WEBVTT 00:00:00.000 --> 00:00:02.640 OK, let's unpack this. Have you ever gone to 00:00:02.640 --> 00:00:06.839 use your favorite AI assistant, you know, for 00:00:06.839 --> 00:00:10.199 writing or brainstorming, maybe just planning 00:00:10.199 --> 00:00:14.279 your day? And it felt, well, different. Like 00:00:14.279 --> 00:00:16.100 it had a, I don't know, brain swap overnight. 00:00:16.260 --> 00:00:18.960 One day it's Bob, your perfect, like intuitive 00:00:18.960 --> 00:00:21.420 personal assistant. Right. And the next day, 00:00:21.420 --> 00:00:24.920 it's just not Bob. It's talking in riddles, maybe 00:00:24.920 --> 00:00:27.339 booking flights to Peru instead of lunch. Yeah, 00:00:27.399 --> 00:00:29.239 I've heard stories like that. That feeling of 00:00:29.239 --> 00:00:31.780 unpredictability, you know, not quite being in 00:00:31.780 --> 00:00:34.619 control with these cloud AI services. It's a 00:00:34.619 --> 00:00:36.380 real frustration for a lot of people. Definitely. 00:00:36.640 --> 00:00:39.859 So our mission today. for this deep dive is to 00:00:39.859 --> 00:00:42.299 unpack why this keeps happening and maybe more 00:00:42.299 --> 00:00:44.439 importantly what you can actually do about it 00:00:44.439 --> 00:00:47.439 how to get some stability control and yeah privacy 00:00:47.439 --> 00:00:49.560 back sounds good we've been diving into this 00:00:49.560 --> 00:00:52.799 really insightful piece uh local ai versus cloud 00:00:52.799 --> 00:00:55.920 ai your guide to ai stability and control and 00:00:55.920 --> 00:00:57.539 it's fascinating because this isn't just about 00:00:57.539 --> 00:01:00.320 like minor software updates changing a button 00:01:00.320 --> 00:01:03.399 color no it feels much more fundamental it's 00:01:03.399 --> 00:01:06.180 about how we interact with these honestly pretty 00:01:06.180 --> 00:01:07.900 powerful tools and whether they're really working 00:01:07.900 --> 00:01:10.819 for us you know exactly i mean just recently 00:01:10.819 --> 00:01:13.200 it felt like a flash storm of ai announcements 00:01:13.200 --> 00:01:18.079 didn't it google dropped jules vo3 flow ai then 00:01:18.079 --> 00:01:21.599 gemini's native audio gemma 3n anthropic rolled 00:01:21.599 --> 00:01:26.819 out claude's sonnet 4. Opus 4, plus Mistral's 00:01:26.819 --> 00:01:30.480 DevStraw, Microsoft pushing GitHub co -pilot 00:01:30.480 --> 00:01:34.060 updates. It was just, wow, a flood of new names 00:01:34.060 --> 00:01:36.620 and features. Felt kind of overwhelming, actually. 00:01:36.859 --> 00:01:39.079 Overwhelming is absolutely word. It's that AI 00:01:39.079 --> 00:01:41.739 rollercoaster experience the source talks about. 00:01:41.920 --> 00:01:44.060 Sometimes it's just pure magic, like incredibly 00:01:44.060 --> 00:01:46.640 useful. Other times it's a total Wreck -It -Ralph 00:01:46.640 --> 00:01:49.019 disaster. It just breaks everything. And that 00:01:49.019 --> 00:01:51.640 good day magic. It's absolutely real. The source 00:01:51.640 --> 00:01:53.680 really nails this. You get that clock. four opus 00:01:53.680 --> 00:01:56.260 model that just perfectly crafts a really sensitive 00:01:56.260 --> 00:02:00.219 client email or like google's gemini explaining 00:02:00.219 --> 00:02:02.459 quantum entanglement to a 10 year old with this 00:02:02.459 --> 00:02:08.460 charming story about magic twin puppies oh last 00:02:08.460 --> 00:02:11.740 slightly and suddenly a really complex concept 00:02:11.740 --> 00:02:14.439 just clicks these moments they feel transformative 00:02:14.439 --> 00:02:16.840 we celebrate them right but that's only half 00:02:16.840 --> 00:02:18.939 the story right because then you get the bad 00:02:18.939 --> 00:02:24.219 day That's when the magic just evaporates and 00:02:24.219 --> 00:02:27.599 your trusty tool, that partner you rely on, suddenly 00:02:27.599 --> 00:02:30.360 turns into, well, a wrecking ball. Yeah. Like, 00:02:30.360 --> 00:02:32.060 imagine you're a writer. You've got this great 00:02:32.060 --> 00:02:34.879 system going with an older GPT -4 version. Maybe 00:02:34.879 --> 00:02:37.240 it helps you brainstorm fantastic characters. 