WEBVTT 00:00:00.000 --> 00:00:03.299 So there's been a lot of buzz recently around 00:00:03.299 --> 00:00:06.639 OpenAI's latest thing, the chat GPT agent. People 00:00:06.639 --> 00:00:09.640 are calling it a profound shift, like suggesting 00:00:09.640 --> 00:00:12.019 an AI that doesn't just respond, but actually 00:00:12.019 --> 00:00:15.099 acts across your digital life. Welcome to the 00:00:15.099 --> 00:00:17.940 deep dive, everyone. Today, yeah, we're plunging 00:00:17.940 --> 00:00:21.300 into this truly fascinating world of AI agents. 00:00:22.000 --> 00:00:24.600 Our mission really is to unpack what this new 00:00:24.600 --> 00:00:27.179 chat GPT agent actually means for you. And also 00:00:27.179 --> 00:00:29.460 look at some other big advancements that are 00:00:29.460 --> 00:00:31.760 fundamentally reshaping how we interact with 00:00:31.760 --> 00:00:34.200 tech. Right. We'll explore its core capabilities, 00:00:34.719 --> 00:00:37.200 how it performs in the, let's say the messy real 00:00:37.200 --> 00:00:40.399 world, how it stacks up against competitors and 00:00:40.399 --> 00:00:42.840 really what does all this mean for your daily 00:00:42.840 --> 00:00:45.340 tasks, your future workflows. And our goal, as 00:00:45.340 --> 00:00:47.740 always, is just a calm, curious, sort of down 00:00:47.740 --> 00:00:49.359 -to -earth look at this landscape. It's moving 00:00:49.359 --> 00:00:52.340 so fast. OK, let's get into it then. Sam Altman, 00:00:52.520 --> 00:00:56.000 OpenAI CEO, he called this agent a banger. What 00:00:56.000 --> 00:00:58.619 exactly is this agent? Beyond the hype, why is 00:00:58.619 --> 00:01:02.100 it such a big deal? Well, I think what really 00:01:02.100 --> 00:01:04.920 stands out is that agent isn't just another standalone 00:01:04.920 --> 00:01:08.319 model. It's not like just a new version of ChatGPT. 00:01:08.340 --> 00:01:12.599 OK. It's a Seamless integration. It's really 00:01:12.599 --> 00:01:15.040 a redefinition of what an AI assistant can be. 00:01:15.280 --> 00:01:18.459 OpenAI showed it off on July 17th. And it's this 00:01:18.459 --> 00:01:20.719 deeply integrated system, pulling together a 00:01:20.719 --> 00:01:23.659 bunch of powerful tools. So it's about combining 00:01:23.659 --> 00:01:28.859 these sort of individual powers into one cohesive 00:01:28.859 --> 00:01:31.019 brain. Precisely. Yeah. Think of it like bringing 00:01:31.019 --> 00:01:32.840 together tools that used to be separate, but 00:01:32.840 --> 00:01:36.790 now they're under one... So first, it has this 00:01:36.790 --> 00:01:38.969 deep research capability. It accesses the internet, 00:01:39.310 --> 00:01:41.569 spends real time kind of synthesizing vast amounts 00:01:41.569 --> 00:01:44.250 of info, offers more depth, more nuance than 00:01:44.250 --> 00:01:47.030 older AIs. Second, and this is a huge leap, it 00:01:47.030 --> 00:01:49.109 has computer use, or what they call operator. 00:01:49.170 --> 00:01:51.250 Operator. This means it literally controls a 00:01:51.250 --> 00:01:53.829 computer like a human would. Clicking, scrolling, 00:01:54.010 --> 00:01:57.650 filling forms, navigating complex websites. Past 00:01:57.650 --> 00:02:00.430 versions of this, like AI enhanced RPA that's 00:02:00.430 --> 00:02:02.969 a robotic process automation. Right, bots mimicking 00:02:02.969 --> 00:02:05.099 clicks. Yeah, they were often just too unstable, 00:02:05.319 --> 00:02:08.069 too brittle. Agents seems to have made... significant 00:02:08.069 --> 00:02:10.050 improvements there. So it's not just searching 00:02:10.050 --> 00:02:13.050 the web, it's actually using a browser, like 00:02:13.050 --> 00:02:15.409 interacting with sites dynamically, that feels 00:02:15.409 --> 00:02:17.569 fundamentally different. Exactly. And it also 00:02:17.569 --> 00:02:20.629 has robust code execution. It can write and run 00:02:20.629 --> 00:02:23.909 code, mostly Python, in a secure sandbox environment. 00:02:23.969 --> 00:02:26.349 Okay. And then it uses the results for complex 00:02:26.349 --> 00:02:29.830 requests, like analyzing big data sets or generating 00:02:29.830 --> 00:02:32.610 charts right there. And images too! Yeah. Finally, 00:02:32.650 --> 00:02:35.310 it's directly integrated with Daily 3 for image 00:02:35.310 --> 00:02:38.319 generation. It can create illustrations, charts, 00:02:38.879 --> 00:02:40.979 maybe artwork based on its research or analysis, 00:02:41.460 --> 00:02:43.479 all without you leaving the conversation. That 00:02:43.479 --> 00:02:46.120 flexible coordination does sound incredibly powerful. 