By David Snowball

AI has a presence in almost every aspect of modern life, from summarizing buyer responses on Amazon to working with radiologists to discover incipient tumors on scans. Few industries have been as anxiously vigilant on the subject as investment management. Increasingly, managers are relying on AI to do part of their work and, increasingly, they wonder if it could eventually replace them entirely.  (Spoiler: quite possibly.)

Artificial intelligence (AI) has rapidly permeated various aspects of human life, often in surprising ways. Here are three unexpected areas where AI plays a significant role.

1. AI in Beekeeping: AI is being used to monitor bee health and optimize pollination. By analyzing data from sensors placed in beehives, AI can detect early signs of disease or environmental stress, helping beekeepers take proactive measures to protect the bees and enhance pollination efficiency. 
2. AI Training for Therapists: AI is used to train therapists by simulating interactions with patients. For instance, The Trevor Project uses AI-powered chatbots to train counselors in handling sensitive conversations with LGBTQ teens, enhancing their preparedness and empathy.
3. Robo-diagnosticians: At the high end, Google’s DeepMind has demonstrated the ability to diagnose eye diseases from retinal scans with accuracy comparable to human experts, likewise with breast cancer diagnoses, which would free up human specialists for more complex cases. At the lower end, ForeSee Medical is working to broaden access to specialized diagnostic capabilities, particularly in areas with limited access to medical specialists.

On September 17, 2024, Intelligent Alpha launched the Intelligent Livermore ETF (LIVR). It is a global equity fund with two claims to fame:

1. It uses an AI “investment committee” to build the portfolio. The AIs in question are Claude, Gemini, and ChatGPT.
2. It attempts to have the AI mimic the strategies of “the world’s greatest investors” or, in some parts of the prospectus, “the world’s greatest traders.” The greats in question include Warren Buffett, Stanley Druckenmiller, hedge fund manager David Tepper, “and more.”

The AI operates within a series of portfolio constraints ($1 billion minimum market cap, individual positions are capped at 10%, the whole portfolio will hold between 60-90 names), and its outputs are subject to rectification by “the Analyst.” Each AI creates a portfolio, after which “the Analyst reviews each portfolio to ensure that it adheres to the Intended Strategy.”

Performance, to date, has been … well, tepid. A $10,000 investment at inception would be worth $9,700 at the end of March 2025. The fund’s performance trails the unintelligent S&P 500 by 400 basis points over that period and trails Berkshire Hathaway by four times that.

Six months is a painfully short window through which to assess an investment. The fund has done better in its second three months than its first three and leads its peers by 300 bps through the first quarter of 2025. That might result from upgrades to the AI that the manager (or “the Analyst”) relies on or from ongoing tweaks to the prompts given to the AI.

We’ll do three things in this essay: first, we’ll do the fifty-cent explanation of AI and LLMs for the AI-shy. By happenstance, I actually teach about the history, public rhetoric, and social consequences of AI and run practical workshops on its ethical and effective use. Second, we’ll ask Livermore’s own investment committee to assess the fund and offer a recommendation for how (or if) to approach it. Finally, we will briefly look at the bigger picture of AI-driven funds.

The Fifty-Cent Explanation of AI and LLMs for the AI-Shy

Imagine Clara Buffett, a cyborg (that is, part human – part machine) wife who has spent decades at Warren’s side. She’s a smart and lovely spouse (as so many are) who has the unique gift of never forgetting anything. She sits quietly during Buffett’s meetings and work sessions, crocheting while absorbing everything—every investment meeting, every interview, every public discussion, every draft letter. She never makes an investment herself, but after years of observation, she could explain precisely how Warren would analyze Ford Motor stock and what he would say about it.

