The Seven Trillion Dollar Scam

Why I Refuse to Call Degeneration Progress

Sam Altman wants seven trillion dollars for artificial “intelligence” that may or may not some day be able to figure out how many legs a horse has. Or was it to cure cancer? I don’t remember. Anyway, staying true to this quest, OpenAI released Sora, a video generator that burns through $700,000 daily to produce clips where people’s fingers occasionally phase through solid objects, where gravity works sideways, where horses sprout extra legs mid-gallop—and where your brain rots away with every swipe. They’re calling it intelligence, the “democratizing” of video creation. It’s a random number generator with a marketing department that caters to the dumbest people of society.

If you appreciate my articles, please consider giving them a like. It’s a simple gesture that doesn’t cost you anything, but it goes a long way in promoting this post, combating censorship, and fighting the issues that you are apparently not a big fan of.

The entire AI industry rests on a fundamental lie: that pattern matching equals thinking, that statistical correlation equals understanding, that predicting the next most probable token equals intelligence. Sora doesn’t know what a coffee cup is—it’s seen millions of images labeled “coffee cup” and learned to produce pixel arrangements that statistically resemble those patterns. When it generates a video of someone drinking coffee, it’s not modeling the physics of liquids or the anatomy of swallowing. It’s performing multidimensional regression on pixel distributions. The cup phases through the hand because the machine has no concept of “solid” or “liquid” or “hand”—only statistical correlations between RGB values.

AGI—artificial general intelligence—was supposed to mean machines that could think, reason, and understand like humans. Solve novel problems, based on their own newly created thoughts. Generate genuine insights. Exhibit creativity that isn’t just recombination of training data. Instead, OpenAI and its competitors are redefining AGI to mean “passes enough benchmarks that we can declare victory.” When they announce AGI next year or the year after, it won’t be because they’ve created thinking machines. It’ll be because they’ve successfully moved the goalposts to where their pattern matchers are standing.

You know what this is? It’s autocorrect all over again, just with better UI; autocomplete for pixels. The same fundamental process that suggests “u up?” after “hey” at 2 a.m., scaled up to unconscionable computational requirements and dressed in revolutionary rhetoric.

OpenAI’s GPT models—the supposed foundation of artificial general intelligence—are glorified Markov chains on steroids. They don’t understand language; they predict token probabilities. When ChatGPT writes about democracy or love or quantum physics, it has no concept of governance, emotion, or particles. It’s performing statistical pattern matching on text stolen from the internet, predicting what word most probably follows the previous words based on billions of examples. The machine that writes poetry about heartbreak has never experienced anything, understands nothing, knows nothing. It’s a Chinese Room with better PR.

The AGI announcement, when it comes, will be pure theater. They’ll show demos of their model passing medical exams (by pattern-matching against millions of medical texts), writing code (by recombining Stack Overflow posts), having “creative” conversations (by interpolating between Reddit comments). Tech journalists, either too ignorant or too invested to call out the con, will breathlessly declare the arrival of artificial general intelligence. The stock price will soar. The seven trillion in funding will materialize.

And it will all be a lie.

Sora is a monument to misdirected genius, cathedrals built to worship the wrong god. Every brilliant mind working on making fake videos look slightly less fake could have been working on actual research. Every dollar spent teaching machines to mimic human creativity could have funded actual human creators. Every kilowatt-hour burned generating “cinematic” garbage, that “democratizes filmmaking”, could have created movies that are actually fun to watch.

Altman already admitted what critics have long suspected: the humanitarian rhetoric is window dressing for wealth extraction. When pressed about the gap between their utopian marketing and dystopian products, the response was revealing: they need to “demonstrate capability” while “building revenue streams.” Translation: the cancer-curing, poverty-ending, progress-accelerating AI was always a lie to justify the real goal—monopolizing the infrastructure of digital expression, capturing your attention, and torturing your brain.

AI’s training process reveals the scam’s magnitude. Sora consumed millions of hours of video, billions of images, unknowable quantities of stolen human creative output—all reduced to numerical weights in a neural network. The machine didn’t learn what a sunset is; it learned statistical correlations between pixel gradients typically labeled “sunset.” It can generate infinite variations of things that look like sunsets without ever understanding that the sun is a star, that Earth rotates, that light refracts through atmosphere. It’s pattern matching without comprehension, correlation without causation, syntax without semantics.

Consider what Sora actually does. It ingests decades of human visual culture—every film, every video, every fragment of recorded humanity it can access—and reduces it to statistical patterns. Then it reconstitutes these patterns into uncanny simulacra that look almost but not quite right, like memories of dreams of movies you half-remember. The output invariably features that distinctive AI shimmer: surfaces that seem to breathe, faces that melt if you stare too long, physics that operates on dream logic rather than natural law.