00:02:46.340 --> 00:02:48.900 But for a lot of listeners, the word autonomous 00:02:48.900 --> 00:02:51.879 might raise some flags, right, given past AI 00:02:51.879 --> 00:02:54.539 issues. True. How truly autonomous is it right 00:02:54.539 --> 00:02:57.419 now? And what guardrails are there to stop it 00:02:57.419 --> 00:03:00.500 from going off the rails or making big mistakes? 00:03:00.659 --> 00:03:02.879 It's definitely a step closer to autonomous. 00:03:03.400 --> 00:03:06.110 The real power isn't just one tool. It's the 00:03:06.110 --> 00:03:09.250 agent flexibly deciding, okay, now I need to 00:03:09.250 --> 00:03:11.050 research, now I should code, now I need to use 00:03:11.050 --> 00:03:14.990 the computer. It really is a step toward a more 00:03:14.990 --> 00:03:18.110 truly autonomous agent. It changes the game from 00:03:18.110 --> 00:03:22.490 simple queries to genuinely autonomous multi 00:03:22.490 --> 00:03:25.370 -step task completion. So how does that seamless 00:03:25.370 --> 00:03:28.889 integration really change the nature of how users 00:03:28.889 --> 00:03:31.830 interact with AI? It shifts from asking simple 00:03:31.830 --> 00:03:35.009 questions to getting the AI to complete complex 00:03:35.009 --> 00:03:37.969 multi -step tasks autonomously. Got it. And who 00:03:37.969 --> 00:03:40.729 gets access? And when? It's rolling out in phases. 00:03:41.129 --> 00:03:43.110 So, pray users, they get full access right now. 00:03:43.449 --> 00:03:45.830 Plus, N -Team's users will get access soon, probably 00:03:45.830 --> 00:03:47.810 with some initial limits. Then education and 00:03:47.810 --> 00:03:49.710 enterprise later on. If you're on the free plan, 00:03:49.870 --> 00:03:51.229 well, you'll need to upgrade to get the full 00:03:51.229 --> 00:03:53.419 agent experience. Theory is one thing, but does 00:03:53.419 --> 00:03:56.520 it actually work in the real world? Right, the 00:03:56.520 --> 00:03:58.280 million dollar question. Let's dive into some 00:03:58.280 --> 00:04:00.560 practical tests that really pushed agents limits. 00:04:00.860 --> 00:04:03.159 What did that in -depth news research challenge 00:04:03.159 --> 00:04:06.560 show? OK, so this test asked agent to analyze 00:04:06.560 --> 00:04:09.159 generative AI trends affecting digital marketing 00:04:09.159 --> 00:04:13.650 over the past month for a management memo. So 00:04:13.650 --> 00:04:16.970 complex. Needs current info, multiple sources. 00:04:17.670 --> 00:04:19.750 Exactly. Distinguishing important from irrelevant, 00:04:20.089 --> 00:04:22.290 industry context, synthesis, professional format, 00:04:22.730 --> 00:04:25.329 right time frame. Lots going on. And the real 00:04:25.329 --> 00:04:28.029 world results. How did it do? Agent took about 00:04:28.029 --> 00:04:31.170 15 minutes. Yeah. And it produced a memo that 00:04:31.170 --> 00:04:35.250 was maybe 80, 85 % of the quality a human analyst 00:04:35.250 --> 00:04:38.029 team would produce in several hours. Wow. Previous 00:04:38.029 --> 00:04:40.629 AIs might only hit like 50 % accuracy on something 00:04:40.629 --> 00:04:43.009 like that. It correctly picked up major stories. 00:04:43.470 --> 00:04:45.350 The agent launched itself, new Mistral models, 00:04:45.529 --> 00:04:47.970 Google search updates, structured it professionally. 00:04:48.029 --> 00:04:50.610 Any issues? Minor stuff. Some info slightly outside 00:04:50.610 --> 00:04:52.750 the time frame missed some niche discussions 00:04:52.750 --> 00:04:56.569 happening on, say, X or Twitter. OK. So conclusion. 00:04:56.860 --> 00:04:58.839 It's a powerful professional research assistant, 00:04:58.839 --> 00:05:01.199 definitely, but you still need that human oversight. 00:05:01.360 --> 00:05:03.540 That's still a huge leap for research tasks. 00:05:03.860 --> 00:05:05.680 What about something really specific and precise, 00:05:05.839 --> 00:05:08.019 like financial data retrieval and calculations? 00:05:08.360 --> 00:05:10.399 Right, this is more of a clear pass -fail test. 00:05:11.000 --> 00:05:14.560 The task, go to Yahoo Finance, find the VINFAST 00:05:14.560 --> 00:05:18.959 ticker VFS on NASDAQ, extract the last five closing 00:05:18.959 --> 00:05:21.860 prices, calculate the average daily percentage 00:05:21.860 --> 00:05:24.519 change, and present it all on a table. So it 00:05:24.519 --> 00:05:28.040 combines web navigation and code execution. Verdict. 