WEBVTT 00:00:00.000 --> 00:00:03.200 So it's February 2026. And if you're like most 00:00:03.200 --> 00:00:05.059 people I talk to, you're probably looking at 00:00:05.059 --> 00:00:08.000 a credit card statement that it looks less like 00:00:08.000 --> 00:00:11.179 a bill and more like a menu. You've got your 00:00:11.179 --> 00:00:14.439 $20 a month for one AI, maybe another 20 for 00:00:14.439 --> 00:00:18.000 some pro model, 15 for an image generator. And 00:00:18.000 --> 00:00:20.820 it just, it adds up. Oh, it adds up fast. But 00:00:20.820 --> 00:00:23.420 here's the thing. While all the big labs are, 00:00:23.460 --> 00:00:25.100 you know, charging admission at the front door, 00:00:25.320 --> 00:00:28.679 there's this parallel reality happening, an underground 00:00:28.679 --> 00:00:32.500 network, so to speak, where the exact same technology 00:00:32.500 --> 00:00:35.240 and talking models like Gemini 3 .0 Pro, Claude 00:00:35.240 --> 00:00:38.560 Opus 4 .6 is just sitting there. Fully unlocked 00:00:38.560 --> 00:00:40.979 for absolutely nothing. It sounds like a scam 00:00:40.979 --> 00:00:42.700 when you say it out loud, doesn't it? Like one 00:00:42.700 --> 00:00:44.420 of those old one weird trick pop -up ads. It 00:00:44.420 --> 00:00:47.020 really does. But it's real. And honestly, it's 00:00:47.020 --> 00:00:49.039 the most exciting shift I've seen in this space 00:00:49.039 --> 00:00:51.179 all year. Yeah. We aren't just talking about 00:00:51.179 --> 00:00:53.259 saving a few bucks. We're talking about getting 00:00:53.259 --> 00:00:56.740 access to tools that are, in some pretty specific 00:00:56.740 --> 00:00:59.320 ways, actually better than what you pay for. 00:00:59.820 --> 00:01:02.399 Welcome to the Deep Dive. Today, we're going 00:01:02.399 --> 00:01:06.260 to explore this free AI revolution. We have a 00:01:06.260 --> 00:01:08.719 whole stack of research here on, I think, four 00:01:08.719 --> 00:01:11.040 specific platforms that are really changing the 00:01:11.040 --> 00:01:13.200 game right now. Yeah, four big ones. And the 00:01:13.200 --> 00:01:15.379 mission for this conversation isn't just to give 00:01:15.379 --> 00:01:18.599 you a list of free stuff. It's to understand 00:01:18.599 --> 00:01:21.840 a new way of working. Because, you know, usually 00:01:21.840 --> 00:01:24.599 when something's free online, you're the product. 00:01:24.840 --> 00:01:28.079 Or the quality is just... Damage, yeah. So I 00:01:28.079 --> 00:01:30.900 have to ask... Is this a deep dive into second 00:01:30.900 --> 00:01:33.799 rate knockoffs or is this actually usable tech? 00:01:34.400 --> 00:01:36.959 That's the skepticism I was expecting. And honestly, 00:01:37.120 --> 00:01:39.420 it's healthy. You should be skeptical. But what 00:01:39.420 --> 00:01:41.060 we're seeing in the research on these platforms, 00:01:41.280 --> 00:01:44.920 Arena AI, Veer, ChatGLM, and Quinn, it's not 00:01:44.920 --> 00:01:47.500 about knockoffs. It's about a fundamental shift 00:01:47.500 --> 00:01:50.040 in how compute is being distributed. We're not 00:01:50.040 --> 00:01:52.000 just talking about saving 50 bucks a month. We're 00:01:52.000 --> 00:01:53.280 talking about something called platform matching. 