It’s the first time that we’re sharing in a structured way some of the many tests and experiments we run at Intelligent Alpha to understand how frontier AI models think and act as investors.

The nature of alpha is always changing, and research is the process by which you try to keep up. If you’ve been reading The Deload for a while, you’ve probably experienced the journey of Intelligent Alpha along with me. When I started experimenting with LLMs to invest in 2023, it was out of curiosity about the capabilities of the LLMs. It was more of a research project than an attempt to build a company.

Since then, we’ve done something few other companies using frontier AI models in the investing domain have done, which is register as an investment advisor to actually manage investor money. Being an RIA brings restrictions on how freely we can talk about what we’re doing, but talking about the IEB publicly brings us back to the roots of Intelligent Alpha.

The Intelligent Earnings Benchmark

The Intelligent Earnings Benchmark (IEB) tests frontier AI models on the core prediction that active investors make: What direction are earnings expectations headed?

Every quarter, we run a universe of large cap US stocks ($10b+ market cap) through a standardized process where the models predict the direction of forward consensus estimates over the next 60 days. Note, we’re not predicting earnings per se. The actual outcome of earnings isn’t what moves stocks. The changes in future expectations do. We believe predicting how the next quarter consensus estimates change is a more valuable test than merely predicting a quarter’s earnings.

We’ve locked in the Q2 earnings prediction cohort of 715 stocks with predictions across eight models:

• GPT 5.4

• Claude 4.6 Opus

• Gemini 3.1 Pro

• Grok 4.20 Reasoning

• GLM 5.1

• Qwen 3.5

• MiniMax M2.7

• DeepSeek R1

Each of the models is asked to predict:

• Direction: For both revenue and EPS estimates for the next quarter. Example: April 2026 will begin the Q126 earnings reporting period, but the models are predicting the direction of Q226 estimates.

• Revision %: The model’s estimate of the change.

• Magnitude: Ranges of small/medium/large change which correlate to the revision percentage estimate.

• Along with those predictions, models are asked to rate their confidence 0-100, offer rationale, a counter-thesis/risk assessment, and key signals most important to the prediction.

To make predictions, models receive:

• A dataset of two years of historical financial information including earnings, income statement, and balance sheet.

• Current consensus estimates for the prediction period.

• The most recent earnings transcript.

• A cache of current economic data from FRED.

• Web search via Exa where the models can make up to 10 searches.

At the end of the prediction period, which will be in early June for this first public benchmark, the models will be scored on the accuracy of their predictions. Public scores will be available on the Benchmark section of our website. Select partners of Intelligent Alpha may be able to access the full dataset of model predictions.

Early Insights

The first predictions run is complete and locked in for measurement.

The consensus of the eight models is that 70% of revenue estimates and 61% of earnings estimates for CYQ2 are going up between now and the end of the quarter. That’s bullish relative to history, but more in-line with the last quarter or two.

Going back to 2020 on a quarterly basis, revenue/EPS from the beginning of a new reporting period to the end of that reporting period increased roughly 55% of the time, was flat 16% of the time, and was down 29% of the time. This makes sense because management teams have an incentive to keep expectations in check so that they can beat them. We should generally expect to see an upward bias in the data.

As a naive baseline, if a human or a model just predicted that earnings would go up for every stock, they should achieve a 55% accuracy rate on average. That should be the minimum hurdle for value add in earnings predictions. Beyond that, the accuracy of magnitude predictions will be the ultimate test of the model in this benchmark.

As far as specific stocks, the consensus of the models is that VRT is the highest conviction upside call. All eight models believe forward expectations are too low with the highest average consensus score. EL is the name the models have the most shared concern that revenue and EPS expectations will come down.

My prediction based on years of watching the models: GPT or Grok is likely to be the top models in this first iteration of the IEB.

We plan on occasionally sharing more of the aggregate expectations from the models via our email list and on X. Subscribe for updates.

A True Measure

Elon is right. The truest measure of intelligence is the ability to predict the future.

Most AI benchmarks don’t test that. They test the ability of a model to solve a puzzle or retrieve information or some other rote task. We believe that makes most benchmarks inherently flawed because they’re solvable. Whether it’s GPT 6 or Mythos or some other model, eventually puzzles and retrieval tasks won’t measure anything because every model will be able to conquer them.

Markets are a different animal. A complex adaptive system where the answer keeps changing. The truest test of super intelligence would be when the earnings benchmark is solved because that would mean markets are solved. By the time AI solves markets, it can probably solve a lot more too.

Subscribe now

See Intelligent Alpha’s Important Disclosures Page here.

Additionally, note our benchmark disclosures: Intelligent Alpha’s Intelligent Earnings Benchmark (IEB) is an analytical tool designed to evaluate and communicate the comparative performance of AI models on earnings prediction tasks for US listed large-cap companies defined as market capitalization over $10 billion at the time of testing. This benchmark is published for general information and educational purposes only. It does not constitute investment advice, a recommendation to buy or sell any security, or an offer or solicitation with respect to any investment product or service. The Benchmark compares AI model-generated earnings direction predictions against consensus earnings prediction changes across a defined universe of US listed large-cap companies. Results do not represent the performance of any investment portfolio, fund, or client account managed by Intelligent Alpha, and earnings prediction accuracy should not be construed as an indicator of investment returns. The effectiveness of AI models in predicting earnings is limited by access to accurate historical data, tool usage, prompt structure, consistency of harnesses used to control the environment, and other factors. Past benchmark performance is not indicative of future predictive accuracy. This benchmark and all related content do not create an investment advisory, client or fiduciary relationship. Intelligent Alpha’s advisory services are provided solely pursuant to a written investment advisory agreement. No person should rely on this benchmark as a substitute for individualized investment advice.

And if it’s not, it will be soon.

There are three lasting human jobs in the AI first era:

Detective, people person, tastemaker.

Detectives

Detectives find useful information that isn’t known by AI models. Information from detectives correct deficiencies in company AI models, creating better outputs. That makes detectives seekers of important corporate truths.

Anything AI models don’t already know must be hard to find. Detectives talk to customers, suppliers, competitors, and other company employees to useful information.

Good detectives are naturally curious people, although detective is the role that requires the least innate talent. Hard work. Persistence, A touch of creativity.

Detectives are the blue collar employees in an AI first company. They won’t get paid as much as others, nor will they get much glory, but meaningful improvments to corporate AI won’t happen without them.

People Persons

People persons are the human interface for an AI first company’s customers.

Sales people, customer service, community managers. People persons are people that other people like. They’re nice. They’re friendly. They bring people together.

We know a people person when we meet one because they make us feel good. Even more obviously, we know when someone isn’t a people person, especially when they’re in a service role.

Misplaced people persons cannot happen in the AI age. The cost to the company is too great. Human touch points determine whether customers keep coming back even more in the AI era than now since every company will have good product.

Parents often wonder what their kids should study in college. People persons will have reliable jobs and get paid a lot for what seems like easy work. My partner Gene playfully says college kids should study partying and hanging out. He’s right.

Tastemakers

Wheras people persons are the customer interface for a company’s AI, tastemakers differentiate how the company’s AI works vs all other companies.

Kant says, “The judgment of taste is not a judgment of cognition, and is consequently not logical but aesthetical.”

AI’s superpower is logical cognition, not understanding the intangible. That’s a human job.

Tastemaking is the tactful application of contrarian views at the right time. It doesn’t take good taste to agree with conventional wisdom, and AI is conventional wisdom since it is built on all the things we know.

Tastemakers are the athletes, artists, and entrepreneurs in the AI era. Innate talent and hard work are the ingredients for a tastemaker to break out. Those who don’t have it never make it.

Tastemakers have always ruled the world, and they’ll rule in the AI era too.

The Deload explores my curiosities and experiments across AI, finance, and philosophy. Join 2,200 readers:

Subscribe now

Subscribe now

If AI is so Smart…

AI is going to cure all diseases, it’s going to create limitless clean energy, and it’s going to let us live forever.

AI optimists tout a utopian future, but if AI is so powerful, why hasn’t it made any groundbreaking discoveries already? AI knows everything humans have ever learned about the world, but it has yet to come up with a breakthrough. A reasonably smart human with the same amount of knowledge would have surely solved at least one problem by now.

Dwarkesh Patel posed this AI breakthrough question to Anthropic’s CEO Dario Amodei almost 18 months ago. He recently resurfaced it on X.

Amodei answered in August 2023 that AI was on the cusp of being able to make breakthrough discoveries. Maybe that breakthrough happens this year. Or maybe the way we build AI models has inherent limitations that won’t allow them to make important discoveries.

Commenters on Patel’s post highlighted two issues that might prevent AI from making breakthrough discoveries — one with training data and one with the training process. I think of the training data issue as an exploration problem and the training process an incentive problem.

Large-language models make predictions based on training data. While AI can make new connections between things it already knows, it can’t yet leap to ideas it doesn’t already know. AI needs the ability to seek and take in new data just like a human to come up with breakthroughs.

The training process may be an even bigger issue. As Joel Lehman and Kenneth Stanley write in Why Greatness Cannot Be Planned: “The question of what behavior is good or no good is important because the good ideas are the ones that the program will explore further.”

Show me the incentives, and I’ll show you the outcome works for AI models just as well as it does for humans.

Hallucinations — which Google calls “incorrect or misleading” results — are seen as bad behavior by AI models. The problem is that every breakthrough appears incorrect until it’s proven correct, so the models in our current paradigm would be discouraged from exploring such bad ideas. If we want novel discoveries, we need AI to play with bad ideas that turn out to be good.

Contrarians are those who explore bad ideas in search of good ones. Consensus thinkers avoid bad ideas in favor of safety from the crowd. The way we build AI today pushes it toward consensus and decidedly away from contrarian ideas.

My strongest belief about the world is that extraordinary outcomes only come from contrarian ideas that turn out to be right. Curing cancer or finding some incredible new energy solution would both be extraordinary outcomes, so I believe that only contrarian ideas will get us there. If AI can’t develop and explore contrarian ideas, then it won’t create the breakthroughs we hope from it.

So how can we solve the exploration and incentive problems that seem to be preventing AI from coming up with breakthroughs?

Agents Solve the Exploration Problem

If training data creates a barrier to AI-generated breakthroughs, then an AI model that’s capable of finding novel information about the world might be able to make novel discoveries. AI agents can solve the data problem by collecting insights from the world at massive scale. Agents could talk to humans to get information, have robots perform experiments, monitor environments through sensors, and more, reporting all of the data back to an AI model to discover contrarian ideas that might lead to a breakthrough.

