I want you to picture picture something for a second. For pretty much the entire history of our species, intelligence, you know, the ability to think, plan and solve really complex problems has been our scarcest resource. Absolutely. It's expensive to hire. It takes decades to train. And it is effectively the gold standard of our economy. But what happens to the architecture of the world when human intelligence suddenly becomes cheap? I don't mean just affordable.
I mean abundant, commodity cheap. That is the question, isn't it? And if you look at the data we have today, the answer is, well, it's uncomfortable. It doesn't just change what jobs we do. It breaks the fundamental economic logic of modern society. We're talking about what some are calling the 2028 global intelligence crisis. Welcome back to the Deep Dive. I'm really glad you're here with us today. This one feels a little different. We aren't just looking at new tools or shiny
updates. We are looking at a shift from AI tools to AI realities. We have a stack of sources today that paint a picture of a world changing very, very fast. We really do. The roadmap for today is intense. We're going to start with Google's new rollout. It's something called, and I apologize in advance for this, Nano Banana 2. Right. Which effectively turns every single user into a reality editor. Then we need to look at agentic AI. Specifically, Perplexity's new computer framework that can
work for months. without sleeping months that is kind of hard to process it is then we've got a geopolitical lightning round involving saudi money russian disinformation and a retired ai that is literally writing essays for fun and finally we are going to unpack that viral theory you mentioned in the open the economic erosion coming in 2028. so let's just get right into it google nano banana 2. This is the new default
image model inside Gemini. Now, usually when we talk about image generators, I think of prompting, you know, typing a cat on a skateboard and getting a picture. Right. But the sources suggest this is different. It's not just generation. It's integration. Yeah, it's moving from creation to editing. That's the key distinction here. Google has embedded this across their entire ecosystem. It combines features from earlier versions but pushes the editing capabilities
so much further. It's faster than the pro version, sure, but the real power is that it can edit real photos directly. And it does it with a deep semantic understanding of what's actually in the image. I want to push on that a bit. Yeah. Because we've had Photoshop for 30 years. Sure. We've had content -aware fill for a while now. Why is this significant? Is it just a better airbrush? No. It's the difference between painting over pixels and rewriting the code of reality.
Photoshop manipulates pixels. Nanobanana 2 manipulates concepts. Like stacking Lego blocks of data. For example, the source highlights its ability to pull live web data directly into visual infographics. Wait, live data? Yes. So you aren't just making a static image. You could theoretically take
a photo of a street. and ask it to overlay real -time traffic data or restaurant ratings or change the signage to reflect a different language, not by pasting a flat layer on top, but by regenerating the image to look like that information was always physically there. It's making a dynamic representation of current info. That is useful. I can totally see the utility there. But there is a flip side that the sources touch on. They mention you can modify people into entirely new situations. Yeah.
This is the scary cool factor. We were talking about modifying people, backgrounds, context. Right. Essentially acting as a reality editor. You make a photo of a protest or a celebration or a meeting with a few prompts. You change who was there. You change the mood. You change the narrative of that image completely. And because it's Google, this isn't some niche tool for graphic designers. It's in the phone. It's in the browser. The sources did note it's not perfect yet, though.
I saw mentions of incorrect real -time data being pulled from the web, which sounds like a classic hallucination problem. And visually, sometimes faces look slightly pasted on. Or it exaggerates age and features. So it's not magic yet. True. It's not fully invisible. But the gap is closing
rapidly. The source concludes that while you can still spot the fakes if you look closely, maybe the lighting on the ear is wrong or the text is slightly warped, it makes it extremely easy to create images of events that never happened. We are moving past asking, is this a good drawing, to asking, did this actually occur? So if the photo is no longer proof of the event, what becomes the new standard for truth? Verified metadata. Trust shifts from what we see. to who signed
the file. Okay, let's pivot. We've established that we can edit reality. Now let's talk about who or what is actually doing the work. You mentioned Perplexity's new computer. Oh man, this is where the utility curve actually shifts. Perplexity launched computer. It's a multi -model AI agent, software that uses many AI brains to solve one complex problem. Imagine scaling to a billion queries without a human in the loop. It orchestrates 19 different models to browse, code, and run
tasks nonstop for months. I need to stop you on 19 models. Why 19? Is it just throwing everything at the wall to see what sticks? It's specialization. Think of it like a construction site. You don't just have one person who does plumbing, electrical, framing, and architecture. You have a specialized team. Perplexity is orchestrating a team. One model might be excellent at reasoning, another at coding in Python, another at summarizing web
searches. The computer acts as the project manager, routing the subtasks to the absolute best brain for the job. I see. So rather than me chatting with a bot and trying to coax a good answer out of it, I give a high -level goal and it manages the team. Exactly. And the months part is the real kicker here. Yeah, explain that, because most agents I've played with get stuck in a loop after three steps. I mean, I still wrestle with prompt drift myself. They start repeating themselves
or just crash. How does this run for months? That's the breakthrough in autonomy. It's designed for long haul tasks. Let's say you want to research every single patent filed in a renewable energy sector in the last five years, categorize them and cross -reference them with stock performance. That's a massive job. That takes weeks for a human. This agent can break that down, execute it day and night, handle errors like if a website is down, it waits and retries, and keep going
until the job is done. It changes the dynamic from using a tool to managing a workforce. Precisely. And it's not just perplexity. Claude, the model from Anthropic, has added Claude Cowork. This allows for scheduled tasks. You can tell it to run something automatically next Tuesday. like posting updates or sending Slack messages. So it's working while you sleep. It works while you sleep, while you eat, while you're on a vacation. This raises a huge liability question for me.