00:01:53.400 --> 00:01:55.680 Platform matching. Let's unpack that term before 00:01:55.680 --> 00:01:57.480 we get into the tools themselves. What does that 00:01:57.480 --> 00:02:00.379 actually mean for, say, the average user? Sure. 00:02:00.620 --> 00:02:03.019 For the last few years, people really identified 00:02:03.019 --> 00:02:05.989 with a brand. You were a chat GPT person or a 00:02:05.989 --> 00:02:07.930 cloud person. Right. You paid your subscription 00:02:07.930 --> 00:02:11.490 and you kind of forced that one model to do everything 00:02:11.490 --> 00:02:14.509 for you. Write code, make pictures, summarize 00:02:14.509 --> 00:02:18.169 your PDFs. Platform matching is the realization 00:02:18.169 --> 00:02:23.030 that in 2026, there is no single God model. This 00:02:23.030 --> 00:02:25.169 underground network lets you use the absolute 00:02:25.169 --> 00:02:28.590 best tool for a specific job without a paywall 00:02:28.590 --> 00:02:31.360 getting in your way. So it's moving from a generalist 00:02:31.360 --> 00:02:34.099 approach to a specialist one. Exactly. And it's 00:02:34.099 --> 00:02:36.520 about redundancy. If your main model hits a rate 00:02:36.520 --> 00:02:38.860 limit or maybe refuses to answer a sensitive 00:02:38.860 --> 00:02:41.340 question, you need a backup. This network gives 00:02:41.340 --> 00:02:43.520 you that. It's building a workflow that doesn't 00:02:43.520 --> 00:02:45.360 break. Okay, let's map this out then. We're going 00:02:45.360 --> 00:02:47.759 to look at four key players today, Arena AI, 00:02:48.099 --> 00:02:50.879 Veer, ChatGLM, and Quinn. By the time we're done, 00:02:50.979 --> 00:02:52.539 the goal is for you to know how you could build 00:02:52.539 --> 00:02:56.400 a full app, generate 4K video, and do deep academic 00:02:56.400 --> 00:02:58.460 -level research. All without pulling out your 00:02:58.460 --> 00:03:01.650 wallet. Not even once. That's the goal. So let's 00:03:01.650 --> 00:03:03.229 start with the philosophy of it. This source 00:03:03.229 --> 00:03:05.289 material talks about this concept of an underground 00:03:05.289 --> 00:03:07.949 network. And it feels like, you know, in 2026, 00:03:08.090 --> 00:03:10.770 the question just isn't, can I afford the best 00:03:10.770 --> 00:03:13.710 AI anymore? That's not. It's, do I know where 00:03:13.710 --> 00:03:16.030 to find it? Right. And the cornerstone of this 00:03:16.030 --> 00:03:19.030 whole network is a platform called Arena AI. 00:03:19.310 --> 00:03:21.710 Right. Now, you might know them from their leaderboards, 00:03:21.870 --> 00:03:24.770 but they just raised a massive amount of money, 00:03:24.849 --> 00:03:29.580 like $150 million to... basically democratize 00:03:29.580 --> 00:03:32.810 access to this stuff. Wow. And they're offering 00:03:32.810 --> 00:03:34.750 a feature that I think completely changes how 00:03:34.750 --> 00:03:36.689 we interact with these models. It's called Battle 00:03:36.689 --> 00:03:38.949 Mode. I love this concept, but I also find it 00:03:38.949 --> 00:03:41.310 a little unsettling. So let's unpack Battle Mode 00:03:41.310 --> 00:03:43.689 because usually we're biased, right? We see a 00:03:43.689 --> 00:03:45.990 logo, we see GPT -5 or whatever, and we just 00:03:45.990 --> 00:03:48.129 assume it's the best. Precisely. We are brand 00:03:48.129 --> 00:03:51.370 snobs. Yeah. Total snobs. If I show you an answer 00:03:51.370 --> 00:03:53.849 and I label it GPT -5, you are statistically 00:03:53.849 --> 00:03:55.930 more likely to rate it higher than the exact 00:03:55.930 --> 00:03:58.870 same answer labeled Llama Open Source. We trust 00:03:58.870 --> 00:04:00.990 the logo. We trust the logo. Yeah. So Arena's 00:04:00.990 --> 00:04:04.319 Battle Mode. It strips the logos away. It gamifies 00:04:04.319 --> 00:04:06.960 the Turing test. So how does that workflow actually 00:04:06.960 --> 00:04:10.080 work? I type in a prompt, and then what happens? 