WEBVTT 00:00:00.000 --> 00:00:02.299 I remember not that long ago trying to get an 00:00:02.299 --> 00:00:05.799 AI image and you'd wait, you know, maybe a minute 00:00:05.799 --> 00:00:09.240 just to get something weird. Yeah. Or janky or 00:00:09.240 --> 00:00:12.900 with misspelled text. You got it fast, but it 00:00:12.900 --> 00:00:16.100 was rarely precise. That entire tradeoff, that 00:00:16.100 --> 00:00:19.370 assumption. Yeah. It feels like it's been completely 00:00:19.370 --> 00:00:21.429 torn up in the last few months. Welcome back 00:00:21.429 --> 00:00:24.050 to the deep dive. You've brought us a really 00:00:24.050 --> 00:00:26.690 great set of sources today. And I think our mission 00:00:26.690 --> 00:00:28.829 is pretty essential. We're going to filter the 00:00:28.829 --> 00:00:31.050 noise to find the signal. We want to cut through 00:00:31.050 --> 00:00:34.329 all the general AI hype and show you what's actually 00:00:34.329 --> 00:00:37.259 being used right now. by people doing serious 00:00:37.259 --> 00:00:40.899 work. And our roadmap for this dive is, well, 00:00:41.000 --> 00:00:43.619 it's focused on three big shifts. First, we're 00:00:43.619 --> 00:00:45.340 going to look at the image generation arms race, 00:00:45.479 --> 00:00:47.700 specifically how these models are getting this 00:00:47.700 --> 00:00:51.960 surgical precision. Yeah. Then we need to decode 00:00:51.960 --> 00:00:54.119 some of the vocabulary. It gets confusing. We 00:00:54.119 --> 00:00:56.399 need to draw a clear line between what a tool 00:00:56.399 --> 00:00:58.979 is, what an LLM is, and this really critical 00:00:58.979 --> 00:01:01.979 idea of an agent. And finally, we're diving into 00:01:01.979 --> 00:01:04.890 a new study from Perplexity and Harvard. And 00:01:04.890 --> 00:01:07.530 it's fascinating because it kind of debunks the 00:01:07.530 --> 00:01:10.609 common myth about how people use AI agents. The 00:01:10.609 --> 00:01:13.090 Robo Butler idea. Exactly. It's not about that 00:01:13.090 --> 00:01:15.370 at all. It's something much more powerful. So 00:01:15.370 --> 00:01:18.049 our goal for you is simple. Walk away from this 00:01:18.049 --> 00:01:20.870 with the core insights you need to, you know, 00:01:20.870 --> 00:01:23.510 move faster and smarter. Okay. Let's unpack this. 00:01:23.569 --> 00:01:25.310 Let's start on the creative side. So the biggest 00:01:25.310 --> 00:01:27.629 headline from our sources was definitely the 00:01:27.629 --> 00:01:32.049 launch of OpenAI's GPT image 1 .5. And this isn't 00:01:32.049 --> 00:01:34.430 just another upgrade. It feels more like a correction 00:01:34.430 --> 00:01:38.269 of all the past failures. For years, the big 00:01:38.269 --> 00:01:39.989 frustration was that you just couldn't trust 00:01:39.989 --> 00:01:43.549 the AI to respect your actual instructions. Exactly. 00:01:43.829 --> 00:01:45.629 I think about those early attempts, you know, 00:01:45.650 --> 00:01:47.269 where you'd ask it to change one little thing, 00:01:47.430 --> 00:01:49.909 like swap a blue car for a red one. And the whole 00:01:49.909 --> 00:01:52.189 image would just melt. Yeah. The background, 00:01:52.329 --> 00:01:54.530 the lighting, the driver's face. Yeah. It would 00:01:54.530 --> 00:01:56.750 all warp. You'd get these weird artifacts. It 00:01:56.750 --> 00:01:59.250 was a mess. It was a distortion nightmare. But 00:01:59.250 --> 00:02:02.170 1 .5 introduces what they're calling high control 00:02:02.170 --> 00:02:05.560 editing. So you can add, remove, or change things 00:02:05.560 --> 00:02:08.219 in the image with surgical precision without 00:02:08.219 --> 00:02:09.800 completely breaking the rest of the picture. 