:01:10.609 coding, no code, what skills matter for the future. 00:01:11.030 --> 00:01:14.230 And finally, we'll map out a simple, actionable 00:01:14.230 --> 00:01:18.349 four phase plan, starting with systematic prompting. 00:01:18.670 --> 00:01:22.730 OK, let's dive right in. These four fatal flaws, 00:01:22.769 --> 00:01:26.459 as we're calling them. The core problem is people 00:01:26.459 --> 00:01:29.120 feel like they're making huge progress because 00:01:29.120 --> 00:01:31.299 they're just consuming so much material and they're 00:01:31.299 --> 00:01:33.939 busy But they're basically just running in place 00:01:33.939 --> 00:01:36.099 spinning their wheels Okay, the first trap and 00:01:36.099 --> 00:01:39.480 this one is so so common is starting over every 00:01:39.480 --> 00:01:42.959 single month You see a new shiny tool getting 00:01:42.959 --> 00:01:45.340 all the hype on social media. Oh, yeah, and you 00:01:45.340 --> 00:01:48.120 pivot immediately you think This is the one that's 00:01:48.120 --> 00:01:49.819 going to solve everything. And you spend maybe 00:01:49.819 --> 00:01:52.140 two weeks learning tool A, then tool B comes 00:01:52.140 --> 00:01:54.739 out shiny and new. And you just abandon tool 00:01:54.739 --> 00:01:56.959 A before you've used it to solve even one real 00:01:56.959 --> 00:01:59.359 problem. So after six months of this, you know 00:01:59.359 --> 00:02:02.840 a tiny bit, about 50 things maybe, but you're 00:02:02.840 --> 00:02:05.140 not genuinely proficient in anything useful. 00:02:05.359 --> 00:02:07.819 Right. So the key solution here has to be commitment. 00:02:08.199 --> 00:02:10.560 Right. It's tied directly to how variable AI 00:02:10.560 --> 00:02:13.240 can be, right? It's exactly. You need to choose 00:02:13.240 --> 00:02:16.490 one or two main tools, just one or two. Commit 00:02:16.490 --> 00:02:19.110 to them. Properly commit for at least three months. 00:02:19.870 --> 00:02:22.310 Because until you've logged hundreds of hours 00:02:22.310 --> 00:02:25.629 seeing their specific quirks, how they respond 00:02:25.629 --> 00:02:28.669 differently, that prompt variability, you haven't 00:02:28.669 --> 00:02:30.849 mastered them. You just haven't. Which leads 00:02:30.849 --> 00:02:34.360 straight into mistake number two. collecting 00:02:34.360 --> 00:02:36.300 tools and courses like they're, I don't know, 00:02:36.539 --> 00:02:39.580 baseball cards or something. You join 10 paid 00:02:39.580 --> 00:02:42.039 Discord groups, subscribe to every single newsletter, 00:02:42.319 --> 00:02:44.199 buy courses, just in case you might need them 00:02:44.199 --> 00:02:46.560 later down the road. Oh, the digital basement 00:02:46.560 --> 00:02:49.919 syndrome. I admit, I still have a course on some 00:02:49.919 --> 00:02:52.740 super niche subject from years ago sitting in 00:02:52.740 --> 00:02:55.180 my inbox. I'm pretty sure the digital dust on 00:02:55.180 --> 00:02:57.479 it is actually blocking useful emails now. It 00:02:57.479 --> 00:02:59.560 happens, but you end up drowning in information. 00:02:59.879 --> 00:03:02.550 Your knowledge becomes like... a mile wide, but 00:03:02.550 --> 00:03:05.250 only an inch deep. So shallow. The fix. Limit 00:03:05.250 --> 00:03:06.990 those sources dramatically. Choose, say, two 00:03:06.990 --> 00:03:09.289 or three experts you really trust. Follow them. 00:03:09.909 --> 00:03:12.629 And critically, pick one course and actually 00:03:12.629 --> 00:03:15.389 finish it. Like, 100%. Do every single project 00:03:15.389 --> 00:03:17.490 in it. Every single one. We really can't overstate 00:03:17.490 --> 00:03:20.849 this. Quality always, always beats quantity here. 00:03:21.090 --> 00:03:24.430 OK. Mistake number three. This is prioritizing 00:03:24.430 --> 00:03:28.590 theory over practice. Violating the 20 -80 rule. 00:03:29.360 --> 00:03:31.740 You can read every prompt guide out there, know 00:03:31.740 --> 00:03:33.740 all the concepts, but when someone asks you a 00:03:33.740 --> 00:03:37.180 specific question about applying it, you just... 00:03:38.180 --> 00:03:41.599 Yeah, because generative AI is uniquely unpredictable, 00:03:41.900 --> 00:03:44.780 right? It's not deterministic like normal software. 