00:04:10.300 --> 00:04:13.360 You type in, say, write a Python script to scrape 00:04:13.360 --> 00:04:16.600 real estate data. Arena then spins up two anonymous 00:04:16.600 --> 00:04:20.060 models side by side, Model A and Model B. They 00:04:20.060 --> 00:04:21.939 both generate the code on your screen at the 00:04:21.939 --> 00:04:24.180 same time. You have no idea who they are. So 00:04:24.180 --> 00:04:25.720 you just look at the code. You look at the code, 00:04:25.819 --> 00:04:27.759 you test it, and you vote for the winner. It 00:04:27.759 --> 00:04:31.019 forces you to value the output. over the label 00:04:31.019 --> 00:04:34.000 it's a blind taste test for intelligence it is 00:04:34.000 --> 00:04:38.220 and once you pick say option a it reveals aha 00:04:38.220 --> 00:04:41.139 that was gemini 3 flash it just removes all the 00:04:41.139 --> 00:04:44.199 marketing fluff okay but beyond the battle arena 00:04:44.199 --> 00:04:46.500 has a direct chat feature that's incredibly practical 00:04:46.500 --> 00:04:48.019 you don't have to battle if you don't want to 00:04:48.019 --> 00:04:50.220 you can just manually pick the heavy hitters 00:04:50.220 --> 00:04:52.899 the guide mentioned using gemini 3 pro specifically 00:04:52.899 --> 00:04:56.139 for messy documents now help me understand the 00:04:56.139 --> 00:04:58.639 distinction here why pro what makes that different 00:04:58.639 --> 00:05:00.959 from the standard free models you see everywhere 00:05:00.959 --> 00:05:03.800 else that's a great question the pro designation 00:05:03.800 --> 00:05:07.600 it usually implies a larger parameter count And 00:05:07.600 --> 00:05:10.019 more importantly, a much larger context window. 00:05:10.300 --> 00:05:12.980 The context window being? Basically, the AI's 00:05:12.980 --> 00:05:15.740 short -term memory. A standard model might remember 00:05:15.740 --> 00:05:18.600 the last 10 pages of your conversation. Gemini 00:05:18.600 --> 00:05:21.879 3 Pro can hold a massive amount of information 00:05:21.879 --> 00:05:25.180 in its brain all at once. So if I have a really 00:05:25.180 --> 00:05:29.399 chaotic PDF, like a 50 -page legal contract that's 00:05:29.399 --> 00:05:32.180 just a wall of text, I need that bigger window. 00:05:32.339 --> 00:05:34.560 Exactly. You upload that to Gemini 3 Pro through 00:05:34.560 --> 00:05:38.230 Arena. You ask it to extract tasks, prioritize 00:05:38.230 --> 00:05:41.430 them with color codes, critical, high, medium, 00:05:41.529 --> 00:05:44.470 and it does it in seconds. If you tried that 00:05:44.470 --> 00:05:46.870 on a smaller model, it would. It would hallucinate 00:05:46.870 --> 00:05:48.589 or just forget the first page by the time it 00:05:48.589 --> 00:05:50.629 got to the last. Or, you know, if you need real 00:05:50.629 --> 00:05:53.209 -time info, you can just toggle on a search -capable 00:05:53.209 --> 00:05:55.480 model like Perplexity. I saw an example in the 00:05:55.480 --> 00:05:58.060 notes about asking for the 10 most relevant AI 00:05:58.060 --> 00:06:00.439 news stories from the past seven days. Right. 00:06:00.500 --> 00:06:02.180 And because it's Perplexity running through this 00:06:02.180 --> 00:06:04.519 interface, it's not just making updates. It's 00:06:04.519 --> 00:06:07.060 searching the live web, finding sources, and 00:06:07.060 --> 00:06:09.459 giving you inline citations. Again, totally free. 00:06:09.680 --> 00:06:11.600 But the part of Arena AI that really stopped 00:06:11.600 --> 00:06:13.860 me was the coding capability. We're not just 00:06:13.860 --> 00:06:15.439 talking about writing a little snippet of Python 00:06:15.439 --> 00:06:18.920 here. No. We're talking about building full web 00:06:18.920 --> 00:06:20.720 apps. This is where it gets really interesting. 