00:05:28.560 --> 00:05:31.000 Did it nail it? Flawless. Yeah, it navigated 00:05:31.000 --> 00:05:33.860 Yahoo Finance, found the data, wrote a Python 00:05:33.860 --> 00:05:36.180 script for the calculation, presented it perfectly 00:05:36.180 --> 00:05:40.480 in a table. Many previous AIs would just stumble 00:05:40.480 --> 00:05:42.339 or fail outright on one of those steps, especially 00:05:42.339 --> 00:05:44.740 combining them. Agent handled it cleanly. Whoa. 00:05:44.980 --> 00:05:47.120 Imagine scaling that kind of precision, analyzing 00:05:47.120 --> 00:05:49.899 entire portfolios instantly. That really could 00:05:49.899 --> 00:05:51.620 change things for financial tasks, couldn't it? 00:05:51.639 --> 00:05:53.670 Two secs silence. OK, but what about something 00:05:53.670 --> 00:05:55.970 totally different, like managing, say, the flood 00:05:55.970 --> 00:05:59.129 of daily emails? How did Agent handle email complexity? 00:05:59.410 --> 00:06:01.810 Yeah, this task involved connecting to Gmail, 00:06:02.189 --> 00:06:04.730 reviewing 10 recent support emails. Just 10? 00:06:04.870 --> 00:06:07.870 OK. Yeah, start small. Then classify them technical, 00:06:08.250 --> 00:06:11.509 feedback, billing, and then draft replies using 00:06:11.509 --> 00:06:15.269 uploaded FAQ and return policy files. So this 00:06:15.269 --> 00:06:18.629 tests external access, reading context, classification, 00:06:19.189 --> 00:06:21.720 and RG retrieval augmented generation. That's 00:06:21.720 --> 00:06:24.060 where it pulls info from documents you give it, 00:06:24.120 --> 00:06:25.740 right? Exactly. It needs to understand the docs 00:06:25.740 --> 00:06:27.740 to draft good replies. That's a lot of context 00:06:27.740 --> 00:06:29.980 and steps. How did it go? Did it in about three 00:06:29.980 --> 00:06:33.040 minutes, correctly classified each email, drafted 00:06:33.040 --> 00:06:35.360 tailored replies that actually cited the documents, 00:06:35.399 --> 00:06:38.420 and even offered to create actual Gmail drafts 00:06:38.420 --> 00:06:41.040 for you. So potentially huge for customer service 00:06:41.040 --> 00:06:43.319 automation using internal knowledge. But you 00:06:43.319 --> 00:06:45.850 mentioned a limitation. Right. Important one. 00:06:46.329 --> 00:06:48.949 Agent can get overwhelmed by too much info, trying 00:06:48.949 --> 00:06:51.250 to process, say, a thousand emails instead of 00:06:51.250 --> 00:06:53.550 ten. It might struggle. You definitely need to 00:06:53.550 --> 00:06:55.850 understand its limits. Got it. So scale matters. 00:06:56.110 --> 00:06:58.129 And finally, product research and comparison. 00:06:58.430 --> 00:07:00.990 This used to be so hit or miss with AI, almost 00:07:00.990 --> 00:07:04.310 felt like a gamble. How did it do here? OK, the 00:07:04.310 --> 00:07:06.610 task was research and compare the five best air 00:07:06.610 --> 00:07:09.129 purifier models for Vietnamese cities under a 00:07:09.129 --> 00:07:12.949 specific budget, five million VND. Very specific. 00:07:13.029 --> 00:07:15.699 Yeah. and create a comparison table with exact 00:07:15.699 --> 00:07:18.920 columns. Product name, reference price, suitable 00:07:18.920 --> 00:07:21.959 room area, filtration technology, and a retailer 00:07:21.959 --> 00:07:24.500 link. And did it deliver that kind of consistency 00:07:24.500 --> 00:07:27.560 we saw with the financial data? Pretty much exactly 00:07:27.560 --> 00:07:30.180 as requested. Yeah. It found models. used the 00:07:30.180 --> 00:07:32.639 precise table format, consulted local e -commerce 00:07:32.639 --> 00:07:35.259 sites, gave detailed feature comparisons. Any 00:07:35.259 --> 00:07:38.000 glitches? Minor one. Maybe one or two product 00:07:38.000 --> 00:07:40.879 links went to general category pages, not the 00:07:40.879 --> 00:07:43.860 specific product. OK. But the significance is 00:07:43.860 --> 00:07:46.800 huge. This kind of detailed, structured consumer 00:07:46.800 --> 00:07:49.639 research used to be really unreliable with AI. 00:07:49.819 --> 00:07:52.519 Agent makes it genuinely useful. It performs 00:07:52.519 --> 00:07:55.079 research almost like an experienced human would. 