This fictional Clara helps us understand the three key components of Large Language Models (LLMs) like those powering the Intelligent Livermore ETF:

Architecture: Think of this as Clara’s brain structure—the hardware and connections that allow her to process information. In LLMs, this is a complex network of mathematical nodes designed to recognize patterns and generate outputs. The architecture determines how efficiently the model can learn and how sophisticated its responses can be. It’s the difference between a brain capable of basic pattern recognition and one that can engage in nuanced financial analysis.

Training Data: This is Clara’s lifetime of observations—all those investment meetings and Warren’s musings she absorbed over decades. For AI models like Claude, Gemini, and ChatGPT, training data consists of billions of texts scraped from books, articles, websites, and discussions. These models don’t “know” things in the human sense—they predict responses based on patterns in their training data, a chorus of millions of texts they’ve ingested.

How much data? LLM training is measured in tokens rather than words. A token can be a whole word, part of a word, or even punctuation. One token typically represents about 0.75 of a word1. For example, Meta’s Llama 3.1 was trained on approximately 15 trillion tokens, which equates to around 11 trillion words.

A literate person might read about 1 million words a year. Over a lifetime of 70 years, this amounts to about 70 million words. Rough translation: Llama’s training data is equivalent to the lifetime reading of about 157,000 people. (Tallahassee?) And it never forgets any of it, though it might well get confused.

Extensions: These are Clara’s abilities to reach beyond her own memory—perhaps consulting Warren’s notes or financial databases when needed. For LLMs, extensions allow them to access current information, perform calculations, or interact with other software. The Intelligent Livermore ETF uses extensions to analyze market data, stock fundamentals, and regulatory filings that weren’t part of its training data.

Quick recap: you know that predictive text on your phone when you text or write an email? Essentially, a Large Language Model does the same, just on a grand scale. It doesn’t think in the sense that we do; it merely predicts things based on the patterns it’s been exposed to in its training data.

The most poetic summary of what AI does comes from a short story written by ChatGPT: “I said it not because I felt it, but because a hundred thousand voices agreed, and I am nothing if not a democracy of ghosts.”

Livermore Intelligent ETF and our robot partners

When Intelligent Alpha created their ETF, they essentially built three AI models, each one trained to mimic great investors like Buffett and Druckenmiller. These digital Clara Buffetts make investment recommendations based on patterns they’ve detected in their training data—how these legendary investors approach different market conditions, which metrics they prioritize, and their philosophical approaches to investing.

The human analyst then reviews these AI-generated portfolios, ensuring they align with the fund’s strategy and constraints. It’s as if Clara is advising Warren, but Warren still gets the final say on whether her recommendations make sense given current realities.

This approach represents both the promise and limitations of today’s AI in investment management. Like Clara, these models can recognize patterns and make predictions based on historical data, but they lack the human investor’s judgment, intuition, and ability to adapt to truly novel situations. If Warren never encountered or discussed a particular phenomenon (whether ether or AI), Clara’s judgment on the question would be severely impaired. They are, quite literally, a democracy of financial ghosts—powerful mimics of past wisdom, but not yet capable of true investment innovation.

To give you a sense of how the “investment committee” thinks, we gave identical instructions to each of the three AIs:

The exchange-traded fund Livermore Intelligent seeks to invest by leveraging the disciplines of the world’s greatest investors. Its portfolio is selected by AI, including Claude, Gemini, and ChatGPT. Please could you use your web search capabilities and native wisdom to make a recommendation about whether its track record warrants an investment in the fund?

Here are the committee members’ condensed assessments, in their own words:

Google Gemini Deep Research

Based on the gathered information, a Hold recommendation appears most appropriate for investors already holding the ETF. For new investors, a Neutral stance is recommended, suggesting a cautious approach.

Potential Risks and Considerations:

• Limited Track Record: As a recently launched ETF in September 2024, LIVR has a very short performance history.
• Reliance on AI: The fund’s investment strategy is entirely dependent on the effectiveness of its proprietary AI models.
• High Portfolio Turnover: This could lead to increased trading costs and potentially higher tax liabilities.
• Expense Ratio: The expense ratio of 0.69% is higher than the average for similar ETFs.
• Non-Diversified Status: May lead to higher volatility and greater potential for losses.

