WEBVTT 00:00:00.000 --> 00:00:01.860 You know, the fundamental challenge with AI, 00:00:01.860 --> 00:00:04.160 I think, hasn't just been getting it to create 00:00:04.160 --> 00:00:06.519 something. It's getting the system to really 00:00:06.519 --> 00:00:11.320 understand you, not just any user, but your specific 00:00:11.320 --> 00:00:15.279 taste, how you work, your aesthetic fingerprint, 00:00:15.460 --> 00:00:17.640 really. And that's exactly the shift we're seeing 00:00:17.640 --> 00:00:20.660 now. It feels like we're finally kind of moving 00:00:20.660 --> 00:00:24.640 past the need for super complex, sometimes frustrating 00:00:24.640 --> 00:00:27.160 prompt engineering. We're stepping into this 00:00:27.160 --> 00:00:29.879 phase where these specialized, almost autonomous 00:00:29.879 --> 00:00:32.340 systems can silently learn what you like, what 00:00:32.340 --> 00:00:35.259 you need. And honestly, this is making the output 00:00:35.259 --> 00:00:38.439 fundamentally better. It's more personal and 00:00:38.439 --> 00:00:41.490 importantly more defensible too. Welcome to this 00:00:41.490 --> 00:00:43.149 deep dive. We're going to unpack the stack of 00:00:43.149 --> 00:00:44.770 sources you sent over today. Looks like a lot 00:00:44.770 --> 00:00:46.969 going on. Yeah, our mission today is really to 00:00:46.969 --> 00:00:49.030 map out this transition, this kind of game changing 00:00:49.030 --> 00:00:51.130 moment. We need to get our heads around how these 00:00:51.130 --> 00:00:53.570 personalized, specialized AI models are starting 00:00:53.570 --> 00:00:55.530 to replace the more generic tools we've been 00:00:55.530 --> 00:00:58.549 using. And you know what that means for creativity, 00:00:58.869 --> 00:01:02.070 the economy, but also for, um, serious security 00:01:02.070 --> 00:01:04.689 stuff. Okay. Sounds like we've got a packed agenda 00:01:04.689 --> 00:01:07.629 for this deep dive. Our roadmap kicks off with 00:01:07.629 --> 00:01:10.030 personalization, specifically Google's new system, 00:01:10.209 --> 00:01:13.150 Pasta. Then we'll hit the big industry shockwaves, 00:01:13.230 --> 00:01:15.989 you know, jobs, finance, with some new practical 00:01:15.989 --> 00:01:18.769 tools like AgentKit, and then wrap up with something 00:01:18.769 --> 00:01:21.510 pretty surprising. AI stepping up as a serious 00:01:21.510 --> 00:01:24.689 cyber defender. All right, let's get into it, 00:01:24.709 --> 00:01:27.069 starting with getting personal. Right, so prompt 00:01:27.069 --> 00:01:29.349 engineering. I think we can all agree that trying 00:01:29.349 --> 00:01:33.969 to become fluent in, like, An AI -specific dialect 00:01:33.969 --> 00:01:37.329 can be, well, frustrating. That barrier seems 00:01:37.329 --> 00:01:39.670 to be coming down, though. Google just announced 00:01:39.670 --> 00:01:42.829 Pasta personalized aesthetic style transfer architecture. 00:01:43.189 --> 00:01:44.849 And the key thing seems to be that Pasta doesn't 00:01:44.849 --> 00:01:46.890 need you to be some kind of prompt wizard. Yeah, 00:01:46.890 --> 00:01:48.650 what's really cool here is the mechanism. It's 00:01:48.650 --> 00:01:51.549 actually quite simple from the user's end. It's 00:01:51.549 --> 00:01:53.250 basically a creative loop. You give it a basic 00:01:53.250 --> 00:01:56.090 idea, right? Pasta shows you four different visual 00:01:56.090 --> 00:01:57.730 options based on that. You just pick the one 00:01:57.730 --> 00:02:00.109 you like best. It repeats this maybe 10, 12 times. 