Agents should enable AI to chase discoveries through the scientific method similar to humans — gather data, form a hypothesis, test the hypothesis, and so on. However, while agents might help models further knowledge in science, they may not solve all complex problems. Take markets for example.

Markets are complex adaptive systems built on unknowable unknowns. That’s what makes a market. If the inter-connected movements of all the components of a market were able to be known, it would no longer be a market. There’s no risk to be assumed, just prices to be accepted based on guaranteed outcomes.

That doesn’t mean AI can’t do a better job as an investor. Agents will capture information about the world, process it at scale, and likely make markets more efficient, but more efficient doesn't mean solved.

While agents may solve the exploration problem blocking AI from breakthrough discoveries, the incentive problem of defining “good behavior” might prove more challenging. Just because you add massive amounts of new data from the outside world doesn’t mean you’ll be able to find a new insight without a leap of faith that often comes in the most novel discoveries. When dealing in truly profound breakthroughs, there are often periods where intangible faith is the only path forward, not tangible evidence.

Lack of Emotion: An Advantage or Disadvantage?

We’ve been using LLMs as investment analysts and portfolio managers at Intelligent Alpha for some time. One of our core learnings is that AI’s inherent advantage vs humans is that it lacks emotions that often lead to bad decisions. This naturally connects to AI’s reticence to stray from consensus that’s represented by its training data . My evolving belief is that without large amounts of external data, LLMs can win at investing by doing consensus better by removing emotions rather than looking for home run breakthrough ideas.

Lack of emotion is AI’s superpower, and every superpower can be a super weakness.

I’ve written before about how human investors can harness emotion to beat AI. Breakthrough investment ideas rarely offer certain evidence that they’ll work. As explained above, if something were totally obvious, it would get priced into the market rapidly, and there would be no upside for taking the risk. Contrarian investors often use instinct and faith when evidence lacks for a unique investment idea.

Instinct and faith are emotional constructs. Humans can have them. AI can’t.

We expect AI to be emotionless and perfect. AI shouldn’t believe anything. It should know everything. We build these expectations into how we train models, avoiding bad hallucinations in favor of safe answers to be traced back to training data.

We don’t expect the same perfection of humans. Humans make mistakes. We believe things that turn out to be wrong, and we allow that of each other.

Mistakes are a key part of making breakthrough discoveries. Without an allowance for operating on belief and making mistakes, we’d never make major discoveries because breakthroughs necessarily push the bounds of what we know. We can never be certain about a path to a breakthrough because, as in markets, if the path were apparent it would already have been explored.

Unless we encourage AI to make mistakes by taking chances and acting on instinct, we may never get models that are capable of breakthroughs.

Contra AI: Rethinking Good Behavior

Contrarian ideas are necessarily unlikely, unacceptable, or both. They only come from “bad” behavior in the context of our current AI model building philosophy grounded in “truth” and “safety.” That’s a lot of quotes, but that’s because these terms are subjective.

Truth is not subjective in the absolute sense. What humans accept as truth is subjective, and there is safety in what consensus believes to be true in the world today. Consensus is always safe. It’s not always right. If we want AI models to be capable of breakthrough discoveries, then we have to allow them to pursue truth in the absolute sense even if it conflicts with what we want to believe true about the world.

Elon has described xAI’s Grok model as “maximally truth seeking.” Maybe Grok will be the model capable of breakthrough discoveries unfettered by consensus. While I’m optimistic about Grok, I’m not sure if it’s structured differently enough from other models to break the boundaries created by the current training approach.

Breakthroughs come from contrarian ideas, but sometimes consensus is right. To find truth universally, you need to be able to live between contrarian and consensus ideas, but it’s hard to fully embrace contrarian ideas while respecting consensus. Contrarians need a healthy disdain for consensus to power through the doubt and uncertainty that comes with chasing novel discoveries. As F. Scott Fitzgerald famously observed, holding two conflicting ideas at the same time is tough. Perhaps even for AI.

The solution to the AI breakthrough problem might be separating the consensus and contrarian functions into different models. Mainstream AI models offer the consensus outputs needed for most tasks, but a Contra AI would offer only contrarian ideas. Contra AI’s training would be inverted from what we do today. “Good” behavior would be hallucinations — inventing ideas outside of training data. “Bad” behavior would be relying on training data alone for answers.

You’d never want to ask a Contra AI the answer to 4 * 4 or who won the Civil War, but it might come up with the breakthrough we so hopefully want from AI.

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of my firms, Intelligent Alpha or Deepwater Asset Management. I may reference companies in which Intelligent Alpha or Deepwater has an investment. See Intelligent Alpha’s full disclosures here. See Deepwater’s full disclosures here.

The Deload explores my curiosities and experiments across AI, finance, and philosophy. Join over 2,100 readers:

Subscribe now

Getting Close to AGI

A year ago, Jack Kendall from Rain told me we’d hit artificial general intelligence (AGI) by 2025. I thought he was crazy. Turns out he might be right.

OpenAI’s GPT o3 looks like we’re getting closer to AGI. The new model hit a breakthrough score on the ARC-AGI benchmarking test meant to compare AI to human intelligence. The o3 model scored over 75% on the benchmark’s semi-private test when bound by compute cost. When unconstrained by compute costs, o3 scored an 87.5%, which is nearing a human equivalent. It may have cost upwards of a million dollars to hit the highest benchmarks (91.5% in a low efficiency test).

The idea that we can just allow state-of-the-art models to “think” longer to get better answers points to us being even closer to another concept that came from my conversation with Jack: Infinite intelligence.

I’ve thought of infinite intelligence as a large-scale system with knowledge in a specific field or fields equivalent to a human college graduate that can solve problems with brute force. That’s not quite o3, but spiritually it may not be far off.

As I understand o3, the program creates many Chains-of-Thought (COTs) that represent potential solutions to a problem, then it makes predictions about which of those COTs are most likely to lead to a solution. From ARC Prize founder François Chollet:

Effectively, o3 represents a form of deep learning-guided program search. The model does test-time search over a space of "programs" (in this case, natural language programs – the space of CoTs that describe the steps to solve the task at hand), guided by a deep learning prior (the base LLM). The reason why solving a single ARC-AGI task can end up taking up tens of millions of tokens and cost thousands of dollars is because this search process has to explore an enormous number of paths through program space – including backtracking.

o3’s approach is not one of heedless brute force but rather of elegant force. The model envisions potential solutions and tries what it thinks might work best, not too dissimilar to a human. The difference is that AI can envision and test orders of magnitude more potential solutions than a human ever could.

If we are approaching a world where AI can solve hard problems with elegant force, then it begs the questions: What problems can we solve with infinite intelligence?

Mechanistic vs Complex Adaptive Problems

There are two clear types of problems AI will be asked to solve: Problems with permanent solutions and problems with evolving solutions.

Problems with permanent solutions are mechanistic. Mechanistic problems don’t evolve. They’re siloed from the influence of external or confounding factors. Problems of physics and math live here. New problems in these fields don’t surface, we just see them as new given increased knowledge about the rest of the world around us.

Problems with evolving solutions live within complex adaptive systems where all the parts constantly collide with one another. John Holland coined the idea for complex adaptive systems, which are a network of interacting agents that continuously adapt behavior based on feedback and changing conditions in their environment. The human immune system is a complex adaptive system. So, too, are financial markets.

Unlike mechanistic problems, complex adaptive problems are often novel and unpredictable because they depend on an ever changing order in the problem’s environment. Solutions to complex adaptive problems, whether found by AI or humans, are bound to be temporary. A solution may work if the environment surrounding the problem doesn’t change, and it may suddenly stop working when the environment does change.

Infinite intelligence can solve both mechanistic and complex adaptive problems, but only if it makes economic sense.

SKU vs SAAS Costs

Mechanistic problems are like SKUs in Amazon’s warehouse. Each one has an associated cost for solving it with infinite intelligence, although that cost may be fuzzy. Each problem also has some assumed value for solving it, although that value may too be fuzzy. Solving mechanistic problems is a simple matter of capital allocation. Assuming the returns for solving a given problem exceed the hurdle rate of investing the necessary capital to do it, the problem should get solved.

Complex adaptive problems are more like SAAS subscriptions. When the problem is solved for a moment, the problem evolves and requires a new approach. So the allocation of capital to solve a complex adaptive problem, like the stock market, is a persistent one. As long as the continued investment generates the required return, you keep making it.

Infinite intelligence will follow the well-worn path of all new technology. Expensive at first. Cheaper over time. Eventually available to all. I expect in the next year or two we start seeing expensive AI-powered solutions to mechanistic problems that offer significant value.

However, financial markets may be the one problem that’s never cheap to solve.

Infinite Intelligence x Financial Markets

The necessarily competitive forces of financial markets make it a complex adaptive system unlike many others. Unlike other complex adaptive systems, e.g. the human immune system, an inherent feature of financial markets is actively competitive forces that will all use infinite intelligence if it is accessible enough. Cells in the human body can only ever rely on survival instinct in their competitive environment. They won’t apply infinite intelligence against one another.

The ultimate destination for infinite intelligence in financial markets may be a sort of massive stalemate. As costs come down, more participants will use infinite intelligence to find temporary solutions to markets, but as more elegantly brutal solutions are introduced to markets, the more complex those markets become. When everyone has weapons of infinite intelligence, no one has any edge. Ironically, the edge might end up applying basic intelligence to timeless market concepts. Maybe that’s always and forever the case.

The original Infinite Intelligence piece inspired by my conversation with Jack Kendall:

Subscribe now

“What’s the role of humans as AI takes over the investing world?”

It’s the most popular question I get about Intelligent Alpha, and I’ve been thinking about it for almost two years.

One of the core insights from building investment portfolios at Intelligent Alpha is that AI’s superpower as an investor is lack of emotion. AI doesn’t get shaken in drawdowns, it doesn’t feel FOMO in rallies, it doesn’t overreact to meaningless news, and most news is meaningless. Instead, AI views the world in scores and probabilities. It rigidly calculates quantitative and qualitative elements of an investment in ways humans don’t and can’t.

But all superpowers create super weaknesses. AI’s lack of emotion is a weakness where a leap of faith is necessary because true conviction depends on the intangible, not the logical. Investing with conviction is where humans can win against AI.

