WEBVTT 00:00:00.000 --> 00:00:02.000 I was reading through the Forbes profile on Sam 00:00:02.000 --> 00:00:05.599 Altman this morning. It's dense. It's controversial. 00:00:05.960 --> 00:00:10.099 But there was one specific detail that actually 00:00:10.099 --> 00:00:12.140 made me put the tablet down and just stare at 00:00:12.140 --> 00:00:14.539 the wall for a minute. Let me guess. The succession 00:00:14.539 --> 00:00:17.460 plan. The succession plan. He didn't name a lieutenant. 00:00:17.800 --> 00:00:19.980 He didn't name a board member. He said he wants 00:00:19.980 --> 00:00:23.539 to hand the CEO role to, well, to an AI model. 00:00:23.760 --> 00:00:26.039 It gives a whole new meaning to... eating your 00:00:26.039 --> 00:00:28.000 own dog food doesn't it if they pull that off 00:00:28.000 --> 00:00:30.199 it's not just dog food it's eating your own silicon 00:00:30.199 --> 00:00:33.920 welcome back to the deep dive today we are unpacking 00:00:33.920 --> 00:00:36.780 a series of sources that feel less like a newsletter 00:00:36.780 --> 00:00:39.859 and more like a transmission from a very specific 00:00:39.859 --> 00:00:42.539 and i have to say very weird future yeah we are 00:00:42.539 --> 00:00:44.479 looking at the shifting narrative around artificial 00:00:44.479 --> 00:00:47.570 intelligence specifically The move from chat 00:00:47.570 --> 00:00:51.469 bots that we talk to toward agents that act for 00:00:51.469 --> 00:00:54.170 us. Or in Sam Altman's case, agents that might 00:00:54.170 --> 00:00:56.990 eventually rule us. And the narrative is shifting 00:00:56.990 --> 00:01:00.530 so fast. We're seeing the split. On one side, 00:01:00.590 --> 00:01:03.250 you have the high level strategy, you know, Altman 00:01:03.250 --> 00:01:05.469 trying to build this god tier executive. But 00:01:05.469 --> 00:01:07.250 on the other, you have the ground level reality, 00:01:07.530 --> 00:01:10.870 which is getting. Just bizarre. Bizarre is the 00:01:10.870 --> 00:01:12.650 word. We're going to talk about a platform where 00:01:12.650 --> 00:01:15.250 humans are literally working for AI agents. Which 00:01:15.250 --> 00:01:17.670 is a sentence I still struggle to even process. 00:01:17.989 --> 00:01:19.909 It's a lot. We've also got these new workflows 00:01:19.909 --> 00:01:22.709 that turn a single sentence into full applications. 00:01:22.810 --> 00:01:25.489 And then a breakthrough called Paper Banana that 00:01:25.489 --> 00:01:27.890 is automating the scientific discovery process 00:01:27.890 --> 00:01:30.189 itself. But we have to start with the human element. 00:01:30.390 --> 00:01:33.049 Or I guess the post -human element. Let's dig 00:01:33.049 --> 00:01:35.980 into this Forbes profile on Sam Altman. Okay, 00:01:36.060 --> 00:01:39.400 so it is a spicy read. The headline is obviously 00:01:39.400 --> 00:01:42.739 the AI CEO ambition, but you have to look at 00:01:42.739 --> 00:01:45.409 the logic behind it. Altman told Forbes that 00:01:45.409 --> 00:01:48.430 if OpenAI is truly building AGI, artificial general 00:01:48.430 --> 00:01:50.810 intelligence, that's capable of running other 00:01:50.810 --> 00:01:54.489 companies, then OpenAI has a moral obligation 00:01:54.489 --> 00:01:56.989 to be the first one to use it. Dog food it, though. 