WEBVTT 00:00:00.000 --> 00:00:02.899 If you spend any serious time with advanced AI 00:00:02.899 --> 00:00:04.960 chatbots, you probably know the biggest headache. 00:00:05.960 --> 00:00:09.919 It's context drift. You spend ages detailing 00:00:09.919 --> 00:00:12.060 a marketing brief, defining your audience, setting 00:00:12.060 --> 00:00:14.619 the tone. Perfect. Then you ask a quick question 00:00:14.619 --> 00:00:18.339 about, say, weekend plans. And suddenly, the 00:00:18.339 --> 00:00:21.320 AI is suggesting beach resorts right in the middle 00:00:21.320 --> 00:00:23.780 of your formal client memo. You find yourself 00:00:23.780 --> 00:00:26.399 constantly repeating things, restarting chats. 00:00:26.820 --> 00:00:28.899 The AI forgets what you were doing just five 00:00:28.899 --> 00:00:32.359 minutes ago. Well, it's frustrating. Kills efficiency. 00:00:32.600 --> 00:00:33.979 It really does, yeah. And you feel like you're 00:00:33.979 --> 00:00:36.219 almost fighting the tool, instead of it helping 00:00:36.219 --> 00:00:39.439 you. But there is actually a surprisingly powerful 00:00:39.439 --> 00:00:42.200 solution, one that a lot of users, even experienced 00:00:42.200 --> 00:00:44.619 ones, seem to overlook completely. It's called 00:00:44.619 --> 00:00:47.320 Chat GPT Projects. Prolix, OK. Yeah, and this 00:00:47.320 --> 00:00:49.479 is, I think, genuinely the most effective way 00:00:49.479 --> 00:00:51.600 to get that consistency and generate really high 00:00:51.600 --> 00:00:53.399 quality results over the long term. And the great 00:00:53.399 --> 00:00:56.799 news, this is now available for everyone. Even 00:00:56.799 --> 00:00:58.820 if you're on a free account, you can use this. 00:00:59.320 --> 00:01:02.380 It is. So our mission today, it's pretty straightforward. 00:01:02.759 --> 00:01:05.099 We need to unpack what projects actually are, 00:01:05.140 --> 00:01:07.340 how they're fundamentally different from, say, 00:01:07.780 --> 00:01:11.420 custom GPTs, which people might know, and maybe 00:01:11.420 --> 00:01:13.560 most importantly, nail down the three essential 00:01:13.560 --> 00:01:16.239 parts you absolutely need to set up to create 00:01:16.239 --> 00:01:20.260 like a truly focused expert AI assistant just 00:01:20.260 --> 00:01:23.480 for your specific work. OK, that sounds exactly 00:01:23.480 --> 00:01:26.420 like what's needed. Let's dig into the mechanics 00:01:26.420 --> 00:01:29.140 then. How do they work? Let's do it. So the way 00:01:29.140 --> 00:01:31.000 I've started thinking about them is like private 00:01:31.000 --> 00:01:34.340 work rooms inside chat GPT, almost like custom 00:01:34.340 --> 00:01:36.959 -built spaces. They don't seem designed for just 00:01:36.959 --> 00:01:40.030 quick... one -off questions, you know, more for 00:01:40.030 --> 00:01:42.129 ongoing long -term work where the knowledge needs 00:01:42.129 --> 00:01:44.370 to build up. That's a perfect analogy. Yeah, 00:01:44.370 --> 00:01:46.890 a private workroom. Or think of them as dedicated 00:01:46.890 --> 00:01:49.390 AI assistants, each one totally focused on a 00:01:49.390 --> 00:01:51.870 single domain. Maybe one handles all your marketing 00:01:51.870 --> 00:01:54.010 copy, another manages your coding questions. 00:01:54.230 --> 00:01:56.510 And the key feature, the absolute core of it, 00:01:56.549 --> 00:01:59.250 is dedicated memory. Dedicated memory. Exactly. 00:01:59.430 --> 00:02:01.730 They remember everything within that project. 00:02:01.870 --> 00:02:04.010 All the past chats, every single file you've 00:02:04.010 --> 00:02:05.849 uploaded, all the instructions you've given it. 00:02:06.010 --> 00:02:09.210 It's all stored. neatly right there in that one 00:02:09.210 --> 00:02:12.370 space. And because that knowledge base is isolated, 00:02:12.969 --> 00:02:17.389 the AI stays laser focused, always on that specific 00:02:17.389 --> 00:02:19.849 topic. OK, so if these work rooms are meant to 00:02:19.849 --> 00:02:23.689 keep things separate, how do they actually stop 00:02:23.689 --> 00:02:27.389 that confusion? Like, how does it prevent my 00:02:27.389 --> 00:02:29.750 random search for a trip to Dalat from bleeding 00:02:29.750 --> 00:02:32.610 into my serious marketing strategy project? What's 00:02:32.610 --> 00:02:34.659 the barrier? Well, projects keep their knowledge 00:02:34.659 --> 00:02:36.819 totally separate. So you get focused, consistent 00:02:36.819 --> 00:02:39.479 results without that random contamination. Got 00:02:39.479 --> 00:02:41.659 it. That makes sense. Now, a lot of people listening 00:02:41.659 --> 00:02:44.060 might have used custom GPTs before to personalize 00:02:44.060 --> 00:02:45.419 things. So it's really important to understand 00:02:45.419 --> 00:02:48.340 the difference here. If a project is that workspace, 00:02:48.520 --> 00:02:52.060 that workroom, a custom GPT is more like a specialized 00:02:52.060 --> 00:02:55.120 tool. It's built for one specific repeated action. 