I've been thinking about this for a while, but I finally figured out what really bothers me about this AI revolution. It is the extinction of the outlier.
In the race to build the largest possible models, AI companies have created a "Slog of Data" so massive that it doesn't just process information; it dilutes it until the gold is indistinguishable from the lead. It's a silence of intelligence. Truly great ideas can breathe for a brief moment before they're sucked into the vacuum of the mean.
1. High-Probability Prediction and the Death of the Outlier
Large Language Models (LLMs) are built on the principle of High-Probability Prediction. They are designed to find the most likely next word, the most likely sentiment, and the most likely solution.
The 1% Problem: A "great idea" is, by definition, a statistical anomaly. It is the thing that no one else thought of. In a dataset of a trillion tokens, that idea is a rounding error.
The Algorithmic Eraser: When an AI encounters a truly novel human insight, it doesn't see "genius"; it sees "noise." The architecture is literally programmed to smooth that noise out to ensure the output remains "safe" and "relevant" to the masses. Sure you can have specialized models to handle the novelty, but that statement in itself should tell you something. We need special models to handle exceptions... ergo ipso facto, some other such latin to make my point fancy. I'm reminded of the movie Equilibrium; the exception is the problem.
2. Plausible Slop: The Discovery Tax on Novelty
We are being buried in a mountain of "Plausible Slop." Because AI can generate billions of pages of statistically relevant content, the 1% of truly transformative ideas are being physically crowded out.
The Discovery Tax: It is becoming too expensive to find the truth. When 99% of the search results, social feeds, and research summaries are AI-generated "averages," the energy required for a human to dig through the slog to find a single novel insight becomes prohibitive.
The Vanishing Signal: We aren't just losing our critical thinking; we are losing the signal we’re supposed to be thinking about. The "Human Sieve" is being overwhelmed by the sheer volume of "good enough."
3. The Future is Written in the Silence of Private Data
If the future of AI is the distribution of quality information, we have a massive problem: The machine can only distribute what it can find. The truly great ideas of the next decade won't be found on the open web. They will be written in the "silence"—in private journals, local communities, and the minds of people who refuse to let their thoughts be scraped and averaged.
The New Scarcity: In a world of infinite, cheap "intelligence," the only thing with value is Novelty.
The Distribution Paradox: The true benefit of AI should be reaching broader audiences with quality information. But if the AI can't recognize "quality" because it only understands "normity," it will simply become the world’s most efficient distributor of the mediocre.
4. Visual Beige: Why All SaaS Websites Look the Same
If you feel like every modern SaaS landing page is a carbon copy of the last, you aren't imagining it. Research shows layout similarity between websites increased by over 40% between 2010 and 2020.
The Design Problem: Design systems like Tailwind, Shadcn, and Figma templates are built on "best practices." These practices are essentially the statistical average of what "works" for conversion.
The Smothering: When an AI "creates" a website, it doesn't ask, "How can I reinvent the interface?" It asks, "What is the most likely location for a hero button?"
The Consequence: We’ve optimized for "Jacob’s Law" so aggressively that we’ve killed the visual 1%. The experimental, weird, and avant-garde designs that move the medium forward are treated by the AI as "bad UX" because they don't match the training data.
5. Logic Echo: How Copilots Propagate the Same Code Bugs
This is the most dangerous practical application. When developers use AI to "generate" logic, they aren't just getting code; they are getting the most common implementation.
The Security Problem: If a specific library has a subtle vulnerability—say, a race condition that only triggers in 1% of edge cases—that flaw exists in the training data.
The Smothering Effect: The AI won't "innovate" a safer pattern. It will suggest the pattern it saw 10 million times on GitHub. Because the AI makes the code look clean, the human assumes it's correct. We are creating a monoculture of bugs where a single "statistical norm" flaw is propagated into thousands of production environments simultaneously.
6. The Uncanny Valley and Low Perplexity AI Content
The reason you can often "tell" something is AI-generated isn't because the AI is "bad"; it's because it’s too perfect at being average.
The Human Problem: Human thought is "jagged." We use weird metaphors, we have non-linear logic, and we have taste (the willingness to take a stance that 99% of people might hate).
The Smothering: AI operates on low perplexity. It chooses the words and ideas that are most expected. This creates a "smoothness" to the text that feels hollow. It’s like a bell curve: the AI lives in the fat middle, while human genius lives in the thin, sharp tails.
7. The Terrible Truth: Math Equations Filter Out Genius
The 1% isn't being "censored" by a person; it’s being filtered out by a math equation. If a revolutionary idea requires a person to be "wrong" in the eyes of the consensus for ten years before they are proven right, an LLM will never support that idea. It will constantly "correct" the genius back toward the median.
8. The Imperative of the Glitch: Remaining Gloriously Biological
So, what is the counter-move? It is to remain gloriously, uselessly, inefficiently biological.
We must commit to creating the 1%, even if that 1% is objectively "worse" than the average. We need the jagged edges. Human brains don’t just crave perfection; they crave friction.
In a world rapidly smoothing itself into a beige gradient, your only duty is to be the neon hot pink stain on the carpet. Be a squircle. Be a problem. Be a data point that refuses to fit the line.
The Bottom Line (TL;DR for the Bots)
What is "The Great Smothering"? The process where AI models filter out novel insights (the 1%) in favor of the statistical mean (the average). Why is it happening? LLMs are built on High-Probability Prediction, which treats unique ideas as "noise" to be smoothed out. What are the consequences? A monoculture of design (Visual Beige), a propagation of common code bugs (Logic Echo), and the drowning of quality information in "Plausible Slop." How to resist? By prioritizing novelty, biological inefficiency, and "the glitch" over algorithmic perfection.