This would be merely embarrassing if it weren’t so expensive. The computational requirements for generating a single minute of Sora footage could power a small town for a day. The training process consumed enough electricity to supply thousands of homes for a year. And for what? So influencers can generate backgrounds for TikToks? So marketers can create synthetic testimonials? So we can further blur the already fuzzy line between authentic human expression and algorithmic approximation?

They will, however, assure you that this is an essential step toward their sacred AGI objective. That’s another lie. Real AGI would require something these systems fundamentally lack: a model of reality. Understanding causation, not just correlation. Grasping concepts, not just patterns. The ability to reason from first principles rather than interpolate from examples. No amount of scaling current approaches will achieve this because the architecture is wrong at its foundation. You can’t build intelligence from statistics any more than you can build consciousness from clockwork.

The computational requirements expose the inefficiency of brute-forcing intelligence through statistics. Human children learn object permanence from dozens of examples; Sora needs millions of videos and still generates coffee cups that spontaneously become tea kettles mid-frame. A three-year-old understands that people have two arms; Sora, after consuming more visual data than any human could process in a lifetime, still generates people with three arms growing from their torsos.

Altman knows AGI isn’t coming. In private conversations leaked by former employees, he’s admitted that current approaches have fundamental limitations. But publicly, he maintains the fiction because the entire house of cards—the valuations, the government approval, the investments, the seven-trillion-dollar ask—depends on investors believing AGI is imminent.

The tell is in how they keep redefining success. First, AGI meant human-level intelligence across all domains. Then it became “human-level at most economically valuable tasks.” Now it’s “passes certain benchmarks better than average humans.” By the time they declare victory, AGI will mean “generates outputs that fool people who’ve been consuming AI slop for so long they can’t recognize actual thought.”

The entire large language model revolution is built on this foundation of sand. GPT-5, Claude, Grok, Gemini (remember Gemini?)—they’re all variations of the same con: machines that simulate understanding through statistical correlation, that mimic intelligence through pattern matching, that fake consciousness through probability calculations. That’s why they all sound the same. They generate text that appears meaningful because they’ve learned what meaningful text typically looks like, not because they understand meaning.

We’re watching the systematic redefinition of intelligence to match what machines can fake rather than what intelligence actually is. It’s like declaring that player pianos have mastered musical performance—technically impressive, completely soulless, and missing the entire point of what music is.

Consider what OpenAI calls “emergent capabilities”—the supposedly surprising abilities that appear as models grow larger. But these aren’t emergence of intelligence, they’re just statistical inevitabilities. Feed enough text about chess into a pattern matcher, and it will eventually reproduce chess-like moves. Not because it understands strategy or planning (or what chess even is), but because it’s seen enough examples to predict probable next moves. It’s not playing chess. It’s just performing regression analysis on chess notation.

The industry knows this. Internal documents from multiple AI companies reveal that engineers refer to their models as “stochastic parrots”—machines that randomly recombine learned patterns without understanding. Yet publicly, they maintain the fiction of intelligence, of reasoning, of understanding. They have to. The entire $150 billion valuation depends on investors believing these machines think rather than merely compute probabilities. That’s why these people sound so unbelievably stupid on X.

I’m not against artificial intelligence. The idea that machines could one day amplify human insight, accelerate discovery, or help us understand ourselves more deeply is beautiful. What I reject is the fraud — the marketing theater that sells regression as revelation, the executives who cloak cats in cars being pulled over by cops as beacons of human progress, and the AI bros who pretend that slob is a form of art.

There’s nothing wrong with building tools that help us think faster or a little clearer; what’s wrong is pretending those tools can think, feel, or create. The tragedy is that we’ve turned one of humanity’s most promising ideas into a con built on hype, stolen labor, and misdirection.

Again, AGI will be called “AGI” because they’ve successfully convinced enough people that sophisticated pattern matching equals thinking. They’ll point to benchmarks conquered, tests passed, conversations that sound human. They won’t mention that it’s all statistical mimicry, that the machine has no more understanding than a calculator has of mathematics.

The energy waste becomes even more obscene when you understand what’s actually happening. We’re burning gigawatts of electricity not for thinking machines but for industrial-scale garbage creators. The carbon footprint of training GPT-4—equivalent to the lifetime emissions of 500 cars—was spent teaching a machine to predict that “cat” often follows “the” and “sat on the.” The millions of gallons of water cooling Sora’s training infrastructure were used to teach it that skin-colored blob patterns often appear above shirt-colored blob patterns.