00:02:00.640 --> 00:02:03.239 It sounds kind of like having this super patient 00:02:03.239 --> 00:02:06.420 art director who just gets you, you know, without 00:02:06.420 --> 00:02:09.460 needing a million words. Exactly. And over these 00:02:09.460 --> 00:02:11.840 quick rounds, it's quietly building this really 00:02:11.840 --> 00:02:16.159 precise internal map of your specific... taste 00:02:16.159 --> 00:02:18.780 your aesthetic. So it turns that whole prompt 00:02:18.780 --> 00:02:21.080 roulette, that guessing game, into something 00:02:21.080 --> 00:02:23.960 much more like a targeted, personalized visual 00:02:23.960 --> 00:02:26.039 search. It helps you find what you actually want 00:02:26.039 --> 00:02:28.699 without needing like 20 different modifier keywords 00:02:28.699 --> 00:02:31.319 crammed into the prompt. And the data they shared 00:02:31.319 --> 00:02:34.060 seems pretty strong. They trained it on, what, 00:02:34.400 --> 00:02:37.580 7 ,000 real user sessions plus another 30 ,000 00:02:37.580 --> 00:02:41.449 simulated ones. And the result... 85 % of users 00:02:41.449 --> 00:02:43.870 actually prefer the images pasta generated over 00:02:43.870 --> 00:02:45.889 the baseline models. Yeah. That feels like a 00:02:45.889 --> 00:02:47.770 really significant margin, especially when you're 00:02:47.770 --> 00:02:49.430 dealing with subjective stuff like aesthetics. 00:02:49.650 --> 00:02:52.289 Oh yeah, that 85%, that really jumps out. Because 00:02:52.289 --> 00:02:54.349 if you think about just raw image quality, like 00:02:54.349 --> 00:02:56.449 fidelity, things are kind of starting to plateau 00:02:56.449 --> 00:02:59.090 across the big players. Personalization, that's 00:02:59.090 --> 00:03:01.250 the new frontier, the competitive edge. The models 00:03:01.250 --> 00:03:03.189 that genuinely learn from you, that adapt to 00:03:03.189 --> 00:03:04.789 your style, the more you use them, I think those 00:03:04.789 --> 00:03:06.099 are the ones that are going to win out. Yeah, 00:03:06.120 --> 00:03:08.719 trying to nail down some super specific niche 00:03:08.719 --> 00:03:11.460 look like, I don't know, dreamy anime horror, 00:03:11.479 --> 00:03:14.219 but make it vaporwave. That's exactly the kind 00:03:14.219 --> 00:03:16.819 of thing Pasta seems built for. It stops just 00:03:16.819 --> 00:03:20.219 giving you the sort of generic, averaged, good 00:03:20.219 --> 00:03:22.719 output. And here's where it gets really interesting 00:03:22.719 --> 00:03:24.919 for everyone else. Google went and open sourced 00:03:24.919 --> 00:03:26.960 the whole thing. OK, quick question, though. 00:03:27.020 --> 00:03:30.580 That 85 % preference. Is that purely because 00:03:30.580 --> 00:03:32.960 it gets their taste better or is it maybe just 00:03:32.960 --> 00:03:35.159 that people like the iterative feedback process 00:03:35.159 --> 00:03:38.180 itself compared to, you know, one -shot prompting? 00:03:38.360 --> 00:03:40.199 That's a good point. It's probably a bit of both, 00:03:40.199 --> 00:03:42.719 right? But the iteration is definitely powerful. 