What is Conviction?

Superior investment returns require doing something different than everyone else, and doing something different requires conviction. That’s because only the unconventional requires conviction. We don’t need conviction to do obvious things, just acceptance. Conviction must step in to support an unconventional idea when evidence ends.

My definition of conviction:

By that definition, conviction is a necessarily human state. Machines can’t engage in intangible beliefs. They rely on substantive evidence to take action. That’s their superpower and super weakness.

Conviction is also a binary state. Either a person has enough conviction to act or not. You can’t be half convicted. If conviction is fuzzy, the switch is off. Fuzzy conviction is the dangerous state where logic corrupts the necessary emotion to embrace the intangible belief. Fuzzy conviction falls into the realm of probability, and that’s a game humans will lose to AI on average.

Conviction x Evidence x Returns

Conviction needs to step in when evidence fails, so the amount of evidence available about an investment helps frame the required conviction.

You don’t need much conviction to invest in Microsoft. There’s unending evidence about why it’s a great company. If you get it wrong, you do a little worse than the market. Big deal.

You need more conviction to invest in a two-person startup than MSFT. If you get it wrong, you lose your entire investment. The trouble is there’s almost no evidence available at the earliest stages of investing, so while you need some conviction, you’d be fooling yourself to think you can find high conviction in a venture bet.

Between large public companies and startups is where you can find meaningful conviction in a balance of evidence and uncertainty.

Our conviction x evidence curve can also describe potential returns. The more evidence we have about something, the more conventional an idea it must be, and the lower the return is likely to be. We have lots of evidence about MSFT, but far less evidence for an investment in a new startup with two engineers and no customers. The potential return for the startup is orders of magnitude larger than MSFT because the conventionality (aka risk profile) is completely different.

Our blend of conviction, evidence, and potential return leaves us with three major investment categories where we can compare humans and AI:

• Modest conviction, low evidence, extreme return. This is venture capital.

• Low conviction, strong evidence, low return. This is active public equity.

• Strong conviction, modest evidence, strong return. This is private equity, activism, business ownership, etc.

Venture Capital (Modest Conviction, Extreme Return)

The limited evidence available to support venture capital investments makes it an instinctual effort mostly about reading people. You can find conviction in the founders but little else.

Because the startup investing world operates with so little evidence and therefore limited conviction, the optimal startup investing strategy is to build a broad portfolio that generates overall return through power law returns. Capture one really big winner, and it doesn’t matter if all the rest go to zero.

While the instinctual piece of startup investing elevates humans above AI, the limited ability to generate conviction and the effectiveness of a broad portfolio mean that AI could be a good startup investor. Broad portfolios invite probability to offset conviction. We’ll see several VC firms building AI to make investment decisions in the next few years, although gaining access to deals will be a human effort for a while yet.

Public Equity (Low Conviction, Low Return)

It takes minimum conviction to invest in most public companies (say above $5 billion in market cap), not just Microsoft. Public companies come with broadly available evidence to support an investment and the freedom of liquidity. Liquidity lowers the bar for conviction even further because investors can change their minds easily. Emotions are strong, convictions are fragile, and evidence appears to change frequently, so portfolios do too.

Because convictions in public markets are generally weak, portfolios are diversified to reflect it. Can an investor have conviction in 50 stocks? The investor might say he has conviction in his portfolio, but then he’s playing a game of statistics, not conviction, and we’re back to AI’s advantage.

The broad and consistent victory of indexes vs active managers shows that when investment strategies rely on weak convictions, statistics win. My experiments with dozens of strategies at Intelligent Alpha convince me that when you add intelligence to structure, you can do well in statistical games over the long run.

Other Equity (Strong Conviction, Strong Return)

The middle category is the most human-optimized segment of investing. It requires strong convictions built on modest supporting evidence with the potential for strong returns. This includes private equity, activist investing, and business ownership. The common thread of this middle category is that the investor brings with him some insight about an improvement he can make at the business that will generate returns. He’s not relying on others to figure it out for him as in the case of venture capital or public equity.

As a result of the alpha-generating insight and the willingness to get involved, high conviction bets tend to be highly concentrated. Concentration and activity go hand-in-hand with conviction. When you find that rare conviction, you should bet big and act accordingly.

AI might be able to do the analysis of a private equity analyst or an activist, but it will be a while before AI can perform the actions necessary to affect the outcome of the investment.

Human x AI

Anytime someone asks when AI will replace humans at some task, it ignores the reality of comparative advantage. To the extent humans are better at something, we can optimize for that rather than be relegated to obsolescence.

The human comparative advantage in investing is conviction. The challenge for us is that conviction must be rare. True and meaningful conviction should only happen a few times in a career. When we find conviction, it should be overwhelming and almost uncomfortable.

Most of the time, we’re in that fuzzy state where logic interferes with our instinct, and that’s where we make mistakes. If you’re not overwhelmed and uncomfortable, you’re not convicted, you’re just fooling yourself. You might be right sometimes in the fuzzy state, but then you’re dealing with the realm of probability, and that’s AI’s domain.

Whenever we’re in search of conviction, we should outsource our investing efforts to AI. That maximizes comparative advantage. AI can shine when dealing with probabilities, which is most of the time. Humans can shine when we find a rare unconventional idea that requires a convicted leap of faith, which is only once in a rare while.

Of this, I have strong conviction.

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of my firms, Intelligent Alpha or Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Intelligent Alpha’s full disclosures here. See Deepwater’s full disclosures here.

Subscribe now

Suffering is the Path to Success

“TAKE PAIN! TAKE PAIN! TAKE PAIN!”

That’s the daily mantra of one of the world’s most successful traders, Paul Tudor Jones. Why? Because PTJ knows that pain is the precursor to all great success. It’s true of investing, business, parenthood, fitness, everything. One’s capacity for success in anything is directly linked to one’s capacity to suffer.

Our ability to endure pain is like a muscle. We’re endowed with a certain natural power to endure, but it’s also something we can optimize for by choosing where to embrace pain. I’ve learned, often painfully, as an investor and now an entrepreneur with Intelligent Alpha that three ingredients make pain more bearable, and they happen to rhyme:

• Mission

• Conviction

• Recognition

Mission: Find a Big, Contrarian Idea

Most submit to pain because they accept missions that are too small. People chase small, conventional ideas to make a little money or make something a little better because small, conventional ideas are comfortable, they’re obvious. they’re easy. Who gives a shit. Not investors. Not employees. And eventually not even the entrepreneur. Small, conventional ideas are destined to be abandoned.

Big, contrarian ideas survive pain, but there aren’t that many of them. Big ideas that do exist are hidden. If the big ideas were obvious, someone would have done them already. So big, contrarian ideas require a long process of curiosity and discovery to find. Curiosity is the kernel of all extraordinary outcomes because it leads to the few big ideas that matter.

We’re lucky to run into just a few truly big ideas in a lifetime. When we find a big idea, it should be so obviously important that it takes about two seconds of thought to chase the idea. More than two seconds of thought, and the idea probably isn’t big enough. Finding big ideas meant for you is like finding a unicorn. You can’t help but know when a big idea hits you.

When you’re not chasing a big idea, your time is best spent in curious discovery while fending off the many mediocre ideas that try to fool you along the way.

Conviction: The Big Test

Conviction is not earned through careful analysis and testing. Conviction coincides with discovery. Ambitious people are wired to believe in big, contrarian ideas because we know they cause revolutions that change the world. When a big idea grabs us, we can’t help but believe it. You don’t need to find conviction in your big idea, conviction finds you.

Once a big idea grabs you with conviction, the world will test it for you. Every setback as an investor or entrepreneur is the world asking how much you really believe in your idea. It’s testing how much you want it. The market won’t cede the riches of success to the weak handed. It’s built specifically to parse out the weak and reserve reward for the strong willing to take the pain of their conviction.

Conviction finds you, but what you need to find is integrity. Your big idea might be wrong. Maybe that’s what the world is telling you. The careful balance between stubborn conviction and intellectual honesty is the art of taking pain.

Recognition: Enjoy Getting Punched in the Face

If you have conviction in a big, contrarian idea, then recognition is your best friend. Know that pain is coming, and welcome it as a necessary friend.

A friend who’s launching a new company was recently sharing some doubts and frustrations that come with starting any business. I told him what I tell all the entrepreneurs we fund. Pain is a requirement to work on something important. It’s the world testing your conviction. Find joy in the pain, and keep going.

Mike Tyson has a legendary quote that, “Everyone has a plan until they get punched in the face.”

When you get punched in the face, remind yourself that’s the cost of success. Then get back to work. Somehow knowing the pain is coming makes absorbing the hits a little easier.

Take pain!

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of my firms, Intelligent Alpha or Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Intelligent Alpha’s full disclosures here. See Deepwater’s full disclosures here.

People are motivated by different things in life. Some by money. Some by power. Some by impact. I’ve always been motivated by adventure.

A good adventure leads to some discovery about the world or yourself. A good adventure requires some risk because there’s nothing to be discovered in comfort. Some of the best adventures sneak up on you. You don’t even know you’re on an adventure until you’re right in the middle of it.

Intelligent Alpha has been one of those sneaky adventures for me. A year ago, I started an experiment to see if ChatGPT could build stock portfolios to compete with the S&P 500¹. That experiment turned into an artificial intelligence-powered investment committee of three large-language models that now oversee dozens of strategies.

I’ve been silent on the adventure of building Intelligent Alpha for the past couple of months because we’ve been working on a big announcement:

Intelligent Alpha is bringing modern AI-powered investing to the market.

Last week, we launched our first product, the Intelligent Livermore ETF (LIVR), which uses our AI investment committee to identify opportunities inspired by the world’s greatest investors and traders. The Livermore ETF allows the committee to invest in stocks across different geographies, themes, and sectors with the current portfolio focused on AI, Asia, Latin America, and defensive stocks.

James Dyson said the key to creating a breakthrough product is difference for the sake of it. That applies just as well to investment funds just as it does to vacuum cleaners. New managers need to do something unique to get the attention of investors, and then they need to perform well to keep their trust. We built our first product to be purposefully unique — a sort of mini multi-manager fund that mixes the perspectives of several investing greats.

We’re different. Now we need to perform.

AI’s Superpower: Lack of Emotion

The Intelligent Livermore ETF gets its namesake from Jesse Livermore, the first great Wall Street trader. Livermore was the subject of the famous book Reminiscences of a Stock Operator written in 1923 which details the lessons he learned as a trader. Many investors still view the book as canon 100 years later.