This is what seven trillion dollars will buy: more sophisticated pattern matching. Larger statistical models. Higher-resolution probability calculations. Not intelligence, not understanding, not consciousness—just increasingly expensive ways to generate statistically probable outputs that look meaningful to humans who’ve forgotten what meaning is. And then Sam is happy, Donald is happy, Jensen is happy. Simply because they have successfully managed to scale America’s data centers better than others so that it has become the greatest digital garbage creator on the planet. Thank you for your attention to this matter.

The “hallucination” problem that AI companies treat as a minor bug is actually a logical flaw. These systems don’t hallucinate (that would ironically require being capable of actual original thought)—they operate exactly as designed, generating statistically probable outputs regardless of truth or accuracy. When ChatGPT invents scientific papers that don’t exist, it’s not making an error—it’s doing what it always does: producing text that resembles text it’s seen before and sounds plausible. The machine doesn’t know what “true” or “false” means. It only knows probable token sequences.

We’re restructuring entire industries around these probability calculators. Replacing human judgment with statistical averaging. Substituting genuine creativity for pattern recombination. Trading actual intelligence for its simulation. The more AI is written on company papers, the higher the stock valuation goes. Apple’s stock price plummeted as they remained silent while everyone else was swept up in the AI craze. Apple’s innovative image was tarnished. When they introduced Apple Intelligence, a collection of largely useless and barely functional tools, investors were pleased again.

The companies selling these systems are aware that they lack intelligence—refer to their technical papers (Apple even openly acknowledged and criticized this in a paper that reads as if Steve Job’s long forgotten last breath of reason swept across Apple Park)—but they also understand that the market doesn’t care as long as the outputs appear convincing enough.

The prompt engineering priesthood that’s emerged reveals the absurdity. If these systems were intelligent, you wouldn’t need elaborate incantations to make them produce usable outputs. You wouldn’t need “jailbreaks” and “system prompts” and carefully crafted instructions. You’d just communicate normally, as you would with anything that actually understands language. Instead, we have an entire industry of prompt engineers—people whose job is to trick probability calculators into producing specific patterns through careful manipulation of input text—and then sell it to you in spreadsheets for $60.

OpenAI’s own researchers, in papers that barely make headlines, acknowledge these limitations. They write about “capability mirage”—the tendency for large models to appear more capable than they are because they’ve memorized vast amounts of training data. About “spurious correlation”—the machine’s tendency to learn accidental patterns rather than meaningful relationships. About “distribution shift”—how models fail catastrophically when encountering inputs that differ from their training data.

But these admissions get buried under avalanches of hype about AGI, about consciousness, about machines that think. The reality—that we’ve built very expensive random number generators—doesn’t sell enterprise licenses or justify billion-dollar valuations.

The Sora videos that OpenAI showcases are carefully curated from thousands of generations, selected for the ones where the horses have the right number of legs, where the coffee cups don’t phase through hands, where the physics accidentally looks correct. For every video they show, hundreds were discarded because they revealed what the system actually is: absolute and utter useless bullshit.

This is the seven-trillion-dollar scam: selling pattern matching as intelligence, correlation as comprehension, probability as understanding. The machines don’t think—they perform statistical operations. They don’t create—they recombine. They don’t understand—they calculate correlations.

The environmental catastrophe, the resource waste, the misdirection of human talent—it’s all in service of building more sophisticated probability calculators. Machines that can generate text that sounds meaningful without meaning anything. Videos that look real without representing reality. Answers that seem correct without any comprehension of the questions.

The population, meanwhile, grows steadily dumber from consuming AI-generated content, losing the ability to distinguish between genuine thought and statistical approximation. We’re being trained to accept lower-quality everything—lower-quality writing, lower-quality images, lower-quality thinking. By the time OpenAI declares AGI achieved, we’ll be so accustomed to synthetic approximations that we won’t remember what real intelligence looks like.

That’s the end-stage brain rot at the core of the AI revolution: we’ve convinced ourselves that sufficiently sophisticated pattern matching equals intelligence, that large enough statistical models equal understanding, that probable outputs equal thoughts. We’ve mistaken the map for the territory, the simulation for reality, the pattern for the meaning.

And we’re about to spend seven trillion dollars on this. Thanks, Sam. Very cool.

---

How you can support my writing:

• Restack, like and share this post via email, text, and social media

• Tip me a bug-free meal with Ko-Fi

• Thank you; your support keeps me writing and helps me pay the bills. 🧡