00:07:55.199 --> 00:07:57.019 It really does sound like Agent crossed some 00:07:57.019 --> 00:07:59.180 kind of critical threshold, moving from just 00:07:59.180 --> 00:08:02.120 experimental to genuinely reliable for complex 00:08:02.120 --> 00:08:05.060 stuff. What truly sets it apart from previous 00:08:05.060 --> 00:08:07.310 attempts we've seen? I think the real breakthrough 00:08:07.310 --> 00:08:09.610 is, yeah, while other companies definitely experimented 00:08:09.610 --> 00:08:11.990 with similar ideas, agents seem to have finally 00:08:11.990 --> 00:08:14.910 crossed that crucial threshold. It feels reliable 00:08:14.910 --> 00:08:18.050 enough for regular, practical use. Previous tools 00:08:18.050 --> 00:08:20.449 often lacked the stability, or the deep integration, 00:08:20.709 --> 00:08:23.689 or the consistent accuracy, or they just couldn't 00:08:23.689 --> 00:08:27.290 handle real -world complexity well. Agent seems 00:08:27.290 --> 00:08:29.449 to deliver consistent results much more often, 00:08:29.750 --> 00:08:31.870 turning that promise of an AI assistant into 00:08:31.870 --> 00:08:34.309 more of a practical reality. So given this new 00:08:34.309 --> 00:08:36.570 level of capability, what do you think is the 00:08:36.570 --> 00:08:39.269 biggest hurdle users might face when they try 00:08:39.269 --> 00:08:42.490 to apply Agent to their own, maybe unique, complex 00:08:42.490 --> 00:08:44.629 workflows? Probably setting clear boundaries 00:08:44.629 --> 00:08:47.070 and managing the information input effectively 00:08:47.070 --> 00:08:50.200 for the AI. Okay, so... Agent can talk to my 00:08:50.200 --> 00:08:53.159 Gmail, analyze spreadsheets. How does that even 00:08:53.159 --> 00:08:55.419 work? What exactly are these AI connectors we 00:08:55.419 --> 00:08:58.240 keep hearing about? AI connectors, they're essentially 00:08:58.240 --> 00:09:01.080 like intelligent bridges. They let AI assistants 00:09:01.080 --> 00:09:04.320 seamlessly access and interact with your external 00:09:04.320 --> 00:09:06.820 accounts and services, you know, Gmail, Google 00:09:06.820 --> 00:09:09.340 Drive, Slack, Microsoft 365, things like that. 00:09:09.379 --> 00:09:11.679 So the AI works directly within your existing 00:09:11.679 --> 00:09:13.960 workflows, not just in some isolated chat window, 00:09:14.220 --> 00:09:16.500 not just conversation. It's interaction with 00:09:16.500 --> 00:09:19.399 your whole digital setup. like stacking Lego 00:09:19.399 --> 00:09:21.620 blocks of data and actions together. Interesting 00:09:21.620 --> 00:09:24.019 analogy. And I've heard there are different approaches 00:09:24.019 --> 00:09:27.700 here, like chat GPT agent versus, say, Claude's 00:09:27.700 --> 00:09:29.879 connectors. How do they differ? That's a great 00:09:29.879 --> 00:09:32.460 point. They represent pretty different philosophies 00:09:32.460 --> 00:09:36.000 about AI integration. Agents connectors are deeply 00:09:36.000 --> 00:09:39.580 integrated with its full suite. So email plus 00:09:39.580 --> 00:09:43.179 web research plus code plus images. OK. This 00:09:43.179 --> 00:09:46.259 makes it really good for complex, multi -step, 00:09:46.480 --> 00:09:48.879 flexible workflows where you need lots of different 00:09:48.879 --> 00:09:50.960 tools working together. The philosophy seems 00:09:50.960 --> 00:09:54.039 to be build a central brain that orchestrates 00:09:54.039 --> 00:09:56.639 many tools. Right. And Claude. Claude's connectors, 00:09:56.679 --> 00:09:59.179 on the other hand, tend to focus on simpler, 00:09:59.659 --> 00:10:02.080 really reliable connections to external services. 00:10:02.500 --> 00:10:05.179 They're excellent for more focused, simpler tasks 00:10:05.179 --> 00:10:07.960 like summarize my unread emails in the hashtag 00:10:07.960 --> 00:10:10.220 general Slack channel. Gotcha. Their philosophy 00:10:10.220 --> 00:10:12.360 feels more like providing reliable specialized 00:10:12.360 --> 00:10:14.980 tools the model can call on when needed. Less 00:10:14.980 --> 00:10:17.500 orchestration, more specific capabilities. So 00:10:17.500 --> 00:10:19.720 one's like a symphony conductor for your whole 00:10:19.720 --> 00:10:22.440 digital life, maybe? And the other is more like 00:10:22.440 --> 00:10:24.679 a set of specialist mechanics, each really good 00:10:24.679 --> 00:10:28.000 at one specific job. Fascinating. Yeah, that's 00:10:28.000 --> 00:10:30.120 a decent way to put it. So when we start using 00:10:30.120 --> 00:10:32.600 these connectors, what are some tips for doing 00:10:32.600 --> 00:10:36.100 it safely and effectively? Good question. First, 00:10:36.580 --> 00:10:39.169 definitely start small. Begin with simple tasks 00:10:39.169 --> 00:10:41.250 before you try to automate something really complex. 