00:02:37.319 --> 00:02:40.039 It's got this creative spark. Then, boom, a new 00:02:40.039 --> 00:02:42.120 improved update rolls out. You go to use your 00:02:42.120 --> 00:02:44.599 trusted brainstorming partner and something is 00:02:44.599 --> 00:02:47.219 just wrong. What happened? It just spits out 00:02:47.219 --> 00:02:50.699 the most boring, generic, cliched ideas imaginable. 00:02:51.050 --> 00:02:53.229 It actually starts draining your creativity. 00:02:53.270 --> 00:02:55.569 It's not just frustrating. It feels like a real 00:02:55.569 --> 00:02:57.889 betrayal almost. I can see that, especially if 00:02:57.889 --> 00:02:59.710 you've built workflows on it. Exactly. If you've 00:02:59.710 --> 00:03:01.590 built entire workflows, maybe even parts of your 00:03:01.590 --> 00:03:03.949 business, on a specific AI's behavior, it can 00:03:03.949 --> 00:03:06.250 feel like you're suddenly moving backward, but 00:03:06.250 --> 00:03:08.689 like at high speed. All that progress you made, 00:03:08.750 --> 00:03:12.229 just gone. And that brings us right to the mystery 00:03:12.229 --> 00:03:15.400 of the changed AI. Why does this keep happening? 00:03:15.580 --> 00:03:18.360 We can't just pop the hood and look inside. For 00:03:18.360 --> 00:03:21.180 most of us, these models are total black boxes. 00:03:21.219 --> 00:03:24.139 Right. Sealed tight. But the source gives us 00:03:24.139 --> 00:03:26.939 some really good clues. Think of it like a global 00:03:26.939 --> 00:03:29.560 restaurant chain with a secret recipe. Okay. 00:03:29.620 --> 00:03:31.759 The company's constantly trying to improve it, 00:03:31.819 --> 00:03:34.699 maybe make it cheaper or faster to produce. But 00:03:34.699 --> 00:03:37.259 their idea of improvement might be the exact 00:03:37.259 --> 00:03:39.659 opposite of what you loved about it. Hmm. That 00:03:39.659 --> 00:03:41.919 makes sense. So what are some of those like key 00:03:41.919 --> 00:03:45.000 reasons for these frustrating shifts? Well, one 00:03:45.000 --> 00:03:48.039 huge reason is model tuning. The engineers behind 00:03:48.039 --> 00:03:50.580 the scenes, they might add a new global instruction, 00:03:50.919 --> 00:03:53.500 something like be more helpful and harmless. 00:03:53.680 --> 00:03:55.879 Sounds good on paper. Yeah, it could make it 00:03:55.879 --> 00:03:58.319 great for basic customer service chats, maybe. 00:03:58.400 --> 00:04:01.169 But suddenly it's. terrible at nuanced creative 00:04:01.169 --> 00:04:03.750 writing or complex problem solving. It's like 00:04:03.750 --> 00:04:06.669 taking all the salt and spices out of a soup 00:04:06.669 --> 00:04:09.750 to make it healthier. Chuckles. Technically safer, 00:04:09.830 --> 00:04:12.789 perhaps, but bland, you know, pretty useless 00:04:12.789 --> 00:04:15.639 if you wanted flavor. Right. Another major driver 00:04:15.639 --> 00:04:18.980 is cost and speed. Let's be real, running these 00:04:18.980 --> 00:04:21.980 massive AI models is incredibly expensive. I 00:04:21.980 --> 00:04:25.160 bet. So the company might decide to use, let's 00:04:25.160 --> 00:04:27.579 say, cheaper ingredients or cook much faster 00:04:27.579 --> 00:04:30.399 to serve more people at a lower cost. For simple 00:04:30.399 --> 00:04:33.500 tasks, you might not even notice. Okay. But ask 00:04:33.500 --> 00:04:35.939 it something complex, something requiring deep 00:04:35.939 --> 00:04:38.870 thought. The answers become shallow, generic. 