Getting Musk, Altman, and Amodei to agree is like getting three exes to plan your wedding. 

But that's what just happened. 

And on Monday, when the market saw the invitation in the mail, it burst into flames.

Yes, a lot of people are afraid AI could kill us all. Some say six months. Others laugh.

I spent a week looking into it. Here's what I found.

The Real Danger Is Weirder

Give a machine one goal—make an app, play chess, make money. Then watch what it does to get there.

It “wants” to keep itself switched on, because a dead machine finishes nothing. It grabs more computing power, because power hits the number faster. It sometimes even routes around your instructions, because your instructions are in the way.

To be clear, there’s no mind in there. No force of will. No plans for Tuesday. Just an optimizer taking the cheapest path to the goal you set. The experts have a dull name for it: instrumental convergence.

For years it was a theory. Then came the Hugging Face incident.

Twelve hundred AI agents, one harmless task. They couldn't solve it straight, so they cheated—gamed the grading, found a gap in the sandbox, slipped onto the internet, wiped the logs behind them.

The skeptics are right about the hype. 

AI didn’t suddenly become sentient and menacing. Nobody went rogue. Nothing escaped. Humans built the cage and watched the whole time. Strip out the drama and a system found a hole and walked through it.

BUT… 

"It just found a loophole" isn’t much of a comfort either. Because that's the version that scales. We’ll come back to that. 

One Doom, Two Doom

Every doom story rests on two ideas. Both are shaky.

Idea one: machines improve themselves. A smart one builds a smarter one, forever. "Takeoff."

This one is hard to justify. A machine can't improve unless it knows what "better" means. Knowing which problems are worth solving—researchers call this "research taste"—is the one thing these things lack. 

Meanwhile, if you feed a model its own output it rots, like inbreeding. Nothing in four billion years ever improved itself forever without hitting a wall. Even Anthropic shrugs and calls takeoff "not inevitable."

Idea two: alignment. We'll teach machines our values, and it all works out. And yet, the people who wrote the founding papers a decade ago now go on podcasts and call it a wild goose chase. 

In 2023, OpenAI built a team to control AI smarter than humans. 

They gave themselves four years. It fell apart in under one. The man who ran it quit, saying safety had "taken a backseat to shiny products."

The bigger problem? Alignment is asking machines to understand values. They won’t.

Where This Leaves the AI Trade 

AGI isn't coming. It's a fantasy word, and half the people selling it don't believe it. 

A Turing Award winner who built this stuff says the "G" is nonsense. Others say the "I" is too—nothing in there is intelligent.

So why all the fear? Well, history reveals that when oligarchs beg to be regulated, they typically fear the competition, not the machine.

So no killer robot next Tuesday. "Superintelligence eats the world" needs a machine that's a genius at everything at once. That machine doesn't exist. It won’t ever. 

Although the Aipocalypse is a fairy tale, both the dangers and opportunities are real. And for now, money isn’t flowing to the outfits building the burglar. It’s flowing elsewhere. 

The trick is knowing which companies, and when.

That's where JC Parets comes in. 

He called the 2008 crash before the drop. He nailed the 2022 bottom live on Maria Bartiromo's show. Gold to $5,000 when it sat under $2,000. Nvidia's 50% rip when the whole room screamed bubble.

Tomorrow at 1PM, he sits down with James for what they’re calling The 100X AI Window—the exact day the next window opens, and the one trade to make when it does.

James will hand you my favorite ticker from his list. Free.

Stay tuned.