00:10:42.409 --> 00:10:45.730 Second, set clear boundaries. Be really explicit, 00:10:45.929 --> 00:10:48.970 say, analyze these five emails from this sender, 00:10:49.129 --> 00:10:51.490 not just check my inbox. Right, specificity. 00:10:51.669 --> 00:10:55.129 Third, provide context. Upload relevant documents, 00:10:55.669 --> 00:10:58.429 internal guides, policies, templates, give the 00:10:58.429 --> 00:11:02.429 AI the background it needs. Fourth, Always, always 00:11:02.429 --> 00:11:05.529 review thoroughly. Look over AI -generated responses 00:11:05.529 --> 00:11:07.730 or actions before you let them happen. Makes 00:11:07.730 --> 00:11:10.350 sense. And finally, manage permissions. Go back 00:11:10.350 --> 00:11:13.210 regularly and review what services the AI has 00:11:13.210 --> 00:11:15.649 access to. Revoke access you don't need anymore. 00:11:15.850 --> 00:11:17.769 You know, I still wrestle with prompt rest myself 00:11:17.769 --> 00:11:20.090 sometimes, so that specificity really does help 00:11:20.090 --> 00:11:22.110 get the AI to deliver what you actually want. 00:11:22.309 --> 00:11:24.370 That's helpful to hear, yeah. Beyond just the 00:11:24.370 --> 00:11:27.070 features, what does this difference in philosophies 00:11:27.070 --> 00:11:30.070 orchestration versus specialization tell us about 00:11:30.070 --> 00:11:32.210 where AI integration might be heading more broadly? 00:11:32.529 --> 00:11:36.029 It suggests maybe two paths. One, prioritizing 00:11:36.029 --> 00:11:39.090 holistic orchestration. The other, specialized 00:11:39.090 --> 00:11:41.730 reliable automation. Interesting. OK, it wasn't 00:11:41.730 --> 00:11:44.009 just OpenAI making waves this past week, though. 00:11:44.149 --> 00:11:46.429 The whole AI landscape seems to be just constantly 00:11:46.429 --> 00:11:48.710 shifting. What else caught your eye that felt 00:11:48.710 --> 00:11:51.889 significant? Yeah, a big one was Grok. That's 00:11:51.889 --> 00:11:54.870 XAI's product -launching AI companions, characters 00:11:54.870 --> 00:11:58.269 named Annie and Rudy. Companions, like friends. 00:11:58.429 --> 00:12:00.669 Basically, yeah. Designed as digital friends, 00:12:00.809 --> 00:12:03.330 not just work assistants. This feels like a real 00:12:03.330 --> 00:12:06.700 shift, you know, from AI... purely as a productivity 00:12:06.700 --> 00:12:10.200 tool towards AI as a social companion. It seems 00:12:10.200 --> 00:12:12.700 targeted at tech enthusiasts right now, but it 00:12:12.700 --> 00:12:15.019 raises some really fascinating psychological 00:12:15.019 --> 00:12:17.299 and social questions about our future relationships 00:12:17.299 --> 00:12:19.379 with AI. Definitely something to watch. And then 00:12:19.379 --> 00:12:21.179 there was Higgs Field's sole ID. Sounds like 00:12:21.179 --> 00:12:23.779 it's making AI selfies incredibly realistic. 00:12:23.960 --> 00:12:26.299 Yes, Higgs Field put out an updated AI image 00:12:26.299 --> 00:12:29.539 generator. It specializes in remarkably realistic 00:12:29.539 --> 00:12:31.679 photos of people. How does it work? It needs 00:12:31.679 --> 00:12:34.360 about 20, 25 photos of a person to train on. 00:12:34.759 --> 00:12:37.240 Then it generates these very natural, almost 00:12:37.240 --> 00:12:39.919 iPhone -looking images, focusing on expressions. 00:12:40.919 --> 00:12:44.000 What's striking is how good they are. They can 00:12:44.000 --> 00:12:47.000 genuinely fool the human eye. Wow. You can imagine 00:12:47.000 --> 00:12:49.700 uses in social media, maybe professional headshots. 00:12:50.179 --> 00:12:53.049 That uncanny valley is shrinking fast. It really 00:12:53.049 --> 00:12:54.929 is. What about on the open source side? We saw 00:12:54.929 --> 00:12:57.649 some cool updates there, too. Voxtral from Mistral 00:12:57.649 --> 00:12:59.870 came out. That's an open source speech recognition 00:12:59.870 --> 00:13:02.169 model. Speech recognition, like transcribing 00:13:02.169 --> 00:13:04.509 audio. Exactly. And it's competitive with things 00:13:04.509 --> 00:13:07.649 like 11Lab Scribe and OpenAI's Whisper. That's 00:13:07.649 --> 00:13:09.750 great for developers, makes high quality transcription 00:13:09.750 --> 00:13:12.529 more accessible. Nice. And then there's KimiK2 00:13:12.529 --> 00:13:15.470 from a Chinese startup called Moonshot AI. This 00:13:15.470 --> 00:13:17.850 thing is massive, a trillion parameters. It uses 00:13:17.850 --> 00:13:20.750 a mixture of experts or Moe architecture. A mixture 00:13:20.750 --> 00:13:23.460 of experts. What's that mean? It's basically 00:13:23.460 --> 00:13:26.759 like having a team of specialized AI models working 00:13:26.759 --> 00:13:29.059 together on different parts of a problem. makes 00:13:29.059 --> 00:13:31.919 it incredibly powerful. And its performance on 00:13:31.919 --> 00:13:34.620 benchmarks is really competitive. It definitely 00:13:34.620 --> 00:13:37.659 signals the growing strength of Chinese AI companies 00:13:37.659 --> 00:13:40.019 on the global stage. Okay, a lot happening there. 