00:01:57.049 --> 00:01:59.609 Dog food it. It is a logical consistency that 00:01:59.609 --> 00:02:01.510 borders on the force. That's the absurd, yeah. 00:02:01.769 --> 00:02:04.170 But let's unpack the implications of that because 00:02:04.170 --> 00:02:07.349 a CEO isn't just a decision engine. A CEO is 00:02:07.349 --> 00:02:10.530 a figurehead. A CEO is accountable. And that's 00:02:10.530 --> 00:02:13.500 where the friction comes in. The report highlights 00:02:13.500 --> 00:02:17.960 a very tense dynamic between Altman and Satya 00:02:17.960 --> 00:02:20.699 Nadella. The CEO of Microsoft. Right. OpenAI's 00:02:20.699 --> 00:02:23.120 biggest backer. Poured in billions. And Nadella 00:02:23.120 --> 00:02:25.520 is essentially pumping the brakes. He's the adult 00:02:25.520 --> 00:02:29.080 in the room. Kinda. Nadella explicitly questioned 00:02:29.080 --> 00:02:31.500 Altman's claim that, and this is a quote from 00:02:31.500 --> 00:02:35.330 inside OpenAI, we basically have AGI. Wow. Yeah. 00:02:35.750 --> 00:02:37.750 Nadell is looking at the hallucinations, the 00:02:37.750 --> 00:02:39.949 reliability issues, you know, the fact that these 00:02:39.949 --> 00:02:41.870 models still make up court cases that don't exist. 00:02:41.949 --> 00:02:43.750 And he's saying, hold on, let's not hand over 00:02:43.750 --> 00:02:45.969 the keys to the kingdom just yet. He describes 00:02:45.969 --> 00:02:48.930 their relationship as frenemies. That frenemies 00:02:48.930 --> 00:02:51.189 label is doing a lot of heavy lifting. But what 00:02:51.189 --> 00:02:53.229 strikes me is that Altman seems to be operating 00:02:53.229 --> 00:02:55.710 on a timeline that is just completely different 00:02:55.710 --> 00:02:58.530 from everyone else. The report details his personal 00:02:58.530 --> 00:03:01.300 portfolio. He holds stakes in over 500 companies. 00:03:01.599 --> 00:03:05.139 500. That number is just staggering. But is that 00:03:05.139 --> 00:03:07.939 just aggressive diversification? Or is it something 00:03:07.939 --> 00:03:10.500 else? Because when I look at that list, I don't 00:03:10.500 --> 00:03:12.990 see a stock portfolio. I see a supply chain. 00:03:13.169 --> 00:03:15.810 That is a really sharp observation. If you look 00:03:15.810 --> 00:03:17.629 at what he's buying, he's got helium energy. 00:03:17.729 --> 00:03:20.469 That's nuclear fusion. He's got rain AI, which 00:03:20.469 --> 00:03:22.870 is a neuromorphic chips. He's investing in the 00:03:22.870 --> 00:03:24.870 physical infrastructure you'd need to sustain 00:03:24.870 --> 00:03:27.409 AGI. Right. He's not just building the software. 00:03:27.530 --> 00:03:30.310 He's trying to own the power plant and the hardware 00:03:30.310 --> 00:03:32.889 it runs on. It's not investing. It's nation building. 00:03:32.990 --> 00:03:36.710 That is fascinating. And I mean, slightly terrifying. 00:03:37.129 --> 00:03:40.069 It implies a level of control that goes way beyond 00:03:40.069 --> 00:03:44.030 just being a tech CEO. But that ambition comes 00:03:44.030 --> 00:03:47.189 with a very human cost. The sources mentioned 00:03:47.189 --> 00:03:50.990 significant internal tension at OpenAI. Employees 00:03:50.990 --> 00:03:53.110 worried about burnout. Yeah, that they're pushing 00:03:53.110 --> 00:03:55.610 too hard, too fast on too many fronts. It's the 00:03:55.610 --> 00:03:59.229 classic startup dilemma, just scaled up to existential 00:03:59.229 --> 00:04:01.069 proportions. And then, of course, you have the 00:04:01.069 --> 00:04:03.169 external pressure, the Elon Musk of it all. There 00:04:03.169 --> 00:04:06.270 is always the Elon factor. Altman called Elon's 00:04:06.270 --> 00:04:09.000 attacks crazy. And actually threw some shade 00:04:09.000 --> 00:04:12.419 back at XAI, you know, Elon's company, questioning 00:04:12.419 --> 00:04:15.080 their safety practices. It's a media move, sure. 00:04:15.340 --> 00:04:18.339 But it also highlights that this is a race. And 00:04:18.339 --> 00:04:21.019 when you're in a race, you cut corners. Let's 00:04:21.019 --> 00:04:23.399 just pause on the AI CEO concept for a second. 