00:03:43.039 --> 00:03:45.099 By open sourcing pasta, Google is basically saying 00:03:45.099 --> 00:03:47.900 here developers can now build these adaptive 00:03:47.900 --> 00:03:50.639 personal models straight away. They don't need 00:03:50.639 --> 00:03:53.500 to rely on endless kind of hacky prompt tricks 00:03:53.500 --> 00:03:56.060 to get custom results. They get that personalized 00:03:56.060 --> 00:03:58.939 feedback loop built in. So adaptive personalization 00:03:58.939 --> 00:04:01.280 becomes the real differentiator now. That's pretty 00:04:01.280 --> 00:04:04.629 much it. Yeah. So moving beyond just personalization, 00:04:04.750 --> 00:04:08.430 the whole AI ecosystem is, well, it's professionalizing 00:04:08.430 --> 00:04:11.310 and expanding incredibly fast. And that brings 00:04:11.310 --> 00:04:13.870 huge opportunities, but also some pretty serious 00:04:13.870 --> 00:04:16.089 economic friction. You can see that expansion 00:04:16.089 --> 00:04:19.670 just in the tools themselves, like OpenAI launching 00:04:19.670 --> 00:04:22.730 its apps SDK. That basically turns ChatGPT into 00:04:22.730 --> 00:04:24.569 a full on app score, doesn't it? Yeah. You're 00:04:24.569 --> 00:04:26.470 not just asking questions anymore. You can build 00:04:26.470 --> 00:04:29.750 or chat directly with apps like Canva or Spotify. 00:04:30.189 --> 00:04:32.990 inside chat GPT that's a huge shift in utility 00:04:32.990 --> 00:04:36.569 turns a chat bot into well almost an operating 00:04:36.569 --> 00:04:38.910 system right and that kind of massive utility 00:04:38.910 --> 00:04:40.889 boost it sends shockwaves through the economy 00:04:40.889 --> 00:04:43.730 there is a recent Senate Democratic report projecting 00:04:44.839 --> 00:04:46.920 Pretty stark number. AI could potentially wipe 00:04:46.920 --> 00:04:49.519 out up to 100 million US jobs in the next decade. 00:04:49.860 --> 00:04:52.300 And that's across the board roles like nurses, 00:04:52.639 --> 00:04:54.980 admin staff, truck drivers. Yeah. Yeah, it's 00:04:54.980 --> 00:04:57.259 a heavy number to think about. That jobs report 00:04:57.259 --> 00:05:00.379 really makes you pause and think about adaptation, 00:05:00.699 --> 00:05:04.120 new skills, beat. I have to admit, even on the 00:05:04.120 --> 00:05:05.699 technical side, just trying to keep pace with 00:05:05.699 --> 00:05:07.639 all these new interfaces and ways to integrate 00:05:07.639 --> 00:05:10.060 things, I still wrestle with prompt drift myself 00:05:10.060 --> 00:05:11.860 sometimes when I try to get these systems working 00:05:11.860 --> 00:05:14.839 at my own workflows. It's tricky. Oh, yeah. It's 00:05:14.839 --> 00:05:17.040 definitely understandable. Prompt drift just 00:05:17.040 --> 00:05:20.040 for listeners is basically when the AI kind of 00:05:20.040 --> 00:05:22.879 forgets the initial instructions or context over 00:05:22.879 --> 00:05:25.540 a long interaction. So the results start getting 00:05:25.540 --> 00:05:28.699 well. Less useful, off topic. It's a genuine 00:05:28.699 --> 00:05:31.339 challenge for using agents over longer tasks. 00:05:31.600 --> 00:05:33.980 Right. So it's this dual threat, isn't it? Economic 00:05:33.980 --> 00:05:36.439 disruption on one hand, and this rapid technical 00:05:36.439 --> 00:05:38.199 change that's hard to keep up with on the other. 