00:02:09.960 --> 00:02:12.319 And that's a huge step toward making assets you 00:02:12.319 --> 00:02:14.439 can actually use in production. And for me, the 00:02:14.439 --> 00:02:16.439 most mind -bending part is the real text support. 00:02:16.659 --> 00:02:19.240 We've all seen that weird, alien -looking gibberish 00:02:19.240 --> 00:02:23.740 text in AI images. Oh, yeah. Now, 1 .5 can render 00:02:23.740 --> 00:02:27.379 dense, small fonts perfectly clear inside an 00:02:27.379 --> 00:02:29.800 image. That was always the ultimate test for 00:02:29.800 --> 00:02:32.479 precision. And it feels like they finally cracked 00:02:32.479 --> 00:02:35.340 it. And that capability, it ties right into reliable 00:02:35.340 --> 00:02:37.860 instruction following, which is the key feature 00:02:37.860 --> 00:02:40.479 that really changes the game for professional 00:02:40.479 --> 00:02:42.780 creative teams. Right. The sources have this 00:02:42.780 --> 00:02:46.319 incredible example asking the AI for a six by 00:02:46.319 --> 00:02:50.800 six grid of 36 totally unique items. And it places 00:02:50.800 --> 00:02:53.560 each object exactly in the right tile. That's 00:02:53.560 --> 00:02:56.259 a true whoa moment. It's not just about making 00:02:56.259 --> 00:02:58.360 a pretty picture anymore. It's about executing 00:02:58.360 --> 00:03:01.530 a complex creative brief flawlessly. On the first 00:03:01.530 --> 00:03:03.710 go. And it does it four times faster than the 00:03:03.710 --> 00:03:05.889 1 .0 version. That's the real breakthrough, right? 00:03:05.949 --> 00:03:08.550 Speed plus precision. Absolutely. And we have 00:03:08.550 --> 00:03:10.949 to look at this strategically. This wasn't some 00:03:10.949 --> 00:03:13.689 casual release. It was a direct shot at Google, 00:03:13.750 --> 00:03:16.289 right? It was clearly rushed to market to counter 00:03:16.289 --> 00:03:19.789 Google's nano banana pro model. We're seeing 00:03:19.789 --> 00:03:22.370 the market for image generation start to segment 00:03:22.370 --> 00:03:25.009 just like software did years ago. So what does 00:03:25.009 --> 00:03:27.349 that segmentation look like? Because for most 00:03:27.349 --> 00:03:29.889 people, they probably just see two. really good 00:03:29.889 --> 00:03:32.389 image generators. Well, they're targeting different 00:03:32.389 --> 00:03:36.210 kinds of users. GTC Image 1 .5 is all about speed 00:03:36.210 --> 00:03:39.229 and accessible control. It's for rapid iteration, 00:03:39.689 --> 00:03:42.509 for brainstorming, for the pro user who needs 00:03:42.509 --> 00:03:45.110 10 quick versions of something for a social post. 00:03:45.289 --> 00:03:48.210 So speed and accessibility. Exactly. The new 00:03:48.210 --> 00:03:50.530 images tab, the pre -built styles, it's all built 00:03:50.530 --> 00:03:53.389 for velocity. Nano Banana Pro, on the other hand, 00:03:53.389 --> 00:03:55.840 is slower. But it's specialized for the high 00:03:55.840 --> 00:03:58.280 -end production pipeline. Okay, so more for enterprise. 