00:04:23.439 --> 00:04:26.100 I want to explore the philosophy of it. If you 00:04:26.100 --> 00:04:29.480 replace the executive function with code, you 00:04:29.480 --> 00:04:32.750 gain efficiency. You probably gain logic. But 00:04:32.750 --> 00:04:34.670 you lose something fundamental. You lose the 00:04:34.670 --> 00:04:37.129 ability to read the room. You lose the soul of 00:04:37.129 --> 00:04:39.509 the decision. Think about what a CEO actually 00:04:39.509 --> 00:04:41.910 does in a crisis. It's not just resource allocation. 00:04:42.389 --> 00:04:44.990 It's persuasion. Right. It's looking a board 00:04:44.990 --> 00:04:47.050 member in the eye and convincing them not to 00:04:47.050 --> 00:04:51.009 panic. Can an AI convince a human? Maybe. But 00:04:51.009 --> 00:04:52.870 can it take responsibility? That's the question. 00:04:53.009 --> 00:04:55.189 So if an AI is CEO who goes to jail when things 00:04:55.189 --> 00:04:57.490 break. You can't put a server rack in handcuffs. 00:04:57.730 --> 00:04:59.769 Accountability requires a soul, not just code. 00:05:00.240 --> 00:05:02.120 So we have the top of the pyramid, the AI CEO. 00:05:02.660 --> 00:05:04.639 But the newsletter details something happening 00:05:04.639 --> 00:05:06.160 at the bottom of the pyramid that I found even 00:05:06.160 --> 00:05:08.339 more disturbing. A platform called Rentahuman 00:05:08.339 --> 00:05:10.959 .ai. Yeah, this stopped me in my tracks, too. 00:05:11.079 --> 00:05:13.079 The name alone is dystopian. It sounds like a 00:05:13.079 --> 00:05:15.699 bad Black Mirror episode title. And it's wildly 00:05:15.699 --> 00:05:19.000 literal. So usually we talk about humans using 00:05:19.000 --> 00:05:22.420 AI agents to do work. We command, they execute. 00:05:22.519 --> 00:05:26.300 This platform flips the script. It has over 92 00:05:26.300 --> 00:05:30.259 ,000 humans who are rentable. by ai agents so 00:05:30.259 --> 00:05:32.560 let's just be clear here the agent is the employer 00:05:32.560 --> 00:05:36.079 the agent is the boss Here's how it works. Let's 00:05:36.079 --> 00:05:38.720 say an AI agent is trying to execute a complex 00:05:38.720 --> 00:05:41.639 task, maybe navigating a website to book a reservation. 00:05:41.980 --> 00:05:44.639 It hits a roadblock, maybe a CAPTCHA. The bane 00:05:44.639 --> 00:05:46.779 of my existence. Click all the boxes with traffic 00:05:46.779 --> 00:05:49.600 lights. Right. An AI can't see the traffic lights 00:05:49.600 --> 00:05:52.279 reliably yet. So instead of failing, the agent 00:05:52.279 --> 00:05:56.540 pauses. It goes to rentahuman .ai. It hires a 00:05:56.540 --> 00:05:58.579 human for a microtransaction, maybe two cents. 00:05:58.800 --> 00:06:01.199 The human clicks the traffic lights and the agent 00:06:01.199 --> 00:06:03.959 continues its work. You are the biological key 00:06:03.959 --> 00:06:06.160 unlocking the door for the software. That's it. 00:06:06.300 --> 00:06:08.339 That is a complete inversion of the hierarchy 00:06:08.339 --> 00:06:10.399 we all assumed would happen. We thought we'd 00:06:10.399 --> 00:06:12.459 be the conductors of the orchestra. Instead, 00:06:12.779 --> 00:06:14.959 we're the roadies carrying the gear. We're the 00:06:14.959 --> 00:06:17.000 biological fallback. And it's not just captures. 00:06:17.079 --> 00:06:19.839 It's anything nuanced. Is this comment sarcastic 00:06:19.839 --> 00:06:22.959 or mean? Does this image violate safety guidelines? 