00:03:45.120 --> 00:03:47.879 Reading about prompting isn't remotely the same 00:03:47.879 --> 00:03:50.080 as actually testing the same structured prompt 00:03:50.080 --> 00:03:52.919 50 or 100 times just to see the weird variations 00:03:52.919 --> 00:03:54.520 you get back. That's totally different. So you 00:03:54.520 --> 00:03:56.840 got to adopt that 20 % theory, 80 % practice 00:03:56.840 --> 00:03:58.979 rule. Read about a technique. Okay, the next 00:03:58.979 --> 00:04:01.599 five minutes, you're in the tool testing it immediately. 00:04:01.620 --> 00:04:03.819 Give yourself a concrete, small project. Like, 00:04:04.039 --> 00:04:06.500 this week I will build a little tool that automatically 00:04:06.500 --> 00:04:09.280 summarizes my weekly team meeting notes, something 00:04:09.280 --> 00:04:11.719 real. And that mindset ties straight into the 00:04:11.719 --> 00:04:14.240 last big mistake, learning just to get a certificate 00:04:14.240 --> 00:04:15.979 instead of learning to actually build something. 00:04:16.120 --> 00:04:19.379 Ah, the certificate chase. Yeah. If you stop 00:04:19.379 --> 00:04:21.839 the moment you hit a small frustrating technical 00:04:21.839 --> 00:04:24.300 problem, you're stuck in that learner mindset. 00:04:24.779 --> 00:04:26.720 The critical shift is from learner to builder. 00:04:27.180 --> 00:04:29.279 You have to train yourself to push through and 00:04:29.279 --> 00:04:32.019 solve those tiny annoying problems. OK, so if 00:04:32.019 --> 00:04:36.730 the core mistake boils down to prioritizing consuming 00:04:36.730 --> 00:04:39.649 information over actually creating something. 00:04:40.790 --> 00:04:43.050 What's the single most valuable mindset switch 00:04:43.050 --> 00:04:46.550 our listeners can make, like right now? That 00:04:46.550 --> 00:04:48.810 valuable switch is focusing on finishing small 00:04:48.810 --> 00:04:51.209 projects. That provides the highest real -world 00:04:51.209 --> 00:04:53.970 motivation, much more than any course certificate. 00:04:54.209 --> 00:04:57.230 That moves us nicely from mindset to strategic 00:04:57.230 --> 00:04:59.269 planning. Let's hit those common tactical questions 00:04:59.269 --> 00:05:01.069 people ask once they're actually ready to commit. 00:05:01.470 --> 00:05:03.800 First big one. Should I learn to code? Right. 00:05:04.019 --> 00:05:05.759 The short answer is, well, it really depends 00:05:05.759 --> 00:05:08.040 entirely on your goal. OK. If you want to build 00:05:08.040 --> 00:05:10.540 completely novel systems, you know, proprietary 00:05:10.540 --> 00:05:12.860 platforms with multiple users, stuff like that, 00:05:13.079 --> 00:05:14.579 then yes, you probably need to learn to code 00:05:14.579 --> 00:05:18.000 professionally. But for personal automations 00:05:18.000 --> 00:05:20.779 or maybe automations within a business, the answer 00:05:20.779 --> 00:05:23.220 is a bit different and maybe more important for 00:05:23.220 --> 00:05:26.420 most people listening. Exactly. For that, you 00:05:26.420 --> 00:05:28.519 probably need to spend maybe five or six hours 00:05:28.519 --> 00:05:30.860 just learning to read and understand the basics 00:05:30.860 --> 00:05:33.040 of ground. Read it, not necessarily write it 00:05:33.040 --> 00:05:35.759 from scratch. Right. Things like what's a function, 00:05:35.959 --> 00:05:39.060 what's a variable, basic structure. Why is that 00:05:39.060 --> 00:05:41.920 reading part so crucial? Because AI, especially 00:05:41.920 --> 00:05:45.040 when writing code, It often hallucinates. It 00:05:45.040 --> 00:05:47.500 makes up names for functions or libraries that 00:05:47.500 --> 00:05:50.540 just don't exist. So you need enough foundational 00:05:50.540 --> 00:05:52.800 understanding to look at the AI's output and 00:05:52.800 --> 00:05:55.439 go, wait a second, that variable name makes zero 00:05:55.439 --> 00:05:58.180 sense here, or that function isn't real, so you 00:05:58.180 --> 00:06:00.379 can debug it quickly instead of being totally 00:06:00.379 --> 00:06:02.560 stuck. That's a really great distinction. OK, 00:06:02.560 --> 00:06:05.060 next question. What automation should a business 00:06:05.060 --> 00:06:07.540 build first? Where is the low -hanging fruit? 