00:03:58.599 --> 00:04:01.780 Think big enterprise features, massive batch 00:04:01.780 --> 00:04:04.479 processing, the kind of absolute fidelity you 00:04:04.479 --> 00:04:07.199 need for a major brand campaign. It's quality 00:04:07.199 --> 00:04:12.759 over just pure speed. Whoa. Just imagine scaling 00:04:12.759 --> 00:04:15.460 that precise instruction following to a billion 00:04:15.460 --> 00:04:18.139 unique queries a day. If you can trust the AI 00:04:18.139 --> 00:04:21.800 to execute 36 unique commands perfectly every 00:04:21.800 --> 00:04:23.879 single time, that just fundamentally changes 00:04:23.879 --> 00:04:26.579 the job for graphic designers everywhere. So 00:04:26.579 --> 00:04:28.579 if I'm a user, what's the bottom line? How do 00:04:28.579 --> 00:04:31.100 I choose between 1 .5 and nano banana? If you 00:04:31.100 --> 00:04:33.139 need speed and quick iteration, you go with 1 00:04:33.139 --> 00:04:35.660 .5. If it's enterprise workflows and production 00:04:35.660 --> 00:04:38.259 quality, you lean toward nano. So we've covered 00:04:38.259 --> 00:04:40.360 the creation side, but all that speed and precision. 00:04:41.129 --> 00:04:43.370 It doesn't mean much if we can't even agree on 00:04:43.370 --> 00:04:45.269 what to call the things we're using. Right. The 00:04:45.269 --> 00:04:47.470 vocabulary is changing so fast and it just creates 00:04:47.470 --> 00:04:50.589 this information fatigue. It really does. People 00:04:50.589 --> 00:04:53.250 feel this massive pressure, this FOMO, like they're 00:04:53.250 --> 00:04:55.170 missing that one key update that's going to change 00:04:55.170 --> 00:04:57.410 everything. Yeah. But the answer isn't to read 00:04:57.410 --> 00:04:59.350 more. It's just to clarify the architecture. 00:04:59.509 --> 00:05:01.250 You only really need to get your head around 00:05:01.250 --> 00:05:03.449 three core categories when people talk about 00:05:03.449 --> 00:05:05.629 AI. Okay. Let's start with the simplest one, 00:05:05.730 --> 00:05:07.829 the most common one. That would be the AI tool. 00:05:08.220 --> 00:05:11.040 A tool is just an app made for one specific task. 00:05:11.220 --> 00:05:14.019 Like image generation. Or video editing or voice 00:05:14.019 --> 00:05:16.680 cloning. They're the single function apps with 00:05:16.680 --> 00:05:18.920 a clean interface. Simple. Okay, next up are 00:05:18.920 --> 00:05:22.040 the engines that power everything. Exactly. The 00:05:22.040 --> 00:05:24.699 LLMs or large language models. What everyone 00:05:24.699 --> 00:05:26.720 just calls chatbots. These are the big predictive 00:05:26.720 --> 00:05:30.319 models. The brains of the operation. Like ChatGPT, 00:05:30.540 --> 00:05:35.810 Claude. Gemini, Grok. Most of the really powerful 00:05:35.810 --> 00:05:38.529 ones are closed source. The company keeps the 00:05:38.529 --> 00:05:40.569 code secret and they run them on their own servers. 00:05:40.910 --> 00:05:43.430 And what's the alternative to those huge closed 00:05:43.430 --> 00:05:46.209 source brains? That's the third category, open 00:05:46.209 --> 00:05:48.290 source models. You find these on platforms like 00:05:48.290 --> 00:05:50.810 Hugging Face. The code is free, it's public, 00:05:50.889 --> 00:05:53.610 and it's mostly used by coders who want to build 00:05:53.610 --> 00:05:56.329 their own custom apps or run AI on their own 00:05:56.329 --> 00:05:58.290 machines. So they can avoid big tech servers. 