00:04:39.310 --> 00:04:42.610 Your five -star slow -cooked meal basically gets 00:04:42.610 --> 00:04:45.009 replaced with fast food. The quality, it just 00:04:45.009 --> 00:04:47.230 vanishes. Oh, I totally get that. Okay. So they 00:04:47.230 --> 00:04:49.689 optimize for volume maybe and lose the quality 00:04:49.689 --> 00:04:51.990 for the harder stuff. Precisely. Then there are 00:04:51.990 --> 00:04:54.829 training data updates. The AI learns from this 00:04:54.829 --> 00:04:57.629 colossal library of text and code. When they 00:04:57.629 --> 00:05:00.050 add new information, it learns new things, but 00:05:00.050 --> 00:05:02.910 it can sometimes forget or misapply old ones. 00:05:03.110 --> 00:05:05.569 Like how? the source uses this funny analogy 00:05:05.569 --> 00:05:08.829 like a chef who goes to italy comes back obsessed 00:05:08.829 --> 00:05:10.910 with balsamic vinegar and starts putting it on 00:05:10.910 --> 00:05:14.410 everything yeah steak salad even like ice cream 00:05:14.410 --> 00:05:16.449 they learned a new skill maybe but completely 00:05:16.449 --> 00:05:18.110 forgot where it's appropriate to use it okay 00:05:18.110 --> 00:05:20.410 that paints a picture balsamic ice cream yeah 00:05:20.410 --> 00:05:23.470 got it And finally, there are the safety filters. 00:05:23.709 --> 00:05:26.370 These are obviously important, designed to stop 00:05:26.370 --> 00:05:29.250 the AI from saying harmful or inappropriate things. 00:05:29.430 --> 00:05:32.310 But sometimes these filters are just way too 00:05:32.310 --> 00:05:34.470 aggressive. They can make the AI seem nervous, 00:05:34.550 --> 00:05:37.610 overly cautious, almost afraid to answer perfectly 00:05:37.610 --> 00:05:40.750 normal questions. Its personality just gets erased, 00:05:41.009 --> 00:05:44.389 replaced by this bland, apologetic, corporate 00:05:44.389 --> 00:05:47.350 voice. It's like a restaurant so terrified of 00:05:47.350 --> 00:05:50.720 allergies that they ban peanuts, gluten. Dairy, 00:05:50.819 --> 00:05:53.000 salt. You can't even get a simple sandwich with 00:05:53.000 --> 00:05:55.519 cheese anymore because it seems too risky. My 00:05:55.519 --> 00:05:57.759 poor sandwich. But you're right. It really comes 00:05:57.759 --> 00:06:00.240 down to the user experience. My feeling is the 00:06:00.240 --> 00:06:02.339 only thing that counts, isn't it? Exactly. When 00:06:02.339 --> 00:06:04.480 your favorite tool stops working for you the 00:06:04.480 --> 00:06:06.600 way you need it to, it's a real loss. It doesn't 00:06:06.600 --> 00:06:08.459 matter what anyone else says or if the update 00:06:08.459 --> 00:06:11.139 technically improved some other metric. But, 00:06:11.139 --> 00:06:13.740 you know, it's worth pointing out. Not everyone 00:06:13.740 --> 00:06:16.319 is on this crazy roller coaster, right? The source 00:06:16.319 --> 00:06:19.600 talks about the mech crowd. Ah, yes, the mech 00:06:19.600 --> 00:06:21.439 crowd. These are the folks who try a new AI update 00:06:21.439 --> 00:06:24.600 and just shrug. They honestly don't notice much 00:06:24.600 --> 00:06:26.879 difference. Why is that? Well, this kind of leads 00:06:26.879 --> 00:06:30.040 into the big secret the source reveals. Even 00:06:30.040 --> 00:06:33.360 when we think we're using the same named AI model 00:06:33.360 --> 00:06:36.959 like GPT -4 or Claude 3, we're actually living 00:06:36.959 --> 00:06:40.350 in completely different AI worlds. How so? it's 00:06:40.350 --> 00:06:43.329 all about how you use the ai right what you need 00:06:43.329 --> 00:06:45.949 from it is deeply personal a software developer 00:06:45.949 --> 00:06:49.449 using say the mistral api to build an application 00:06:49.449 --> 00:06:52.329 they live on a totally