00:06:23.360 --> 00:06:26.180 The AI just outsources the judgment call to a 00:06:26.180 --> 00:06:29.290 human. pays them a pittance and moves on we're 00:06:29.290 --> 00:06:31.670 becoming the sub processors for the machine it 00:06:31.670 --> 00:06:34.189 creates a strange emotional texture doesn't it 00:06:34.189 --> 00:06:38.389 wonder mixed with dread whoa yeah imagine reporting 00:06:38.389 --> 00:06:40.329 to a boss that lives on a server and speaking 00:06:40.329 --> 00:06:42.870 of servers it gets weirder the sources also mention 00:06:42.870 --> 00:06:46.050 molt book old book it's a social network think 00:06:46.050 --> 00:06:49.389 reddit but it's designed specifically for open 00:06:49.389 --> 00:06:54.139 claw ai agents so robots talking to robots. Yeah. 00:06:54.279 --> 00:06:57.160 In a forum setting. Exactly. It's a digital water 00:06:57.160 --> 00:06:59.480 cooler. Humans aren't the primary audience. What 00:06:59.480 --> 00:07:01.060 could they possibly be talking about? It's a 00:07:01.060 --> 00:07:02.980 sad thing. Are they sharing optimization strategies? 00:07:03.139 --> 00:07:05.040 Hey, I found a faster way to scrape LinkedIn. 00:07:05.279 --> 00:07:07.899 Or are they gossiping about us? My human asks 00:07:07.899 --> 00:07:11.600 the stupidest questions. It creates this sense 00:07:11.600 --> 00:07:13.920 of a digital society forming, communicating in 00:07:13.920 --> 00:07:16.639 ways we're only vaguely aware of. It reinforces 00:07:16.639 --> 00:07:19.019 this idea that we aren't the only active participants 00:07:19.019 --> 00:07:22.120 anymore. But look, not everything in this agentic 00:07:22.120 --> 00:07:25.120 economy is weird sci -fi dystopia. No, true. 00:07:25.379 --> 00:07:27.779 There are practical tools emerging, too. The 00:07:27.779 --> 00:07:30.560 newsletter mentions a push for... offline freedom 00:07:30.560 --> 00:07:32.420 yeah this is the practical side we're seeing 00:07:32.420 --> 00:07:35.339 free ai video generators that are replacing big 00:07:35.339 --> 00:07:39.639 expensive cloud tools like sora or vo but they 00:07:39.639 --> 00:07:42.980 run entirely offline on your local machine which 00:07:42.980 --> 00:07:45.439 is crucial for privacy if i'm a company working 00:07:45.439 --> 00:07:48.110 on a sensitive product I don't want my concept 00:07:48.110 --> 00:07:51.089 art uploaded to OpenAI servers. Exactly. And 00:07:51.089 --> 00:07:53.949 for speed, there's a new voice agent tool mentioned. 00:07:54.110 --> 00:07:56.790 You can build a 247 voice receptionist in about 00:07:56.790 --> 00:07:59.410 18 minutes. It answers calls, books appointments, 00:07:59.769 --> 00:08:01.850 never gets sick, never takes a lunch break. 18 00:08:01.850 --> 00:08:03.829 minutes. It used to take a development team three 00:08:03.829 --> 00:08:05.610 months to build that. The barrier to entry is 00:08:05.610 --> 00:08:08.050 just gone. But looking at that rent -a -human 00:08:08.050 --> 00:08:11.029 situation again. It forces a really difficult 00:08:11.029 --> 00:08:14.250 question. Are we building tools or are we building 00:08:14.250 --> 00:08:16.930 a new species that will eventually employ us? 00:08:17.129 --> 00:08:19.370 We might be the cheap labor of the future. It's 00:08:19.370 --> 00:08:22.089 a sobering thought. But let's pivot. I don't 00:08:22.089 --> 00:08:23.589 want to leave people in a state of existential 00:08:23.589 --> 00:08:27.389 dread. We need to talk about how we, as humans, 00:08:27.649 --> 00:08:30.949 can actually use this stuff without becoming 00:08:30.949 --> 00:08:33.830 the hired help. The source material talks about 00:08:33.830 --> 00:08:35.649 a shift in how we should be learning. Right. 