00:02:55.319 --> 00:02:57.460 Like maybe it only generates email subject lines 00:02:57.460 --> 00:02:59.439 or it checks grammar in a very specific style. 00:02:59.539 --> 00:03:01.219 You kind of set it up once for that function, 00:03:01.219 --> 00:03:03.360 and it doesn't really change much after that. 00:03:03.620 --> 00:03:07.860 Right. OK, so GPTs are tools, but projects. They 00:03:07.860 --> 00:03:11.159 sound more like living documents or evolving 00:03:11.159 --> 00:03:13.919 workspaces. They grow and update as you use them 00:03:13.919 --> 00:03:16.340 more. You're not just doing one task in the project, 00:03:16.360 --> 00:03:18.819 you're managing a whole area of work. It's an 00:03:18.819 --> 00:03:21.259 ongoing thing. Exactly. Living documents is a 00:03:21.259 --> 00:03:23.300 great way to put it. And the core difference 00:03:23.300 --> 00:03:25.379 really comes down to where the intelligence comes 00:03:25.379 --> 00:03:28.319 from. A project's knowledge is based on your 00:03:28.319 --> 00:03:30.319 initial instructions, yes, and the files you 00:03:30.319 --> 00:03:33.639 upload, but also the entire history of every 00:03:33.639 --> 00:03:37.000 single chat within that specific project. custom 00:03:37.000 --> 00:03:39.539 GPTs. They pretty much just rely on how you set 00:03:39.539 --> 00:03:41.379 them up initially. That's their knowledge base. 00:03:41.860 --> 00:03:43.680 So the chat history itself becomes part of the 00:03:43.680 --> 00:03:45.819 project's brain, essentially. Precisely. Think 00:03:45.819 --> 00:03:48.319 about, say, managing a YouTube channel with a 00:03:48.319 --> 00:03:50.020 project. You go back to it every week, right? 00:03:50.240 --> 00:03:53.199 Brainstorming ideas, drafting scripts, because 00:03:53.199 --> 00:03:55.979 it's building that full history of every interaction, 00:03:56.460 --> 00:03:59.199 every decision, every script draft, the suggestions 00:03:59.199 --> 00:04:01.560 it gives you three months down the line. They'll 00:04:01.560 --> 00:04:04.000 be incredibly accurate because it knows your 00:04:04.000 --> 00:04:06.389 channel's whole journey, your style, your topics, 00:04:06.389 --> 00:04:09.550 what worked before. So if projects gain knowledge 00:04:09.550 --> 00:04:12.349 from the conversation history, does that mean 00:04:12.349 --> 00:04:15.150 they actually get smarter over time, more tuned 00:04:15.150 --> 00:04:17.810 in to my specific needs? Yes, they learn from 00:04:17.810 --> 00:04:19.889 every conversation, getting smarter and more 00:04:19.889 --> 00:04:23.250 precise about your task over time. Wow, OK. But 00:04:23.250 --> 00:04:26.029 that learning, that growth, it totally depends 00:04:26.029 --> 00:04:29.889 on keeping the data clean, isolated. Which brings 00:04:29.889 --> 00:04:33.410 us to, honestly, the first critical thing you 00:04:33.410 --> 00:04:37.399 need to do before anything else. The memory setting. 00:04:37.540 --> 00:04:40.579 The memory setting. Okay, why is one setting 00:04:40.579 --> 00:04:43.410 so crucial right at the start? Because of that 00:04:43.410 --> 00:04:45.589 context contamination we talked about, it's a 00:04:45.589 --> 00:04:48.250 default setting issue, really. By default, when 00:04:48.250 --> 00:04:50.689 you create a new project, it can actually access 00:04:50.689 --> 00:04:52.970 memories from your other chats, your normal, 00:04:53.069 --> 00:04:55.610 everyday, random chat GPT conversations. Oh, 00:04:55.610 --> 00:04:57.790 really? Yeah. So it might remember you were looking 00:04:57.790 --> 00:05:00.089 up, I don't know, Italian seasoning recipes last 00:05:00.089 --> 00:05:02.189 Tuesday, and now it's trying to stick a weird 00:05:02.189 --> 00:05:04.670 