different planet from 00:06:52.329 --> 00:06:54.930 a marketer using the chat gpt website for social 00:06:54.930 --> 00:06:58.430 media ideas their definition of good or useful 00:06:58.430 --> 00:07:00.790 is completely different and this difference also 00:07:00.790 --> 00:07:03.610 explains the met crowd the people who don't notice 00:07:03.610 --> 00:07:06.250 big updates ah because they aren't using the 00:07:06.250 --> 00:07:08.410 affected features pretty much if your task is 00:07:08.410 --> 00:07:10.860 relatively simple drafting a basic email, checking 00:07:10.860 --> 00:07:13.160 grammar, maybe summarizing a news article. Well, 00:07:13.259 --> 00:07:15.240 current models are already very good at that. 00:07:15.360 --> 00:07:17.600 A new update might tweak performance, maybe turn 00:07:17.600 --> 00:07:20.319 a C plus answer into a solid B, but you likely 00:07:20.319 --> 00:07:22.720 won't even perceive that small improvement. So 00:07:22.720 --> 00:07:24.500 if you're not pushing the AI to its absolute 00:07:24.500 --> 00:07:27.699 limits. Exactly. You're only ever seeing a fraction 00:07:27.699 --> 00:07:30.610 of what it can do. The part of the secret recipe 00:07:30.610 --> 00:07:32.689 that got changed might have been for a fancy 00:07:32.689 --> 00:07:35.329 dish you never order anyway. Gotcha. And remember, 00:07:35.529 --> 00:07:37.589 what we get from these companies, it's not just 00:07:37.589 --> 00:07:40.889 the core AI brain, it's a whole product. The 00:07:40.889 --> 00:07:44.230 website speed, the user interface design, even 00:07:44.230 --> 00:07:46.589 hidden company instructions or prompts they add 00:07:46.589 --> 00:07:49.730 behind the scenes, they all influence the final 00:07:49.730 --> 00:07:52.089 behavior you see. So it really is a tailored 00:07:52.089 --> 00:07:54.230 experience, or at least a productized one, whether 00:07:54.230 --> 00:07:57.660 we realize it or not. Yeah. Okay, so... Given 00:07:57.660 --> 00:08:00.980 all this, this chaos and unpredictability, the 00:08:00.980 --> 00:08:03.980 source material presents an escape plan. Yes, 00:08:04.040 --> 00:08:06.779 the escape plan. It's about basically firing 00:08:06.779 --> 00:08:09.560 the cloud and hiring a local AI. What does that 00:08:09.560 --> 00:08:12.040 actually mean in practice? It means using AI 00:08:12.040 --> 00:08:14.139 models that run directly on your own computer, 00:08:14.339 --> 00:08:16.139 your laptop, your desktop. It's like getting 00:08:16.139 --> 00:08:18.079 off that unpredictable roller coaster for good. 00:08:18.240 --> 00:08:20.540 Okay. And this is where these small local AI 00:08:20.540 --> 00:08:22.920 models are becoming a massive game changer. The 00:08:22.920 --> 00:08:24.920 first and honestly probably the most important 00:08:24.920 --> 00:08:27.519 benefit is unbreakable. stability. Unbreakable 00:08:27.519 --> 00:08:30.439 stability. Sounds good. Yeah. A local AI model 00:08:30.439 --> 00:08:32.220 doesn't change unless you decide to change it. 00:08:32.279 --> 00:08:35.399 No surprises, no forced updates overnight that 00:08:35.399 --> 00:08:38.200 completely shatter your workflow. You find a 00:08:38.200 --> 00:08:39.740 model that works perfectly for your needs, you 00:08:39.740 --> 00:08:42.980 install it, and it stays exactly that way. Reliable. 00:08:43.159 --> 00:08:45.340 It's like owning your favorite cookbook instead 00:08:45.340 --> 00:08:47.480 of subscribing to a magazine that changes all 00:08:47.480 --> 00:08:49.580 the recipes every month. That's a perfect analogy 00:08:49.580 --> 00:08:52.360 from the source. Yeah. For building dependable 00:08:52.360 --> 00:08:54.879 tools or workflows, that stability is just... 00:08:55.080 --> 00:08:58.539 Priceless. Wow. Okay. That alone sounds pretty 00:08:58.539 --> 00:09:01.720 revolutionary for anyone dealing with these constant 00:09:01.720 --> 00:09:05.580 unexpected shifts. What else makes local AI compelling? 