00:06:20.920 --> 00:06:23.000 They have a split -screen builder. So you select 00:06:23.000 --> 00:06:25.300 the model. The source recommends Cloud Opus 4 00:06:25.300 --> 00:06:28.100 .5 syncing for this. And you just described the 00:06:28.100 --> 00:06:29.819 app. The example they gave was something called 00:06:29.819 --> 00:06:34.120 Task Atlas. Task Atlas. It sounds intense, but 00:06:34.120 --> 00:06:36.800 walk me through what the user is doing. Are they 00:06:36.800 --> 00:06:39.519 writing code? No. And that's the beautiful part. 00:06:39.620 --> 00:06:41.160 The user didn't write a single line of code. 00:06:41.300 --> 00:06:44.050 They went to Arena. pick the model, and just 00:06:44.050 --> 00:06:46.889 describe their dream interface. I want a zoomable 00:06:46.889 --> 00:06:49.889 canvas for my tasks. I want animated cards. I 00:06:49.889 --> 00:06:52.149 want smart clustering. Make the buttons neon 00:06:52.149 --> 00:06:54.569 green. It just builds it. In real time. Yeah. 00:06:54.689 --> 00:06:57.750 The screen splits in half. On the left, the AI 00:06:57.750 --> 00:07:01.810 is writing the HTML, CSS, JavaScript. On the 00:07:01.810 --> 00:07:03.589 right, the app is actually running. You can click 00:07:03.589 --> 00:07:06.050 the buttons, test the drag and drop. Wow. And 00:07:06.050 --> 00:07:07.470 if you don't like the color of the buttons, you 00:07:07.470 --> 00:07:09.449 just say, make them look more cyberpunk, and 00:07:09.449 --> 00:07:12.019 the code just rewrites itself instantly. This 00:07:12.019 --> 00:07:14.360 feels like it bridges that gap between a chatbot 00:07:14.360 --> 00:07:17.319 and an actual IDE and integrated development 00:07:17.319 --> 00:07:20.040 environment. But usually with these web -based 00:07:20.040 --> 00:07:22.399 builders, you're trapped. You can't take the 00:07:22.399 --> 00:07:24.980 code with you. That's the kicker. You can download 00:07:24.980 --> 00:07:28.660 the entire project as a ZIP file. The guide even 00:07:28.660 --> 00:07:31.300 suggests using a free host like Hostinger to 00:07:31.300 --> 00:07:34.240 deploy it. So you go from a text prompt to a 00:07:34.240 --> 00:07:37.379 live public URL without writing a line of code. 00:07:37.939 --> 00:07:40.879 It moves from playing with AI... to actually 00:07:40.879 --> 00:07:43.740 shipping a product. That is wild. It really makes 00:07:43.740 --> 00:07:45.720 me wonder about that psychological shift we touched 00:07:45.720 --> 00:07:48.079 on. If we're constantly in battle mode, choosing 00:07:48.079 --> 00:07:51.360 the best output blindly, does that mean our loyalty 00:07:51.360 --> 00:07:53.939 to these big tech companies is just dead? I think 00:07:53.939 --> 00:07:56.439 it just democratizes quality. It forces the models 00:07:56.439 --> 00:07:59.240 to actually be good, not just famous. If a smaller 00:07:59.240 --> 00:08:01.240 open source model gives you a better answer, 00:08:01.339 --> 00:08:03.480 it wins. The user doesn't care about the logo. 00:08:03.930 --> 00:08:05.709 They care about the result. That's a great segue 00:08:05.709 --> 00:08:07.709 to our next platform because this one seems to 00:08:07.709 --> 00:08:09.670 be all about removing barriers. It's called Veer. 00:08:09.829 --> 00:08:12.949 And the headline here is frictionless. Veer is 00:08:12.949 --> 00:08:16.810 the antidote to subscription fatigue and login 00:08:16.810 --> 00:08:19.269 fatigue. You know that feeling when you just 00:08:19.269 --> 00:08:20.889 want to make a quick image, but you have to log 00:08:20.889 --> 00:08:23.069 in, find your two factor code. Then you realize 00:08:23.069 --> 00:08:26.329 you're out of credits. Yes. Veer has none of 00:08:26.329 --> 00:08:29.709 that. No registration, no credit card, no usage 00:08:29.709 --> 00:08:32.929 caps. That sounds surprisingly risky for them. 00:08:32.990 --> 00:08:35.289 How are they preventing abuse? And honestly, 00:08:35.450 --> 00:08:38.990 if it's that open, is the quality actually any 00:08:38.990 --> 