00:05:58.470 --> 00:06:00.589 Precisely. Now let's get to the term that's really 00:06:00.589 --> 00:06:02.430 exploding, the one that connects all this together, 00:06:02.629 --> 00:06:06.759 the AI agent. Right. An agent is basically a 00:06:06.759 --> 00:06:08.980 smart software layer you build on top of the 00:06:08.980 --> 00:06:12.019 LLM brain. It's a system that manages and runs 00:06:12.019 --> 00:06:14.240 a whole multi -step workflow for you. So the 00:06:14.240 --> 00:06:16.720 LLM is the brain. The tools are the hands and 00:06:16.720 --> 00:06:18.500 the agent is like the personal assistant running 00:06:18.500 --> 00:06:21.720 the whole show. Okay. So an agent is built to 00:06:21.720 --> 00:06:25.500 take complex step -by -step actions like replying 00:06:25.500 --> 00:06:28.360 to an email, pulling out the key details, making 00:06:28.360 --> 00:06:31.160 a PDF summary, and then scheduling a follow -up 00:06:31.160 --> 00:06:34.370 call. All in one automated flow without you having 00:06:34.370 --> 00:06:36.970 to do each step. It solves that friction point 00:06:36.970 --> 00:06:39.959 we've all felt. We call it prompt drift. You 00:06:39.959 --> 00:06:41.720 know, when you try to chain six different single 00:06:41.720 --> 00:06:44.259 task tools together, the instructions start to 00:06:44.259 --> 00:06:47.639 get muddled. Oh, absolutely. I still wrestle 00:06:47.639 --> 00:06:49.600 with prompt drift myself. When you're trying 00:06:49.600 --> 00:06:51.379 to combine these different tools into one perfect 00:06:51.379 --> 00:06:53.639 workflow. It's hard. Yeah, it feels like you're 00:06:53.639 --> 00:06:56.620 trying to give complex errands to a very literal 00:06:56.620 --> 00:06:58.920 six -year -old. And that's why the agent concept 00:06:58.920 --> 00:07:01.060 is so critical. It handles the handoff between 00:07:01.060 --> 00:07:03.959 all those tasks. And for the user, just understanding 00:07:03.959 --> 00:07:07.019 this structure is the shortcut. FOMO is basically 00:07:07.019 --> 00:07:09.459 fake unless you're actually building the AI models 00:07:09.459 --> 00:07:11.800 yourself. For the rest of us, it's just about 00:07:11.800 --> 00:07:13.519 how the agent layer lets you automate things 00:07:13.519 --> 00:07:17.120 smarter. So why is the agent concept becoming 00:07:17.120 --> 00:07:20.279 so critical right now? Because agents automate 00:07:20.279 --> 00:07:23.420 entire workflows, and that saves way more time 00:07:23.420 --> 00:07:26.319 than single task tools ever could. That focus 00:07:26.319 --> 00:07:29.379 on workflow automation. It leads us perfectly 00:07:29.379 --> 00:07:31.540 into our third segment, because this is about 00:07:31.540 --> 00:07:33.500 using real data to cut through the hype. Right. 00:07:33.949 --> 00:07:35.810 The popular story, you know, the one from sci 00:07:35.810 --> 00:07:39.790 -fi, is the AI is the robo -butler. The assistant 00:07:39.790 --> 00:07:41.689 that books your flights, orders your groceries, 00:07:41.870 --> 00:07:44.069 handles all the little chores. And that is absolutely 00:07:44.069 --> 00:07:46.189 not what the data says is the highest value use. 00:07:46.310 --> 00:07:48.329 The perplexity in Harvard's study you shared 00:07:48.329 --> 00:07:50.449 is so interesting because it just debunks that 00:07:50.449 --> 00:07:53.069 whole myth. So what's the core finding? The core 00:07:53.069 --> 00:07:55.589 finding is that users are using AI agents to 00:07:55.589 --> 00:07:58.240 augment cognitive labor. To think better. To 00:07:58.240 --> 00:08:00.920 think better, not just to delegate chores. They're 00:08:00.920 --> 00:08:03.680 using these tools to expand their own intellectual 00:08:03.680 --> 00:08:07.100 bandwidth. And that