Over 50% of the web is synthetic. When AI trains on AI, generic mediocrity is guaranteed.
Welcome back to Business Tech Culture.
Over the last few drops, we’ve laid out how modern builders survive in an era of infinite technological acceleration. We locked down The Enemy Moat, bypassed legacy gatekeepers in The Format Rebellion, and realized that when generation is free, you need The Curation Monopoly.
Today, we need to talk about why the machine is actively breaking down—and why that is the single biggest opportunity for modern culture architects.
Silicon Valley executives and B2B marketing teams are still operating under a massive delusion. They believe that if they just throw more compute at LLMs and generate 100x more blog posts, ad copy, and code, they will win the content war.
They are driving directly off a cliff. We have officially entered the era of Model Collapse.
The Photocopy of a Photocopy
In a sharp analysis on the recursion of synthetic web data, full-stack developer Alex Williams broke down the quiet disaster happening behind the scenes: Model Collapse Is Already Happening, We Just Pretend It Isn’t.
As Williams points out, over 50% of the open web is now AI-generated. Because frontier models are constantly scraping the internet for training data, they are no longer learning from human thought—they are training on the synthetic outputs of previous models.
It is an ouroboros eating its own tail.
When an AI model trains on synthetic data, it suffers from progressive degradation. It loses the "tails" of the distribution—the strange, rare, hyper-creative, eccentric patterns that make human culture rich. It converges toward a bland, sterile statistical mean.
Think of it like making a photocopy of a photocopy. The first copy looks sharp. By the tenth copy, the contrast is washed out, the detail is gone, and everything looks like gray mush.
When you read a piece of corporate content today and feel an instant "ick" without finding a specific error, you are feeling model collapse in real-time. It is the sound of an algorithm sanding down every unique edge until nothing remains.
The Scaling Trap
Corporate PR teams and traditional SaaS founders are trapped in what Williams calls the "Scaling Trap."
When their generic AI content fails to drive actual community engagement, their response isn't to fix the soul of the message—it's to double down on volume. They generate more slop, push more automated campaigns, and flood the timeline with synthetic noise.
They think scale solves quality problems. It doesn't. It amplifies the degradation.
If your entire brand strategy relies on using AI to pump out volume, you are voluntarily turning your company into a sterile commodity. You are racing to the statistical center, where price goes to zero and brand loyalty disappears entirely.
The Luxury of the Edge
This is where the modern builder wins.
When the entire internet is being pulled toward a bland, homogenized mean, the edges of the distribution become the ultimate luxury asset.
AI cannot replicate:
The weird, un-smooth inside jokes of a tight group chat.
The raw, polarizing opinions that make corporate tourists uncomfortable.
The hyper-specific lore forged in the trenches of real execution.
The jagged, unpolished "Imperfection Premium" of a human who actually has skin in the game.
Model collapse proves our core thesis once and for all: Code is cheap. Culture is lethal.
Stop trying to out-generate the machine. Let the corporate suits drown their audiences in synthetic photocopy slop. Your job is to stay weird, protect your taste, and build for the human beings who are desperately searching for something real.
Keep the seams exposed,
Ash