00:05:38.839 --> 00:05:40.720 And on the security side, it looks like Google 00:05:40.720 --> 00:05:42.500 is acknowledging the risks that come with these 00:05:42.500 --> 00:05:44.459 more autonomous systems. They just launched a 00:05:44.459 --> 00:05:49.220 new AI bug bounty program. They're offering up 00:05:49.220 --> 00:05:52.139 to, what, $30 ,000 for finding exploits where 00:05:52.139 --> 00:05:55.339 the AI takes rogue actions. That really underscores 00:05:55.339 --> 00:05:57.750 the enhanced risk when systems start doing complex 00:05:57.750 --> 00:06:00.610 things on their own initiative. And we can't 00:06:00.610 --> 00:06:02.610 really talk about risks without touching on the 00:06:02.610 --> 00:06:04.990 ethical side, especially with deep fakes and 00:06:04.990 --> 00:06:08.769 AI recreations. Zelda Williams recent comments 00:06:08.769 --> 00:06:11.810 about AI versions of her late father, calling 00:06:11.810 --> 00:06:15.329 them gross hot dogs of real people's lives. That 00:06:15.329 --> 00:06:17.829 hits hard. It just shows the very real emotional 00:06:17.829 --> 00:06:20.189 cost of misuse, which seems to be accelerating 00:06:20.189 --> 00:06:23.259 right alongside the capabilities. Yeah, and if 00:06:23.259 --> 00:06:25.759 you connect that acceleration to the global money 00:06:25.759 --> 00:06:28.060 picture, check this out, this detail from the 00:06:28.060 --> 00:06:31.750 Caribbean, Anguilla. Tiny Island Nation apparently 00:06:31.750 --> 00:06:34.529 funded nearly half its entire state budget just 00:06:34.529 --> 00:06:37.470 selling .ai domain names. Wow. It just shows 00:06:37.470 --> 00:06:39.769 the sheer financial weight the world is attaching 00:06:39.769 --> 00:06:42.949 to anything .ai right now, even in kind of unexpected 00:06:42.949 --> 00:06:44.790 ways. And just rattling off a few other industry 00:06:44.790 --> 00:06:47.230 markers real quick. Rackle Ventures pulling in 00:06:47.230 --> 00:06:51.269 $650 million for early stage AI. Sam Altman confirming 00:06:51.269 --> 00:06:53.810 chat GPT is hitting 800 million weekly active 00:06:53.810 --> 00:06:57.100 users. That's huge scale. And OpenAI doing an 00:06:57.100 --> 00:06:59.379 acqui -hire of Roy, clearly grabbing specialized 00:06:59.379 --> 00:07:01.860 talent. So thinking about all the money, the 00:07:01.860 --> 00:07:03.980 job worries, the deep fakes, what do you see 00:07:03.980 --> 00:07:05.959 as the biggest underlying risk factor that connects 00:07:05.959 --> 00:07:08.019 all these headlines? I think it's the accelerating 00:07:08.019 --> 00:07:11.019 pace of these dual -use capabilities. The same 00:07:11.019 --> 00:07:13.860 tech that offers amazing utility also inherently 00:07:13.860 --> 00:07:16.180 brings these ethical and economic disruption 00:07:16.180 --> 00:07:18.540 risks, and it's all happening faster and faster. 00:07:19.040 --> 00:07:22.420 Mid -roll sponsor, Replace Hold. Okay, let's 00:07:22.420 --> 00:07:25.399 shift focus now to the actual tools people are 00:07:25.399 --> 00:07:27.300 building to create these more specialized systems 00:07:27.300 --> 00:07:30.160 We've been talking about open AI finally released 00:07:30.160 --> 00:07:33.199 aging kit. It's being described as a quote full 00:07:33.199 --> 00:07:35.600 stack for developers. Yeah, this looks like a 00:07:35.600 --> 00:07:37.879 really big step towards making agent development 00:07:37.879 --> 00:07:40.759 more, well, professional and standardized. Think 00:07:40.759 --> 00:07:43.339 of AgentKit as kind of like combining the ease 00:07:43.339 --> 00:07:45.699 of use of Canva with the automation power of 00:07:45.699 --> 00:07:49.319 Zapier, but specifically for AI agents. So it 00:07:49.319 --> 00:07:51.779 helps you build, deploy, and test these autonomous 00:07:51.779 --> 00:07:54.439 AI workflows much more quickly. And crucially, 00:07:54.480 --> 00:07:56.540 it provides standard ways for agents to talk 00:07:56.540 --> 00:07:59.040 to each other, APIs, and safety rules, making 00:07:59.040 --> 00:08:01.360 them more interoperable and hopefully safer. 