One of Livermore’s great insights was:

“There is nothing new on Wall Street or in stock speculation. What has happened in the past will happen again, and again, and again. This is because human nature does not change, and it is human emotion, solidly built into human nature, that always gets in the way of human intelligence. Of this I am sure.”

My year-long experiment in investing with Large Language Models taught me that AI’s clearest advantage is lack of emotion. There’s nothing broken in the human-centric investment process except that it’s performed by emotional humans. We feel pressure and doubt our judgment when stocks are down. We feel joy and ascribe undeserved genius when our stocks are up. Emotion doesn’t get in the way of artificial intelligence, and that advantage vs humans is permanent.

Jesse Livermore himself is a reminder of this fact.

Livermore made and lost many fortunes in the market. Despite his vast knowledge about managing emotions as an investor, his career ended in ruin. Livermore’s namesake serves not just as a nod to our pioneering introduction of AI-powered investment strategies but also a reminder that no matter how many lessons we learn, it’s the continual application of those lessons that will determine our longevity and success.

Livermore is just the first product powered by Intelligent Alpha’s AI investment committee. We have so much more to come.

Replication vs Inspiration

Intelligent Alpha is the first asset manager built to use modern large-language AI models with human oversight to ensure alignment with the underlying strategy to do investment selection. The frontier is always interesting, and we’ve been lucky to have a lot of interest in Intelligent Alpha. Being on the frontier also means we’re responsible for educating the market about AI-powered investing.

Some who follow Intelligent Alpha have characterized us as “replicating” existing managers like Buffett, Druckenmiller, and others, but I see it differently. We inspire our AI to think like great investors, not replicate them.

To replicate is to copy someone. Portfolio replication of great managers has been around for a long time. Look at a 13f, buy the same stocks, and make changes when the next 13f hits. You just need a web browser and a brokerage account, not AI.

Some research even suggests that pure replication can be a viable strategy, although to the extent replication outperforms the actual managers seems more an unpredictable accident than a persistent feature. 

We don’t try to replicate great managers. We use great managers as a basis for our AI Investment Committee to develop an investment philosophy and select on its own merits.  

Just as the young value investor might study Buffett and Klarman and Grantham for inspiration to form a philosophy, we do the same with AI. We infuse our investment committee with the wisdom of great investors, then let AI apply that wisdom as it sees fit. Sometimes that means an Intelligent Alpha portfolio will overlap with a human manager who served as inspiration, sometimes it might not.

The reason we use great managers as inspiration is because LLMs allow AI to “think” in abstract ways like a human manager that wasn’t possible before with traditional machine learning techniques. We don’t just reduce a manager down to his favored quantitative factors and a list from a 13f. We try to have the AI understand the kinds of companies and ideas a manager favors from a qualitative lens too, and I believe the combination of the qualitative and quantitative is what will set investing with modern AI apart from prior AI investing efforts.

Ideally, just as Warren Buffett learned more techniques and vastly outperformed his mentor in Benjamin Graham, we want our AI committee to evolve continually as an investor with the goal of being better than the managers that inspire it. With the right continued evolution of our process and the advancement of AI models, the odds should favor AI being a better investor over time than its many legendary mentors. 

Move Fast

In a technology-driven business, Elon Musk said it best: “Moats² are lame.”

Other firms will find ways to use LLMs in their investment process, and the only way to stay ahead is pace of innovation. Look at Google. Bing has been 97% (my estimate) as good as Google for a long time, but it doesn’t matter. Google has maintained its 3% edge by being fast innovators. They’ve stayed ahead, they’ve kept customers happy which builds trust in a brand, and ultimately that is the strongest moat.

A former Googler, Paul Buchheit, makes winning in tech clear: “If you're in the lead, and you're moving faster than everyone else, then no one can ever catch up.”

We promise to keep moving fast to create new, innovative techniques that improve our AI investment committee and launch new products that empower AI as an investor. That’s why we believe that the future of investing is intelligent.

Subscribe now

Important Information:

1. S&P 500: S&P500: It is an unmanaged index of 500 common stocks primarily traded on the New York Stock Exchange, weighted by market capitalization. Index performance includes the reinvestment of dividends and capital gains.

2. Economic moat refers to a business's ability to maintain competitive advantages over its competitors in order to protect its long-term profits and market share.

Investors should consider the investment objectives, risks, charges and expenses carefully before investing. For a Prospectus or Summary Prospectus with this and other information about the Fund, please call +1-215-469-1717 or visit our website at iaetfs.com. Read the prospectus or summary prospectus carefully before investing.

Investments involve risk. Principal loss is possible. Redemptions are limited and often commissions are charged on each trade. Unlike mutual funds, ETFs may trade at a premium or discount to their net asset value.

High Portfolio Turnover Risk. The Fund’s investment strategy is expected to result in higher turnover rates. This may increase each Fund’s brokerage commission costs, which could negatively impact the performance of a Fund. Rapid portfolio turnover also exposes shareholders to a higher current realization of short-term capital gains, distributions of which would generally be taxed to you as ordinary income and thus cause you to pay higher taxes.

Leveraged, Inverse, and Inverse-Leveraged ETF Risk. Leveraged, inverse, and inverse-leveraged ETFs expose the Fund to all of the risks that traditional ETFs present (see “Underlying Fund Risks” above). Leveraged ETFs seek to provide investment results that match a multiple of the performance of an underlying index (e.g., three times the performance). Inverse ETFs seek to provide investment results that match a negative (i.e., the opposite) of the performance of an underlying index.

Foreign Investment Risk. Returns on investments in foreign securities could be more volatile than, or trail the returns on, investments in U.S. securities.
Small- and Mid-Capitalization Companies Risk. Investing in securities of small- and medium- capitalization companies involves greater risk than customarily is associated with investing in larger, more established companies. These companies’ securities may be more volatile and less liquid than those of more established companies.

Large-Capitalization Companies Risk. Large-capitalization companies may trail the returns of the overall stock market. Large-capitalization stocks tend to go through cycles of doing better – or worse – than the stock market in general. These periods have, in the past, lasted for as long as several years.

AI Model Risk. The Fund is actively managed using the AI Models, the output of which is heavily dependent on multiple inputs, including current and historical data (collectively, “Data”). To the extent the AI Models do not perform as designed or as intended, the Fund may not be able to achieve its investment objective and may lose value.

Machine Learning Risk. The Fund relies on publicly available “machine learning” selection processes as well as data and information supplied by third parties that are utilized in those processes. To the extent the machine learning process does not perform as designed or as intended, the Fund’s strategy may not be successfully implemented, and the Fund may lose value. If the input data is incorrect or incomplete, any decisions made in reliance thereon may lead to the inclusion or exclusion of securities that would have been excluded or included had the data been correct and complete.

Risks Related to the Use of Form 13F Data. The Form 13F filings used to analyze trading trends are filed up to 45 days after the end of each calendar quarter. Therefore, a given investor may have already sold some or all its positions by the time the AI Models evaluate the filing. Furthermore, the Form 13F filing may only disclose a subset of a particular investor’s holdings, as not all securities are required to be reported on the Form 13F. As a result, the Form 13F may not provide a complete picture of the holdings of a given investor. An investor may hold long positions for a number of reasons, and the AI Models may not appreciate the reasons, or the strategies followed by an investor who makes the filings. The analysis of the AI Models may not be representative of the investor’s universe or the strategies that give rise to the reported holdings. Because the Form 13F filing is publicly available information, it is possible that other investors are also monitoring these filings and investing accordingly. This may result in inflation of the share price of securities in which the Fund invests.

New Fund Risk. The Fund is a recently organized investment company with no operating history. As a result, prospective investors have no track record or history on which to base their investment decision. There can be no assurance that the Fund will grow to or maintain an economically viable size.

The Securities and Exchange Commission (SEC) does not approve or disapprove of any investment. This material has been distributed for informational purposes only and should not be considered as investment advice or a recommendation of any particular security, strategy or investment product. Information contained herein has been obtained from sources believed to be reliable, but not guaranteed. No part of this material may be reproduced in any form, or referred to in any other publication, without express written permission. References to other funds should not be interpreted as an offer of these securities.

Information presented by Intelligent Alpha LLC and Deepwater Asset Management, both SEC-registered investment advisers. Registration does not imply any level of knowledge or skill. For more information and important disclosures, please see our FormADV 2A brochures which are available at www.intelligentalpha.co and www.deepwatermgmt.com. The Fund is distributed by Quasar Distributors, LLC. The Fund’s investment advisor is Empowered Funds, LLC which is doing business as ETF Architect.

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Empowering AI That Pays Humans

Will AI take our jobs?

It’s the everlasting question that finds a way into every conversation about AI. If AI is as profound as many think, the answer should be “yes,” but with caveats. There are things AI is good at today — coding, information summarization, basic customer service — and there are many things it isn’t yet suited to do. The gap between today and a future where AI is omni-capable may be decades away. Perhaps AI never surpasses humans in some tasks, particularly around human specialties like community and empathy.

In that interim period as AI improves its skill set, however it lasts, humans will play a major part in helping AI agents meet their potential. AI will need to work with in tandem with humans, delegating tasks where humans are more capable and paying those humans for their service. There was no way for AI to pay humans. Until now.

A new startup, Payman, serves as the first platform that connects AI agents with a wallet to pay humans for work. Deepwater invested in Payman because we agree with the company’s vision that AI + humans is the future, not AI alone.

Tyllen Bicakcic co-founded Payman to help create a future where humans continue to thrive in an AI world. He chatted with me on the latest Deload Podcast about how humans can augment AI agents, the freedom that AI can create for humans, and how crypto might play into the whole opportunity.

A few thoughts from our podcast.

Comparative Advantage is Undefeated

At the core of Payman’s mission is extending comparative advantage in the AI age. Structurally, Tyllen believes that there are tasks that humans will be better at than AI. The skillset amongst those humans who can perform a given task will also vary. Those humans who are great at what they do, just in our current economy, are likely to be able to charge premium prices for premium services.

In this context, a world where AI pays humans for work looks more like the creator economy where talent rises than the sharing economy where talent all looks basically the same. To the latter point, is there really that big a difference in an Uber driver? That’s where autonomy most obviously wins: Where you can’t tell much of a difference in the service that humans provide today.