00:08:41.210 good? They're relying on the efficiency of the 00:08:41.210 --> 00:08:43.429 open source models they run. It's all designed 00:08:43.429 --> 00:08:46.080 for speed. The source highlights the text -to 00:08:46.080 --> 00:08:48.580 -image quality mode. You type a frog riding a 00:08:48.580 --> 00:08:50.899 bicycle through New York City, and boom, high 00:08:50.899 --> 00:08:53.340 quality, no watermark, ready to go. I have to 00:08:53.340 --> 00:08:55.679 admit, I get prompt drift sometimes. I'll start 00:08:55.679 --> 00:08:57.899 with one idea, and then I get bogged down in 00:08:57.899 --> 00:09:00.259 all the settings on other apps, and I lose the 00:09:00.259 --> 00:09:02.360 spark. This sounds like it captures that initial 00:09:02.360 --> 00:09:04.259 creative burst. That's the value proposition. 00:09:04.890 --> 00:09:07.070 Speed. It's about getting into that flow state. 00:09:07.389 --> 00:09:09.370 And it's not just static images. They have an 00:09:09.370 --> 00:09:11.049 image -to -video feature. You can take an old 00:09:11.049 --> 00:09:13.929 photo, upload it, and the AI animates it. It'll 00:09:13.929 --> 00:09:15.809 make the character wave or the wind blow through 00:09:15.809 --> 00:09:18.230 the trees. And there's something called the context 00:09:18.230 --> 00:09:21.269 editor. Yeah, this is for style transfer. Imagine 00:09:21.269 --> 00:09:23.570 you have a photo of your street, but you want 00:09:23.570 --> 00:09:27.250 it to look like a 1980s anime or an oil painting 00:09:27.250 --> 00:09:30.850 or some vintage film. Okay. You upload it to 00:09:30.850 --> 00:09:33.549 the context editor, click a button. And it just 00:09:33.549 --> 00:09:36.470 transforms the visual style while keeping the 00:09:36.470 --> 00:09:38.789 geometry of the street exactly the same. It really 00:09:38.789 --> 00:09:41.210 feels like Veer is designed for that creator 00:09:41.210 --> 00:09:46.519 economy pace. It's fast, iterative. Disposable, 00:09:46.559 --> 00:09:49.320 even. It raises an interesting point, yeah. With 00:09:49.320 --> 00:09:52.320 no accounts and no login, it implies a shift 00:09:52.320 --> 00:09:54.059 toward, like you said, disposable creativity. 00:09:54.539 --> 00:09:57.200 You're not building a curated portfolio on Veer. 00:09:57.440 --> 00:09:59.659 You're just treating creation as a stream. You 00:09:59.659 --> 00:10:01.740 generate 50 images, grab the one that works, 00:10:01.779 --> 00:10:03.679 and then you close the tab. It's gone. It's ephemeral. 00:10:03.799 --> 00:10:05.820 Okay, let's pivot a bit. We've talked about coding 00:10:05.820 --> 00:10:07.360 and creative stuff, but what about the office? 00:10:07.620 --> 00:10:10.200 What about all the people drowning in PowerPoints 00:10:10.200 --> 00:10:14.110 and PDFs? This brings us to ChatJLM. Right. And 00:10:14.110 --> 00:10:15.629 we have to address the elephant in the room here. 00:10:16.129 --> 00:10:18.070 ChatGLM is a Chinese platform. It's built by 00:10:18.070 --> 00:10:20.950 Zippo AI. And in the past, a lot of Western users 00:10:20.950 --> 00:10:24.830 have been skeptical of non -Western models. And 00:10:24.830 --> 00:10:26.990 there are valid concerns about data privacy. 00:10:27.129 --> 00:10:28.929 You probably don't want to upload state secrets 00:10:28.929 --> 00:10:31.070 to a free web interface no matter who owns it. 00:10:31.129 --> 00:10:33.710 Fair point. But the source calls it the office 00:10:33.710 --> 00:10:37.169 powerhouse. Yeah. Why? What is it doing that 00:10:37.169 --> 00:10:39.690 the U .S. models aren't? U .S. models are obsessed 00:10:39.690 --> 00:10:42.519 with chat. ChatGLM is obsessed with deliverables. 00:10:42.620 --> 00:10:45.159 The killer feature here is the PPT generator. 