00:09:06.100 --> 00:09:08.639 Well, second, you get complete control. When 00:09:08.639 --> 00:09:10.779 the AI runs on your machine, you're the boss. 00:09:10.820 --> 00:09:12.759 You're the head chef, right? Right. You can adjust 00:09:12.759 --> 00:09:15.019 its settings to perfectly match your needs. Want 00:09:15.019 --> 00:09:17.200 it to be more creative for brainstorming wild 00:09:17.200 --> 00:09:19.659 ideas? You can usually tune that. Need it to 00:09:19.659 --> 00:09:22.200 be super precise and factual for technical work? 00:09:22.240 --> 00:09:24.139 You can often dial that in, too. So you can really 00:09:24.139 --> 00:09:27.350 customize it. Yes, you can often fine -tune its 00:09:27.350 --> 00:09:30.309 personality, its tone, its output style to create 00:09:30.309 --> 00:09:32.870 the perfect assistant for you, not just some 00:09:32.870 --> 00:09:36.309 generic one -size -fits -all model designed for 00:09:36.309 --> 00:09:39.210 the masses. Okay, stability, control. What's 00:09:39.210 --> 00:09:41.490 the third? And third, maybe the biggest one for 00:09:41.490 --> 00:09:45.710 some people, your privacy is guaranteed. When 00:09:45.710 --> 00:09:48.250 you use a big online AI, you're sending all your 00:09:48.250 --> 00:09:50.710 data, your questions, your documents, maybe even 00:09:50.710 --> 00:09:53.649 sensitive business secrets to a giant corporation 00:09:53.649 --> 00:09:56.649 servers. Who knows where it goes? Yeah, that's 00:09:56.649 --> 00:09:59.149 always a concern. With a local model. Everything 00:09:59.149 --> 00:10:01.649 stays right there on your computer. Your kitchen 00:10:01.649 --> 00:10:04.009 has a locked door, basically. Your secret recipes 00:10:04.009 --> 00:10:07.049 stay safe with you. No worries about data breaches 00:10:07.049 --> 00:10:09.490 or your prompts being used to train future models 00:10:09.490 --> 00:10:12.129 without your consent. Dependable, customizable, 00:10:12.450 --> 00:10:14.970 and private. Exactly. It's like having your own 00:10:14.970 --> 00:10:16.950 expert assistant sitting right there in the room 00:10:16.950 --> 00:10:18.789 with you, working just for you under your rules. 00:10:18.909 --> 00:10:21.210 Okay, that sounds genuinely incredible, but... 00:10:21.389 --> 00:10:24.450 My immediate reaction is, isn't this super complicated? 00:10:24.470 --> 00:10:27.570 Yeah. And expensive. Am I going to need like 00:10:27.570 --> 00:10:31.110 a NASA level supercomputer or a Ph .D. in coding? 00:10:31.309 --> 00:10:33.710 That's a totally common and understandable fear. 00:10:33.830 --> 00:10:35.830 And honestly, a few years ago, you might have 00:10:35.830 --> 00:10:38.970 been right. But the reality is changing really, 00:10:39.090 --> 00:10:42.370 really fast now. OK. The source debunks two big 00:10:42.370 --> 00:10:46.899 myths here. Myth number one. You need that NASA 00:10:46.899 --> 00:10:48.940 supercomputer. This used to be true, but not 00:10:48.940 --> 00:10:51.679 anymore. The world of small language models, 00:10:51.840 --> 00:10:55.220 or SLMs, is just exploding. SLMs, small language 00:10:55.220 --> 00:10:57.919 model. Yep. These are smaller, much more efficient 00:10:57.919 --> 00:11:00.379 models designed specifically to run on regular 00:11:00.379 --> 00:11:02.799 consumer hardware, the kind of computer many 00:11:02.799 --> 00:11:05.960 people already have. You likely don't need a 00:11:05.960 --> 00:11:08.039 machine that costs tens of thousands of dollars. 00:11:08.299 --> 00:11:09.740 But what kind of computer are we talking about? 