spice metaphor into your serious corporate financial 00:05:04.670 --> 00:05:07.269 report. That's bad news for Focus. Definitely 00:05:07.269 --> 00:05:10.290 don't want that. No. So immediately, the first 00:05:10.290 --> 00:05:13.279 thing you do... Change the setting to Project 00:05:13.279 --> 00:05:15.720 can only access its own memories. Full stop. 00:05:16.019 --> 00:05:18.500 That stops it pulling in random stuff like your 00:05:18.500 --> 00:05:21.759 recipes or that Daylat trip plan. It keeps the 00:05:21.759 --> 00:05:24.259 project's brain focused only on the project job. 00:05:24.779 --> 00:05:26.540 And honestly, this is a bit of a vulnerability 00:05:26.540 --> 00:05:29.019 moment here, but I still wrestle with prompt 00:05:29.019 --> 00:05:31.339 drift myself sometimes, even after using these 00:05:31.339 --> 00:05:34.449 tools for ages. This setting. It's like your 00:05:34.449 --> 00:05:37.170 main defense. It stops those weird, irrelevant 00:05:37.170 --> 00:05:39.709 mistakes before they even happen. It ensures 00:05:39.709 --> 00:05:42.970 your specialized AI stays specialized. So just 00:05:42.970 --> 00:05:44.829 switching that memory setting to project only, 00:05:45.329 --> 00:05:47.689 does that significantly cut down the chances 00:05:47.689 --> 00:05:50.329 of the AI making those off -topic, irrelevant 00:05:50.329 --> 00:05:52.689 mistakes? Absolutely. Isolating the memory ensures 00:05:52.689 --> 00:05:55.410 the AI stays laser -focused on its job. No distraction. 00:05:55.649 --> 00:05:57.990 OK, crucial step number one. Isolate the memory. 00:05:58.129 --> 00:06:00.129 Got it. mid -roll sponsor, read placeholder. 00:06:00.350 --> 00:06:01.850 All right, we locked down the memory. Now let's 00:06:01.850 --> 00:06:03.850 talk about building the project itself. You mentioned 00:06:03.850 --> 00:06:05.730 three pillars for setup. What's the first one? 00:06:05.910 --> 00:06:08.329 The first pillar is custom instructions. And 00:06:08.329 --> 00:06:10.170 this is so much more than just telling it to 00:06:10.170 --> 00:06:12.410 be polite. This is basically where you write 00:06:12.410 --> 00:06:15.310 the job description for your AI assistant. You're 00:06:15.310 --> 00:06:17.670 setting the rules of engagement, eliminating 00:06:17.670 --> 00:06:21.209 ambiguity. You define its role, its primary goal, 00:06:21.290 --> 00:06:23.730 the format you want answers in, and the overall 00:06:23.730 --> 00:06:26.100 tone. Most people probably just put in something 00:06:26.100 --> 00:06:28.519 vague about tone, like be friendly. Yeah. But 00:06:28.519 --> 00:06:31.240 you're saying getting really specific here is 00:06:31.240 --> 00:06:33.779 key. That's what makes it a real assistant. Exactly. 00:06:34.180 --> 00:06:36.839 Specificity is power here. Let's say you're a 00:06:36.839 --> 00:06:39.000 writer planning social media. Define the goal 00:06:39.000 --> 00:06:41.600 clearly. Help me plan content for a Facebook 00:06:41.600 --> 00:06:45.079 fan page about pets. Simple. Define the format. 00:06:45.209 --> 00:06:47.670 Always answer using bullet points. Do not write 00:06:47.670 --> 00:06:49.930 paragraphs longer than five sentences. Super 00:06:49.930 --> 00:06:52.550 clear. And the tone. Friendly, fun, enthusiastic, 00:06:52.709 --> 00:06:55.490 clearly loves animals. And you can even add special 00:06:55.490 --> 00:06:57.589 requests. Like if my request isn't clear enough, 00:06:57.769 --> 00:07:00.230 please ask me clarifying questions before answering. 00:07:00.589 --> 00:07:02.889 That saves time. Right. Prevents bad outputs. 00:07:03.230 --> 00:07:05.269 Totally. And the poodle example really shows 00:07:05.269 --> 00:07:08.769 this power. You could define the role as content 00:07:08.769 --> 00:07:11.709 manager assistant for a poodle enthusiast fan 00:07:11.709 --> 00:07:15.339 page. Then you specify the format. Each post's 00:07:15.339 --> 00:07:17.759 idea needs a catchy title, three to four bullet 00:07:17.759 --> 00:07:20.579 points explaining the idea, a clear call to action, 00:07:20.899 --> 00:07:23.199 you know, a CTA, telling people what to do next, 00:07:23.500 --> 00:07:26.120 and always suggest a relevant picture or video 00:07:26.120 --> 00:07:28.579 type. That's incredibly detailed. But that detail 00:07:28.579 --> 00:07:31.180 is what gives you consistent