00:10:45.639 --> 00:10:47.980 You know how much time we all waste formatting 00:10:47.980 --> 00:10:50.620 slides? Too much. It's the bane of my existence. 00:10:50.960 --> 00:10:53.980 With ChatGLM, you toggle on AI slides, and you 00:10:53.980 --> 00:10:56.399 just say, create a professional quarterly review 00:10:56.399 --> 00:10:58.940 deck for a marketing agency. It asks you a few 00:10:58.940 --> 00:11:01.080 questions about tone and structure, and then 00:11:01.080 --> 00:11:03.059 it generates a fully formatted PowerPoint or 00:11:03.059 --> 00:11:07.120 PDF. Layouts, visuals, charts. It's just done. 00:11:07.259 --> 00:11:09.740 But are they good? Usually AI slides look like 00:11:09.740 --> 00:11:12.320 something a high schooler made in 1999. They're 00:11:12.320 --> 00:11:15.019 surprisingly competent. They're structured correctly. 00:11:15.320 --> 00:11:17.100 Yeah. And more importantly, they're editable. 00:11:17.179 --> 00:11:19.799 It gets you maybe 80 % of the way there in 10 00:11:19.799 --> 00:11:23.139 seconds. It turns a three -hour task into a 20 00:11:23.139 --> 00:11:25.240 -minute polished job. And it handles documents, 00:11:25.500 --> 00:11:27.919 too. Yeah. It's got an AI reading feature. You 00:11:27.919 --> 00:11:30.360 can upload huge PDFs. We're talking 80 -plus 00:11:30.360 --> 00:11:33.240 pages, like dense research papers. Yeah. It ingests 00:11:33.240 --> 00:11:35.519 them and basically turns into a conversational 00:11:35.519 --> 00:11:37.759 search engine for that specific document. So 00:11:37.759 --> 00:11:40.600 you don't scan for keywords. No. You just ask 00:11:40.600 --> 00:11:43.159 it, what are the conclusions about variable X 00:11:43.159 --> 00:11:46.820 on page 40? And it pinpoints it for you. There 00:11:46.820 --> 00:11:48.559 was one specific example in the notes that I 00:11:48.559 --> 00:11:50.259 thought was really cool. The precision editing 00:11:50.259 --> 00:11:53.440 on images. The blue hoodie example. Can you explain 00:11:53.440 --> 00:11:55.500 how that works technically? Oh, I love this. 00:11:55.840 --> 00:11:57.980 So usually if you want to change an image with 00:11:57.980 --> 00:12:01.340 AI, say change the hoodie to blue, the AI regenerates 00:12:01.340 --> 00:12:03.259 the whole thing. The guy's face changes. The 00:12:03.259 --> 00:12:06.120 background shifts. It ruins the photo. Right. 00:12:06.179 --> 00:12:07.960 It just rolls the dice again on the entire scene. 00:12:08.620 --> 00:12:12.019 ChatGLM does object segmentation. It identifies 00:12:12.019 --> 00:12:14.559 the pixels that are the hoodie. It masks them. 00:12:14.940 --> 00:12:17.879 Changes only those. pixels to blue, and leave 00:12:17.879 --> 00:12:20.539 the face and the background 100 % original. It's 00:12:20.539 --> 00:12:24.019 surgical. So looking at ChatGLM, does this replace 00:12:24.019 --> 00:12:27.360 the need for a separate office suite? Or is it 00:12:27.360 --> 00:12:29.720 something else? I think of it as an intelligent 00:12:29.720 --> 00:12:32.740 layer on top of the office suite. You might still 00:12:32.740 --> 00:12:35.799 need PowerPoint to do the final polish, but ChatGLM 00:12:35.799 --> 00:12:38.059 does all the heavy lifting. It automates that 00:12:38.059 --> 00:12:40.220 tedious formatting work so you can focus on the 00:12:40.220 --> 00:12:42.360 actual story. We're going to take a quick break, 00:12:42.440 --> 00:12:44.240 but when we come back, we're going to look at 00:12:44.240 --> 00:12:46.820 the heavy lifter. the tool for deep work that 00:12:46.820 --> 00:12:49.100 might just replace the junior research analyst. 