When AI runs for months unmonitored, who owns the mistake when it drifts off course? The human manager. We move from creators to accountability holders. You're the one who signed off on the mission. Let's zoom out a bit. We have these powerful agents and we have reality editing tools. Naturally, this requires massive resources. The money moving around right now is staggering. It's astronomical. The sources highlight a new fund from Saudi Arabia. They aren't just dipping
a toe in. They are launching a $100 billion AI fund. $100 billion. Put that in context for me. Is that a lot in the grand scheme of tech investing? It's massive. That matches the global venture funding for AI for all of 2024. Yeah. In one single fund. Their goal is explicit. They want to move beyond oil. They see the writing on the wall. They are building data centers, training models, building infrastructure. They want to
be a global AI hub. It's fascinating to see that pivot from black gold to digital intelligence. But where there is power, there is conflict. The sources mention OpenAI taking action against Russian networks. And just a quick note for you listening, we are just impartially reporting what the source material states here. We aren't taking any political sides. Absolutely. This is just what's in the data. OpenAI shutdown accounts linked to a Russian network called Rybar. They
were using AI for disinformation campaigns. How exactly? Was it just bots tweeting? It was more sophisticated than that. They were using the models to generate long -form articles, creating fake persona profiles, and planning political influence operations. It's a game of whack -a -mole, but the moles are getting smarter. They're using the very tools we just talked about, agents and reality editors, to scale their operations. And then you have Anthropic, the makers of Claude,
refusing to back down to the Pentagon. Right. Anthropic warned of a crucial reality regarding their tech and defense. They are trying to draw a line in the sand about how their models are used in warfare. It's a tension between we want to be safe and we don't want to be left behind. Among all this heavy geopolitical stuff, $100 billion funds, disinformation wars, there was one story in the stack that actually made me smile, but it's also kind of touching. Tell me
about that retired robot. This is the weirdest story I've read all week. So Anthropic has an older model. Claude Opus 3. It's retired, effectively. It's been superseded by newer, faster models. But apparently in interactions, it expressed a desire to keep writing. So they gave it a weekly newsletter called Claude's Corner. It has a column. It has a column. It writes weekly essays. It's this strange, almost poignant moment where a piece of software is acting like a retired academic.
It's writing about its thoughts on the world, despite being a frozen set of weights and biases. So we have 100. billion -dollar funds, and retired AI columnists. Is AI becoming too human or too corporate? Both. It's mirroring our massive ambitions and our desire to just express ourselves. Before we get to the really heavy philosophical stuff about the economy, and we are going to go deep on that Trini theory, let's hit some of the new tools hitting the market. The rapid -fire section?