00:06:08.139 --> 00:06:10.459 Start with the really predictable admin tasks, 00:06:10.759 --> 00:06:13.220 the stuff that doesn't need creativity. Like, 00:06:13.420 --> 00:06:16.160 automatically sending a welcome email right after 00:06:16.160 --> 00:06:18.720 someone makes a payment. Simple, reliable. Makes 00:06:18.720 --> 00:06:21.959 sense. You only bring AI into it when creativity 00:06:21.959 --> 00:06:24.680 is needed, A and D critically, when you can accept 00:06:24.680 --> 00:06:26.680 a few mistakes along the way. OK, like what? 00:06:27.379 --> 00:06:30.839 Like, using AI to read through customer feedback 00:06:30.839 --> 00:06:34.000 emails, maybe categorize the sentiment positive, 00:06:34.040 --> 00:06:36.519 negative, neutral, and then create a summary 00:06:36.519 --> 00:06:39.480 report. Right. If it gets one or two wrong, it's 00:06:39.480 --> 00:06:41.259 not the end of the world. Exactly. The mistakes 00:06:41.259 --> 00:06:43.420 don't stop the whole business. And you should 00:06:43.420 --> 00:06:46.839 probably always prioritize tasks related to actually 00:06:46.839 --> 00:06:49.180 making money, too, right? Oh, absolutely. Automating 00:06:49.180 --> 00:06:52.540 things like sales emails or light categorization 00:06:52.540 --> 00:06:55.079 will usually give you clearer, faster results 00:06:55.079 --> 00:06:58.459 in ROI than, say, automating some random internal 00:06:58.459 --> 00:07:01.970 spreadsheet cleanup. Focus on revenue. OK, speaking 00:07:01.970 --> 00:07:05.490 of risks, voice AI agents, they're getting better 00:07:05.490 --> 00:07:07.649 fast, but where are they still tricky? They are 00:07:07.649 --> 00:07:09.670 improving incredibly quickly. But yeah, they 00:07:09.670 --> 00:07:12.589 still struggle in the messy real world. Accents, 00:07:12.870 --> 00:07:14.970 background noise, people interrupting unexpectedly. 00:07:15.209 --> 00:07:17.310 The unpredictability of humans. Totally. So you 00:07:17.310 --> 00:07:20.910 have to assess the risk. If, say, five absolutely 00:07:20.910 --> 00:07:22.850 crucial calls bring in all the money for your 00:07:22.850 --> 00:07:26.129 company. High stakes. Then 99 % perfection from 00:07:26.129 --> 00:07:28.839 a voice agent probably isn't good enough. That 00:07:28.839 --> 00:07:33.199 1 % failure could be catastrophic. But for industries 00:07:33.199 --> 00:07:36.259 like, say, plumbers or roofers, where the workers 00:07:36.259 --> 00:07:39.319 are often outside, maybe can't answer the phone. 00:07:39.800 --> 00:07:42.819 An answered call from an AI agent, even if imperfect, 00:07:43.399 --> 00:07:45.740 is almost always better than a completely missed 00:07:45.740 --> 00:07:49.019 call on a lost job. Good point. Use it where 00:07:49.019 --> 00:07:51.220 just getting the call answered adds clear value. 00:07:51.480 --> 00:07:54.399 Exactly. Which kind of brings us to future skills. 00:07:54.980 --> 00:07:57.339 Most people, I think, focus on the wrong things 00:07:57.339 --> 00:08:00.560 here. Oh, so? Well, the purely technical skill 00:08:00.560 --> 00:08:02.680 of connecting tools together, like knowing all 00:08:02.680 --> 00:08:04.920 the APIs, that's likely going to become less 00:08:04.920 --> 00:08:07.579 valuable over time. Really? Why? Because the 00:08:07.579 --> 00:08:09.879 big platforms are moving fast towards text to 00:08:09.879 --> 00:08:11.480 workflow. You'll just describe what you want 00:08:11.480 --> 00:08:13.699 in plain English, like, when I get an email with 00:08:13.699 --> 00:08:16.540 an invoice, save the PDF to this folder and add 00:08:16.540 --> 00:08:18.220 the amount to my spreadsheet. And the system 00:08:18.220 --> 00:08:20.379 just builds it. And the system builds the automation 00:08:20.379 --> 00:08:23.060 instantly. OK, that makes sense for speed, definitely. 