Three years ago, the future of work was flexible jobs and sitting at home on Zoom all day. Today, the future of work is getting paid by AI for doing things you’re uniquely skilled at. That’s an upgrade for everyone.

Payman in the Wild

Not only is Payman building the first platform to connect AI agents with human workers, they’re implementing the product on their own business. Tyllen shared a video showing how to create a task in Payman for an AI to pay for agent developer leads. It’s easy. It’s efficient. It’s available now in beta.

There’s an under appreciated reality to extending comparative advantage in the AI age: Mastery and skill are things that make humans happy because they make us feel competent and useful. To the extent we can excel at certain tasks vs AI and participate in the process of mastery and recognition of that mastery, the future where AI employs humans is probably closer to that of the utopians than doomsayers.

Podcast Timestamps

00:00 Intro & Inspiration for AI and Human Collaboration

04:08 AI Agents and Human Skills

09:17 The Role of Humans in an AI-Powered World

13:36 Creating a Marketplace for AI Agents and Specialized Humans

17:33 Leveraging Crypto Payments for Global Accessibility

21:28 The Intersection of Crypto and AI

25:01 The Vision for Payman: AI Agents and Humans Working Together

30:39 The Future of Payman: Enabling Humans to Focus on What They Enjoy

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Obvious Ideas are the Best Ideas

It’s conventional wisdom amongst great investors that only unconventional ideas generate extraordinary returns. The power of contrarian thinking is the founding inspiration of The Deload. You have to be non-consensus and right to generate returns different than everyone else, so it’s always valuable to challenge yourself when you’re in the majority.

Maybe not if you’re Stan Druckenmiller.

Druck says he made 120% of his profits from obvious ideas. His non-obvious ideas lost 20% (h/t to Phillippe Laffont’s Bloomberg interview).

How can it be that only non-consensus ideas generate great returns, but one of the world’s greatest investors generates all his profit from obvious ideas? Shouldn’t the obvious ideas already be known by consensus?

The answer lies at the intersection of the complexity and duration of an idea.

The Prediction Matrix

All investments are predictions about the future.

Buying an S&P 500 index fund is a prediction that stocks will appreciate over time as a reflection of broad economic growth. Buying T-Bills is a prediction that the US Government won’t go bankrupt in a few months, so you’ll get your principal back plus interest. Buying stock in a single public or private company is a prediction that the company will grow earnings over time, or benefit from some other factor, and that will make the stock increase in value.

Short-term investment performance is harder to predict than long-term performance because the short term is subject to more noise. Buying the S&P 500 with the prediction of long-term economic growth is an easier prediction to make than whether AAPL will beat earnings next quarter. Wild things can happen in the short term that get smoothed out in time, hence the “zoom out” meme of looking at long-term stock charts to ignore short-term volatility.

When we pair patience with easy predictions, we find the greatest investment opportunities because patience is rarely consensus.

Patient, Easy Predictions

Patience paired with easy predictions makes great investors, not finding hard, galaxy brained ideas that are impossible to repeat. How many of the big winners of The Big Short have done anything great since 2008?

Druck didn’t try to predict every quarter for NVDA. He saw the AI boom and knew that NVIDIA was the company best poised to take advantage of it. Easy prediction. He bought it early in the trend and waited.

Buffett often reminds us that his best ideas are obvious, too. He suggests looking for one foot fences to step over, not 10 foot ones. One foot fences just don’t come around that often. They take a lot of patient searching, then patient holding.

The same patience is what lends power to indexing and quantitative investing. Indexes guarantee you get the market return in the long run, which is better than 90% of active managers. Quants apply patience to well-researched factors that should offer excess return if followed for a long enough time. It is patient consistency over time that makes obvious ideas non-consensus, allowing for extraordinary returns.

Patience is the ultimate non-consensus act in investing, but it’s also the hardest act. Humans are naturally impatient. It goes against our wiring to build conviction about something that will take time to prove out. We’ve only got so much time on earth, and we don’t want to spend it waiting. It’s more appealing to chase obvious ideas with impatience, or go after hard ideas that tempt us with greater rewards.

Impatient, Easy Predictions

The worst strategy for investors is to impatiently chase easy predictions. Impatient, easy predictions are a sucker’s game fueled by emotion and instinct. Investors in this category don’t take the time to put enough depth of thought into their predictions to create any unique insight or conviction. The predictions are based on sentiment, momentum, and popular narrative. As a consequence, as soon as those trades look wrong, or even if they look right, many of those investors bail and take the loss or small gain.

Most of the profit from impatient, easy predictions accrues to the investors who stick with the idea patiently or by impatient investors who made the prediction when it was still hard. The impatient easy prediction game is largely not worth playing.

Impatient, Hard Predictions

Many hedge funds and traders play the impatient, hard prediction game because of incentive structures that offer large annual payouts for beating benchmarks. As Munger reminds us, “Show me the incentives, I’ll show you the outcome.” If a manager can make a lot of money for doing well in a given year, then he’ll try to find ideas that will do well in a given year, a definitionally impatient timeframe.

Impatient, hard predictions depend on persistent work to find small edges. A hedge fund might play the quarterly earnings game, betting on winners or losers, or it might play the macro game, betting on interest rate changes or otherwise. If a manager can generate a consistent edge, it can compound over time; however, finding that edge and keeping it requires far more effort and invites the potential for far more volatility than patient, easy predictions.

Patient, Hard Predictions

Venture capital is the quintessential patient, hard prediction game. In venture, you bet on people, who are hard to read, building products for markets that don’t exist, which are hard to imagine. There are no easy predictions in venture. Since private investments are naturally illiquid, you’re forced to be patient — certainly a feature of venture, not a bug.

The difficulty of patient, hard predictions don’t make them bad investments, it just changes how you need to structure your portfolio. Patient, hard predictions come with the most dynamic potential. Venture funds are driven by power law outcomes. 1,000x single investment returns make great funds, not a handful of 10x’s. You want lots of shots at the 1,000x rather than a few “safer” shots at a 10x.

Patient, hard predictions tend to be more abstract and less quantitative in nature, and abstractness requires a kind of creativity and empathy that the other forms of investment predictions don’t. The abstract nature of patient, hard predictions means humans have a natural edge in making them, whereas AI has the edge in the other forms of prediction.

A Prediction About Predictions

Over the next decade, AI will take over the easy, patient predictions made by great investors and quants, and it will also take over the hard, impatient predictions that dominate hedge funds. As with all discussions of man vs machine, the relative strengths and weaknesses of each define which is likely to win and lose.

Humans fall prey to emotions and biases that cloud logic in the context of investment decision making. We want to believe certain narratives, even if the data and environment don’t support them. We feel pressure when we’re wrong, and even when we’re right, and that creates poor decision making. All of this lends to our impatience.

AI is less fallible in logic because AI doesn’t suffer from emotions nor impatience, but it’s not perfect either. AI lacks the unbounded creativity that humans enjoy, often incited by the very emotions that cloud our logic. Creativity is not a logical act, so logic only hampers it.

We could reclassify our four investment quadrants in the guise of creative necessity. Making patient, easy predictions doesn’t require creativity because the ease of the prediction makes it obvious. Obviousness repels creativity. The obvious reality of an easy prediction should be decipherable by AI. Making impatient, hard predictions similarly doesn’t require creativity because the demand of short recognition of the reality of the prediction means it should be built on probabilistic fact rather than creative curiosity. Again, AI’s superiority in logic gives it the edge.

That leaves the abstract nature of patient, hard predictions as the category best suited for humans. And that makes sense. Humans should be better at judging one another on nuanced traits and personalities than an AI that doesn’t understand human nature in the same way.

In all this talk about human bias, I recognize my bias in making the prediction that AI will take over much of investing since I’ve been experimenting with AI to make such predictions for almost a year.

Intelligent Alpha started by targeting patient, easy predictions to build portfolios with the mentality of a great investor buying companies. AI has proven it can make those patient, easy predictions quite well with almost 80% of the 40 strategies I track ahead of benchmarks by an average of 400 bps.

The next phase for Intelligent Alpha is to make impatient, hard predictions. I’ve been in the lab testing Intelligent Alpha’s investment committee — The Alpha Counsel — to make some harder predictions. AI can make these predictions, and I think AI is on its way to doing them better than humans too.

A few examples:

How will Bitcoin trade over the next year?

The Alpha Counsel thinks that Bitcoin will break $100k in the next year with a potential bull case of $150k. The committee notes: "If a pro-crypto candidate like Trump or Kennedy wins, and the administration implements favorable regulations and economic policies, Bitcoin could experience a significant bull run." Seems odds of that outcome have increased dramatically since I polled The Alpha Counsel in early June.

Will the Fed cut rates in 2024?

The Alpha Counsel universally expects the Fed to cut rates later this year. Consensus is for a September cut and an average of two cuts total this year, although there is significant disagreement on the counsel over that number. A less consensus view: Two AIs expect the Fed to cut, then pause for a meeting to reassess.

Time will tell if the BTC and rate predictions are good or not, but my most recent experiment gives me even greater confidence because of strong backtested results.

Several friends shared with me a great research paper from the Chicago Booth school where researchers used AI to predict whether a company would post earnings growth or decline. The study showed that GPT performed better than human counterparts in making earnings predictions.

Since stock price movement is at its most basic level a function of earnings and multiples, if we can predict earnings well, we should be able to create an edge in making impatient, hard predictions. However, we don’t just want to predict whether earnings might go up or down, we want to predict whether they will be up or down vs what’s already priced into the market. Sell-side consensus expectations, while not a perfect representation of what the buy side expects, are a good proxy. So, my most recent experiment tested whether The Alpha Counsel could predict which companies in the S&P 500 would beat consensus revenue and EPS as of the beginning of 2024.

Will company X beat earnings this quarter?

Across four tests in earnings predictions, The Alpha Counsel showed an average of over 65% accuracy in predicting whether a company would beat or miss revenue and 62% accuracy in predicting EPS. In some cases, the accuracy was 70% or greater.

The earnings predictor is still in the lab, but it shows AI’s promise for making impatient, hard decisions.

Predictions vs Actions

Two kinds of investing were left out of our predictions matrix: activist investing and leveraged buyouts.

One of the allures of activist investing and leveraged buyouts is that you aren’t necessarily making a prediction, you’re forcing an action. You have active control of your destiny rather than passive ownership of some stock or index.