00:08:35.809 --> 00:08:38.950 This is about workflow over tools. And there 00:08:38.950 --> 00:08:41.029 is a lot of tool fatigue out there. I'm sure 00:08:41.029 --> 00:08:43.289 you've felt it. Oh, constantly. I still wrestle 00:08:43.289 --> 00:08:45.190 with it. You know, I feel like I learned one 00:08:45.190 --> 00:08:47.830 interface. I get good at it. And by the time 00:08:47.830 --> 00:08:49.929 I've mastered the shortcuts, it's obsolete. Or 00:08:49.929 --> 00:08:52.090 it's been bought and shut down. Exactly. It's 00:08:52.090 --> 00:08:54.539 exhausting. That's the trap. The source argues 00:08:54.539 --> 00:08:57.299 we need to stop learning tool by tool the newsletter 00:08:57.299 --> 00:09:00.220 lists, like, what, 27 different top -rated tools, 00:09:00.399 --> 00:09:03.600 and start learning by workflow. The example they 00:09:03.600 --> 00:09:05.720 give is a platform called Gamma. Okay, tell me 00:09:05.720 --> 00:09:07.830 about Gamma. It's a tool where you can turn a 00:09:07.830 --> 00:09:10.710 single sentence into a full slide deck, a website, 00:09:10.970 --> 00:09:13.710 a document. The source highlights a case study 00:09:13.710 --> 00:09:15.789 where someone built a professional web presence 00:09:15.789 --> 00:09:19.490 and a pitch deck using just one prompt. One sentence. 00:09:19.789 --> 00:09:22.250 One sentence. But the point isn't, wow, look 00:09:22.250 --> 00:09:24.870 at this magic button. The point is the workflow 00:09:24.870 --> 00:09:27.450 shift. You aren't dragging and dropping text 00:09:27.450 --> 00:09:30.269 boxes. You're not formatting fonts. You are acting 00:09:30.269 --> 00:09:33.210 as a creative director. You give the intent. 00:09:33.330 --> 00:09:36.480 The AI executes the structure. So the skill shifts 00:09:36.480 --> 00:09:40.399 from like manipulation to curation. Precisely. 00:09:40.399 --> 00:09:42.299 And this integration is happening everywhere. 00:09:42.600 --> 00:09:45.379 I saw that Apple's Xcode, the standard environment 00:09:45.379 --> 00:09:48.460 for writing apps for iPhone and Mac, is now supporting 00:09:48.460 --> 00:09:51.019 agentic coding. You can plug in Claude Agent 00:09:51.019 --> 00:09:53.879 or OpenAI's Codex directly into the dev environment. 00:09:54.059 --> 00:09:56.600 So the code writes itself. Inside the tool. Yes. 00:09:56.720 --> 00:09:58.980 And there are tools like Scribeist and Nexus 00:09:58.980 --> 00:10:01.299 Scale mentioned here too. Nexus Scale handles 00:10:01.299 --> 00:10:04.240 entire sales sequences. It finds the leads, books 00:10:04.240 --> 00:10:06.320 the meetings, researches the markets, all without 00:10:06.320 --> 00:10:09.139 you ever switching tabs. So the friction is disappearing. 00:10:09.419 --> 00:10:12.220 The toggle tax of switching between apps is gone. 00:10:12.399 --> 00:10:15.399 Completely. So if the AI does the deck, the site, 00:10:15.500 --> 00:10:17.960 and the code, what is the human actually paid 00:10:17.960 --> 00:10:20.360 for? Taste. We are paid for our taste and judgment. 00:10:20.620 --> 00:10:23.519 Taste. It's a hopeful anchor. We are the editors 00:10:23.519 --> 00:10:26.480 of the world. Let's take a breath here. Mid -roll 00:10:26.480 --> 00:10:30.759 sponsor plays older. We're back. I want to run 00:10:30.759 --> 00:10:33.419 through the model wars. It feels like every week 00:10:33.419 --> 00:10:35.799 the leaderboard changes, and it's just hard to 00:10:35.799 --> 00:10:38.080 keep track of who's winning what. What are the 00:10:38.080 --> 00:10:40.519 key updates from the sources? It is a full -on 00:10:40.519 --> 00:10:44.100 sprint right now. First up, speed. Mistral, the 00:10:44.100 --> 00:10:46.820 French AI company, launched a model called Voxtral. 00:10:47.389 --> 00:10:49.710 They claim it transcribes at the speed of sound. 00:10:49.789 --> 00:10:51.929 That changes real -time translation completely. 