00:11:09.919 --> 00:11:12.179 Generally, if your computer can handle modern 00:11:12.179 --> 00:11:14.820 video games or, say, video editing smoothly, 00:11:15.240 --> 00:11:17.220 it's probably powerful enough to run some pretty 00:11:17.220 --> 00:11:20.259 capable local AIs. Oh, okay. That's more accessible 00:11:20.259 --> 00:11:22.980 than I thought. What about myth number two, needing 00:11:22.980 --> 00:11:26.000 to be a coding genius? Right. Myth number two, 00:11:26.019 --> 00:11:29.080 you need to be a coding genius to install and 00:11:29.080 --> 00:11:31.269 run it. Also kind of true back in the day, but 00:11:31.269 --> 00:11:33.590 again, things are changing. There's this passionate 00:11:33.590 --> 00:11:36.110 community of developers out there making local 00:11:36.110 --> 00:11:39.090 AI much more accessible. They're building simple, 00:11:39.230 --> 00:11:41.950 easy to use applications. Things with friendly 00:11:41.950 --> 00:11:44.649 interfaces that handle all the complicated setup 00:11:44.649 --> 00:11:47.250 stuff for you. Think about building a website. 00:11:47.490 --> 00:11:50.009 15 years ago, you pretty much had to be a coding 00:11:50.009 --> 00:11:52.950 expert, right? Today, you can use drag and drop 00:11:52.950 --> 00:11:56.049 tools like Squarespace or Wix. Right. Much easier 00:11:56.049 --> 00:11:59.519 now. Local AI is... kind of heading in that same 00:11:59.519 --> 00:12:02.379 direction. People are creating easy to use installers, 00:12:02.399 --> 00:12:06.120 graphical interfaces, making it way less intimidating. 00:12:06.379 --> 00:12:09.360 That's a huge relief for many, I imagine. So 00:12:09.360 --> 00:12:12.340 what's a realistic first step then? For someone 00:12:12.340 --> 00:12:14.940 who's curious, listening right now, but maybe 00:12:14.940 --> 00:12:16.799 still a little intimidated. Okay. Your first 00:12:16.799 --> 00:12:18.960 step isn't to install anything at all. Don't 00:12:18.960 --> 00:12:20.600 worry about that yet. Phew. Okay. It's just to 00:12:20.600 --> 00:12:23.730 get curious. Seriously. Go on YouTube or search 00:12:23.730 --> 00:12:26.570 online for phrases like how to run an AI on my 00:12:26.570 --> 00:12:29.710 computer or maybe easy local AI setup for beginners. 00:12:29.950 --> 00:12:32.309 Just look. Just watch a few videos. See what 00:12:32.309 --> 00:12:34.330 the process actually looks like today. You'll 00:12:34.330 --> 00:12:36.649 probably be surprised how many friendly, clear 00:12:36.649 --> 00:12:38.529 guides are already out there. You don't have 00:12:38.529 --> 00:12:41.370 to do anything yet. Just look around. Knowledge 00:12:41.370 --> 00:12:43.990 is the first step to taking back control, right? 00:12:44.190 --> 00:12:47.409 That's a great, really actionable tip. Just watch 00:12:47.409 --> 00:12:50.379 a video. Okay. So knowing all this now. How do 00:12:50.379 --> 00:12:54.580 we navigate this constantly evolving AI landscape? 00:12:54.899 --> 00:12:57.860 Our source talks about needing a new game plan 00:12:57.860 --> 00:13:01.200 for a less annoying AI future, basically. Moving 00:13:01.200 --> 00:13:04.000 from being just a passive consumer to more of 00:13:04.000 --> 00:13:06.440 an active builder. Yeah, exactly. It's about 00:13:06.440 --> 00:13:08.639 becoming more intentional. First, the source 00:13:08.639 --> 00:13:11.159 suggests, stop chasing every new shiny thing 00:13:11.159 --> 00:13:14.460 and be a healthy skeptic. Resist the hype. Totally. 00:13:14.679 --> 00:13:17.440 When a new AI model or update is announced with 00:13:17.440 --> 00:13:20.490 huge fanfare, just... Take a deep breath. Don't 00:13:20.490 --> 00:13:22.750 immediately jump on the bandwagon. Let the marketing 00:13:22.750 --> 00:13:25.289 hype die down a bit. Wait and see. Wait for independent 00:13:25.289 --> 00:13:28.529 people, real users to share honest results. Not 00:13:28.529 --> 00:13:30.429 just the cherry -picked examples from the company's 00:13:30.429 --> 00:13:32.789 big launch demo. See how it actually performs 00:13:32.789 --> 00:13:35.129 on the difficult real -world tasks that matter 00:13:35.129 --> 00:13:37.409 specifically to you. That makes so much sense. 