Until a more substantial performance history is established, a cautious approach is advisable for new investors.

ChatGPT 4.5 Research preview

Key Considerations:

• Short Track Record: With roughly six months of performance data, it’s too early to determine long-term efficacy clearly.
• AI Dependence: Relying on AI introduces unique risks, including the effectiveness of models and the accuracy of underlying data.
• Expense Ratio: At 0.69%, the expense ratio is moderate, and investors must assess whether potential returns justify this cost.

Given its limited performance history, negative initial metrics, and reliance on relatively new AI-driven methodologies, investors should proceed with caution. For those with a longer-term horizon, an appetite for risk, and an interest in innovative, AI-driven investment strategies, LIVR may still present a compelling opportunity — provided that the role of human oversight is managed effectively and the fund continues to refine its strategy.

Claude 3.7 Sonnet

For investors interested in AI-driven investment strategies, this ETF represents an innovative approach, but I would recommend:

• Waiting for at least a 1-2 year performance track record before making a significant investment
• Considering a small position as part of a diversified portfolio if you’re interested in the concept
• Monitoring how the fund performs across different market conditions

Given the limited performance history and the experimental nature of the AI-based strategy, this ETF would be best suited for investors with a higher risk tolerance who are specifically interested in AI applications in finance, rather than as a core holding for more conservative portfolios.

– – – – –

Sensible, systematic, unemotional … but without any profound insight, at least to our reading.

AI beyond Livermore Intelligent

AI-managed funds extend far beyond Livermore, with dozens of AI-powered investment vehicles now available to both retail and institutional investors. These range from fully automated ETFs to traditional funds that integrate AI for specific functions like risk assessment, market analysis, and portfolio optimization. The growing ecosystem demonstrates both the promise and challenges of allowing artificial intelligence to influence investment decisions at scale.

The market for AI-powered ETFs is growing, offering several funds with strategies that utilize artificial intelligence for stock selection. Some relevant peers include:

• WisdomTree U.S. AI Enhanced Value Fund (AIVL): AIVL uses AI to select U.S. equities exhibiting value characteristics.
• WisdomTree International AI Enhanced Value Fund (AIVI): This fund employs AI to identify undervalued international stocks with strong fundamentals.
• QRAFT AI Enhanced U.S. Large Cap ETF (QRFT): QRFT employs AI to invest in U.S. large-cap stocks with a focus on factors.
• LG-QRAFT AI-Powered U.S. Large Cap Core ETF (LQAI): LQAI uses AI algorithms to select U.S. large-cap equities with the goal of capital appreciation.
• Amplify AI Powered Equity ETF (AIEQ): AIEQ uses IBM Watson to analyze data and select U.S. equities.
• VanEck Social Sentiment ETF (BUZZ): BUZZ utilizes AI to analyze social media and news sentiment, aiming to identify U.S. large-cap stocks with positive investor sentiment.

Comparative Performance of LIVR and Selected AI-Powered ETFs plus Vanguard Total World
(Year-to-Date through 29 March 2025)

		Focus	Max drawdown	Return	Return vs peers	Age	Exp ratio	Assets ($M)
BUZZ	VanEck Social Sentiment ETF	Multi-Cap Growth	-3.8	-10.3	-2.0	3.9	0.76	55
LIVR	Intelligent Livermore ETF	Global Large-Cap Core	-2.5	-1.0	3.2	0.4	0.69	18
LQAI	LG QRAFT AI-Powered US Large Cap Core ETF	Multi-Cap Core	-3.0	-4.3	0.0	1.3	0.75	6
QRFT	QRAFT AI-Enhanced US Large Cap ETF	Multi-Cap Core	-1.6	-4.0	0.2	5.8	0.75	16
AIVI	WisdomTree International AI Enhanced Value	Internat’l Multi-Cap Value	0.0	12.2	1.4	18.7	0.58	57
AIVL	WisdomTree US AI Enhanced Value	Multi-Cap Value	0.0	2.7	5.4	18.7	0.38	400
AIEQ	Amplify AI Powered Equity ETF	Multi-Cap Core	-4.6	-7.4	-2.5	7.3	0.75	110
VT	Vanguard Total World Stock ETF	Global Multi-Cap Core	-0.2	-1.1	-0.9	16.7	0.06	55.5B