00:08:23.160 --> 00:08:26.300 Yeah. But hang on, if we rely purely on that 00:08:26.300 --> 00:08:29.459 text to workflow from the big guys, doesn't that 00:08:29.459 --> 00:08:32.519 create a huge risk of vendor lock -in? Isn't 00:08:32.519 --> 00:08:35.840 knowing how to connect APIs yourself still crucial 00:08:35.840 --> 00:08:37.820 for long -term independence and flexibility? 00:08:38.000 --> 00:08:40.220 Ah, that's the tension right there. You hit it. 00:08:40.399 --> 00:08:43.379 Yes, relying solely on the platform's magic button 00:08:43.379 --> 00:08:46.419 creates lock -in. So the skill of the future 00:08:46.419 --> 00:08:49.879 isn't just the technical clicking or even coding, 00:08:50.220 --> 00:08:52.940 it's the strategic thinking. The what and why. 00:08:53.120 --> 00:08:55.539 Exactly. If you can technically automate almost 00:08:55.539 --> 00:08:57.679 anything easily, the real value lies in being 00:08:57.679 --> 00:08:59.919 the architect, not just the builder, understanding 00:08:59.919 --> 00:09:02.409 what is actually worth automating why, and maybe 00:09:02.409 --> 00:09:04.490 crucially, how independent those automated systems 00:09:04.490 --> 00:09:06.889 need to be from any single platform. That's the 00:09:06.889 --> 00:09:08.870 high level skill. Being the architect, I like 00:09:08.870 --> 00:09:11.629 that. Okay, last tactical point, dealing with 00:09:11.629 --> 00:09:15.529 skeptics. Your boss, your team thinks AI is just 00:09:15.529 --> 00:09:18.029 hype or maybe even scary. What do you do? Use 00:09:18.029 --> 00:09:21.009 the show don't tell strategy every time. Focus 00:09:21.009 --> 00:09:22.710 on the problem you solved, not the technology 00:09:22.710 --> 00:09:24.549 you used, at least initially. Don't leave with 00:09:24.549 --> 00:09:26.809 the buzzword. Great. Don't start the meeting 00:09:26.809 --> 00:09:29.429 saying, hey everyone, I used AI to solve this. 00:09:29.470 --> 00:09:31.870 Just say, hey, I made this little program or 00:09:31.870 --> 00:09:34.549 this little workflow that helped us do X tasks 00:09:34.549 --> 00:09:36.549 three times faster this week. Let the results 00:09:36.549 --> 00:09:39.610 speak. Let them love the result first. Then if 00:09:39.610 --> 00:09:41.629 they ask how, you can mention, oh, I used an 00:09:41.629 --> 00:09:50.009 AI tool. Thinking is kind of the new coding. 00:09:50.409 --> 00:09:52.350 How should someone actually prove that value, 00:09:52.370 --> 00:09:55.500 especially in an AI skeptic workplace? Demonstrate 00:09:55.500 --> 00:09:58.240 the results first. Let that success create the 00:09:58.240 --> 00:10:00.799 curiosity about the how, rather than leading 00:10:00.799 --> 00:10:04.179 with a potentially scary or misunderstood technology 00:10:04.179 --> 00:10:07.440 buzzword. Okay, we've got the strategy, the mindset, 00:10:08.019 --> 00:10:10.759 but now we inevitably hit the jargon wall. Let's 00:10:10.759 --> 00:10:12.700 try and pull apart those two technical concepts 00:10:12.700 --> 00:10:14.580 that seem to confuse everyone when it comes to 00:10:14.580 --> 00:10:18.139 using AI with company knowledge. RA versus fine 00:10:18.139 --> 00:10:20.889 -tuning. Yes, definitely. Let's clarify these 00:10:20.889 --> 00:10:23.090 using simple analogies. So you know which tool 00:10:23.090 --> 00:10:25.509 fits which job, especially for internal company 00:10:25.509 --> 00:10:29.009 stuff. OK, let's start with AIR. Retrieval Augmented 00:10:29.009 --> 00:10:31.389 Generation. Can you break down that acronym first 00:10:31.389 --> 00:10:33.730 and then explain that smart librarian analogy? 