It creates this fascinating tension, doesn't it? On one side, you have this massive grassroots movement triggered by a $25 million donation. People are actually canceling subscriptions in protest. trying to stop the machine. And at the exact same time. At literally the same time, on the other side of the country, you have a $10 billion construction project breaking ground.
Right. It's the ultimate contrast. You have the human layer trying to pull the emergency brake, and then you have the industrial layer hitting the gas harder than we have ever seen. Hello and welcome back to the Deep Dive. I've been sitting with this stack of sources you sent over for a few hours now, and I have to be honest, the signal -to -noise ratio in the AI space this week has been overwhelming. It feels that way. It feels like the industry is splitting in two
directions at once. We have a cultural reckoning happening in the software and a physical explosion happening in the hardware. It really is an inflection point. I mean, usually when we sit down, we're obsessing over the models. Is it smarter? Is it faster? But today's sources paint a much, much messier picture. To really get what's happening this week, we have to look at three specific pillars. Okay, walk us through the roadmap. First, we have to talk about the cultural backlash.
We're seeing the rise of this quit GPT movement. This is activism finally colliding with algorithms. Second, the infrastructure explosion. We aren't talking about code here. We're talking concrete, chips, and power grids, the physical reality of the cloud. And third, despite all that noise, the capability leap is still happening. We have new tools from Alibaba and OpenAI that fundamentally change how we actually work. Let's start with
the friction, the cultural backlash. For a long time, the public sentiment around AI was just pure awe. Look what this thing can do. But reading these notes on QuitGPT, this feels different. This isn't just people being afraid of the Terminator. This is organized economic pressure. What sparked this? It is different because it's political. The catalyst here wasn't a software bug or a bad answer. It was a $25 million donation to Magia Inc. by Greg Brockman, the president of
OpenAI, and his wife. Okay, let's pause on that. So we have a direct financial link from the leadership of the leading AI company to a very polarizing political campaign. Exactly. And that donation basically lit the match. But if you look at the source material, it's not just that one check. It's a compounding effect. You have the donation, but then you also have this new partnership between. OpenAI and IC immigration and customs enforcement.
They're using ChatGPT4 for hiring and for administrative processing. Which for a significant portion of the tech workforce and that early adopter user base represents a massive red line. A huge red line. Remember, a lot of the people who pay for ChatGPT Plus are in that creative tech adjacent demographic. They feel a sense of ownership over the tool. So when they see the leadership funding
one political side. and then the technology being deployed by ICE, it just feels like a betrayal of that original open and beneficial mission. And Scott Galloway is involved now. I saw his name in the notes. Yeah, the NYU professor. He's very good at capturing the zeitgeist of these things. He went viral urging people to resist and unsubscribe. Right. He's framing it not just as a boycott, but as a resistance against big tech tools fueling specific political agendas.
And it's resonating. The sources say there are about 17 ,000 pledges on the QuitGPT site so far. Okay, 17 ,000. Let's look at the numbers. On a human level, that's... That's a stadium full of angry people. That's significant. It is. But ChatGPT has something like 900 million weekly users. Does 17 ,000 cancellations actually matter to a company like OpenAI? Or is this just screaming into the void? Financially, no. The
math is brutal. Only about 5 or 6 % of those 900 million users are even paid subscribers. So even if they lost 100 ,000 subscribers, the revenue hit is just a rounding error. It won't stop the lights from coming on. OpenAI is burning cash on compute, not relying on $20 subscriptions to survive. So if it doesn't hurt the wallet, is it just performative? I wouldn't call it performative. I'd call it a signal. A cultural signal. OK. You have to remember, OpenAI isn't just fighting
for customers. They are fighting for talent. The engineers who build these models are often very politically active, very ethics focused. If the brand becomes toxic. If working for OpenAI becomes synonymous with funding the bad guys in the eyes of their peers, that hurts them way more than losing some subscription revenue. That's a great point. The talent war is the real war. Exactly. If the smartest 25 year old researcher refuses to sign the offer letter because of the
brand reputation, that's the actual cost. Precisely. And it's not just consumers versus companies anymore. The sources highlight something fascinating. The companies are starting to fight each other politically, too. It's a full -on war of the pigs. I saw that bit about Anthropic. They frame themselves as the safety -first, constitutional AI alternative. What are they doing? They just funded a $20 million super piece. $80 million?
Specifically to push for AI regulation. That is not a small amount of change for a policy push. No. And it's directly opposing the political interests backed by open eyes. Fifty million dollar push. So you have these two. massive entities who are competitors in code now becoming competitors in the political lobby. They're literally spending tens of millions of dollars to shape the laws that will govern the other. It really highlights that AI isn't just a tool anymore. It's a political
battlefield. You can't just build it and they will come. You have to build it, you have to lobby for it, and you have to defend it from your own user base. That's the reality of 2026. So just to cap this segment off, does this consumer pressure actually work in an age of enterprise dominance? It creates a morale crisis, which creates a tenant crisis, which eventually creates a product crisis. So, yes, it matters. Interesting.