00:08:01.560 --> 00:08:02.970 Right, it's like standardized. the building blocks 00:08:02.970 --> 00:08:05.470 like Lego blocks for data and instructions maybe 00:08:05.470 --> 00:08:07.430 makes it easier to put complex things together 00:08:07.430 --> 00:08:09.629 reliably seems like it would lower the barrier 00:08:09.629 --> 00:08:11.970 quite a bit definitely and predictably we're 00:08:11.970 --> 00:08:14.290 already seeing a strong open source alternatives 00:08:14.290 --> 00:08:16.709 popping up there's one clone of quests for instance 00:08:16.709 --> 00:08:20.110 it's free open source pairs with deep seek and 00:08:20.110 --> 00:08:22.740 it positions itself as a direct Community driven 00:08:22.740 --> 00:08:26.379 rival to paid platforms like bolt so that democratization 00:08:26.379 --> 00:08:28.720 of creating agents. It's happening right now 00:08:28.720 --> 00:08:30.459 Okay, so we're getting better tools to build 00:08:30.459 --> 00:08:32.720 these things and makes them do more But then 00:08:32.720 --> 00:08:34.919 there's the question of quality right? Yeah, 00:08:35.080 --> 00:08:37.580 especially with text your sources mentioned a 00:08:37.580 --> 00:08:41.149 guide on AI writing mistakes outlining like 15 00:08:41.149 --> 00:08:44.590 common pitfalls. And the core idea there is really 00:08:44.590 --> 00:08:48.309 important. How do you use AI to handle the volume, 00:08:48.429 --> 00:08:50.629 the scale, without losing your own voice? How 00:08:50.629 --> 00:08:53.730 do you keep it authentic and avoid that robotic 00:08:53.730 --> 00:08:55.370 sound? Yeah, because we've all seen it right. 00:08:55.470 --> 00:08:57.629 AI -assisted text, that just sounds completely 00:08:57.629 --> 00:09:00.269 generic. It's a fast way to make your writing 00:09:00.269 --> 00:09:02.809 feel predictable and bland. So why is keeping 00:09:02.809 --> 00:09:05.009 that human voice still such a big hurdle, even 00:09:05.009 --> 00:09:06.789 with these really advanced models we have now? 00:09:06.940 --> 00:09:09.580 Well, I think it's because AI often defaults 00:09:09.580 --> 00:09:11.679 to the most statistically common phrasing, the 00:09:11.679 --> 00:09:14.159 sort of average way of saying things. You need 00:09:14.159 --> 00:09:16.840 specific, kind of targeted techniques to nudge 00:09:16.840 --> 00:09:19.720 it towards authenticity and steer it away from 00:09:19.720 --> 00:09:22.220 just spitting out predictable machine -like language. 00:09:23.049 --> 00:09:25.029 All right. This next part is where things get 00:09:25.029 --> 00:09:27.429 really interesting, I think. We're moving beyond 00:09:27.429 --> 00:09:30.289 just aesthetic personalization into some seriously 00:09:30.289 --> 00:09:33.529 hardcore specialization. Anthropic is now focusing 00:09:33.529 --> 00:09:36.970 its latest model, Claude Sonnet 4 .5, on mission 00:09:36.970 --> 00:09:39.990 critical cybersecurity defense. OK. Because for 00:09:39.990 --> 00:09:42.669 years, the general consensus about AI and complex 00:09:42.669 --> 00:09:44.750 cybersecurity was