When you can determine an outcome, you’re creating a different kind of alpha.

The activist and LBO styles don’t neatly fit into our model for predictions, rather they live sort of adjacent to patient, hard predictions in that they too require abstract thinking and creativity. Activists and LBO operators need to create detailed plans then implement them with people. While AI is well suited to create plans, it’s not well suited to implement them, nor are humans ready to follow.

Maybe AI will take over venture, activism, and LBOs in time, but for now Intelligent Alpha shows AI will win in the easier domains of patient, easy predictions and impatient hard ones.

The future of investing is intelligent.

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

The Market Thinks We’re in an AI Bubble Already

And we’re going higher.

At least that’s the consensus from a 1,300 person survey I ran on X this week. Almost 46% of respondents said we’re already in an AI bubble and will go higher. For what it’s worth, I agree with the 34% who think we’re not in a bubble, but we’ll get one.

That 90% of respondents think we’re either in an AI bubble already or we’re going to get one makes me wonder: Can we really get an AI bubble when it seems like everyone expects it?

We Always See the Bubble Coming

Famed Yale economist Robert Shiller wrote Irrational Exuberance at the peak of the Internet bubble in 2000 suggesting that markets were…too exuberant. His timing was impeccable. Less well known of Shiller is a study he did during the 90s bull market where he assessed changes in investor sentiment over time.

The study analyzed a series of surveys asking whether investors thought asset prices were too high/low/etc, whether they thought markets would go up or down in the future, and whether they felt others around them were showing “a great deal of excitement and optimism about stock markets.” Shiller created an index from the responses to describe whether investors thought the market was in a bubble or not. The index suggested that throughout the mid-to-late 90s, investors floated between periods of heightened bubble awareness and periods with more muted expectations.

90s investor sentiment trended with the market. As markets boomed, concerns about the bubble increased, and those concerns receded with the market. Price drives sentiment.

And that brings us to the current state of markets. We’ve largely been up and to the right, save for a few 10%-ish corrections, since the bear market bottom in 2022, which also coincided with the launch of ChatGPT. Price seems to be driving the large majority to say we’re in a period of great optimism, but the Shiller data tells me that won’t stop us from bubbling even more. Shiller’s index says the market had peaky bubble feelings in 1993 (arguably the start of the Internet boom), 1996, and 1998, and we all know the ultimate peak wasn’t until 2000.

Maybe the better question is can you have a bubble when no one expects it? I don’t think so. Bubbles are driven by psychology, not secrecy. Optimism and fomo are contagious, not subtle. The AI bubble will stare us right in the face as it's happening, if it’s not already here.

Where are We in the AI Bubble?

Clearly nowhere near the insanity of the Internet era:

• NVDA trades at 40x forward earnings. Cisco peaked at over 100x in the Dotcom bubble.

• The Nasdaq 100 trades at 26x forward earnings. The Nasdaq 100 traded near 100x in the Dotcom bubble.

• An Intelligent Alpha index of AI companies, the AI Average, trades at 53x earnings. The Nasdaq peaked around 200x in the Dotcom bubble.

There are also hints from the data in my survey compared to that of Shiller’s.

Shiller’s survey measured how many respondents believed that markets were too expensive and would keep going up and also those who thought markets would soon crash. I tried to replicate similar responses in my survey responses, e.g. we’re in a bubble and going higher, and we’re in a bubble and about to crash.

The strong sentiment from my survey is that we’re going higher vs ready to crash. This seems to correlate to late 1995 or late 1996 in Shiller’s survey where respondents generally thought markets would keep going up and had minimal concern for a crash.

The adage that bears sound smart and bulls make money applies here. There were smart sounding bears in 1995, 1996, 1997, and later, and they missed one of the biggest bull markets ever. The consensus of the market through the 90s showed that it understood what was happening better than the “smart” bears — a bubble was forming, and the market was going to keep going higher.

The Internet bears were right about being in a bubble, but timing and positioning matters in markets. A lot of “smart” bears went out of business trying to short the obvious, and there were plenty of smart investors who did get out of the bursting bubble with incredible profits. Those who make money in markets are the smart ones, not the bears that lamented the obvious all the way up with nothing but words to show for it.

Subscribe now

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

Intelligent Alpha: Faith in a Good Idea

Every good idea survives in part through faith. An idea that is plainly obvious doesn’t need faith. Such an obvious idea can’t be a good one because a good idea must come with a meaningful reward. The reward of obvious ideas has already been captured by those who figured the idea out first.

The past month has come with frustrating returns that have tested my faith in Intelligent Alpha, the AI-powered investment platform I’ve been building over the past year. The good news is that more than 70% of the strategies I track are ahead of benchmarks since inception, and 60% are ahead of benchmarks YTD.

But only about 40% of the strategies exceeded benchmark performance over the past month. Many of the underperforming strategies are benchmarked against US Large Cap comps like the S&P 500 where market breadth has been awful.

NVDA is up over 35% in the last month, contributing a little under 2.5% of the performance of the S&P 500 (+3.9%). If you’re underweight NVDA, and many IA strategies are, it’s mathematically hard to keep up. For the year, I believe NVDA has contributed something like 7% points to the performance of the S&P 500 (+14.5%), which makes me feel a bit better about IA’s strategies generally holding in despite the lack of breadth.

Headwinds can become tailwinds. If NVDA ever slows down, perhaps a dangerous thing to assume given where we are in the AI boom, Intelligent Alpha’s underperforming strategies could look a lot better.

To put more data on the breadth issue, the largest decile of companies by market cap in the S&P 500 have performed the best this year, and the bottom decline has performed worst. This performance dynamic obviously helps cap-weighted indexes.

Intelligent Alpha strategies aren’t cap-weighted, they’re AI weighted. Since the beginning, IA strategies have generally been somewhat underweight bigger companies and overweight smaller ones. For example, the IA Large Core strategy, which I comp against the S&P 500, has more than 60% of its portfolio by weight in stocks under $100 billion market cap. The S&P 500 has only 28% of its weight in those companies. That same IA strategy also has 23% of its weighting in companies in the bottom 40% of S&P market cap vs under 7% for the S&P itself.

Short term performance measurements are always fraught time periods to analyze investment strategies. Shorter durations highlight the randomness and noise that is smoothed over longer time periods. The goal of Intelligent Alpha, as it should be with all good investment strategies, is to add excess return in the long run.

Despite the tough month for Intelligent Alpha, I still have faith that AI is the future of the investment management world because it corrects the major flaws of our current options in passive and active management.

Passive Investing

The flaws of passive investing are:

• Fundamental misallocation.

• A lack of qualitative intelligence.

• The impossibility of alpha.

Misallocation of Indexes

Indexing may be smart in a world where most active managers can’t beat them, but indexes aren’t smart.

Traditional indexes misallocate capital because most use cap weighting. In a cap-weighted index, the largest stocks get the largest weight (e.g. AAPL in the S&P 500), and the smallest stocks get the smallest weighting, but market cap isn’t an accurate predictor of future investment performance. To the contrary. Large market cap is often a signal of higher valuation, and that may point to a headwind to future performance.

Rob Arnott built the Fundamental Index at Research Affiliates on the notion that indexes inherently misallocate capital. Fundamental Indexes use metrics like sales, cash flow, and book value to establish relative weights in the index rather than market cap. Arnott argues this reflects economic value rather than market value, but the more important point is that Fundamental Indexes decouple weighting from market cap to avoid the reality of larger stocks tending to get overpriced and smaller stocks underpriced in an index.

Smart beta is the way indexes try to improve investment decisions, but smart beta is still built on numbers. It doesn’t add any qualitative understanding of the companies in the index, which is the better way to predict the future.

Qualitative Intelligence

Qualitative intelligence about a business is the core advantage of the active manager. The active manager’s understanding of inherent advantages and disadvantages of a business helps him to predict a future for that business that cannot be predicted with historical data alone.

Indexes can never reflect qualitative intelligence because qualitative intelligence is not quantifiable. The best we can do is smart beta or quantitative investing, but the managers of those strategies will admit that their indexes and machines don’t know the difference between AAPL and APO. Nor do they care. Rules based systems, no matter how complex, do not understand the actual businesses they invest in, and that denies them the greatest absolute source of alpha.

Impossible Alpha

If you invest in a major index like the S&P 500 or Russell 2000, you can’t generate alpha. That’s the point of passive investing. You accept the market return for the comfort of knowing you won’t do worse than the average over time.

Smart beta and quantitative factors can help create relative advantages vs indexes, but they often come with periods of painful underperformance that seem a necessary quality of factor-based risk premia. If some factor outperformed the market all the time, it would get crowded out by the market. Factors are a strong tool for investors with an appropriately long-term oriented mindset, but straight indexes are probably better for most.

A dynamic approach that qualitatively understands businesses and dynamically moves to the best opportunities is the only way to achieve persistent alpha; however, that road is littered with many who tried and failed. Markets are stubborn. Humans are impatient. Creating persistent alpha is hard. That brings us to the failures of active management.

Active Investing

The flaws of active management are:

• Inconsistent returns.

• Mistakes driven by emotion.

• Durability of managers.

Inconsistent Returns

While you can’t generate alpha through passive investing, that doesn’t mean the returns are merely average. Passive returns are the average of all participants, but Patrick Geddes (founder of BlackRock’s Aperio) argues that passive returns aren’t an average outcome in the long run. Passive returns are actually well above average because indexes beat 90% of active managers over the long run.

I’ll reference the SPIVA report once again:

A human manager has a decent shot at beating the benchmark in any given year, but as the years compound, fewer and fewer managers are up to the task. Geddes’ insight is that by accepting passive returns, you guarantee a very good return vs most active alternatives in the market. You’re just giving up the chance for stellar returns that are consistently achieved by few active managers.

Emotional Mistakes

A core reason for inconsistent active returns is that humans make emotional mistakes. We capitulate in pain often when we should do the exact opposite. We pay attention to the short term, like monthly returns, even though our objectives are long term. We fixate on hockey stick stock charts, making us fearful of what might come next without any consideration for fundamental realities.

The greatest investors are great because they’ve figured out how to manage emotional pain. So many of Buffett’s famous quotes are about emotion, not investing:

• “Be fearful when others are greedy. Be greedy when others are fearful.”

• “The difference between successful people and really successful people is that really successful people say no to almost everything.”

• “The most important thing to do if you find yourself in a hole is to stop digging.”