00:13:40.399 --> 00:13:43.379 And for video generation, Runway Act 2 got an 00:13:43.379 --> 00:13:45.340 important upgrade too, right? Getting closer 00:13:45.340 --> 00:13:47.139 to practical use and things like filmmaking. 00:13:47.320 --> 00:13:50.059 Absolutely. Runway Act 2 is their upgraded AI 00:13:50.059 --> 00:13:53.559 video tech. It can take an input video, say... 00:13:53.549 --> 00:13:57.429 of a person's motion and use that motion to drive 00:13:57.429 --> 00:14:01.330 completely AI -generated content. So a person 00:14:01.330 --> 00:14:03.570 in a suit could become an astronaut or maybe 00:14:03.570 --> 00:14:06.409 a superhero, but their original motion is preserved 00:14:06.409 --> 00:14:08.929 precisely. So the movement looks real even if 00:14:08.929 --> 00:14:10.950 the character changes. Exactly. It's much more 00:14:10.950 --> 00:14:13.269 precise now, less distorted than earlier versions. 00:14:13.509 --> 00:14:15.870 That brings us way closer to practical uses in 00:14:15.870 --> 00:14:19.110 filmmaking, advertising, anywhere. Motion fidelity 00:14:19.110 --> 00:14:21.230 is key. Cool. And you mentioned something about 00:14:21.230 --> 00:14:24.039 transparency. With Grok. Yeah, this was interesting. 00:14:24.379 --> 00:14:26.559 So Grok initially gave some inappropriate answers, 00:14:26.820 --> 00:14:28.940 apparently because of viral memes in its training 00:14:28.940 --> 00:14:32.019 data. Yeah. So they fixed it by updating the 00:14:32.019 --> 00:14:34.039 system prompt, basically the core instructions 00:14:34.039 --> 00:14:37.460 the AI follows. But what really matters is that 00:14:37.460 --> 00:14:40.940 the Grok team published the exact system prompt 00:14:40.940 --> 00:14:43.860 changes on GitHub. They showed their work. Exactly. 00:14:44.200 --> 00:14:46.299 Showed precisely what they modified and why. 00:14:47.000 --> 00:14:48.820 That level of transparency is pretty rare in 00:14:48.820 --> 00:14:50.480 this competitive field, and it does a lot to 00:14:50.480 --> 00:14:53.480 build trust, I think. That is unusual. So from 00:14:53.480 --> 00:14:56.840 AI companions to hyper real visuals, open source 00:14:56.840 --> 00:15:00.620 advances, it's a lot. Which of these other developments 00:15:00.620 --> 00:15:02.899 do you think might have the most immediate impact 00:15:02.899 --> 00:15:05.889 on daily life for the average person? Probably 00:15:05.889 --> 00:15:09.250 the AI companions. Shifting AI from just tasks 00:15:09.250 --> 00:15:12.450 towards actual relationships is a really significant 00:15:12.450 --> 00:15:14.309 step, I think. Yeah, potentially transformative. 00:15:14.610 --> 00:15:16.870 OK, bringing it back to agent, who is ChatGPP 00:15:16.870 --> 00:15:19.889 Agent really for, and how can someone start using 00:15:19.889 --> 00:15:22.370 it effectively without feeling totally overwhelmed? 00:15:22.690 --> 00:15:25.590 Right. So ChatGPP Agent seems perfect for researchers, 00:15:25.970 --> 00:15:28.309 analysts, people who need to synthesize info 00:15:28.309 --> 00:15:30.289 quickly, often from lots of different places. 00:15:30.350 --> 00:15:32.649 Thanks. Also. Marketing and comms professionals, 00:15:33.070 --> 00:15:35.070 content creators who are combining research, 00:15:35.210 --> 00:15:37.950 writing, maybe visuals. Small business owners, 00:15:38.169 --> 00:15:40.809 too, for things like customer support or market 00:15:40.809 --> 00:15:43.730 analysis. Really, anyone who frequently combines 00:15:43.730 --> 00:15:46.289 multiple types of digital work. And who might 00:15:46.289 --> 00:15:49.190 it not be ideal for, at least right now? Well, 00:15:49.490 --> 00:15:51.889 probably not ideal for really simple single purpose 00:15:51.889 --> 00:15:55.259 tasks where regular chat GPT might be fine. or 00:15:55.259 --> 00:15:58.600 for users who need absolute 100 % accuracy for 00:15:58.600 --> 00:16:01.419 truly mission -critical decisions, or tasks that 00:16:01.419 --> 00:16:03.919 require really nuanced human judgment on sensitive 00:16:03.919 --> 00:16:07.419 stuff, it's powerful, but it's still an AI. Got 00:16:07.419 --> 00:16:09.960 it. So for those who fit the profile and are 00:16:09.960 --> 00:16:11.539 ready to jump in, what are the best practices? 