Two things to put on your radar: 

1. Frontier AI labs have a big problem. It’s only going to get worse. And…
   
   
2. Investors should pay close attention. Most won’t. And they’ll learn too late. 

The problem? All AI models train on the same internet. Same threads. Same answers. Same slop. And then they regurgitate it back onto the internet. 

Now the internet is filling up with machine output, and the machines are training on that. 

Here's the dirty little secret… 

Models learn by counting what shows up most often. Common things get reinforced. Rare things show up only a handful of times, so the model eventually forgets them.

Researchers at Oxford and Cambridge proved it in Nature in 2024. They trained each generation of a model on text written by the one before it. 

Nine generations in, the models collapsed. A prompt about medieval church towers produced gibberish—a looping list of jackrabbits with different colored tails.

Here’s how it’s already hurting investors using AI to hunt for the next edge… 

The Tails Vanish First 

You ask a chatbot about a stock. It gives you the consensus, scraped from what everyone's already written. Then you read the forums, which quote the chatbots. Then the news, which now runs AI summaries of the forums. 

Every source is a copy of the last. You think you're gathering information. You're circling one answer, drifting further from the company with every click.

The tails vanish first. 

The odd fact, the minority view, the one angle that says a stock is mispriced—averaged out of existence. Soon enough, you’re talking about Nvidia and the AI starts talking about jackrabbits. 

The obvious response is to out-think the machines. Good luck. The better response? Feed on what they can't. 

This problem will inevitably change how everyone invests. Those who look for edges now will win the biggest.

Three ways to do it.

One: Look Where the Machine Can't

The machine only knows what's been written down. This makes observing the real world 100x more important. 

Philip Fisher called it scuttlebutt, based on the barrel of drinking water on old ships where sailors gathered to trade rumors—the talk you pick up by asking around.

Peter Lynch found one of his best winners because his wife liked a product she bought at the grocery checkout. The tried-and-true is even more valuable now. 

The method: Leverage the problem itself. Ask AI what everyone believes. It hands you the consensus in seconds. That used to be the hard part. Now it’s free. 

Then go check that one assumption in the physical world. You're not gathering blind anymore. You're aiming a single real-world look at the exact crack in the consensus.

Two: Buy What AI Makes Worse 

What is AI making worse as it grows bigger? There are a lot of them. Here are three. 

Cybersecurity. AI makes attacks cheaper and faster, so defense becomes non-optional spend. This is a direct, growing bill. We’re still early. 

Specialist data. The next frontier is specialist data because that’s what’s getting cancelled out. Labs are paying doctors, lawyers, physicists, and coders to produce fresh reasoning—not answers, but the messy work of getting to an answer. Businesses are springing up to broker that: connect a lab that needs 10,000 hours of radiologist reasoning to the radiologists. The bleeding-edge players here are private, but it’s a real market, moving fast. 

Deepfake financial fraud defense. Fake-CEO voice calls authorizing wire transfers are happening now, today, with real losses. Every company that moves money will need verification. This is early, real, and underappreciated. 

Three: Buy What the Giants Can't

A $100 billion fund can't touch a $300 million company. Its own buying would spike the price, so it skips the company no matter how good it is. Index funds must sell whatever drops out of the index. 

For decades this didn't help you—researching a tiny company took more time than it was worth. 

This is where research is actually most valuable. Because although the research cost collapsed for everyone, the size wall didn't move.

So the newest tool on Wall Street hands its biggest gift to the smallest investor. That almost never happens.

All three are the same bet—find the ground the giants and the algorithms have abandoned. The real world they can't observe, the messes they create, and the small stocks they're too big to own.