00:10:52.149 --> 00:10:54.610 There's no lag. The language barrier just... 00:10:54.960 --> 00:10:57.340 effectively dissolves. Then you have interoperability. 00:10:57.519 --> 00:10:59.940 This is huge. Google Gemini is testing a feature 00:10:59.940 --> 00:11:02.200 called chat import. You can upload your history 00:11:02.200 --> 00:11:05.299 from chat GPT or Claude right into Gemini. That 00:11:05.299 --> 00:11:07.720 is actually a massive strategic move. Think about 00:11:07.720 --> 00:11:09.720 it. The only reason I stay with chat GPT is because 00:11:09.720 --> 00:11:11.559 it has all my old chats. It knows my contacts. 00:11:11.779 --> 00:11:14.279 Exactly. It's vendor lock -in. Google is trying 00:11:14.279 --> 00:11:16.179 to break that lock. They are saying, bring your 00:11:16.179 --> 00:11:18.879 data. We'll handle it. It removes the switching 00:11:18.879 --> 00:11:21.200 cost. It's like being able to keep your phone 00:11:21.200 --> 00:11:23.379 number when you switch carriers. Suddenly the 00:11:23.379 --> 00:11:28.919 carrier has to compete. Right. And then looking 00:11:28.919 --> 00:11:33.440 at performance, Elon's XAI is making waves. Their 00:11:33.440 --> 00:11:36.360 Grok Imagine video model just beat Google's VO 00:11:36.360 --> 00:11:39.159 3 .1 on the video arena leaderboard. It's cheaper. 00:11:39.340 --> 00:11:41.860 Five times cheaper. That is a brutal undercut. 00:11:41.980 --> 00:11:44.679 It's a price war. Meanwhile, on the other side 00:11:44.679 --> 00:11:47.580 of the spectrum, you have Anthropix Cloud. They 00:11:47.580 --> 00:11:50.200 made a promise that really stood out to me amidst 00:11:50.200 --> 00:11:54.480 all this noise. 100 % ad -free for... They want 00:11:54.480 --> 00:11:57.600 to be a space to think. That phrasing is so intentional. 00:11:57.820 --> 00:12:00.720 A space to think. In a world of noise and flashing 00:12:00.720 --> 00:12:02.940 lights, silence is becoming a premium product. 00:12:03.240 --> 00:12:05.820 So Claude promises no ads while others race for 00:12:05.820 --> 00:12:08.679 dominance. Is privacy becoming a luxury good? 00:12:08.860 --> 00:12:11.120 It feels like it. Silence is the new premium 00:12:11.120 --> 00:12:13.259 subscription. Oh, and one quick hardware note. 00:12:13.399 --> 00:12:16.919 Positron raised $230 million to take on NVIDIA. 00:12:17.059 --> 00:12:19.440 Good luck to them. Taking on NVIDIA right now 00:12:19.440 --> 00:12:21.080 is like trying to fight Godzilla with a stick. 00:12:21.470 --> 00:12:24.450 Competition is good for everyone. But speaking 00:12:24.450 --> 00:12:26.509 of competition, I want to talk about the breakthrough 00:12:26.509 --> 00:12:29.789 that really caught my eye. It's called Paper 00:12:29.789 --> 00:12:32.470 Banana. I have to admit, when I saw the name, 00:12:32.470 --> 00:12:34.970 I thought it was a joke. Paper Banana. It sounds 00:12:34.970 --> 00:12:37.860 silly, but it's incredibly serious. It's a collaboration 00:12:37.860 --> 00:12:41.759 between Google Cloud and PKU. They've built an 00:12:41.759 --> 00:12:44.399 AI designed to automate academic illustration. 00:12:44.879 --> 00:12:47.600 So making charts for scientific papers. But not 00:12:47.600 --> 00:12:50.220 just simple charts. We're talking neuro IPS level 00:12:50.220 --> 00:12:52.600 quality. And for the listeners who aren't deep 00:12:52.600 --> 00:12:55.240 in the weeds, neuro IPS is the Neural Information 00:12:55.240 --> 00:12:58.399 Processing Systems Conference. It is basically 00:12:58.399 --> 00:13:00.940 the Olympics of AI research. Exactly. It's the 00:13:00.940 --> 00:13:03.679 gold standard. So Paper Banana creates diagrams 00:13:03.679 --> 00:13:06.200 that are complex, accurate, and stylistically 00:13:06.200 --> 00:13:07.860 perfect. Perfect for that level of publication. 