So while the culture is fighting over donations and hiring policies and the lobbyists are fighting over regulation, the actual physical machine is just getting bigger. Oh, yeah. I want to pivot to the infrastructure piece because the numbers here are staggering. The hardware reality. This is the stuff you can kick. I was looking at this note about meta in Indiana. $10 billion for a data center. It's monumental. Meta just broke ground on this. And data center feels like too
small a word. Right. It's a campus. 13 buildings, 4 ,000 construction jobs. This is a physical manifestation of their next -gen ambitions. When Zuckerberg says they are building the future, they are literally pouring the concrete for it in Indiana. It changes how you visualize the cloud, doesn't it? We tend to think of the cloud as this ethereal thing floating above us. Right. But it's actually 13 massive buildings in Indiana sucking up electricity. And that electricity
part is becoming the friction point. The power demands are so high that these companies are having to play good neighbor in a very, very expensive way. I saw that in the notes. Anthropic, Microsoft, and OpenAI. are all promising to pay the extra power bills. They have to. Imagine you live in a town, a data center moves in, and suddenly your residential electricity rate goes up 20 % because demand spiked. That's a political nightmare. It is. So they are preemptively saying,
we will cover the difference. It shows you how desperate they are to keep these things running without local backlash. They're essentially paying rent on the grid itself. It's almost like a pollution tax, but for electron consumption. Exactly. And speaking of desperate need for speed, we have to talk about this OpenAI and Cerebras deal. This connects the hardware to the actual user experience. This is Codex Spark. Right. OpenAI
launched Codex Spark. It's a lighter version of GPT -5 .3 designed specifically for real -time coding. But the moment of wonder here isn't the software. It's the chip. It's the chip powering it. The Cerebras chip. I saw the Spex 4T transistors. Four trillion. Can we contextualize that? Because trillion is one of those numbers that human brains aren't good at visualizing. What does a four trillion transistor chip actually allow you to do that a normal chip doesn't? Speed. Specifically,
inference speed. So training is teaching the AI. That takes months. Inference is when you ask it a question and it answers. Right. That's the daily usage. Usually there's a lag. You type, the bubbles spin, the text streams out. With a chip this size, the memory bandwidth is so high that the latency basically vanishes. So it moves from email correspondence to instant messaging. It moves to thought speed. If you're a coder using Codex Spark. The AI isn't pausing
to think. It's typing with you. It predicts your next logic block before you even finish the syntax. Wow. It changes the psychological relationship with the tool. It feels less like a tool and more like an extension of your own fingers. That's incredible. And this ties into that valuation for Modal Labs, right? Exactly. Modal Labs is raising at a $2 .5 billion valuation. They focus entirely on inference infrastructure, basically the plumbing that allows these massive models
to run quickly and cheaply. The market is betting heavily that this is the year we stop just training models and start running them at massive scale. So are we underestimating the physical footprint, the concrete and electricity required for digital intelligence? Absolutely. Software is now driving the largest industrial construction boom in recent history. Which brings us to the tools themselves. We built the data centers. We fought the political battles. Yeah. What can we actually do with this
stuff today that we couldn't do yesterday? This is the fun part. The capabilities are shifting from just chatting to doing. And the big story here is Alibaba. Quinn Image 2 .0. Catchy name, right? Rolls right off the tongue. The note here says it's a PowerPoint native model. What does that mean exactly? Okay, think about how most image generators work right now. Mid -journey, daily. You ask for a picture, you get a picture. It's art. But if you try to put text on it, it
looks like alien hieroglyphics. If you try to make it a specific resolution for a slide deck, you have to crop it. Right. It's creative. But it's not productive. Exactly. Quen changes that. It merges generation and editing in one shot. It handles 2048 by 2048 resolution. So no more stitching tiles together. And crucially, it has high text fidelity. So you can actually read it. You can tell it. Make me a slide about Q3 earnings with a bar graph and the title growth
vector in Arial font. And it just does it. So it's an AI competitor to Canva. not just an AI computer to an artist. That's the promise. It's moving the workflow. You aren't just generating assets that you then have to take into Photoshop to fix. You are producing a final asset right in the chat. That is a significant shift for enterprise users. I know I struggle with that, constantly getting the image right, but having
the text look ridiculous. Right. And while Alibaba is fixing images, Anthropic is coming for your Office suite. They just unlocked file tools for free users. Excel, PowerPoint, Word, no sh**. Previously, you had to pay to have Claude analyze these. Now it's open to everyone. That is a big democratization move. Suddenly, having an AI analyst go through your spreadsheets isn't a premium feature. It's a baseline expectation.