really points to where the 00:08:07.100 --> 00:08:09.600 real productivity gains are happening. Can you 00:08:09.600 --> 00:08:11.060 give some examples? What kind of high -level 00:08:11.060 --> 00:08:13.860 tasks were they actually doing? Well, over half 00:08:13.860 --> 00:08:16.680 of the queries were for tasks that involved synthesizing 00:08:16.680 --> 00:08:19.699 huge amounts of complex information. Things like 00:08:19.699 --> 00:08:22.459 summarizing really long documents. Like pulling 00:08:22.459 --> 00:08:25.199 the core arguments from a 50 -page report. Exactly. 00:08:25.199 --> 00:08:27.459 Or editing and structuring technical reports. 00:08:27.470 --> 00:08:30.350 managing complex research workflows, and getting 00:08:30.350 --> 00:08:32.970 high -level coursework help. So these are tasks 00:08:32.970 --> 00:08:36.149 that need judgment And integration. Yeah. Not 00:08:36.149 --> 00:08:39.049 just grabbing facts. They're using the agent 00:08:39.049 --> 00:08:42.149 as a cognitive accelerant. And the user demographics 00:08:42.149 --> 00:08:45.330 tracked with this perfectly. The main users weren't 00:08:45.330 --> 00:08:47.490 just everyday consumers. They're knowledge workers. 00:08:47.730 --> 00:08:50.690 Tech folks, marketing strategists, finance professionals, 00:08:51.230 --> 00:08:54.049 academics doing literature reviews, people who 00:08:54.049 --> 00:08:56.789 have to process information at an inhuman speed. 00:08:57.190 --> 00:08:58.629 And there was a pattern there, right? Yeah. A 00:08:58.629 --> 00:09:01.340 very clear predictor. A higher education level 00:09:01.340 --> 00:09:04.740 and a higher GDP correlated directly with more 00:09:04.740 --> 00:09:07.639 agent usage. And it suggests this kind of graduation 00:09:07.639 --> 00:09:11.019 process. What do you mean by that? Well, users 00:09:11.019 --> 00:09:13.120 might start with light stuff, like planning a 00:09:13.120 --> 00:09:15.500 trip. But as soon as they realize the agent can 00:09:15.500 --> 00:09:18.080 handle complex thought, they quickly shift to 00:09:18.080 --> 00:09:20.600 these deeper cognitive tasks. That makes sense. 00:09:20.740 --> 00:09:22.659 The more complex your daily work is, the more 00:09:22.659 --> 00:09:25.230 value you get from an agent right away. But, 00:09:25.269 --> 00:09:27.269 and this is important, we have to be critical 00:09:27.269 --> 00:09:30.649 of the data source. The study was based on Perplexity's 00:09:30.649 --> 00:09:33.070 users. Which is a research -focused platform. 00:09:33.389 --> 00:09:36.110 Right. So their user base is already skewed toward 00:09:36.110 --> 00:09:38.929 more academic and professional queries than, 00:09:39.009 --> 00:09:41.889 say, a general user of standard chat GPT. So 00:09:41.889 --> 00:09:44.190 does this study accurately represent general 00:09:44.190 --> 00:09:46.799 user behavior? It might not represent the average 00:09:46.799 --> 00:09:49.519 consumer, but it clearly shows advanced cognitive 00:09:49.519 --> 00:09:52.360 tasks are already the highest value use for people 00:09:52.360 --> 00:09:54.779 actively using agents. All right, let's shift 00:09:54.779 --> 00:09:57.259 gears. Time for a rapid fire summary of some 00:09:57.259 --> 00:10:00.980 of the most impactful recent news and strategic 00:10:00.980 --> 00:10:03.340 shifts, the stuff you need to know. Let's start 00:10:03.340 --> 00:10:06.139 with something a bit more cultural. Let's talk 00:10:06.139 --> 00:10:09.000 about slop. Slop. It's officially been named 00:10:09.000 --> 00:10:12.830 the 2025 word of the year. And slop is the term 00:10:12.830 --> 00:10:16.870 for that low quality, high volume, AI generated 00:10:16.870 --> 00:10:19.809 content that's just flooding everything. Search 00:10:19.809 --> 00:10:22.610 engines, social media feeds, even book markets. 