00:12:49.480 --> 00:12:52.399 Stay with us. Welcome back. We've covered the 00:12:52.399 --> 00:12:55.840 coder, which was Arena, the creative, Veer, and 00:12:55.840 --> 00:12:58.220 the office assistant, ChatGLM. Now we need to 00:12:58.220 --> 00:13:00.179 talk about the heavy lister. The source calls 00:13:00.179 --> 00:13:02.539 this one industrial strength. We're talking about 00:13:02.539 --> 00:13:05.500 Quinn from Alibaba. Quinn is fascinating because 00:13:05.500 --> 00:13:07.659 it's really aiming for deep work. This is not 00:13:07.659 --> 00:13:09.759 just for chatting. The standout feature here 00:13:09.759 --> 00:13:12.240 is called projects. How is that different from 00:13:12.240 --> 00:13:14.879 a normal chat history? Most AIs have a history 00:13:14.879 --> 00:13:17.860 tab. It's about context and isolation. A chat 00:13:17.860 --> 00:13:21.820 history is just a linear list. A project in Quen 00:13:21.820 --> 00:13:24.559 is a dedicated workspace with its own memory 00:13:24.559 --> 00:13:26.940 instructions. So let's say you're a freelancer. 00:13:27.080 --> 00:13:30.019 You have client A who likes a formal tone and 00:13:30.019 --> 00:13:32.419 British spelling. You have client B who likes 00:13:32.419 --> 00:13:34.580 slang and emojis. That's a nightmare to keep 00:13:34.580 --> 00:13:36.879 straight in your head. Exactly. So you create 00:13:36.879 --> 00:13:39.299 a project for each of them. You upload the brand 00:13:39.299 --> 00:13:42.360 guidelines and the tone of voice right into the 00:13:42.360 --> 00:13:45.059 project settings. Now, every time you step into 00:13:45.059 --> 00:13:47.820 the Client A project, Quinn becomes that employee. 00:13:48.200 --> 00:13:49.940 You don't have to remind it of anything. It's 00:13:49.940 --> 00:13:53.360 all isolated context. That's crucial for freelancers 00:13:53.360 --> 00:13:55.879 or anyone juggling multiple roles. But the feature 00:13:55.879 --> 00:13:58.139 that really caught my eye was deep research. 00:13:58.539 --> 00:14:01.340 We hear that term a lot, but what does Ken actually 00:14:01.340 --> 00:14:04.000 do that's different? There's a step change. You're 00:14:04.000 --> 00:14:05.840 not just asking a question and getting a paragraph 00:14:05.840 --> 00:14:07.779 back. You give it a prompt like, what are the 00:14:07.779 --> 00:14:10.490 five best foods for longevity? And you set it 00:14:10.490 --> 00:14:12.649 to advanced mode. It doesn't just answer you. 00:14:12.690 --> 00:14:15.690 It goes off and conducts a 15 to 20 minute agentic 00:14:15.690 --> 00:14:18.490 search. Wait, it takes 20 minutes. In the age 00:14:18.490 --> 00:14:20.570 of instant gratification, that feels like a bug. 00:14:20.730 --> 00:14:23.429 It's a feature. Think about it. If you asked 00:14:23.429 --> 00:14:26.070 a human researcher that question, would you want 00:14:26.070 --> 00:14:28.870 the answer in three seconds? No. No. You'd want 00:14:28.870 --> 00:14:31.289 them to dig. So in those 20 minutes, Quinn is 00:14:31.289 --> 00:14:33.990 performing agentic search. It's breaking your 00:14:33.990 --> 00:14:35.629 question down into sub -questions. It's going 00:14:35.629 --> 00:14:38.470 out to Google, Scholar, PubMed, News Archives. 00:14:38.830 --> 00:14:41.289 It's reading dozens of papers, cross -referencing 00:14:41.289 --> 00:14:43.549 them, throwing out the bad ones and synthesizing 00:14:43.549 --> 00:14:46.230 the good ones. So it's doing the work of a junior 00:14:46.230 --> 00:14:49.009 research analyst. Yes. And what it comes back 00:14:49.009 --> 00:14:51.730 with is not a chat bubble. It's a 12 -page report 00:14:51.730 --> 00:14:55.470 with citations, executive summaries, links to 00:14:55.470 --> 00:14:57.730 the sources. That is incredible. But it's also 00:14:57.730 --> 00:15:00.289 a little terrifying. If I can get a 12 -page 00:15:00.289 --> 00:15:02.889 report on longevity in 20 minutes for free, am 00:15:02.889 --> 00:15:05.830 I even a researcher anymore? I think your role 00:15:05.830 --> 00:15:08.570 shifts. You stop being the hunter, the person 00:15:08.570 --> 00:15:10.809 digging through pages of Google results, and 00:15:10.809 --> 00:15:13.149 you become the editor. You become the verifier. 