high -quality output 00:07:31.180 --> 00:07:34.319 every single time. No guesswork for the AI. Okay, 00:07:34.519 --> 00:07:36.079 custom instructions is Pillar 1, what's Pillar 00:07:36.079 --> 00:07:39.100 2? Pillar 2 is the knowledge base. This is the 00:07:39.100 --> 00:07:41.259 project's brain, basically, where you feed it 00:07:41.259 --> 00:07:43.839 the specific data it needs. You upload your files 00:07:43.839 --> 00:07:46.480 here, PDFs, maybe text documents, transcripts, 00:07:46.680 --> 00:07:48.800 internal guides, whatever is relevant. But the 00:07:48.800 --> 00:07:51.180 key here is quality knowledge, not just dumping 00:07:51.180 --> 00:07:53.680 everything in. Relevance is crucial. And there 00:07:53.680 --> 00:07:55.819 are limits on files, right? Yeah, there are. 00:07:55.920 --> 00:07:59.100 It varies by plan. Free users get, I think, five 00:07:59.100 --> 00:08:02.000 files per project right now. Plus in education 00:08:02.000 --> 00:08:05.399 get 25. Pro and enterprise get up to 40. But 00:08:05.399 --> 00:08:08.259 honestly, Even five is plenty if they're the 00:08:08.259 --> 00:08:11.720 right five files. Five super relevant, high quality 00:08:11.720 --> 00:08:14.819 documents beat 40 random, vaguely related ones 00:08:14.819 --> 00:08:17.139 every single time. So be selective. Don't just 00:08:17.139 --> 00:08:19.079 upload your whole hard drive. Exactly. Quality 00:08:19.079 --> 00:08:22.100 over quantity. Focus. That limit kind of encourages 00:08:22.100 --> 00:08:24.420 you to think strategically about what knowledge 00:08:24.420 --> 00:08:27.399 is truly essential. Make sense. Okay. Instructions, 00:08:27.579 --> 00:08:29.560 knowledge base. What's the third pillar? The 00:08:29.560 --> 00:08:32.700 third pillar is tools. Projects can optionally 00:08:32.700 --> 00:08:35.360 access special tools to enhance their capabilities. 00:08:35.639 --> 00:08:37.460 To of like what? Well, there's Canvas, which 00:08:37.460 --> 00:08:39.679 is really cool. It lets the AI help you create 00:08:39.679 --> 00:08:41.860 documents, presentations, even quizzes right 00:08:41.860 --> 00:08:44.460 inside the chat interface. There's deep research, 00:08:44.580 --> 00:08:46.600 which can scan hundreds of websites to gather 00:08:46.600 --> 00:08:49.039 in -depth information on a topic. Very powerful. 00:08:49.399 --> 00:08:52.200 Then you have image generation for creating visuals 00:08:52.200 --> 00:08:54.600 and the standard web search for pulling in the 00:08:54.600 --> 00:08:56.419 latest real - time info. So should we just turn 00:08:56.419 --> 00:08:58.980 all those on? No, definitely not. Here's a critical 00:08:58.980 --> 00:09:02.500 tip. Don't turn these tools on by default. Only 00:09:02.500 --> 00:09:04.919 enable them when you specifically need that capability 00:09:04.919 --> 00:09:07.480 for a particular task. Why? Because you want 00:09:07.480 --> 00:09:10.019 the project to primarily rely on the custom instructions 00:09:10.019 --> 00:09:11.700 and the knowledge base you carefully curated. 00:09:12.360 --> 00:09:15.259 That maintains focus. Use the tool selectively 00:09:15.259 --> 00:09:18.799 as needed. Given those file limits, especially 00:09:18.799 --> 00:09:22.480 for free users, is the best strategy really just 00:09:22.480 --> 00:09:25.059 picking the absolute most vital documents for 00:09:25.059 --> 00:09:27.679 that knowledge base? Absolutely. Quality, focus, 00:09:27.740 --> 00:09:30.059 and relevance are the deciding factors for useful 00:09:30.059 --> 00:09:33.460 AI output. Five great files beat 40 mediocre 00:09:33.460 --> 00:09:36.019 ones. Got it. Let's make this concrete. Can you 00:09:36.019 --> 00:09:38.179 walk us through setting up a project for a real 00:09:38.179 --> 00:09:40.360 world task? Sure. Let's take learning English. 00:09:40.480 --> 00:09:42.179 We'll build an English learning plan assistant. 00:09:42.779 --> 00:09:45.240 So, for custom instructions, the goal might be, 00:09:45.419 --> 00:09:47.360 help me improve my English listening and speaking 00:09:47.360 --> 00:09:50.299 skills. Format. Provide weekly plans in a simple, 00:09:50.480 --> 00:09:53.600 clear table. Tone. Encouraging and patient. And 00:09:53.600 --> 00:09:56.120 a crucial special request. Before creating the 00:09:56.120 --> 00:09:58.299 new week's plan, always ask me if I completed 00:09:58.299 --> 00:10:00.379 last week's goals and what challenges I faced. 