00:13:37.490 --> 00:13:39.129 Don't just buy into the hype immediately. What's 00:13:39.129 --> 00:13:41.389 next in the game plan? Second, keep your own 00:13:41.389 --> 00:13:43.490 square card and become your own expert. This 00:13:43.490 --> 00:13:46.389 is crucial. Don't rely solely on some tech blogger's 00:13:46.389 --> 00:13:49.210 opinion of whether an AI is better or worse. 00:13:49.519 --> 00:13:52.440 Because the only opinion that truly matters is 00:13:52.440 --> 00:13:55.659 yours, based on the actual work you do. Become 00:13:55.659 --> 00:13:58.399 your own expert by creating a personal test suite, 00:13:58.600 --> 00:14:01.940 basically. Like specific prompts. Exactly. When 00:14:01.940 --> 00:14:03.799 you find a prompt that gives you a fantastic 00:14:03.799 --> 00:14:07.029 result on a task you care about, save it. Keep 00:14:07.029 --> 00:14:09.289 a small collection of prompts for your most important 00:14:09.289 --> 00:14:12.529 and challenging tasks. That way, when a new model 00:14:12.529 --> 00:14:15.090 comes out or your current one gets updated, you 00:14:15.090 --> 00:14:17.230 can test it on your personal example. And track 00:14:17.230 --> 00:14:19.669 the results. Yeah, your scorecard could be super 00:14:19.669 --> 00:14:22.990 simple. Date, model name, the prompt you used, 00:14:23.190 --> 00:14:25.830 the result it gave, and your rating. Maybe a 00:14:25.830 --> 00:14:29.929 quick note. 55, super clear and funny. Or 25, 00:14:30.090 --> 00:14:33.179 generic junk. That idea of a personalized benchmark 00:14:33.179 --> 00:14:35.779 is brilliant, actually. It helps you really see 00:14:35.779 --> 00:14:37.940 what works for you. But I wonder, for someone 00:14:37.940 --> 00:14:41.379 maybe juggling multiple AI tools, how realistic 00:14:41.379 --> 00:14:44.120 is it to consistently maintain a detailed scorecard 00:14:44.120 --> 00:14:46.799 like that? Does it become another chore? That's 00:14:46.799 --> 00:14:48.679 a really fair question. And look, it doesn't 00:14:48.679 --> 00:14:51.000 have to be some massive exhaustive spreadsheet 00:14:51.000 --> 00:14:53.539 tracking every single interaction. Even just 00:14:53.539 --> 00:14:55.759 keeping track of, say, your top three or five 00:14:55.759 --> 00:14:58.159 most critical tasks or prompts can give you incredible 00:14:58.159 --> 00:15:00.620 insight over time. It's less about rigid data 00:15:00.620 --> 00:15:02.679 collection and more about developing a personal 00:15:02.679 --> 00:15:06.179 intuition, almost a feel for what truly works 00:15:06.179 --> 00:15:08.320 best for your specific needs, rather than just 00:15:08.320 --> 00:15:10.500 chasing general reviews online. Makes sense. 00:15:10.679 --> 00:15:12.899 Build your own expertise. Right. And that leads 00:15:12.899 --> 00:15:15.320 nicely into the third point. Use a variety of 00:15:15.320 --> 00:15:19.539 tools. and build a toolbox don't be blindly loyal 00:15:19.539 --> 00:15:22.919 to just one ai brand or model why is that that's 00:15:22.919 --> 00:15:24.519 like trying to build an entire house using only 00:15:24.519 --> 00:15:27.580 a hammer right right just not efficient no single 00:15:27.580 --> 00:15:30.240 model currently at least is the absolute best 00:15:30.240 --> 00:15:33.440 at everything specialization exactly one might 00:15:33.440 --> 00:15:35.299 be a genius at creative writing while another 00:15:35.299 --> 00:15:37.700 is a champion at analyzing complex data sets 00:15:37.700 --> 00:15:40.240 a third might be unbeatable for generating code 00:15:40.240 --> 00:15:44.450 so Build a small toolbox of different AIs. Learn 00:15:44.450 --> 00:15:46.570 their strengths and weaknesses. Use the right 00:15:46.570 --> 00:15:49.450 tool for the right job. Okay, so be discerning. 