00:15:13.350 --> 00:15:15.870 Your value isn't in finding the wrong information. 00:15:16.169 --> 00:15:18.850 It's in synthesizing the AI's report and figuring 00:15:18.850 --> 00:15:20.990 out if it's actually true and relevant. You move 00:15:20.990 --> 00:15:23.789 up the value chain. From gathering to synthesizing. 00:15:23.809 --> 00:15:25.909 Precisely. And on the complete opposite end of 00:15:25.909 --> 00:15:28.470 the spectrum, Quinn is also doing high -end video. 00:15:28.690 --> 00:15:30.809 This was the moment that made me go, whoa. I 00:15:30.809 --> 00:15:33.230 mean, we've all seen AI video. But Quinn is doing 00:15:33.230 --> 00:15:35.470 video with native audio. Audio, too. The example 00:15:35.470 --> 00:15:37.950 in the source is so vivid, you prompt for a samurai 00:15:37.950 --> 00:15:41.720 action sequence. Shaky camera, motion blur. Swords 00:15:41.720 --> 00:15:45.000 clashing. Yeah. And the AI generates the video, 00:15:45.080 --> 00:15:47.539 but also the sound of the metal striking, the 00:15:47.539 --> 00:15:51.179 sound of arcane lightning crackling. It's multisensory. 00:15:51.240 --> 00:15:53.700 And this is all free. Currently, yes. Unlimited 00:15:53.700 --> 00:15:56.139 generation, no watermarks. So let's zoom out. 00:15:56.570 --> 00:15:59.850 We've looked at Arena AI, Veer, ChatGLM, and 00:15:59.850 --> 00:16:02.330 Quinn. It's a lot of firepower. It's massive. 00:16:02.389 --> 00:16:04.850 And if we just recap, you've got Arena as the 00:16:04.850 --> 00:16:07.070 decision engine using battle mode to write code 00:16:07.070 --> 00:16:09.710 and pick the best models. You've got Veer for 00:16:09.710 --> 00:16:12.950 that instant frictionless creativity. ChatGLM 00:16:12.950 --> 00:16:15.409 is your corporate weapon for all those PDFs and 00:16:15.409 --> 00:16:18.809 slides. And Quinn is for the really deep dives. 00:16:19.240 --> 00:16:21.779 cinematic production. The big idea here seems 00:16:21.779 --> 00:16:24.240 to be that in 2026, having a budget isn't the 00:16:24.240 --> 00:16:27.259 competitive advantage anymore. No. The advantage 00:16:27.259 --> 00:16:30.519 is knowledge. The advantage is strategy. It's 00:16:30.519 --> 00:16:33.720 about redundancy. The person who relies on one 00:16:33.720 --> 00:16:37.279 $20 subscription is fragile. If that model goes 00:16:37.279 --> 00:16:39.799 down or gets dumber, they're stuck. The person 00:16:39.799 --> 00:16:42.789 who knows this underground network. They're anti 00:16:42.789 --> 00:16:45.289 -fragile. They can pivot. They can match the 00:16:45.289 --> 00:16:47.789 platform to the task. That's a powerful place 00:16:47.789 --> 00:16:50.210 to land. So here's our challenge to you listening. 00:16:50.429 --> 00:16:53.490 Pick one task you usually pay for. Maybe it's 00:16:53.490 --> 00:16:55.490 a stock image. Maybe it's summarizing a document. 00:16:55.730 --> 00:16:58.090 And just try one of these free tools instead. 00:16:58.370 --> 00:17:01.210 Just one. Absolutely. Go build an app on Arena. 00:17:01.409 --> 00:17:03.690 Yeah. Generate a frog on a bike on Veer. Just 00:17:03.690 --> 00:17:05.769 test the waters. Because the tools became free. 00:17:05.910 --> 00:17:08.430 The strategy is now your job. Couldn't have said 00:17:08.430 --> 00:17:10.720 it better myself. Thanks for diving in with us. 00:17:10.759 --> 00:17:11.700 We'll catch you in the next one.