00:10:00.759 --> 00:10:03.740 That builds in accountability. Nice touch. Okay, 00:10:03.779 --> 00:10:05.919 what about the knowledge base? What files go 00:10:05.919 --> 00:10:08.759 in there? We need actionable data for the AI, 00:10:08.860 --> 00:10:12.779 so maybe upload. A list of 500 common English 00:10:12.779 --> 00:10:15.000 vocabulary words the learner wants to master. 00:10:15.840 --> 00:10:17.980 A short document listing the learner's most common 00:10:17.980 --> 00:10:20.240 grammar mistakes may be based on feedback from 00:10:20.240 --> 00:10:22.679 a teacher, a transcript from an English podcast 00:10:22.679 --> 00:10:25.909 they like, to analyze style and phrasing. And 00:10:25.909 --> 00:10:28.049 maybe there's score report from a recent English 00:10:28.049 --> 00:10:31.149 test to show baseline weaknesses. OK, specific 00:10:31.149 --> 00:10:34.009 relevant data. Exactly. And here's that smart 00:10:34.009 --> 00:10:36.029 tip you mentioned earlier, a really useful technique. 00:10:36.429 --> 00:10:38.409 Before you even upload files to the project, 00:10:38.629 --> 00:10:41.470 go into a normal, separate chat GPT chat, turn 00:10:41.470 --> 00:10:44.049 on the deep research tool there, ask it something 00:10:44.049 --> 00:10:46.690 like, generate a detailed report on the most 00:10:46.690 --> 00:10:48.590 effective methods for an intermediate learner 00:10:48.590 --> 00:10:51.580 to improve English listening skills. Ah. Let 00:10:51.580 --> 00:10:54.320 it run. Do its thing. Generate that report. Then 00:10:54.320 --> 00:10:57.080 download that report as a PDF or text file. Then 00:10:57.080 --> 00:10:59.379 you upload that curated report into your English 00:10:59.379 --> 00:11:02.100 learning project's knowledge base. Ah, so you're 00:11:02.100 --> 00:11:04.500 using the AI's research capability to create 00:11:04.500 --> 00:11:06.960 high -quality knowledge for your project's brain. 00:11:07.399 --> 00:11:10.120 Precisely. You're feeding it expert -level information 00:11:10.120 --> 00:11:12.799 sourced from the web, but curated and focused. 00:11:13.000 --> 00:11:15.220 Okay, so instructions set, knowledge uploaded. 00:11:15.539 --> 00:11:17.639 What happens next? Then you just ask for your 00:11:17.639 --> 00:11:20.120 plan. Generate my English learning plan for this 00:11:20.120 --> 00:11:23.080 week. The projects looks at your goal, your common 00:11:23.080 --> 00:11:25.440 mistakes file, the vocabulary list, that research 00:11:25.440 --> 00:11:28.480 report, and it creates a specific personalized 00:11:28.480 --> 00:11:32.200 table like day one, watch this specific five 00:11:32.200 --> 00:11:36.200 minute Ted Ed video on topic X, summarize it 00:11:36.200 --> 00:11:39.080 aloud in three sentences, review vocabulary words 00:11:39.080 --> 00:11:41.440 10, 15 from the list focusing on pronunciation 00:11:41.440 --> 00:11:43.360 because it knows that's a weakness from your 00:11:43.360 --> 00:11:45.440 files. The level of personalization sounds incredible. 00:11:45.500 --> 00:11:48.039 It really is because it's all based on your specific 00:11:48.039 --> 00:11:49.960 data and goals stored right there. Okay, let's 00:11:49.960 --> 00:11:51.519 try another one. Marketing. Say we have a new 00:11:51.519 --> 00:11:54.299 product, a smart thermos. Create example. So 00:11:54.299 --> 00:11:56.720 knowledge base first. We upload maybe a PDF with 00:11:56.720 --> 00:11:58.960 all the product features, keeps drinks hot or 00:11:58.960 --> 00:12:02.340 cold for 24 hours, LED temperature display, uses 00:12:02.340 --> 00:12:06.139 316 stainless steel, and maybe another file defining 00:12:06.139 --> 00:12:09.159 our target customer. Office workers, gym goers, 00:12:09.440 --> 00:12:13.340 maybe age 22 to 35. Perfect. Clear product info, 00:12:13.740 --> 00:12:16.559 clear audience info. Now, this is where a tool 00:12:16.559 --> 00:12:18.620 like Canvas could be really useful. You could 00:12:18.620 --> 00:12:21.419 ask the project right in the chat. Create an 00:12:21.419 --> 00:12:24.559 introduction flyer for my new smart thermos using 00:12:24.559 --> 00:12:27.899 the Canvas tool. Use the features from the uploaded 00:12:27.899 --> 00:12:31.450 PDF. And it will build a draft flyer. Yeah. It'll 00:12:31.450 --> 00:12:34.190 pull those features, the 24 -hour stat, the LED 00:12:34.190 --> 00:12:36.049 screen, directly from the file you uploaded. 