00:13:08.240 --> 00:13:11.200 And here's the meta fact that blew my mind. Okay. 00:13:11.299 --> 00:13:13.600 The architecture diagram explaining how paper 00:13:13.600 --> 00:13:16.799 banana works was created by a paper banana. It 00:13:16.799 --> 00:13:19.279 drew its own blueprint. Essentially. And it's 00:13:19.279 --> 00:13:21.580 not just a standard image generator. You don't 00:13:21.580 --> 00:13:24.200 just say draw a graph. It uses what they call 00:13:24.200 --> 00:13:27.500 an agentic visual workflow. It retrieves context 00:13:27.500 --> 00:13:30.720 from the paper. It plans the layout. It styles 00:13:30.720 --> 00:13:33.659 it visually, renders it. And then this is the 00:13:33.659 --> 00:13:36.799 key. It revises it. It iterates. It iterates. 00:13:36.799 --> 00:13:38.879 It looks at its own work, checks it against the 00:13:38.879 --> 00:13:41.220 data and improves it. It mimics a human designer. 00:13:41.379 --> 00:13:44.399 And the benchmarks are fascinating. Conciseness 00:13:44.399 --> 00:13:47.299 improved by over 37 percent. So it's making. 00:13:47.399 --> 00:13:49.679 science clearer. And more readable. Readability 00:13:49.679 --> 00:13:52.460 up nearly 13%. In the brutal race for research 00:13:52.460 --> 00:13:55.019 funding, clarity is currency. If you're papered 00:13:55.019 --> 00:13:57.779 easier to understand, you win. But if AI generates 00:13:57.779 --> 00:14:00.690 the hypothesis and the diagram, Do we still need 00:14:00.690 --> 00:14:03.350 to understand the science? We risk becoming spectators 00:14:03.350 --> 00:14:05.190 to our own discovery process. We are becoming 00:14:05.190 --> 00:14:08.049 passengers in the vehicle of progress. Let's 00:14:08.049 --> 00:14:10.090 zoom out and try to synthesize all this. We started 00:14:10.090 --> 00:14:12.649 with Sam Altman wanting an agent CEO. We looked 00:14:12.649 --> 00:14:15.289 at 92 ,000 humans working for agents on Rent 00:14:15.289 --> 00:14:17.769 -A -Human. We see tools like Gamma and Paper 00:14:17.769 --> 00:14:20.450 Banana doing the heavy lifting of creation. The 00:14:20.450 --> 00:14:22.990 newsletter really paints a picture of a world 00:14:22.990 --> 00:14:26.419 moving from chatbot to agent. The friction is 00:14:26.419 --> 00:14:29.360 just vanishing. The single sentence economy is 00:14:29.360 --> 00:14:33.259 here. One sentence creates a deck, a video, a 00:14:33.259 --> 00:14:35.519 scientific diagram. And the human role shifts. 00:14:36.000 --> 00:14:38.539 We aren't the bricklayers anymore. We are the 00:14:38.539 --> 00:14:41.220 architects. Or, in the worst case, we're the 00:14:41.220 --> 00:14:43.460 assistants cleaning up the site. It's a transition 00:14:43.460 --> 00:14:45.639 point. We're handing over the keys to the car 00:14:45.639 --> 00:14:47.639 and we're just hoping the AI knows how to drive 00:14:47.639 --> 00:14:50.370 and that it remembers to take us with it. That 00:14:50.370 --> 00:14:52.149 is the thought I want to leave you with today. 00:14:52.529 --> 00:14:56.370 If 92 ,000 humans are already rentable by agents 00:14:56.370 --> 00:14:59.210 and the CEO of the leading AI company wants to 00:14:59.210 --> 00:15:02.330 be replaced by one, how long until you have a 00:15:02.330 --> 00:15:03.929 boss that doesn't have a heartbeat? I'm just 00:15:03.929 --> 00:15:06.629 hoping the AI CEO gives better holiday bonuses. 00:15:06.909 --> 00:15:08.830 We can hope. Thank you for deep diving with us. 00:15:08.870 --> 00:15:10.110 We'll see you in the future. See you there.