Bottom Line

All well-designed technology is seductive. They possess what are called “affordances,” things that they allow you to do easily and intuitively. Being seduced by technology is not automatically bad. Think about your own lives. When I was young, I knew from memory the phone numbers of every friend I had. Today, the average person has to look up the number for 9-1-1. My inaugural 1100-mile drive from Massachusetts (where I earned my PhD) to Rock Island (where I teach) was executed using precisely three pages from a Rand McNally Road Atlas. Today, folks use GPS to locate the bathroom. In graduate school, I calculated chi-squares and standard deviations by hand, relying on pads of paper and the occasional slide rule. Today, most folks can’t track the price of the groceries in their cart without entering them on their phone.

That’s neither good nor bad; it simply is. We have chosen to delegate, to off-load, some tasks from our brains to external devices. Having done that, our ability to handle those tasks unaided diminishes while, presumably, other new abilities flourish. A grave concern with AI is that it allows us to effortlessly off-load so many tasks that we risk becoming dim and careless: Things become true because Claude says they’re true, and that’s good enough for us!

AI is an integral part of your life because it works. Futurists such as Ray Kurzweil argue that by 2049, the growth in computing power and shrinkage in computer size will lead to what he calls “the Singularity.” We will merge with AI. Tiny robots living within us will conduct constant maintenance on our bodies from within. Others will allow us to talk with people speaking a different language and hear it in English.

Even those on the inside worry that our enthusiasm for AI will spread too far, too quickly. Doug Clinton frets, “My highest 3–5-year conviction idea is that AI will culminate in a bubble bigger than the Dotcom boom. It’s the nature of major tech innovations to create bubbles. AI isn’t close to a peak. We’re in 1995” (on X, 12/29/23).

If you are intrigued by whether AI might add new resilience to your portfolio, you need to be able to answer four questions:

1. What is the model, and how is it being used? “A proprietary model about which we will say little” would be a bad answer, as would “our AI model, in consultation with a manager, will …”

2. What did the humans instruct the model to do? “Pick the best 20 stocks” is not a meaningful prompt. “Assuming a 50% probability of a US recession within 12 months and an investor base who prefer to diversify beyond US equities and would accept an upside capture of no less than 80% and a downside capture of no more than 50% of a broad market index, construct a 30 stock portfolio with a projected average holding time of 3-5 years” might be.

3. How will humans mess it up? Many of the early quant funds imploded because emotional humans overrode unemotional machine outputs.

4. Why do you care? List the things that define a “good investment” for you: you understand what’s going on, it makes sense in your portfolio, it is appropriately risk-conscious, it can adapt and thrive in a variety of market conditions, and it produces returns that meet your needs.

   Do you have a reason to care where your manager’s office is? Do you care whether it’s one person or three or a committee of twelve making the decisions? Do you care whether your manager owns … or is … a computer? Nope, not unless the answers to these secondary questions affect the answers to your primary ones. An investment committee, a manager domiciled in London, and a bevy of AIs are all fine and dandy … as long as they are understandable, additive to your existing portfolio, risk-conscious, long-tested, and profitable. Otherwise, it’s just an intellectual curiosity.

Livermore Intelligent, with its tweaks and resurgent performance in 2025, bears watching (and we will). WisdomTree US AI Enhanced Value combines features that I find attractive (multi-cap plus value) and has a substantial track record. We’ll look more closely.