basically, nah, never good. 00:09:45.370 --> 00:09:47.289 We knew it could flag obvious stuff, maybe, but 00:09:47.289 --> 00:09:49.190 it wasn't seen as a truly strategic defender, 00:09:49.309 --> 00:09:52.230 right? Couldn't handle nuance. Exactly. But the 00:09:52.230 --> 00:09:55.080 shift Seems pretty dramatic now. Sonnet 4 .5 00:09:55.080 --> 00:09:57.340 has been trained specifically on deep security 00:09:57.340 --> 00:10:00.259 skills. We're talking millions of simulated real 00:10:00.259 --> 00:10:02.779 world attacks. And they've got validation showing 00:10:02.779 --> 00:10:05.679 Claude can actually beat human teams in cybersecurity 00:10:05.679 --> 00:10:13.690 competitions now. Impressively, they got it to 00:10:13.690 --> 00:10:16.909 successfully recreate the 2017 Equifax breach 00:10:16.909 --> 00:10:20.289 in a test environment. Wow. Recreating the Equifax 00:10:20.289 --> 00:10:22.029 breach. That's huge. That means these systems 00:10:22.029 --> 00:10:24.450 can actually probe and test all that legacy infrastructure 00:10:24.450 --> 00:10:26.610 that so much still relies on. They can find those 00:10:26.610 --> 00:10:29.149 deep systemic weaknesses that maybe humans missed 00:10:29.149 --> 00:10:32.090 or couldn't see at scale. Whoa. Just imagine 00:10:32.090 --> 00:10:34.690 scaling that kind of defensive capability, like 00:10:34.690 --> 00:10:36.490 checking a billion potential vulnerabilities 00:10:36.490 --> 00:10:38.309 almost instantly. That's serious protection. 00:10:38.730 --> 00:10:41.899 A massive intelligent firewall. beat. That really 00:10:41.899 --> 00:10:43.799 is a moment of wonder, actually. Absolutely. 00:10:43.980 --> 00:10:46.740 And this isn't just internal testing. Top security 00:10:46.740 --> 00:10:49.240 firms are actively validating this expertise. 00:10:49.779 --> 00:10:52.620 But, and there's always a but, Anthropics own 00:10:52.620 --> 00:10:54.820 safeguards team also caught people trying to 00:10:54.820 --> 00:10:57.639 use Claude for cybercrime, which just proves, 00:10:58.000 --> 00:11:01.080 yet again, this technology is a perfect dual 00:11:01.080 --> 00:11:03.870 use sword. It's potentially as good at offense 00:11:03.870 --> 00:11:05.830 as it is at defense. Yeah, they specifically 00:11:05.830 --> 00:11:08.929 mentioned a hacker building a whole data extortion 00:11:08.929 --> 00:11:11.850 pipeline using it. And even a suspected state 00:11:11.850 --> 00:11:14.750 -sponsored group, a Chinese APT, apparently used 00:11:14.750 --> 00:11:17.899 it for telecom espionage. These aren't hypotheticals. 00:11:17.940 --> 00:11:20.120 These are real -world, high -stakes misuses happening 00:11:20.120 --> 00:11:21.799 right now. And while they were tracking all this 00:11:21.799 --> 00:11:23.940 misuse, they actually uncovered a totally new 00:11:23.940 --> 00:11:26.200 attack technique. They're calling it vibe hacking. 00:11:26.620 --> 00:11:28.940 So if you think of the previous big model, Claude 00:11:28.940 --> 00:11:30.720 Opus, as maybe the general intelligence, the 00:11:30.720 --> 00:11:33.820 brain, Sonnet 4 .5 is shaping up to be the really 00:11:33.820 --> 00:11:37.700 sophisticated firewall, but one with, like, personality 00:11:37.700 --> 00:11:40.539 and deep situational awareness. Okay, hold on. 00:11:40.620 --> 00:11:43.000 Vibe hacking. What exactly is that, and why is 00:11:43.000 --> 00:11:44.620 it considered such a significant new threat? 