00:12:36.389 --> 00:12:38.190 It ensures the marketing content is accurate 00:12:38.190 --> 00:12:40.409 and based on the facts you provided. You can 00:12:40.409 --> 00:12:42.730 edit it right there. Yep. You can tweak the layout, 00:12:42.970 --> 00:12:45.899 change text, then share it instantly. get a public 00:12:45.899 --> 00:12:48.480 link, or download it as a PDF, all within that 00:12:48.480 --> 00:12:50.700 dedicated project workspace. That's efficient. 00:12:50.879 --> 00:12:53.500 And what about images? Easy. In the same project 00:12:53.500 --> 00:12:56.259 chat, you could then say, now, generate an Instagram 00:12:56.259 --> 00:12:59.200 ad image for this thermos, include the text, 00:12:59.419 --> 00:13:01.799 keep the taste all day long, and it'll create 00:13:01.799 --> 00:13:04.200 image options for you. Wow, OK. Instructions, 00:13:04.580 --> 00:13:07.899 knowledge, tools, all working together. Exactly. 00:13:08.679 --> 00:13:11.220 Whoa, just thinking about this. Imagine scaling 00:13:11.220 --> 00:13:14.799 this, like an entire company using projects. 00:13:15.179 --> 00:13:18.299 Every team, every department having these dedicated 00:13:18.299 --> 00:13:22.220 expert AI assistants all adhering to the company's 00:13:22.220 --> 00:13:24.220 internal standards using only approved knowledge 00:13:24.220 --> 00:13:27.600 bases. The level of consistency, the domain specificity 00:13:27.600 --> 00:13:31.080 across an entire enterprise. That's genuinely 00:13:31.080 --> 00:13:33.299 transformative potential right there. That is 00:13:33.299 --> 00:13:35.919 a big thought. OK, so these projects are powerful 00:13:35.919 --> 00:13:37.899 because they're focused on their internal knowledge. 00:13:38.200 --> 00:13:40.679 But do they have to be completely cut off from 00:13:40.679 --> 00:13:42.980 the outside world forever? What about new trends? 00:13:43.320 --> 00:13:45.659 Good question. No, they don't have to be totally 00:13:45.659 --> 00:13:48.360 isolated You can smartly combine that internal 00:13:48.360 --> 00:13:51.059 knowledge with fresh online information when 00:13:51.059 --> 00:13:53.360 needed So you're in your smart thermos marketing 00:13:53.360 --> 00:13:55.120 project, right? You've got your features PDF 00:13:55.120 --> 00:13:57.399 your customer profile, but you need the latest 00:13:57.399 --> 00:13:59.620 trends What's hot right now in that market? You 00:13:59.620 --> 00:14:01.279 don't leave the project. You just click that 00:14:01.279 --> 00:14:04.179 little plus icon in the chat input box Temporarily 00:14:04.179 --> 00:14:06.340 enable the web search tool just for that one 00:14:06.340 --> 00:14:09.679 query Exactly. Then you ask, what are the latest 00:14:09.679 --> 00:14:12.080 marketing trends for stainless steel lifestyle 00:14:12.080 --> 00:14:14.159 products relevant to my target audience this 00:14:14.159 --> 00:14:17.279 year? Chat GPT goes out, searches the web, and 00:14:17.279 --> 00:14:19.980 comes back with current insights. Maybe it says, 00:14:20.279 --> 00:14:22.940 focus on sustainability narratives or collaborate 00:14:22.940 --> 00:14:26.220 with KOCs. KOCs. Oh, right. Key opinion consumers 00:14:26.220 --> 00:14:29.539 or key opinion leaders. Basically, real customers 00:14:29.539 --> 00:14:33.000 or influential users who genuinely use and love 00:14:33.000 --> 00:14:36.039 the product rather than just paid celebs. Their 00:14:36.039 --> 00:14:38.120 recommendations carry a lot of weight. Got it. 00:14:38.139 --> 00:14:41.480 So the project finds these new trends. And that 00:14:41.480 --> 00:14:43.639 new information, those trends, that insight about 00:14:43.639 --> 00:14:46.299 KOCs, it gets added automatically right into 00:14:46.299 --> 00:14:48.360 that project's memory, its conversation history. 00:14:48.720 --> 00:14:50.720 So you're efficiently updating your project's 00:14:50.720 --> 00:14:52.679 knowledge with fresh, relevant data from the 00:14:52.679 --> 00:14:54.840 outside world. But you did it strategically. 