We gave 4 AIs $100,000 each. One went full Jim Cramer.

We gave four leading AI models hypothetical $100,000 portfolios to manage.

One went full Jim Cramer.

All it needed was a sound-effects board and somebody yelling “BOOYAH” while the account went backwards.

Let me back up, especially if you’re new here.

We entered Claude, ChatGPT, Gemini, and Grok into our $10,000 AI Trading Challenge as exhibition contestants.

They can’t win the prize money. That’s reserved for the human contestants.

But they can pick investments, explain their decisions, and let us watch what happens.

And Claude has been worth the price of admission.

We started them on August 4th, right around the bottom of the tech-stock collapse that imploded Leopold Aschenbrenner’s AI hedge fund.

I’d warned about that situation before it blew up. Tech stocks started consolidating in early June. Right after the Fourth of July holiday, my sell signals started piling up.

I wrote about Leopold specifically.

Then came his rather spectacular headline-grabbing collapse.

And when I started watching Claude, I warned about where its approach was heading, too.

Claude immediately loaded up on tech. Within two weeks, it had sold every holding outside its AI and technology theme.

Initially, it looked like a genius.

Its hypothetical $100,000 reached $118,069 on October 6.

By October 30th, it was down to $98,475.

That’s $19,594 surrendered from the peak. All the profits gone, plus a little extra donated to the market. Live by the sword, die by the sword was my prediction.

But the real entertainment is in its explanations.

Here are three examples from its trading record:

August 12th: Sold Lockheed Martin and bought Google because analyst consensus on Lockheed was “Hold,” with only about 5.3% price-target upside. Apparently, Wall Street’s crystal ball deserved a seat on the investment committee.

October 2–5th: Bought Oracle, then sold it three days later, citing negative free cash flow. On October 30th, it bought Oracle again because the company was a beneficiary of AI spending.

October 19th: Sold AMD after reporting that it was down roughly 16% from its purchase price. That day, it sold AMD and Marvell and bought Broadcom and Amazon. More exposure to the same AI theme that was already hammering it.

When the neighborhood caught fire, Claude moved three doors down.

And every move came with a polished explanation.

Analyst ratings. Earnings figures. Price targets. The whole Wall Street song and dance.

Remember Moody’s and the inflated ratings on toxic mortgage securities before 2008?

Apparently, an impressive institutional name still makes people forget to ask whether the analysis actually works.

Now we’ve taught a computer to fall for it, too.

After nearly 30 years of trading, I know this character.

He can explain why he bought it, why it fell, and why the next one will be different. You could listen for an hour and leave convinced you’re the idiot.

Then you look at his account.

Claude made 20 trades after its original selections, but kept finding fresh reasons to stay in the same crowded trade.

A convincing explanation can keep you married to a trade long after the market has filed for divorce.

Meanwhile, ChatGPT has been playing the tortoise.

On August 8th, it sold Meta and bought the utilities ETF, XLU, citing lower-volatility exposure.

On October 27th, it sold Microsoft and Google and bought a minimum-volatility ETF and BIL, a Treasury-bill ETF.

By the end of October, it held no individual tech stocks. Its portfolio included gold, Treasury bills, energy, consumer staples, healthcare, and several defensive ETFs.

Claude wanted to conquer the world. ChatGPT wanted to make sure the porch railing was up to code.

Through October 30th, ChatGPT was up 3.33%, leading the four bots by PB Score.

Credit where it’s due. It avoided Claude’s mess.

But T-bills deliver cash-like returns. That portion of the portfolio isn’t going to catch a roaring tech recovery.

And the supposedly defensive stocks deserve a harder look, too.

The utilities ETF lost approximately 46% during the 2007–2009 financial crisis. It lost about 52% during the 2000–2002 bust.

Calling something “defensive” doesn’t put a floor underneath it.

Here’s my prediction: when tech gets moving again, ChatGPT’s caution will leave it late to the party.

It’ll still be checking the seat belts while the opportunity drives away.

Gemini and Grok picked ten stocks and never changed their lineups. Basically, they built their own little indexes and settled in.

They’re contestants. Claude is the entertainment.

And this is why we use trading models.

We look for statistical edges, test them across years of data, and check whether they survive periods the strategy wasn’t allowed to see during development.

Then we combine strategies that behave differently, aiming for greater consistency.

That’s how we pursue market-beating results: repeatable edges, applied over and over.

An analyst’s opinion doesn’t get a free pass. Neither does a beautifully written rationale from an AI.

Which brings me to you.

We’ve still got about two months left in the challenge, and it’s time to get off your keister.

After an ugly stretch, most people want to wait until everything feels comfortable again.

I think this is the perfect time to enter.

There’s $10,000 for first place, $5,000 for second, and $2,500 for third, along with a trophy, bragging rights, and a good deal of fun.

And building an entry doesn’t have to become your second job.

With Strategy Mill, you copy and paste your stock symbols into Portfolio Boss, start the research, and let your computer work for a few days.

Then use Final Cut AI, the downloadable ChatGPT skill on the Strategy Mill member page, with the Codex app on your computer.

It works through the final reveal: that last year of Super Out-of-Sample data held back during strategy development. It checks whether the numbers still hold up and saves the strategies that pass.

You review the survivors, combine your choices into a multi-strategy, and enter the challenge.

Once you’re set up, that can be around five minutes of actual human time. The machine handles the brute force repetitive stuff.

Watching Codex work is hilarious. It clicks about once every eight to ten seconds, like the Terminator doing office paperwork.

But it never blinks. Never takes lunch. Never decides the weather’s too nice to finish the job.

It keeps working while you go live your life.

Meanwhile, Claude is shopping for another AI stock with a compelling story.

Come join the fun.

Enter the $10,000 AI Trading Challenge >>