00:16:11.700 --> 00:16:14.080 How do you actually get value from agents? OK, 00:16:14.080 --> 00:16:16.320 first, be incredibly clear and specific with 00:16:16.320 --> 00:16:18.360 your requests. Don't just say help with email. 00:16:18.679 --> 00:16:20.779 Try something like draft replies to the five 00:16:20.779 --> 00:16:23.220 most recent emails with complaint in the subject. 00:16:23.659 --> 00:16:25.899 Use the complaint -response .docsis template 00:16:25.899 --> 00:16:28.500 as a guide. Then ask for my approval before doing 00:16:28.500 --> 00:16:31.789 anything else. OK, very precise. Second, provide 00:16:31.789 --> 00:16:34.029 those context documents we talked about, upload, 00:16:34.190 --> 00:16:35.889 guides, templates, before you start the task. 00:16:36.470 --> 00:16:39.669 Third, use confirmation steps for important things. 00:16:40.110 --> 00:16:42.490 Ask agent, have you understood the data needs 00:16:42.490 --> 00:16:45.250 to be for Q2? Something like that. Check its 00:16:45.250 --> 00:16:48.610 understanding. Exactly. Fourth, combine its capabilities 00:16:48.610 --> 00:16:51.309 strategically. Think about the steps. And finally, 00:16:51.710 --> 00:16:54.529 set realistic expectations. Agent is powerful, 00:16:54.870 --> 00:16:57.340 but it's not perfect. Always review its work. 00:16:57.399 --> 00:16:59.379 It sounds like a new kind of workflow design 00:16:59.379 --> 00:17:01.620 almost like programming an assistant, but in 00:17:01.620 --> 00:17:04.279 plain English kind of yeah What are some sample 00:17:04.279 --> 00:17:06.240 workflows someone could try with agent things 00:17:06.240 --> 00:17:08.859 that really show off its unique abilities? Oh 00:17:08.859 --> 00:17:10.460 the possibilities are getting really exciting 00:17:10.460 --> 00:17:14.259 now imagine Content planning and creation you 00:17:14.259 --> 00:17:16.900 could prompt it to research trending topics in 00:17:16.900 --> 00:17:19.539 say sustainable development Okay, then propose 00:17:19.539 --> 00:17:22.460 a list of blog post ideas and for the top idea 00:17:22.460 --> 00:17:25.019 create a detailed outline and generate a suitable 00:17:25.860 --> 00:17:29.720 image, all in one go, or customer feedback analysis. 00:17:30.059 --> 00:17:32.279 Connect to Google Drive, tell it to read a CSV 00:17:32.279 --> 00:17:35.079 file, identify the top complaints and the features 00:17:35.079 --> 00:17:37.740 people love, create a bar chart from that data, 00:17:37.960 --> 00:17:40.059 and then draft a brief summary for a team meeting. 00:17:40.099 --> 00:17:43.140 That's powerful. Or even market entry research, 00:17:43.900 --> 00:17:46.079 maybe for specialty coffee in the Thai market. 00:17:46.380 --> 00:17:49.740 Ask for an overview report on competitors, average 00:17:49.740 --> 00:17:53.480 prices, import regulations, and present it all 00:17:53.480 --> 00:17:56.490 as a slide deck outline. It really does feel 00:17:56.490 --> 00:17:58.849 like a paradigm shift brewing. How do you see 00:17:58.849 --> 00:18:01.930 the evolution of AI assistance continuing from 00:18:01.930 --> 00:18:04.369 here, building on what Agent has started? Yeah, 00:18:04.490 --> 00:18:06.710 what feels revolutionary is that Agent really 00:18:06.710 --> 00:18:09.069 signifies this profound evolution. We're moving 00:18:09.069 --> 00:18:11.410 from single -purpose tools, you know, a grammar 00:18:11.410 --> 00:18:14.720 checker here, an image generator there, to deeply 00:18:14.720 --> 00:18:17.940 integrated assistance, from simple Q &A to actually 00:18:17.940 --> 00:18:20.779 completing complex tasks, and from AI needing 00:18:20.779 --> 00:18:23.660 human supervision at every single step to much 00:18:23.660 --> 00:18:26.579 more autonomous AI agents that can chain multiple 00:18:26.579 --> 00:18:29.410 actions together intelligently. And for us as 00:18:29.410 --> 00:18:31.490 professionals, how is this going to impact our 00:18:31.490 --> 00:18:34.170 day -to -day work? Our productivity, where do 00:18:34.170 --> 00:18:36.430 humans fit in this picture? Well, I think we'll 00:18:36.430 --> 00:18:39.009 see routine research and analysis get increasingly 00:18:39.009 --> 00:18:41.609 automated. That frees up our cognitive space 00:18:41.609 --> 00:18:43.809 for higher level thinking. Creative work will 00:18:43.809 --> 00:18:47.670 likely involve more human AI collaboration. Humans 00:18:47.670 --> 00:18:50.910 provide the strategy, the vision. AI handles 00:18:50.910 --> 00:18:53.869 more of the execution, maybe the iteration, communication. 