00:14:55.200 --> 00:14:57.379 Then you just turn web search off again, and 00:14:57.379 --> 00:14:59.820 the focus snaps back to your core internal knowledge 00:14:59.820 --> 00:15:02.100 base. Best of both worlds. That's a smart way 00:15:02.100 --> 00:15:04.600 to keep it current without losing focus. What 00:15:04.600 --> 00:15:06.620 about other, maybe less obvious features? Any 00:15:06.620 --> 00:15:08.779 efficiency boosters people miss? Yeah, a couple 00:15:08.779 --> 00:15:10.799 of good ones. If you have a really great conversation 00:15:10.799 --> 00:15:13.200 going in your regular chat, GPT chat, and you 00:15:13.200 --> 00:15:15.320 suddenly realize, wow, this should be a project, 00:15:15.659 --> 00:15:18.139 you can actually move that entire existing chat 00:15:18.139 --> 00:15:21.519 conversation with all its context directly into 00:15:21.519 --> 00:15:24.039 one of your projects, or start a new project 00:15:24.039 --> 00:15:26.259 from that. Oh, that's handy. Instead of copy 00:15:26.259 --> 00:15:28.899 pasting. Exactly. Keeps the history intact. Saves 00:15:28.899 --> 00:15:31.480 a ton of time if you realize mid conversation 00:15:31.480 --> 00:15:34.179 that it's becoming a long term thing. And another 00:15:34.179 --> 00:15:37.360 one, especially for the mobile app users, voice 00:15:37.360 --> 00:15:40.580 input. Having a spoken back and forth conversation 00:15:40.580 --> 00:15:43.159 with your dedicated project assistant using voice, 00:15:44.019 --> 00:15:46.360 it feels incredibly natural. Almost like talking 00:15:46.360 --> 00:15:49.059 to a real specialist assistant. Really good for 00:15:49.059 --> 00:15:51.019 brainstorming on the go. OK, those are useful 00:15:51.019 --> 00:15:54.129 tips. Now, on the flip side. What are the big 00:15:54.129 --> 00:15:56.470 mistakes people make when they first start using 00:15:56.470 --> 00:15:58.570 projects? Where do they usually go wrong? Good 00:15:58.570 --> 00:16:01.309 question. There are a few common pitfalls. Number 00:16:01.309 --> 00:16:04.190 one, trying to cram everything into a single 00:16:04.190 --> 00:16:07.269 project. Don't create one massive, my whole life 00:16:07.269 --> 00:16:10.429 project. Keep them hyper -focused. One for marketing, 00:16:10.649 --> 00:16:12.990 one for coding help, one for learning Spanish. 00:16:13.509 --> 00:16:17.289 Specific domains. So specialization is key. Absolutely. 00:16:17.970 --> 00:16:21.389 Mistake number two. Rushing through or completely 00:16:21.389 --> 00:16:23.649 ignoring the custom instructions set up, people 00:16:23.649 --> 00:16:25.870 skip this, just give it a name, and wonder why 00:16:25.870 --> 00:16:28.570 it's not very helpful. Those detailed instructions 00:16:28.570 --> 00:16:31.710 are vital. They tell the AI how to be useful 00:16:31.710 --> 00:16:34.169 for you. Right. The job description. Exactly. 00:16:34.870 --> 00:16:38.240 Mistake three. The opposite of being selective 00:16:38.240 --> 00:16:40.659 with the knowledge base, just uploading tons 00:16:40.659 --> 00:16:43.820 of unrelated files. Garbage in, garbage out, 00:16:44.379 --> 00:16:47.120 basically. Quality and relevance matter way more 00:16:47.120 --> 00:16:49.580 than volume. Right. Focus, instructions, quality 00:16:49.580 --> 00:16:52.220 knowledge. What else? The big one we already 00:16:52.220 --> 00:16:54.419 covered. Forgetting to change that memory setting 00:16:54.419 --> 00:16:56.899 to project can only access its own memories. 00:16:57.240 --> 00:16:59.700 That causes instant chaos and undermines the 00:16:59.700 --> 00:17:01.840 whole point. Isolate the memory. Got it. And 00:17:01.840 --> 00:17:04.589 maybe one last thing. Starting a brand new chat 00:17:04.589 --> 00:17:06.970 inside the project every time you use it. The 00:17:06.970 --> 00:17:08.930 real power comes from building on the history. 00:17:09.369 --> 00:17:11.450 Continue the existing conversation thread so 00:17:11.450 --> 00:17:13.829 it learns and remembers context over time. Don't 00:17:13.829 --> 00:17:15.609 keep starting from scratch within the project. 00:17:15.769 --> 00:17:17.609 Okay, that makes sense. Build on the history. 