00:10:22.809 --> 00:10:24.669 It's a real challenge if you're trying to find 00:10:24.669 --> 00:10:26.870 quality information. You just wade through noise. 00:10:27.169 --> 00:10:30.610 And on the tool side, ChatGPT just launched skills 00:10:30.610 --> 00:10:34.129 for specific tasks. This is basically them mirroring 00:10:34.129 --> 00:10:36.450 the kind of targeted functionality that cloud 00:10:36.450 --> 00:10:38.570 users have had for a while. So you can tell the 00:10:38.570 --> 00:10:40.629 model what kind of tasks to optimize for. Right. 00:10:40.970 --> 00:10:44.110 Competition is driving feature parody. And speaking 00:10:44.110 --> 00:10:46.509 of competition, there's a big Google rumor to 00:10:46.509 --> 00:10:49.490 watch. What's that? Their deep mind lead basically 00:10:49.490 --> 00:10:52.049 told people to go bookmark the hugging face page, 00:10:52.350 --> 00:10:54.830 strongly hinting that Gemma 4 is coming very 00:10:54.830 --> 00:10:57.129 soon. That could be a major open source release 00:10:57.129 --> 00:10:59.190 that shifts the whole landscape. And a great 00:10:59.190 --> 00:11:02.190 example of a new app using these models is DoorDash 00:11:02.190 --> 00:11:04.960 Zesty. Oh, I saw this. It's a social app for 00:11:04.960 --> 00:11:07.240 finding restaurants, but it uses these really 00:11:07.240 --> 00:11:10.279 specific natural language queries. Things like 00:11:10.279 --> 00:11:13.019 a low key dinner for introverts with excellent 00:11:13.019 --> 00:11:15.779 lighting. That's a great practical use of an 00:11:15.779 --> 00:11:18.960 LLM to navigate real world data. OK, now let's 00:11:18.960 --> 00:11:20.740 talk about some serious infrastructure moves. 00:11:21.659 --> 00:11:26.320 NVIDIA made a quiet. but huge acquisition. ZMD. 00:11:26.519 --> 00:11:29.059 SkedMD. They're the company behind the Slurm 00:11:29.059 --> 00:11:31.779 workload manager, which is the open source scheduler 00:11:31.779 --> 00:11:33.899 that is absolutely critical for running massive 00:11:33.899 --> 00:11:36.220 AI data centers. That's a heavy acquisition. 00:11:36.580 --> 00:11:38.779 Resource scheduling, deciding what data gets 00:11:38.779 --> 00:11:41.559 processed on which chip and when. That's the 00:11:41.559 --> 00:11:44.620 plumbing of AI training. Yeah. It just strengthens 00:11:44.620 --> 00:11:47.299 NVIDIA's already insane control over the entire 00:11:47.299 --> 00:11:50.179 compute stack from the chip itself all the way 00:11:50.179 --> 00:11:52.419 up to the training software. It's about owning 00:11:52.419 --> 00:11:55.600 the operating system of the AI data center. And 00:11:55.600 --> 00:11:58.139 speaking of control tactics, look at this subtle 00:11:58.139 --> 00:12:00.399 move from OpenAI. What are they doing? They're 00:12:00.399 --> 00:12:03.179 now defaulting free users to the less capable 00:12:03.179 --> 00:12:06.700 GPT 5 .2 instant model. If you want the much 00:12:06.700 --> 00:12:08.279 better performance, you have to manually switch 00:12:08.279 --> 00:12:10.960 over. Ah, that's a classic platform tactic. Make 00:12:10.960 --> 00:12:13.080 the free tier a little less convenient to nudge 00:12:13.080 --> 00:12:16.379 people toward paid plans. And finally, two quick 00:12:16.379 --> 00:12:19.600 utility tools worth checking out. Google's CC. 00:12:20.159 --> 00:12:22.220 It gives you a personalized briefing every morning, 00:12:22.340 --> 00:12:24.600 pulling from your Gmail, calendar, and drive. 