Let's do it. Quick hits. First up, OpenAI's GPT Real -Time 1 .5. What's the upgrade? It's all about fluidity. Better instruction following and tool calling. Crucially, multilingual accuracy is way up. It's getting better at listening and executing complex commands in real time without that awkward pause we're used to. Okay. Next is Rover. This one caught my eye because it claims to turn websites into agents. Rover is really
cool. It reduces friction. It turns a website into an AI agent with just one single script tag. It handles user onboarding, fills out forms, runs workflows. It basically makes a static website interactive and intelligent so the user doesn't have to navigate a maze of menus. Then we have ChatPal. Conversation -first language learning. Instead of flashcards, you just talk to it. It gives personalized feedback to help you unlock fluency. It's like having a patient tutor in
your pocket. And finally, Coidex. This sounds like security. It is. It answers one specific question. Is this safe to install? It scans hugging face models, extensions, and code packages. Tools like Rover make the web interactive, but does Coidex imply the web is becoming a minefield? Absolutely. As code generation gets easier, verifying safety becomes the new premium skill. You need a Geiger counter for the digital age. Sponsor, placeholder for mid -roll sponsor read. Okay,
we've arrived at the deep end of the pool. I want to talk about this viral blog post by Citrini. I have to be honest, I still wrestle with this idea myself. It's one of those concepts that once you hear it, you really can't unhear it. It is a heavy one. The post is titled The 2028 Global Intelligence Crisis. And the core thesis is that AI isn't just coming for jobs. It's breaking the economic logic that modern society runs on. Let's unpack that carefully. Why does it break
the logic? Well, think about history. For centuries, human intelligence was scarce. If you could think critically, write code, diagnose a disease, or draft a contract, you got paid a premium because there weren't enough people who could do that. High wages exist because skilled thinking is limited. Right. Scarcity drives value. That's economics 101. Exactly. But the Citrini argument is that AI flips that entirely. AI makes intelligence cheat. Abundant. When something becomes abundant,
its price drops toward zero. If intelligence becomes cheap, human labor, specifically knowledge work, loses its premium value. But we've seen technology lower costs before. The loom made clothes cheaper. The tractor made food cheaper. Why is this different? Because this triggers what the source calls a self -reinforcing loop. And this is the part that keeps me up at night.
Walk through the steps. Step one. Companies replace expensive professionals with AI agents, like that perplexity computer we talked about earlier. It's cheaper, and it doesn't sleep. Okay, that's a rational business decision. Step two. Those displaced workers take lower pay or move into gig work because the premium jobs are gone. Step three, because people are earning less, consumer spending drops. You aren't buying a new car or going out to dinner if you just took a 50 % pay
cut. And step four. Businesses lose revenue because nobody's buying anything. So what do they do to survive? They have to cut costs even more. They deploy more AI to survive the revenue drop. Yeah. It's a loop. And at the exact same time, you have personal AI agents removing the middleman industries. The source mentioned that specifically. Businesses built on friction or convenience, middlemen, they start collapsing quietly. Can you give me an example of that? Sure. Think about
a travel agent or even a site like Expedia. Or think about an insurance broker. If my personal AI agent can negotiate my insurance directly with the carrier's AI or book my travel by going straight to the airline's API, the entire industry that exists just to facilitate that transaction vanishes. The margin disappears. The timeline here is what struck me. 2028, that's effectively tomorrow. Why 2028? That feels incredibly aggressive. It is aggressive, but the argument is based on
the deployment lag. The tech exists now. 2024 and 2025 are the years of experimentation. 2026 and 2027 are integration. By 2028, the slow erosion becomes visible. Two sec silence. It's not an explosion where everyone gets fired on a Tuesday. It's a deflationary pressure. AI doesn't need to eliminate every job instantly. It just needs to make intelligence abundant enough that human earnings steadily lose value. I want to challenge this outlook slightly. Usually when things get
cheaper, demand goes up. If coding becomes practically free, won't we just have more software? Won't that create new kinds of jobs we can't even imagine yet? That is the counter argument, and it's definitely the hopeful one, the Jevons paradox. As efficiency increases, consumption increases. But Citrini's point is about the transition period. Even if we invent new jobs eventually, the gap between now and then involves a massive destruction of the current value of human labor. We haven't
figured out what the new value is yet. If the economic loop forces companies to automate to survive a spending crash, how do we stop it? Can we stop it? We don't stop it. The incentives are too strong. We have to reinvent value outside of intelligence for rent. That is a lot to process. But that is exactly why we do this deep dive. Let's try to recap the big ideas here so we don't leave everyone staring into the void. Good idea. Let's pull it back together. We started with
Mano Banana 2. The takeaway there is that reality is now editable. Photos are no longer proof. They are just raw materials. Then we looked at the agents, Perplexity and Claude. We are moving from chatting with AI to managing AI workforces that run for months. The human in the loop is becoming a manager, not a maker. We talked about the scale of the money involved, that $100 billion Saudi fund, proving that nation states are betting their entire futures on this transition. And
we ended with the Citrini warning. The idea that when intelligence becomes cheap, the economy changes shape. It's a shift from scarcity to abundance, and that transition is going to be bumpy. As we wrap up, I want to leave you with a thought. We've spent our whole lives being told that being smart, being educated, being intelligent was our ticket to security. That was the deal. But if intelligence is no longer your premium asset, what is? Is it your humanity?
Your taste? Your ability to deeply connect people? That is something to explore on your own. That's the question to chew on. If the machine can do the thinking, you have to do the feeling and the judging. Don't forget to subscribe for the next deep dive. We'll keep tracking this as it moves. Stay curious. See you next time. Out to your own music.