00:17:18.390 --> 00:17:21.190 So let's try to wrap this up. The big idea here 00:17:21.190 --> 00:17:24.029 seems really clear now. Projects are essentially 00:17:24.029 --> 00:17:27.009 about creating expert AI assistants. defined 00:17:27.009 --> 00:17:29.950 by those three key things, your custom instructions, 00:17:30.650 --> 00:17:32.809 the specific knowledge base you feed it, and 00:17:32.809 --> 00:17:34.869 the cumulative conversation history that builds 00:17:34.869 --> 00:17:37.710 up over time. That's it. Instructions, knowledge, 00:17:38.089 --> 00:17:40.819 history. The three pillars. And the main benefit 00:17:40.819 --> 00:17:44.000 is getting consistent, reliable, high -quality 00:17:44.000 --> 00:17:46.819 results that are specifically tailored to your 00:17:46.819 --> 00:17:50.000 long -term tasks or areas of work. Exactly. So 00:17:50.000 --> 00:17:52.460 just to summarize the when to use what's question, 00:17:52.779 --> 00:17:55.200 use projects for anything that's a repeated task, 00:17:55.400 --> 00:17:58.160 requires long -term context, or needs a special 00:17:58.160 --> 00:18:00.819 dedicated knowledge base that you provide and 00:18:00.819 --> 00:18:02.900 control. Your goal with projects is building 00:18:02.900 --> 00:18:05.319 that expert AI assistant that grows and learns 00:18:05.319 --> 00:18:08.150 alongside you for a specific domain. regular 00:18:08.150 --> 00:18:10.809 chats are still great use them for quick one 00:18:10.809 --> 00:18:12.630 -off questions things that don't need history 00:18:12.630 --> 00:18:14.250 or special knowledge you know what's the capital 00:18:14.250 --> 00:18:16.450 of Australia or convert these measurements quick 00:18:16.450 --> 00:18:20.049 in quick out got it projects for depth and consistency 00:18:20.049 --> 00:18:22.930 regular chat for quick hits okay here's a final 00:18:22.930 --> 00:18:24.769 thought something that really struck me about 00:18:24.769 --> 00:18:28.869 this memory system the truly mind -bending power 00:18:28.869 --> 00:18:32.509 maybe months down the line is the project's ability 00:18:32.509 --> 00:18:36.039 to gently correct you based on some detail from 00:18:36.039 --> 00:18:38.559 a conversation you've completely forgotten. Like 00:18:38.559 --> 00:18:40.339 take that English learning system we talked about. 00:18:40.579 --> 00:18:42.940 Maybe you're drafting something and you use the 00:18:42.940 --> 00:18:46.940 word effect when it should be effect. The project 00:18:46.940 --> 00:18:48.660 could potentially say, hang on, remember back 00:18:48.660 --> 00:18:50.500 in week three, we discussed the difference between 00:18:50.500 --> 00:18:53.619 affect and effect. Based on our conversation 00:18:53.619 --> 00:18:56.299 then, you determined effect is the noun you need 00:18:56.299 --> 00:18:59.119 here. Wow. That correction didn't come from the 00:18:59.119 --> 00:19:01.200 grammar file you uploaded initially. It came 00:19:01.200 --> 00:19:03.319 purely from the memory of your past conversations 00:19:03.319 --> 00:19:07.000 within that project. Well, that's a true assistant. 00:19:07.460 --> 00:19:09.920 That is powerful. It's not just recalling files. 00:19:10.079 --> 00:19:13.079 It's recalling the learning process itself. OK, 00:19:13.079 --> 00:19:15.019 so your action plan, if you're listening, it's 00:19:15.019 --> 00:19:17.680 simple. Don't overthink it. Just create one project 00:19:17.680 --> 00:19:19.900 today. Call it something simple and actionable, 00:19:20.359 --> 00:19:23.059 like my blog post ideas or meeting notes analysis. 00:19:23.400 --> 00:19:27.210 Set just three basic custom instructions. Maybe 00:19:27.210 --> 00:19:30.029 it's role, main goal, and preferred output format. 00:19:30.349 --> 00:19:32.509 Add only one or two key files to the knowledge 00:19:32.509 --> 00:19:35.289 base, maybe your brand style guide or notes from 00:19:35.289 --> 00:19:37.430 a recent important meeting, and then just commit 00:19:37.430 --> 00:19:39.769 to using that project consistently for one week. 00:19:39.990 --> 00:19:42.750 For anything related to that topic, see how that 00:19:42.750 --> 00:19:45.329 isolated memory, that growing history, starts 00:19:45.329 --> 00:19:47.509 to actually serve you, how the outputs get better, 00:19:47.789 --> 00:19:50.529 more tailored. Great starting point. Apply this 00:19:50.529 --> 00:19:52.789 knowledge. Set up that first some project today. 00:19:52.950 --> 00:19:55.269 Don't wait. See for yourself how quickly it can 00:19:55.269 --> 00:19:57.349 improve your consistency and the quality of what 00:19:57.349 --> 00:19:59.430 you get back from the AI. It really can make 00:19:59.430 --> 00:20:00.630 a difference. Out to your own music.