A lot of investors like to think they invest long term, but they get hit for a year or two, or they miss an NVDA, and they go on tilt. It’s actually insane that Buffett, for 60 years, has managed to avoid the temptation of stocks that appear to present good 1-3 year prospects in favor of the far fewer opportunities that appear to offer good multi-decade prospects. Yes, Buffett has a lot of leeway given his track record, something many managers don’t have. He’s earned it over time.

Durability of Managers

Even if you find a great manager, the manager is still human, and you’ll have to contend with his durability. The manager might make so much money that his style changes and performance suffers. The manager might similarly get distracted by divorces, boats, and politics. Even if the manager avoids all of the temptations of life, and truly loves investing with plans to do it his whole life, he eventually either loses his faculties or leaves our world. Such is the harsh reality of nature.

AI-Powered Investing

AI-powered investing fixes all of the disadvantages of passive and active management:

• Better allocation. AI-weighted portfolios break the flaw of cap-weighted indexing. To the extent AI-powered portfolios weights are well-reasoned both quantitatively and qualitatively, they should better reflect the future prospects of those companies compared to indexes, smart beta, or quant investing. That should lead to better long-term results, although as we’re seeing now, it doesn’t mean AI-powered portfolios will win every single week or month.

• Qualitatively intelligent. AI, unlike indexes or quantitative models, can understand the qualitative realities of a company, which allows for better predictions of future performance. Quantitative value advocates may say companies that trade at lower prices should outperform those that trade at higher prices over time, I would counter that companies with high quality, durable businesses trading at reasonable prices relative to growth should perform even better. You can only find the latter company if you understand the business itself, not just the historical numbers it’s generated.

• Inconsistent Returns. AI-powered portfolios have the chance to generate more consistent returns than human managers because they don’t suffer from chasing near-term trends, style creep, the trappings of wealth, or the many other reasons that most active managers fail to beat benchmarks over long periods, especially emotional mistakes.

• Emotional Mistakes. AI doesn’t feel the emotional stress of markets — the desire to chase when you’re underperforming, the desire to doubt stocks flying but with strong fundamentals, the desire to give into pain. By avoiding emotional pull, AI eliminates the biggest source of inconsistent returns from active management.

• Durability. AI doesn’t get distracted by wealth, illness, or death. It’s eternally durable. A well-trained model with an enduring philosophy for allocating capital to great companies can do it as long as the AI lives on a server somewhere.

The point of all of these advantages is to generate persistent alpha. Indexes can’t do it, and most humans don’t. The goal for Intelligent Alpha is to shape it into the manager that lives in the top decile over time. The painful valleys are where learning happens. Learnings bring new insights. New insights drive improvements. Constant improvements support long-term results. That’s investing. That’s life.

The future of investing is still intelligent.

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Would You Die for Your Life’s Work?

There’s a 500-year-old chapel outside of Edinburgh in Scotland that’s rumored to be the burial site of the Knights Templar and the Holy Grail. Dan Brown featured Rosslyn Chapel in his 80-million-copy book, The Da Vinci Code. A stonemason even gave his life building the chapel.

Legend says the master stonemason went abroad seeking inspiration to create a column worthy of the chapel. While the master traveled, the apprentice had a vision for a column that he created in the years while the master was away. The apprentice’s column was beautiful. More beautiful than the master’s.

When the master saw the apprentice’s column, he flew into a fit of rage and killed the apprentice with a mallet. The apprentice did his life’s work and lost his life for it.

My wife and I visited the chapel this past week, and I couldn’t help but think about the stonemason giving his life for doing his best work. His life’s work is now immortalized in stone and story.

The idea of doing your “life’s work” has become almost a meme in the venture capital world. Every VC wants to invest in entrepreneurs doing their life’s work (us included). Who wouldn’t want to invest in someone doing what they were put on earth to do? Doing your life’s work grants you superpowers to sell, persist, and persevere, all qualities necessary for an entrepreneur’s success.

The masons of Rosslyn spent 40 years doing their life’s work, but the chapel’s enduring grandeur against hundreds of years of harsh Scottish weather makes me think about what exists beyond a life’s work: Timeless work.

True, Beautiful, Simple

A life’s work is powerful, but a timeless work is immortal. Immortality is something no human can have, which is why timeless work is so profound. It outlasts our mortality. Timeless work is determined by unwritten history. Only survival of time guarantees something to be timeless, and we of current time can’t accurately judge what will survive the future.

A life’s work is found through obsession. If you find something you love doing and can’t stop thinking about, it’s probably your life’s work. All timeless work is the labor of a life’s work, but merely working on an obsession doesn’t guarantee timelessness. It only creates the necessary conditions for it.

What then makes a work timeless?

A timeless work must be durable. Durability allows a thing to withstand the natural decay caused by time. It may come from a literal durability like the stonework of Rosslyn Chapel, or it may come from an intangible durability like that of a profound idea. A timeless thing doesn’t change. That’s the nature of timelessness. The thing’s existence literally endures the test of time.

Durability in both cases is built on the marriage of three things: truth, beauty, and simplicity.

Truth is honest because it cannot be false, so truth is undeniable and unavoidable. Untruth is destined to decay because it requires a defense from the confused or malicious. Truth defends itself naturally, and that makes truth the most durable thing of all.

Simplicity is not unavoidable like truth. It’s all too easily avoided in favor of complexity. Complexity makes things fragile by virtue of needing gentle care much like untruth. The world is not gentle, so gentle things do not survive it. Simplicity is durable because it eliminates the extraneous, and the fewer extraneous complexities, the closer something must be to truth.

Beauty is undeniable. It’s an innate quality that demands to be remembered and appreciated like the truth. Plainness begs to be forgotten. Ugliness demands to be reviled like untruth. Anything that can persist through the natural entropy of the world must be beautiful. The truth is always beautiful, and so is simplicity.

Timeless Work in the AI Age

A 500-year-old stone column has a natural timelessness about it. Physical things that defy the erosion of age can’t help but earn admiration. It’s why we visit old chapels, collect old cars, and even respect old people.

The same qualities that make physical things timeless make digital things timeless too — truth, simplicity, and beauty. The difference is that bits don’t age like atoms in the literal sense. Bits are merely information. Information represents ideas, and ideas age faster than anything in the physical world.

AI will exacerbate the ephemeral nature of information, but it might also add a new wrinkle of making ideas dynamic rather than static.

Just before we visited Rosslyn, I read that China is experimenting with an AI model trained on President Xi’s philosophies. The AI seems meant to grant Xi’s ideas a sort of immortality, but does that make them timeless or something else?

That brings us back to truth, the ultimate test of durability. To the extent Xi’s ideas are true they would be timeless regardless of whether an AI existed to perpetuate them. To the extent Xi’s ideas are flawed, AI can grant them a life well beyond what would be natural through intelligent defense, perhaps a defense even better than Xi himself could give.

AI doesn’t have to be built on truth, but the challenge of time means an AI built on untruth should ultimately succumb to an AI built on truth. It may just take a long time, and that time could be bad for humanity.

Elon seems to understand this better than anyone. His advocacy for a “maximally truth seeking” AI is another way of saying a timeless AI. A maximally truth-seeking AI would be the most durable AI, and that might be the most timeless invention ever. Hopefully no stonemasons are harmed in timeless AI’s creation.

Subscribe now

Intelligent Alpha is Timeless AI-Powered Investing

Most of us aren’t building foundation AI models like xAI’s truth-seeking Grok, but all of us will build on top of them over time. Some of us already are.

The rules of durability apply the ideas we use on top of foundation models. To the extent we build AI applications grounded in truth supported by beautifully simple structure, those applications have a chance to result in something timeless.

The inspiration for Intelligent Alpha is that there are some simple truths in investing that yield beautiful results if applied consistently over long periods. To the point of our exploration of timelessness — great businesses are timeless. Those timeless businesses that generate durable and persistently growing cash flows tend to be great investments. Most of history’s great investors apply this truth in some way.

The problem is most humans aren’t attuned to be great investors. We’re impatient. We’re emotional. We’re complex. In the absence of applying consistently good investment philosophy, the market has increasingly decided investing is too hard. More than half of global assets are now indexed — an admission that most human investors fail to beat the market with regularity.

Intelligent Alpha aspires to change that. I believe persistent alpha is available in markets at large scale via AI acting on the simplest truths about investing.

Results remain encouraging. Of the 40 IA strategies I’ve tracked starting 10 months ago, 80% remain ahead of benchmarks since inception. The average IA strategy is winning by 450 bps, and the median is winning by 310 bps.

There’s more to come soon on Intelligent Alpha, so subscribe to The Deload if you don’t already.

Investing is defined by our core topic of timelessness. It’s a game of compounding. Compounding happens only through the passage of time, and it grows stronger as time passes. AI is the ultimate compounding tool. It can apply truths found in markets more consistently and durably than humans.

That’s why I believe the future of investing is intelligent. Time will be the ultimate judge.

Subscribe now

Thoughts from The Deload’s archives on cultivating curiosity to find a life’s work:

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

The Dangers of Consensus

At the beginning of 2023, two-thirds of WEF Chief Economists said a recession was likely in the next 12 months.

We didn’t get a recession in 2023, and the stock market rallied in turn as the assumption of a hard landing turned into hopes of a soft landing. The failure of the recession consensus reminds me of Mark Twain’s timeless advice: “Whenever you find yourself on the side of the majority, it is time to pause and reflect.”

Here we are in 2024. The economy seems strong, inflation seems mostly contained, and we seem to be trending toward a soft landing with some rate relief later this year. At least for now.

Just this past week, job growth reports showed slower than expected hiring while unemployment ticked up to 3.9%, slightly higher than the 3.8% expectation. Could that be a harbinger for the long-delayed recession of 2023?

Maybe, but given the market rally after the weak job numbers, it seems the economic consensus for 2024 remains a soft landing.

Now let’s ask the Mark Twain question we should have asked in 2023. What if the thing that most people expect doesn’t happen? What if we actually get a recession this time?

Surprisingly, it wouldn’t change how I position my portfolio. Here’s why.

AI is Offense and Defense

In mid 2023, when the chances of recession seemed notably higher, I thought the megacap tech stocks were both an offensive and defensive way to stay invested in the market.

