“Other than that, we try not to really mess with the portfolios,” Clinton told CoinDesk. “As a human, I’ll sometimes look at the portfolios and think ‘Oh, this pick seems like a terrible idea.’ Other times I’ll see something really interesting and try to understand the logic. It’s kind of fun.”

The process involves the three AIs explaining their reasoning to Clinton. Not only does it help him ascertain that the investments are aligned with the portfolio’s goals, but he says that models provide better portfolios when they’re forced to explain why they like specific stocks.

It often happens for the AIs to disagree. And their way of thinking changes as updates get rolled out. “It used to be the case that Claude was the most contrarian model in terms of the outputs, when we first started testing,” Clinton said. “Now I would say it’s ChatGPT.” And while Clinton has tested other AIs such as Grok or Lama AI, keeping the investment committee down to three AIs has proved to be the most efficient set-up.

Investors can gain exposure to Intelligent Alpha’s strategy through an exchange-traded fund, the Intelligent Livermore ETF, which launched in September and uses AI to build a global equity portfolio. More such funds are on the way, Clinton said.

For the Livermore ETF, every financial quarter the models review world events and try to make predictions for the next three to six months. Five or six areas of opportunities are then identified (following the investment philosophies of the greats like Druckenmiller) and the portfolio gets built around these sectors.

Having competing philosophies means the portfolio usually ends up being quite balanced. “In many cases they're looking at idiosyncratic opportunities,” Clinton said. “We haven't seen big issues where [the investment philosophies] are at odds, but even then, it would be like hedging.” The AIs themselves make the decisions on how to weigh the various philosophies found in the portfolio, depending on the areas they’re the most confident in.

“AI has been, at least so far, really good at seeing forward,” Clinton said. “Right before we launched, it made a big bet on Asian stocks, specifically Chinese stocks, and that was right before [billionaire hedge fund manager] David Tepper went on CNBC in September and said that China was his biggest bet, that they were bringing out the bazooka for stimulus. And you know, Chinese stocks went crazy.”

Another memorable trade: chipmaker giant Nvidia has been AI’s top pick since the experiment began in summer of 2023. “Back then, I was like, ‘Oh, my God.’ Nvidia had run so much at that point,” Clinton said. “But it's up now like 400% from the moment the AI picked it.” The lesson in there, he says, is humans will react to charts emotionally, whereas AI “just doesn't care. It says ‘No, this is going to go higher.’”

Not that every bet has been a slam dunk, but so far, the mistakes have been on the margin, according to Clinton. The AI is building a good track record on macro events especially, he said. For one thing, it predicted that former President Donald Trump would be re-elected.

One of the reasons Intelligent Alpha doesn’t focus too much on crypto is simply lack of data. Their trades may have happened on-chain, but there’s no easy way to go back and find the kind of trading setups and investment philosophies used by famous crypto investors like Cobie or GCR. Most of the time, all you can do is go off of their posts on X — and it’s hard to know whether the posts reflect reality.

That being said, the crypto community’s reliance on X means that Grok could end up playing a role in Intelligent Alpha’s triumvirate someday for crypto purposes, Clinton mused, since that model is trained and fine-tuned with data from the social media platform.

“The question that we're exploring here is, what can we do with AI that would maybe be unique and different and stand out a little bit,” Clinton said. “To find a unique way to use AI to identify breakout crypto projects, that would be a really cool way to use the tech.”