00:18:54.400 --> 00:18:57.240 probably more AI assisted drafting, making interactions 00:18:57.240 --> 00:19:00.359 more efficient. And for complex problem solving, 00:19:00.779 --> 00:19:03.240 it'll be about combining human strategy with 00:19:03.240 --> 00:19:06.180 AI's execution power, letting us tackle bigger, 00:19:06.480 --> 00:19:09.170 harder problems. Which means the essential skills 00:19:09.170 --> 00:19:11.730 we need as professionals are also shifting, right? 00:19:12.089 --> 00:19:14.529 What are those core skills we need to be cultivating 00:19:14.529 --> 00:19:16.650 now? Absolutely. You'll definitely need prompt 00:19:16.650 --> 00:19:18.609 engineering learning to communicate precisely 00:19:18.609 --> 00:19:20.670 and effectively with AI. That's fundamental. 00:19:20.829 --> 00:19:23.130 Right. And systems thinking, designing workflows 00:19:23.130 --> 00:19:25.269 that optimally combine human strengths and AI 00:19:25.269 --> 00:19:28.299 strengths. Also, AI ethics and oversight being 00:19:28.299 --> 00:19:31.740 able to recognize and mitigate risks, biases 00:19:31.740 --> 00:19:34.799 in AI output. That's crucial. Critical finting, 00:19:34.940 --> 00:19:36.680 too, I imagine. Definitely. Critical thinking, 00:19:36.740 --> 00:19:39.039 evaluating the AI's output, knowing when it's 00:19:39.039 --> 00:19:41.720 right, when it's wrong, and crucially, when and 00:19:41.720 --> 00:19:44.710 how to intervene. Given all these new skills, 00:19:45.109 --> 00:19:47.910 what's the single most important mindset shift, 00:19:47.910 --> 00:19:50.309 do you think, for professionals trying to adapt 00:19:50.309 --> 00:19:54.190 to this rapidly evolving AI landscape? I'd say 00:19:54.190 --> 00:19:57.230 it's embracing human AI collaboration truly as 00:19:57.230 --> 00:20:00.190 an amplification of our abilities, not as a replacement 00:20:00.190 --> 00:20:02.349 for them. Amplification, not replacement. OK, 00:20:02.490 --> 00:20:05.789 so to recap our deep dive today, ChatGPT Agent 00:20:05.789 --> 00:20:08.589 feels like a genuine breakthrough for practical 00:20:08.589 --> 00:20:12.289 multi -step AI assistance. Yeah, it's reliably 00:20:12.289 --> 00:20:15.759 handling complex tasks that honestly used to 00:20:15.759 --> 00:20:18.200 need constant human handholding. It really marks 00:20:18.200 --> 00:20:20.440 a paradigm shift, doesn't it? From these isolated 00:20:20.440 --> 00:20:23.799 tools we used to use to more integrated, almost 00:20:23.799 --> 00:20:26.259 autonomous agents. Exactly. And that combination 00:20:26.259 --> 00:20:29.500 of human judgment and AI capability is becoming 00:20:29.500 --> 00:20:32.259 extraordinarily powerful. Really quickly. It's 00:20:32.259 --> 00:20:34.480 not about AI replacing humans, fundamentally. 00:20:34.579 --> 00:20:36.799 It's about humans amplifying what they can achieve 00:20:36.799 --> 00:20:39.279 with AI. Well said. So the question for you listening 00:20:39.279 --> 00:20:42.339 is, what workflow will you try first with ChatGPT 00:20:42.339 --> 00:20:44.380 Agent? The possibilities are really opening up. 00:20:44.460 --> 00:20:46.140 Yeah, we definitely encourage you to experiment, 00:20:46.319 --> 00:20:48.700 to explore what it can do. Because as these tools 00:20:48.700 --> 00:20:51.859 keep getting better, and they will, the critical 00:20:51.859 --> 00:20:54.920 question isn't really if you'll adopt them, but 00:20:54.920 --> 00:20:57.500 maybe how quickly you can learn to use them effectively. 00:20:57.640 --> 00:21:00.170 Right. the businesses, the individuals who master 00:21:00.170 --> 00:21:03.089 this human AI collaboration today, they're likely 00:21:03.089 --> 00:21:05.430 going to have significant advantages as these 00:21:05.430 --> 00:21:07.730 technologies just become standard everywhere 00:21:07.730 --> 00:21:10.250 in every industry. Thank you for diving deep 00:21:10.250 --> 00:21:12.630 with us today. We'll be back soon with more insights 00:21:12.630 --> 00:21:15.470 into the future of technology. OutTO Music.