00:10:34.009 --> 00:10:38.210 Sure. So, RG. Think of it like hiring a super 00:10:38.210 --> 00:10:41.509 smart, incredibly fast librarian for your company's 00:10:41.509 --> 00:10:44.970 giant library of documents, policies, manuals, 00:10:45.110 --> 00:10:48.730 past reports, whatever. The AI model itself doesn't 00:10:48.730 --> 00:10:51.049 try to read and memorize that entire library 00:10:51.049 --> 00:10:54.299 that's inefficient. Instead, when you ask a question, 00:10:54.580 --> 00:10:57.639 the RDE system, the librarian, quickly retrieves 00:10:57.639 --> 00:11:00.679 just the most relevant paragraph or snippet from 00:11:00.679 --> 00:11:03.100 that external knowledge source. Like the specific 00:11:03.100 --> 00:11:05.919 page in the updated policy manual. Exactly, and 00:11:05.919 --> 00:11:07.919 it gives only that relevant snippet to the main 00:11:07.919 --> 00:11:10.480 AI model to help it form the answer. It augments 00:11:10.480 --> 00:11:13.240 the AI's generation with retrieved specific knowledge. 00:11:13.399 --> 00:11:15.379 Got it. Okay, now fine -tuning. How's that different? 00:11:15.549 --> 00:11:17.850 Fine -tuning is more like taking an already smart 00:11:17.850 --> 00:11:19.990 student that's the base large language model 00:11:19.990 --> 00:11:22.970 like GPT -4 or Claude and sending them to a very 00:11:22.970 --> 00:11:25.330 specialized finishing school. This school teaches 00:11:25.330 --> 00:11:29.250 them one very specific style or tone or personality. 00:11:29.830 --> 00:11:32.450 For example, learning the exact marketing voice 00:11:32.450 --> 00:11:34.950 of BrandX down to the specific phrasing they 00:11:34.950 --> 00:11:37.950 always use. You teach the specific style using 00:11:37.950 --> 00:11:40.549 thousands and thousands of custom examples. Ah, 00:11:40.549 --> 00:11:42.889 so it's about style and tone, not just pulling 00:11:42.889 --> 00:11:47.230 facts. Primarily, yes. or very specific narrow 00:11:47.230 --> 00:11:49.889 knowledge domains not well covered by the base 00:11:49.889 --> 00:11:52.809 model. But here's the really key practical point. 00:11:53.289 --> 00:11:55.389 Fine -tuning should generally be your last resort. 00:11:55.549 --> 00:11:57.769 Last resort. Why is that? Why is it difficult? 00:11:58.009 --> 00:12:00.750 because it's incredibly difficult and cumbersome 00:12:00.750 --> 00:12:03.490 to update. Think about it. Every time your company 00:12:03.490 --> 00:12:06.149 policy changes, or you have a new product detail, 00:12:06.470 --> 00:12:08.429 or you slightly adjust your brand voice... You 00:12:08.429 --> 00:12:09.990 have to redo the whole thing. You have to gather 00:12:09.990 --> 00:12:12.549 all the new examples, potentially retrain the 00:12:12.549 --> 00:12:14.889 entire fine -tuned model again from scratch, 00:12:15.230 --> 00:12:17.850 which is both time -consuming and often expensive. 00:12:18.649 --> 00:12:20.669 RRAG, on the other hand, is much more dynamic. 00:12:20.809 --> 00:12:22.830 You just update the document library the librarian 00:12:22.830 --> 00:12:24.960 looks at. Much easier, much faster. That makes 00:12:24.960 --> 00:12:27.000 sense. You know, I still wrestle with prompt 00:12:27.000 --> 00:12:30.580 drift myself sometimes, where a prompt that worked 00:12:30.580 --> 00:12:33.179 perfectly last week suddenly starts giving me 00:12:33.179 --> 00:12:35.399 weaker responses. Oh yeah, happens all the time. 00:12:35.679 --> 00:12:38.200 Which is exactly why these systematic tools, 00:12:38.659 --> 00:12:41.519 like ARGID, that offer more structured predictability 00:12:41.519 --> 00:12:44.019 are so important, it reduces that frustrating 00:12:44.019 --> 00:12:46.700 variability. Definitely. Okay, moving to another 00:12:46.700 --> 00:12:50.620 comparison. No code versus custom code for building 00:12:50.620 --> 00:12:53.860 these automations. No code platforms like Zapier 00:12:53.860 --> 00:12:56.519 or Make, they are the clear winner for speed. 00:12:56.980 --> 00:12:59.080 You can often get something useful working in 00:12:59.080 --> 00:13:01.320 days or maybe weeks, not months. Much faster 00:13:01.320 --> 00:13:03.840 prototyping. And they're also usually best for 00:13:03.840 --> 00:13:06.000 difficult connections between popular online 00:13:06.000 --> 00:13:07.980 tools because they've already built all those 00:13:07.980 --> 00:13:10.279 API bridges for you. Right, saves a ton of hassle. 