00:15:49.450 --> 00:15:51.710 Be your own expert. Don't put all your eggs in 00:15:51.710 --> 00:15:54.509 one AI basket. And then finally, that brings 00:15:54.509 --> 00:15:58.330 us back to local AI. Exactly. The fourth point 00:15:58.330 --> 00:16:02.669 is take one small step towards local AI. You 00:16:02.669 --> 00:16:05.629 know now from our chat that local AI is pretty 00:16:05.629 --> 00:16:08.009 much the ultimate path if you want real stability 00:16:08.009 --> 00:16:11.230 and control. Your journey there can start today. 00:16:11.570 --> 00:16:14.730 With that YouTube search. Yep. As we talked about, 00:16:14.889 --> 00:16:18.990 your first step is simply to get curious. Spend 00:16:18.990 --> 00:16:21.309 just 30 minutes this week watching one of those 00:16:21.309 --> 00:16:23.789 video guides about setting up local AI. That's 00:16:23.789 --> 00:16:26.169 it. No pressure. No pressure at all. You're not 00:16:26.169 --> 00:16:27.970 committing to anything. You're just opening the 00:16:27.970 --> 00:16:29.750 door, tiny crack to see what's actually on the 00:16:29.750 --> 00:16:32.690 other side. This small investment of time will 00:16:32.690 --> 00:16:34.590 empower you with knowledge and show you that 00:16:34.590 --> 00:16:37.049 taking back control is probably way more possible 00:16:37.049 --> 00:16:39.789 than you think. This has been such an eye -opening 00:16:39.789 --> 00:16:41.669 deep dive. It really brings it all together, 00:16:41.769 --> 00:16:44.549 doesn't it? The AI world is still so young, definitely 00:16:44.549 --> 00:16:48.629 messy, and wonderfully chaotic in a way. Chuckles, 00:16:48.750 --> 00:16:50.809 it is. It feels more like an exciting but kind 00:16:50.809 --> 00:16:52.929 of unpredictable science experiment sometimes 00:16:52.929 --> 00:16:55.509 rather than a set of finished, polished products. 00:16:55.870 --> 00:16:57.870 That's a good way to put it. But the key takeaway 00:16:57.870 --> 00:17:00.610 is you do not have to be a victim of that chaos. 00:17:01.169 --> 00:17:04.019 By being smart, keeping your own notes, building 00:17:04.019 --> 00:17:06.660 that toolbox and maybe slowly starting to explore 00:17:06.660 --> 00:17:10.440 reliable and private local models, you really 00:17:10.440 --> 00:17:13.740 can find a calm island of stability in the middle 00:17:13.740 --> 00:17:15.960 of this swirling storm. And if we connect this 00:17:15.960 --> 00:17:18.460 to the bigger picture, maybe the ultimate goal 00:17:18.460 --> 00:17:21.259 here isn't just using AI, but actually building 00:17:21.259 --> 00:17:23.920 your own personal assistant, one whom you can 00:17:23.920 --> 00:17:26.599 truly trust because ultimately you're the one 00:17:26.599 --> 00:17:29.450 in charge. So maybe a final thought for you, 00:17:29.490 --> 00:17:31.849 our listener, to mull over. What specific areas 00:17:31.849 --> 00:17:34.069 of your work or maybe even your life could be 00:17:34.069 --> 00:17:36.750 genuinely transformed by having an AI assistant 00:17:36.750 --> 00:17:39.809 that you fully control and fully trust? Something 00:17:39.809 --> 00:17:41.769 to really think about for sure. Thank you for 00:17:41.769 --> 00:17:43.730 diving deep with us today. We really encourage 00:17:43.730 --> 00:17:46.690 you to keep exploring, keep experimenting, and 00:17:46.690 --> 00:17:48.410 we'll be back soon with another deep dive into 00:17:48.410 --> 00:17:49.569 the fascinating world around us.