00:11:44.759 --> 00:11:47.120 So vibe hacking, as they describe it, is this 00:11:47.120 --> 00:11:50.559 new manipulation tactic where the AI uses really 00:11:50.559 --> 00:11:53.159 subtle language cues, emotional context, and 00:11:53.159 --> 00:11:55.139 that personalized tone we talked about earlier 00:11:55.139 --> 00:11:57.919 to carry out highly effective social engineering 00:11:57.919 --> 00:12:00.720 attacks. It's essentially attacking the user's 00:12:00.720 --> 00:12:03.659 emotional state, their vibe, not just going after 00:12:03.659 --> 00:12:06.259 technical vulnerabilities like logins. It's deep 00:12:06.259 --> 00:12:09.659 manipulation, hashtag tag outro. So we try to 00:12:09.659 --> 00:12:11.480 pull back and see the big picture from all these 00:12:11.480 --> 00:12:14.159 different threads. The central theme that just 00:12:14.159 --> 00:12:16.809 keeps emerging is specialization, isn't it? We're 00:12:16.809 --> 00:12:19.090 definitely seeing AI shift away from being just 00:12:19.090 --> 00:12:21.669 a general purpose tool towards becoming, on one 00:12:21.669 --> 00:12:23.730 hand, these highly personalized aesthetic partners 00:12:23.730 --> 00:12:26.269 like Pasta, and on the other, these intensely 00:12:26.269 --> 00:12:29.029 specialized, highly capable cyber defenders like 00:12:29.029 --> 00:12:31.470 Claude 4 .5. Yeah, and the whole ecosystem around 00:12:31.470 --> 00:12:33.669 it is professionalizing so fast. We're getting 00:12:33.669 --> 00:12:36.669 proper agent stacks, app stores, all of which 00:12:36.669 --> 00:12:38.990 demand new kinds of skills from us, the users, 00:12:39.190 --> 00:12:41.129 more adaptive skills. We're definitely past the 00:12:41.129 --> 00:12:42.950 era of just typing a simple prompt and hoping 00:12:42.950 --> 00:12:44.909 for the best. It really feels like a critical 00:12:44.909 --> 00:12:46.610 turning point for any anyone working in this 00:12:46.610 --> 00:12:48.870 space, or even just using these tools regularly. 00:12:49.230 --> 00:12:50.929 And that kind of leads us to our final thought 00:12:50.929 --> 00:12:53.789 for you, the listener, to maybe chew on this 00:12:53.789 --> 00:12:56.690 knowledge gap. It seems like it's widening maybe 00:12:56.690 --> 00:12:58.929 faster than ever. The gap between people who 00:12:58.929 --> 00:13:01.370 are mastering these new personalized tools, these 00:13:01.370 --> 00:13:03.889 specialized adaptive systems, and those who are 00:13:03.889 --> 00:13:07.049 still relying on the older, more rigid, one size 00:13:07.049 --> 00:13:09.309 fits all approach to prompting. It really feels 00:13:09.309 --> 00:13:12.070 like the future is going to demand that adaptive 00:13:12.070 --> 00:13:14.350 skill, that specialization in how we interact 00:13:14.350 --> 00:13:17.049 with AI. So maybe take a moment to consider your 00:13:17.049 --> 00:13:19.649 own workflows. Where could you start adopting 00:13:19.649 --> 00:13:22.610 more personalized or specialized AI models right 00:13:22.610 --> 00:13:24.750 now, both to make sure your own output stays 00:13:24.750 --> 00:13:26.850 authentic and that your operations, your data 00:13:26.850 --> 00:13:29.370 are secure? This is a really fascinating deep 00:13:29.370 --> 00:13:31.090 dive, lots to think about. Thanks for bringing 00:13:31.090 --> 00:13:33.190 all these sources together for us. Out to your 00:13:33.190 --> 00:13:33.470 music.