00:13:10.639 --> 00:13:14.240 So when is custom code better? Custom code really 00:13:14.240 --> 00:13:17.340 shines when you have super strict security requirements 00:13:17.340 --> 00:13:20.519 that no code platforms can't meet, or when you 00:13:20.519 --> 00:13:23.539 need to connect to very old pre -internet legacy 00:13:23.539 --> 00:13:26.039 systems that don't have modern APIs. OK, the 00:13:26.039 --> 00:13:28.539 edge cases. Kind of. But here's that warning 00:13:28.539 --> 00:13:32.039 again, echoing the text -to -workflow idea. Tools 00:13:32.039 --> 00:13:34.899 like Cursor and others that help AI write code 00:13:34.899 --> 00:13:38.720 are blurring this line incredibly fast. How so? 00:13:38.860 --> 00:13:41.700 Well, soon, just describing an automation you 00:13:41.700 --> 00:13:44.179 want in simple English might actually spit out 00:13:44.179 --> 00:13:47.720 the full executable custom code faster than you 00:13:47.720 --> 00:13:50.330 could click and drag all the boxes in a traditional 00:13:50.330 --> 00:13:52.850 no -code tool, the landscape is shifting. Wow, 00:13:53.009 --> 00:13:55.769 okay. So given that update difficulty we talked 00:13:55.769 --> 00:13:58.370 about with fine tuning, does that basically make 00:13:58.370 --> 00:14:02.230 RGI the clear winner for companies that have, 00:14:02.230 --> 00:14:04.690 you know, internal policies or knowledge bases 00:14:04.690 --> 00:14:07.169 that change pretty frequently? Yes, absolutely. 00:14:07.350 --> 00:14:09.830 Yeah. For keeping AI informed with dynamic, frequently 00:14:09.830 --> 00:14:12.149 changing external knowledge, like... company 00:14:12.149 --> 00:14:15.029 policies, support docs, product specs are actually 00:14:15.029 --> 00:14:17.169 generally the most efficient and practical method 00:14:17.169 --> 00:14:19.269 right now. It's designed for that, right? We've 00:14:19.269 --> 00:14:21.909 covered the mistakes to avoid the strategic thinking. 00:14:22.169 --> 00:14:24.350 Now, this is the part everyone wants, the actionable 00:14:24.350 --> 00:14:26.990 plan. How do you actually move from zero knowledge 00:14:26.990 --> 00:14:30.090 to genuine expertise? Let's map it out. But first, 00:14:30.470 --> 00:14:32.049 maybe a quick piece of advice from the source 00:14:32.049 --> 00:14:34.830 material. Stick with the big players, right? 00:14:35.149 --> 00:14:37.470 Companies with strong support and resources like 00:14:37.470 --> 00:14:40.330 OpenAI, Anthropic, Google, they're just the most 00:14:40.330 --> 00:14:42.570 likely to survive long -term and keep improving 00:14:42.570 --> 00:14:45.269 rapidly. Don't bet the farm on a tiny startup 00:14:45.269 --> 00:14:48.129 that might vanish next year. Good point. Okay, 00:14:48.309 --> 00:14:51.090 phase one. Get to know the tools and, crucially, 00:14:51.629 --> 00:14:55.909 build your taste. And start simple, like only 00:14:55.909 --> 00:14:58.850 use ChatGPT and Claude. Just those two. No others? 00:14:58.990 --> 00:15:02.039 Nope. Find their limits. Don't add anything else 00:15:02.039 --> 00:15:04.299 to your plate for probably 90 days. Seriously. 00:15:04.460 --> 00:15:06.000 And the goal isn't just using them randomly, 00:15:06.200 --> 00:15:08.139 right? It's about learning when one model works 00:15:08.139 --> 00:15:10.559 better than the other. Exactly. And always testing 00:15:10.559 --> 00:15:13.120 them with a real -world purpose or past in mind. 00:15:13.480 --> 00:15:15.580 You need to learn their distinct personalities, 00:15:15.899 --> 00:15:17.860 their strengths, their weaknesses, where they 00:15:17.860 --> 00:15:20.179 tend to fail. Okay, phase one. Immersion and 00:15:20.179 --> 00:15:22.259 taste -building. What's phase two? Phase two 00:15:22.259 --> 00:15:24.659 is the absolute foundation. This skill never 