Founder, R360 · "Street Sense," MarketWatch
Room 1 of 5: Prediction
Stop.
Before you read another word, I need you to understand something.
What you are about to do is not "learn about AI."
You are about to touch it.
Not a chatbot. Not a demo. The actual mechanism, the thing underneath the magic trick that every tech CEO hopes you never look at too closely.
Room 1: Prediction
Here is the entire secret of artificial intelligence:
It guesses the next word.
That's it. That is the whole trick. A four-hundred-billion-dollar industry, built on a machine that does one thing: it reads what came before, and it guesses what comes next.
Don't believe me. Try it yourself. See that sentence below? Type something. Anything. Change it to whatever you want. Watch the bars move.
Those bars are the model's mind, laid bare. Each bar is a word it might say next, and the height tells you how confident it is. The model isn't thinking. It's betting.
Now see that dial? That's called "temperature." Turn it left, and one word takes over. The model becomes certain. Predictable. Boring. Correct. This is your bank's chatbot.
Turn it right. The bars flatten. Every word has a chance. The model becomes creative, which is the industry's polite word for "reckless." This is the AI that writes your kid's college essay in six seconds.
Here's what should bother you: the mechanism is identical either way. The model doesn't become smarter when it's confident or dumber when it's creative. It's running the same math.
So the model predicts the next word. But how does it know which words matter?
Room 2: Attention
Read this sentence: "The bank by the river was steep."
What kind of bank? You knew instantly. A riverbank.
But a machine doesn't "know" anything. It can't see a river. It's never been outside. So how does it figure out that "bank" means dirt and rocks, not money and vaults?
It cheats. It draws invisible lines between words, connecting them with weights that tell it which words to pay attention to. "Bank" is strongly connected to "river." So "bank" means the geographic kind.
You're about to prove this by breaking it. See those lines connecting the words? Grab the one between "bank" and "river." Drag it down.
Watch the predictions change. "Eroded" and "muddy" vanish. "Fees" and "interest" appear. You just turned a riverbank into a financial institution, by severing a single connection.
The model doesn't understand the word "bank." It understands which other words "bank" is connected to, right now, in this sentence. Change the connections, change the meaning.
This is attention. This is how ChatGPT "understands" language.
It doesn't. It draws a map. A brilliant map gives you brilliant answers. A bad map gives you confidently wrong ones. And the model has no idea which kind of map it drew.
Now you can see what the model pays attention to. But where did it learn to draw these maps?
Room 3: Training
Nobody taught it. Nobody could teach it, because nobody sat down and wrote rules for every word in the English language.
Instead, they did something faster and crazier.
They fed it the internet. Billions of pages. News. Medical journals. Legal briefs. Reddit threads. Romance novels. Earnings calls. Everything ever written and posted and forgotten.
The toggles below represent categories of training data. They're all on right now. Start turning them off.
Turn off everything except "Legal filings." Read the response. The model sounds like a lawyer. Not a bad lawyer. A convincing lawyer. It uses words like "pursuant" and "fiduciary obligations."
Now turn off everything except "Social media." The model sounds like your nephew's Twitter feed. Emoji. Slang. Opinions delivered with the confidence of someone who has never been wrong because they've never been specific.
The model became its training data. Completely. When you changed the training data, you changed the model's entire personality.
So training data shapes the model. What happens when you make it bigger?
Room 4: Scale
Not better. Not faster. Bigger. More parameters. More connections.
The answer should be simple: it gets a little better. Like making an engine bigger makes a car a little faster. Linear. Predictable. Boring.
That is not what happens.
Grab the scrubber below. Drag it left: that's a tiny model, one million parameters. It can complete a sentence. That's about it. Now drag it right. Slowly.
Did you see it? "Reason about a problem" didn't fade in gradually. It popped. One moment the model couldn't reason. The next moment it could. Nobody added a "reasoning module." They just made it bigger, and reasoning appeared.
This is a staircase, not a ramp.
At the far right, the last line lights up in amber: "Do something unexpected." At sufficient scale, the model does things its creators did not train it to do and cannot fully explain. This is called an "emergent capability." It is not marketing. It is why the people building these systems are simultaneously excited and terrified.
So bigger is better. Until it isn't.
Room 5: Limits
Last room. This is the one that matters most.
Everything you've seen so far tells you what the model can do. This room tells you how it fails.
And it fails in the most dangerous way possible.
It fails silently.
Look at the widget below. Four constraints are already loaded: four rules the model has been told to follow. "Be factual." "Be creative." "Be concise." "Cite sources."
Now look at the failure indicator. One constraint has already been quietly dropped. "Cite sources": the model stopped doing it. It didn't tell you. It didn't flag an error. It just stopped.
And the output still reads fine.
Try removing a constraint. Take one off. Watch the failure indicator go green: the model can handle fewer rules without dropping any.
Now add a fifth. "Match tone." Watch what happens.
Two constraints dropped. The output still sounds fluent. Still reads like it was written by a professional. The kind of output that would sail through a boardroom review. But two of the rules you explicitly gave it have been silently abandoned.
Add a sixth. A seventh. Watch the gap between what the model appears to do and what it actually does widen with every constraint you add.
The people who understand this will use AI brilliantly.
The people who don't will use it confidently.
Which is worse.
You now understand more about how AI actually works than 95% of the people making decisions about it.
Not because you memorized jargon. Because you touched the mechanism. You turned the dials. You broke the connections. You watched the model become its training data and fail in silence.
That understanding doesn't expire.
The technology will change. The companies will change. The headlines will change. But the five things you just learned: prediction, attention, training, scale, limits. Those are the bones. Everything else is skin.
Charlie Garcia writes about what all of this means for your money, your companies, and your family. Every week. Specific. No jargon. No hype.
The kind of analysis that has generated seven-figure returns for the people who read it first.
If the last twenty minutes were worth your time, the next twenty weeks will be worth considerably more.
Charlie Garcia has advised six U.S. presidents of both parties.
He founded R360 and serves as its Chief AI Officer. Nearly 200 families, average net worth $600 million, 80% first-generation entrepreneurs who built what they have themselves.
He writes Street Sense for MarketWatch and Capital Mischief, a top 15 finance publication among more than 10,000 on Substack, read in 138 countries. He is editor in chief of The Night Owl, R360's private briefing on global markets. He manages no one's money and sells no products.
He writes about what this technology does to capital, to companies, and to families.