Right. Exactly. But, and there is always a but in this industry, with these new capabilities comes a darker side. We have to talk about what happened to Google's Gemini. The stealth attack. I read this twice because I wasn't sure I understood it. This wasn't a hack in the traditional sense, right? They didn't break a password. No, they didn't break in. They invited themselves in. It's called a model copying campaign or model distillation. Walk us through how that works.
So Google's Gemini is a trillion dollar brain. It's huge, expensive, and smart. A competitor or a bad actor wants that intelligence but doesn't have the money to train it. Okay. So they hit Gemini with over 100 ,000 very specific complex prompts. Explain quantum physics. Write a legal brief. Debug this code. And they just record all the answers? Record the answers. Then they take those high -quality answers and feed them
into their own much smaller, cheaper model. They are teaching their cheap students using Google's expensive textbook. Wow. Effectively, they are siphoning off the intelligence of the master model to create a clone. That sounds incredibly difficult to police. It is. Google is calling it theft. They are warning startups that if it can happen to Gemini, it can happen to anyone. But the legal framework here is nonexistent. Is asking a question theft? Is learning from
an answer copyright infringement? It raises this massive question about what intellectual property even looks like when you can steal it just by asking questions. If I can distill your secret sauce just by talking to your bot, do you even have a moat? That is the billion -dollar question. And it connects back to that feeling of vulnerability. We are building these massive systems, but the edges are porous. Speaking of vulnerability, reading through these sources made me think about
my own usage. I have to admit, I still wrestle with what they call confident guessing. Oh, hallucinations. Yeah, but specifically in high -stakes stuff, like I'll use AI for a contract review or a compliance check, and it sounds so sure of itself. But there's that nagging fear. Is it making this up? The source material mentioned a safety prompt rule to help with this. Yes, this is a crucial takeaway
for you listening. The source mentioned a specific framework for stopping that confident guessing. Essentially, when you are doing something high stakes, checking a legal doc, looking for citations, you have to explicitly instruct the model on the negative constraint. Meaning what? You tell it. If you do not find the specific clause in the text provided, state that you cannot find it, do not infer, do not guess. It sounds so simple. But we forget to do it. Yeah. We trust
the chat interface too much. We treat it like a conversation where it's rude to be that blunt. But you have to. We need to force it to prove its claim. Absolutely. We have to move from being passive users, hey, tell me about this, to active supervisors. Analyze this. Improve your work. So with things like Quinn merging editing and generation, what does this actually mean for creative workflows? We are moving from generating ideas to producing final assets without leaving
the chat interface. It's a brave new world. It certainly is. We're going to take a very quick break to thank our partners who help keep this deep dive running. And when we come back, we're going to try to synthesize all this, the politics, the concrete, and the code into one big picture. Stay with us. Okay, we're back. We've covered a lot of ground today. From the cancellation parties of QuitGPT to the massive construction sites in Indiana to the model thieves stealing
intelligence prompt by prompt. Yeah. When you look at this entire stack of stories, what is the through line for you? I think we are seeing a massive divergence. That's the word that keeps coming to mind. Avergence between what and what? Between the human layer and the machine layer. On one hand, you have the human layer pushing back. You have the quit GPT movement. You have the political maneuvering, the ethical concerns about ICE and funding. It's messy. It's emotional.
It's loud. Right. It's people saying, wait, stop. Does this align with our values? And on the other hand. On the other hand, the machine layer is just accelerating with total indifference to that noise. Ten billion dollar chips, massive data centers, models that can clone each other. Yeah. The technology is no longer just software on a screen. It is becoming a physical industrial force and a political lightning rod simultaneously. It's almost like the technology has gained enough
momentum that it's. It's decoupling from the public sentiment. The boycott is happening, but the cement trucks are still pouring. Exactly. The train has left the station, and now we are just arguing about who gets to sit in the conductor's chair. That is a sobering thought. The scale of the investment, the sheer physical weight of it, makes it very hard to stop or even steer. It does. And if I can leave you with one final thing that's been sticking in my brain, it's
that story about the model copying. The Gemini theft. Yeah. Think about it. If a trillion -dollar company like Google, with all its defenses, can have its intelligence stolen via 100 ,000 text prompts, what does intellectual property even mean in the age of fluid intelligence? If knowledge can be decanted from one machine to another just by asking questions, the economics of this whole industry might be more fragile than we think. That is a question I think we're going to be
wrestling with for the next decade. I think so, too. If you want to protect your own workflows, especially if you're using these tools for contracts or anything legal, I highly recommend you check out that safety prompt rule we mentioned. We'll have the details and the specific phrasing in the show notes. It's a small step, but it might save you from a very confident, very wrong answer. Worth a click. Thanks for diving Dean with us today. It's a complex world out there, but hey,
at least we're figuring it out together. See you next time. Take care, everyone.