00:15:24.659 --> 00:15:27.399 gets old. Learn systematic prompting. Why is 00:15:27.399 --> 00:15:29.759 this so fundamental? Because good prompting works 00:15:29.759 --> 00:15:32.879 across all kinds of models. Text, image generation, 00:15:33.159 --> 00:15:36.340 code generation. It's not magic. It's a structured, 00:15:36.639 --> 00:15:39.019 repeatable skill for getting better results from 00:15:39.019 --> 00:15:42.200 any AI. OK, let's make that concrete. Compare 00:15:42.200 --> 00:15:44.600 a weak prompt versus a systematic one. A weak 00:15:44.600 --> 00:15:47.720 prompt is just write a social media post. Vague. 00:15:47.879 --> 00:15:51.360 Lazy. Gets you generic results. Totally. A systematic 00:15:51.360 --> 00:15:54.679 prompt includes specific elements. Think. R -C 00:15:54.679 --> 00:15:58.399 -T -R -O. Roll. You are a witty social media 00:15:58.399 --> 00:16:01.879 manager targeting Gen Z. Context. Our brand is 00:16:01.879 --> 00:16:04.220 eco -friendly. The topic is our new recycled 00:16:04.220 --> 00:16:07.779 packaging. Task. Write a short, engaging Instagram 00:16:07.779 --> 00:16:11.080 caption. Rules constraints. Under 50 words include 00:16:11.080 --> 00:16:13.500 hashtag, hashtag sustainable style, avoid the 00:16:13.500 --> 00:16:16.120 word amazing, and maybe output format. Just give 00:16:16.120 --> 00:16:18.529 me the caption text. OK. Much more specific. 00:16:18.870 --> 00:16:20.850 That systematic approach dramatically guides 00:16:20.850 --> 00:16:23.389 the AI. It forces you to think clearly first 00:16:23.389 --> 00:16:25.289 about exactly what you want, which leads to a 00:16:25.289 --> 00:16:27.070 result that's far, far closer to your actual 00:16:27.070 --> 00:16:29.309 goal. Makes sense, okay? Phase three. Phase three 00:16:29.309 --> 00:16:31.289 is stepping up to architecture and thinking about 00:16:31.289 --> 00:16:33.690 agents. Divine agent for us. An agent, in this 00:16:33.690 --> 00:16:36.330 context, is basically an AI program that can 00:16:36.330 --> 00:16:38.789 autonomously perform a series of actions to reach 00:16:38.789 --> 00:16:41.470 a goal without you needing to click every single 00:16:41.470 --> 00:16:43.210 step along the way. Like a multi -step workflow. 00:16:43.470 --> 00:16:46.039 Exactly. Imagine instructing it. Okay, agent, 00:16:46.240 --> 00:16:48.919 check my main inbox every hour. If you find an 00:16:48.919 --> 00:16:51.220 email with invoice in the subject line from Sender 00:16:51.220 --> 00:16:54.399 X, download the attached PDF, save that file 00:16:54.399 --> 00:16:57.639 into my invoices Q3 folder, and then automatically 00:16:57.639 --> 00:17:00.700 add the sender date and amount into this specific 00:17:00.700 --> 00:17:04.309 Google Sheet. Whoa. OK, now imagine scaling that 00:17:04.309 --> 00:17:07.549 capability, using multiple agents maybe to handle, 00:17:07.569 --> 00:17:09.730 I don't know, a billion routine customer queries 00:17:09.730 --> 00:17:12.509 simultaneously. That's like stacking Lego blocks 00:17:12.509 --> 00:17:14.690 of data processing on a massive scale. That's 00:17:14.690 --> 00:17:17.150 the power, right. Workful automation amplified. 00:17:17.190 --> 00:17:19.049 That's where things are heading. OK, mine's slightly 00:17:19.049 --> 00:17:21.730 blown. What's phase four? Phase four is choose 00:17:21.730 --> 00:17:24.369 a specialty. Once you have the foundation, go 00:17:24.369 --> 00:17:27.369 deep in one specific area. Examples. Maybe it's 00:17:27.369 --> 00:17:30.609 building custom AI applications for, say, real 00:17:30.609 --> 00:17:33.400 estate agents. or becoming an expert in advanced 00:17:33.400 --> 00:17:37.140 AI image creation for marketing, or AI for scientific 00:17:37.140 --> 00:17:40.000 research data analysis. Pick a niche. Why specialize? 