00:12:24.960 --> 00:12:27.860 It's the ultimate catch -me -up tool. And Okara. 00:12:27.980 --> 00:12:30.220 And Okara, which lets you chat with a whole bunch 00:12:30.220 --> 00:12:32.720 of different open -source models, Lama, Quen, 00:12:32.759 --> 00:12:36.139 DeepSeek, all from one single app. So what's 00:12:36.139 --> 00:12:38.500 the biggest infrastructure implication of that 00:12:38.500 --> 00:12:41.000 NVIDIA acquisition? It really just strengthens 00:12:41.000 --> 00:12:43.379 NVIDIA's control over scheduling massive data 00:12:43.379 --> 00:12:45.980 center workloads. Okay, let's pull all this together. 00:12:46.159 --> 00:12:48.679 Let's synthesize the core insights from this 00:12:48.679 --> 00:12:50.899 deep dive for you. We covered a lot, but there 00:12:50.899 --> 00:12:53.600 are really three major takeaways to hold on to 00:12:53.600 --> 00:12:56.519 as things keep changing. Go for it. First, image 00:12:56.519 --> 00:12:58.600 models have hit a critical point of maturity. 00:12:58.940 --> 00:13:02.200 The kind of control and speed we see in GPT image 00:13:02.200 --> 00:13:06.179 1 .5 means these tools are now not just optional 00:13:06.179 --> 00:13:08.679 for any kind of rapid creative work. They've 00:13:08.679 --> 00:13:10.480 solved the precision problem. They've solved 00:13:10.480 --> 00:13:13.860 the precision problem. Second, the real value 00:13:13.860 --> 00:13:16.759 of AI agents isn't handling your chores. The 00:13:16.759 --> 00:13:19.320 Harvard study is pretty clear. Agents are being 00:13:19.320 --> 00:13:22.019 used to augment cognitive labor. They help you 00:13:22.019 --> 00:13:24.779 think, research, and synthesize information better 00:13:24.779 --> 00:13:29.799 and faster. And third, the key to navigating 00:13:29.799 --> 00:13:32.220 all this noise is just understanding the architecture. 00:13:32.600 --> 00:13:34.860 Knowing the difference between a single task 00:13:34.860 --> 00:13:37.879 tool, the LLM engine, and the workflow managing 00:13:37.879 --> 00:13:40.919 agent. That's the essential shortcut. And if 00:13:40.919 --> 00:13:43.779 we take that study's findings seriously. That 00:13:43.779 --> 00:13:47.000 agent usage links so strongly to higher education 00:13:47.000 --> 00:13:50.080 and higher GDP. That brings up a pretty provocative 00:13:50.080 --> 00:13:52.480 question for the future. Yeah. If these agents 00:13:52.480 --> 00:13:54.820 are primarily tools for enhancing high level 00:13:54.820 --> 00:13:57.240 thought, for accelerating the work of the already 00:13:57.240 --> 00:14:00.960 well -educated. What changes in education or 00:14:00.960 --> 00:14:02.940 training or accessibility do we need to make 00:14:02.940 --> 00:14:05.679 sure their true power benefits society broadly 00:14:05.679 --> 00:14:07.879 and not just, you know, the top tier of knowledge 00:14:07.879 --> 00:14:09.620 workers? That's the challenge for the next five 00:14:09.620 --> 00:14:12.039 years. Keep exploring those edges of knowledge. 00:14:12.279 --> 00:14:14.080 Thank you for sharing your sources and for diving 00:14:14.080 --> 00:14:15.799 deep with us today. We'll talk to you next time.