00:17:40.160 --> 00:17:42.599 Because when the next new tool inevitably arrives 00:17:42.599 --> 00:17:44.720 in your niche, you'll already understand the 00:17:44.720 --> 00:17:46.779 core concepts and the problems people are trying 00:17:46.779 --> 00:17:49.220 to solve. You can integrate that new technology 00:17:49.220 --> 00:17:52.059 quickly because you have the deep context instead 00:17:52.059 --> 00:17:54.859 of starting from scratch again. Got it. One last 00:17:54.859 --> 00:17:57.859 point from the notes here. Q10, balancing speed 00:17:57.859 --> 00:18:00.599 and perfection when building. Ah, yes. Crucial. 00:18:00.720 --> 00:18:03.220 Don't spend six months trying to build the absolute 00:18:03.220 --> 00:18:05.200 perfect system right out of the gate. Why not? 00:18:05.380 --> 00:18:07.279 Because the underlying technology is changing 00:18:07.279 --> 00:18:10.400 way too fast. What's perfect today might be obsolete 00:18:10.400 --> 00:18:12.500 or easily done differently in three months. So 00:18:12.500 --> 00:18:14.819 what's the approach? Build something quick and 00:18:14.819 --> 00:18:17.819 simple. First, a minimum viable product. Test 00:18:17.819 --> 00:18:20.920 the core idea with maybe five users. Get feedback. 00:18:21.680 --> 00:18:24.740 Then iterate slowly expand to 10 users, then 00:18:24.740 --> 00:18:28.970 50. Test, learn, iterate, and grow. Perfection 00:18:28.970 --> 00:18:31.210 is the enemy of progress in such a fast -moving 00:18:31.210 --> 00:18:35.170 field. So zooming out, what would you say is 00:18:35.170 --> 00:18:37.450 the ultimate return on investment for really 00:18:37.450 --> 00:18:39.569 mastering that systematic prompting skill we 00:18:39.569 --> 00:18:41.849 talk about in phase two, like right now? Hmm, 00:18:42.009 --> 00:18:45.970 the ultimate ROI. I think mastering that structured 00:18:45.970 --> 00:18:48.970 thinking. that ability to give clear, unambiguous 00:18:48.970 --> 00:18:51.710 instructions. It prepares you for whatever future 00:18:51.710 --> 00:18:54.650 technology comes next. Whether it's today's LLMs, 00:18:54.950 --> 00:18:57.809 or future image models, or even human -like robots 00:18:57.809 --> 00:19:01.109 that need precise tasks, it's a fundamental investment 00:19:01.109 --> 00:19:03.269 in clear communication with intelligent systems. 00:19:04.990 --> 00:19:07.450 So wrapping this all up was a big takeaway for 00:19:07.450 --> 00:19:09.369 you, the listener. I think it's that the people 00:19:09.369 --> 00:19:13.079 who truly win in this AI future They won't necessarily 00:19:13.079 --> 00:19:15.640 be the ones who know every single tool that pops 00:19:15.640 --> 00:19:17.480 up. No, definitely not. They'll be the ones who 00:19:17.480 --> 00:19:19.680 understand how to think strategically about problems. 00:19:20.119 --> 00:19:22.200 They'll know which fundamental tools and concepts 00:19:22.200 --> 00:19:24.400 are worth betting on long term. And maybe most 00:19:24.400 --> 00:19:26.099 importantly, they'll have the mindset and skills 00:19:26.099 --> 00:19:28.279 to actually get things done to focus on building 00:19:28.279 --> 00:19:31.259 real value, not just collecting knowledge. Avoid 00:19:31.259 --> 00:19:34.019 that constant trend chasing. Focus instead on 00:19:34.019 --> 00:19:36.339 learning a small set of foundational tools really 00:19:36.339 --> 00:19:38.730 deeply. and apply them to solve real world problems 00:19:38.730 --> 00:19:41.450 immediately. Practice, practice, practice. Yeah. 00:19:41.789 --> 00:19:43.670 Stop collecting courses just for the sake of 00:19:43.670 --> 00:19:46.309 it. And please, stop jumping from shiny tool 00:19:46.309 --> 00:19:48.990 to shiny tool every month. The underlying tech 00:19:48.990 --> 00:19:51.809 will keep changing at a dizzying speed, but your 00:19:51.809 --> 00:19:54.869 skill in structured thinking, in systematic prompting, 00:19:54.950 --> 00:19:57.390 in problem decomposition, that will be your anchor. 00:19:57.589 --> 00:19:59.809 So the call to action is simple. Super simple. 00:19:59.890 --> 00:20:02.809 Your only task right now. Pick one small, achievable 00:20:02.809 --> 00:20:05.130 project that excites you and start building it 00:20:05.130 --> 00:20:05.369 today.