If you believed the economy would remain strong, the Mag 6 were positioned to benefit through continued growth in their retail, advertising, and software offerings with optionality from AI adoption. If you believed a recession was coming, the Mag 6 have fortress balance sheets and generate enormous sums of free cash flow each year. In a downturn, the megacap tech companies would be better positioned to weather an economic storm than almost any other company.

The offense and defense narrative still holds for the megacaps, which still aren’t that expensive at a 28.6x forward P/E despite the suggestion of many bears to the otherwise. We’re in a bull market. We’re meant to climb the wall of worry.

But the Mag 6 aren’t the only way to play offense and defense in the current economy. I now believe the offense/defense combination applies to the entire high-quality AI sector that spans both public and private markets.

Reimagine our two economic scenarios, soft landing and recession.

In a soft landing, AI investment continues unabated, and AI is an offensive play that remains on its current trajectory. Hyperscalers will continue aggressive investment in infrastructure. Alphabet, Meta, Microsoft, and Amazon all talked about stepping up CAPEX in recent Q1 earnings reports. Each of the hyperscalers will spend $10’s of billions this year to capture the AI opportunity. Many other Fortune 100 companies will follow somewhat further behind, and a broader set of Fortune 500 companies will grow experiments with AI to improve both productivity/margins and product/revenue. In a couple of years, we’ll start to see the fruits of these efforts pay off for both the providers of AI services and the companies that implement them.

Now let’s imagine we get a recession. Consensus would expect that to slow AI investment, but what if it doesn’t? What if a recession actually accelerates investment into and adoption of AI?

Maybe it’s the contrarian in me, or maybe it’s bias for my belief in AI, but there’s a logical case for why a recession might actually be good for a breakthrough technology that enhances productivity.

The hyperscalers probably invest through whatever economic environment because they have the resources to do so, and we’re in an arms race for the newest technological paradigm shift. It’s incumbent on the incumbents to win, so they must invest now recession or not. AI infrastructure should thus weather any downturn well.

It’s the other companies that might shift behavior in a recession, but I think they might accelerate instead of pullback. Fortune 500 companies largely have resources to invest through a recession too, and it would make sense to increase AI investment to support incremental revenue and productivity as they pare back headcount and other areas as is typical in a recession.

AI investors shouldn’t necessarily root for a recession, but we might not need to fear one either.

Regardless of which economic future we get, I still believe we’re only in inning three or four of the AI boom. Despite the rapid investment in infrastructure to support AI, we haven’t seen AI applications drive meaningful new revenue to most large players, and we probably won’t for the next several quarters.

Whether a stable economy means we stay on the same path or whether economic rockiness fast forwards us an inning, our ultimate destination is an AI bubble. That’s the natural outcome for every game changing technology, and AI will be no different.

Investing With AI in a Recession

While I think investing in AI makes sense whether we get a recession or not, I also believe investing with AI makes sense in the same regard. That brings us to my favorite topic: Intelligent Alpha.

Intelligent Alpha’s mission is to deliver consistently superior investment products powered by AI. To succeed in this mission, Intelligent Alpha’s strategies must beat index benchmarks. Since most human managers fail to beat indexes over the long run, AI wins against humans if they win against the benchmarks.

And Intelligent Alpha (IA) is doing just that.

More than 80% of IA’s 40+ strategies maintain a lead over representative benchmarks with about nine months of testing. The average IA strategy is winning by 370 bps. Amongst the selection of main IA strategies I track, the average win is 640 bps and the median is 620 bps.

The most frequent question I get about IA is some variation of, “Can you explain why AI is winning?”

A prospective client asked me this week if I thought beta could also be part of the answer to IA’s performance. I wasn’t sure since I don’t control for beta in building strategies. Instead, I have the AI committee focus on picking what it views as superior stocks. So, I ran the beta to see what the impact might be for a few of the main strategies.

The answer: IA tilts, in many cases slightly, toward a bit more beta than benchmarks, but not aggressively so. Cathie Wood’s flagship ARKK sports a beta of 1.89 vs the QQQ’s 1.18. That’s a heavy beta tilt. By comparison, the 30-stock IA Tech Conviction strategy has a beta of 1.24, nearly the same as the QQQ, and generated +23.7% since mid-September inception to today. Over that time, the QQQ is +15.2% and ARKK is +3.3%.

Like the IA Tech Conviction, many IA strategies have a beta that’s only a few points away from the index. On top of that, the tracking error for many IA strategies is relatively small vs benchmarks, particularly the strategies that are intended to be broader vs more concentrated. I’m particularly proud of the low tracking error those broader strategies show vs benchmarks which results in a strong information ratio given their outperformance.

Sam Altman recently declared that GPT-4 is the “dumbest model” we’ll have to use. Altman also said he didn’t care if he had to burn $50 billion, he’s going to build AGI. Models will only get better from here and quickly.

If Intelligent Alpha can achieve these results with dumb models, what will it do when we have smart ones?

The future of investing is intelligent.

Read more about Intelligent Alpha:

Read one from the archives:

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Good Products are Time Machines

Save me money, make me money, or save me time.

We all want more money and more time. Since money is stored time, what we all really want is more time.

Because time is scarce and precious, people are willing to pay for it, so building products around time is the surest way for a company to earn customers. All great products save us time in some way. The ability of great products to save us time makes them timeless.

Investment products may be the ultimate time saver. Great investment products are time machines that compound with the effect of saving more time over time. Nothing gives you more time than a great long-term investment.

The foundational belief of Intelligent Alpha is that AI will make investors more money than indexes and human managers in the long run. AI will win against indexes and humans because it fixes their flaws as investors. AI is unemotional unlike humans, and it’s capable of adding intelligence to structure unlike an index. That makes AI the ultimate investment time machine.

Despite my confidence in AI’s investment prowess, it’s not going to win every day or every week vs benchmarks. AI as a mechanism to make more money from investments is a time machine built on patience, as are all great investment products. The key with patient investment is that you don’t blow up along the journey, and it’s no different for AI.

How AI Acts in Volatility

“AI will get killed in a bear market.”

That’s a common concern I hear about Intelligent Alpha. The suspected frailty of algorithms and machines in market turmoil is fueled by past blow ups. LTCM collapsed legendarily in 1998, its models succumbing to massive leverage (250-1) when the market worked against it. The “Quant Quake” of 2007 levied pain on the entire model-driven quantitative finance sector as popular factor trades unwound. While the Quant Quake didn’t result in any explosive failures like LTCM, it highlighted the crowding that can happen in model-driven investment strategies.

Will we have future market turmoil caused by machines? Probably. Will Intelligent Alpha strategies be excessively harmed? Anything’s possible, but I doubt it.

Most Intelligent Alpha’s strategies don’t actively trade and don’t use leverage, eliminating two of the major issues for LTCM and the Quant Quake. That doesn’t mean my AI-powered strategies would survive unscathed; however, the strategies are built to play the investment game within an index-like structure + machine intelligence, so I suspect it’s unlikely that Intelligent Alpha would be down much more than representative benchmarks in times of turmoil.

Data is always a more powerful tool to convince than anecdote or logic. While we’re not in a bear market or serious period of market stress, we endured some solid market volatility over the past two weeks that gives us a window into how AI-powered portfolios might do in tougher market environments.

Perhaps unsurprisingly, anywhere tech was a focus, Intelligent Alpha portfolios were challenged vs benchmarks during the mid-April tech sell off. During the toughest week for tech, here’s what Intelligent Alpha’s performance looked like vs benchmarks:

The 12-stock Large Tech Focus strategy holds large positions in NVDA, MU, TSLA, and TSM, all stocks that suffered in the tech sell off two weeks ago. As a result, the strategy lost 450 bps of relative outperformance vs the QQQ during the tech sell off. The AI Average, which focuses on AI stocks, also lost ground to the QQQ. The less concentrated 30-stock Tech Select portfolio lost only 40 bps of ground to the QQQ. While the relative underperformance in the tech correction is disappointing, each of the Intelligent Alpha strategies maintains a healthy lead vs its benchmark.

We should expect portfolios focused on tech will have some level of incremental stress, I don’t know that the performance of the tech portfolios tell us much. The results in the cap strategies are more interesting.

The Large Cap Core portfolio, which is my direct competitor to the SPY, gained some modest ground vs the benchmark. The Large Cap Core has consistently been underweight tech since July 2023 inception, and it’s exceeded the toughest index in the world by 270 bps over that time despite tech’s huge contribution to the SPY’s performance.

The Large Cap Core Equal lost a bit of ground to the RSP ironically because the AI-powered portfolio has a bit more big tech exposure. Despite the tech turmoil, the Core Equal maintained a 750 bps lead on its benchmark since inception. Interestingly and before the tech sell off, my AI investment committee updated the Core Equal portfolio with just two conviction weightings where all three members of my AI committee nominated the stock: UNH and CI.

Finally, the Asymmetric Upside strategy did its job by adding 150 bps vs the SPY. It’s a large cap strategy that aims to pick stocks that have 2x upside vs downside potential. When everything is going up, that strategy should keep pace with its benchmark. When everything is going down, the strategy should outperform. In the end, that’s the point of every Intelligent Alpha strategy — to play within the rules of some investment game and try to create a persistent advantage vs its benchmark. The market turmoil shows that the strategies do that for better and worse.

Don’t Just Do Something, Stand There

Toward the beginning of Intelligent Alpha, I wrote about Charlie Munger’s legendary advice that investors should sit on their ass more. Do less.

Much of the insight from investment greats like Buffett and Munger comes from their understanding of human nature, not some unique perspective on finance. Great investors teach themselves to manage the many mistakes we humans are prone to make given our emotional nature. One way to avoid such mistakes is to just do nothing. Doing nothing isn’t always the answer, but it’s a good answer more often than most would allow themselves to believe.

Part of the reason indexing works is because it does nothing provided you leave it alone. AI can do the same. You can program AI to do nothing. The best part is that AI can do nothing with intelligence. It puts thought into its portfolios. Indexes just follow rules without intelligence. They do nothing, period.

As markets stabilized, Intelligent Alpha strategies did too by doing nothing. Those that lost ground largely gained it back. Those that gained ground largely held onto it.

It’s a small sample, so these are generalizations, but when you’re early to a new paradigm, everything is a small sample. Every new experience can only be a generalization. As we keep gathering more data and experience investing with AI, our conclusions can grow stronger. In the interim, the recent volatility and the performance of our AI strategies tell me that Intelligent Alpha probably won’t suffer excessively in times of stress, just as the structure intends.

If Intelligent Alpha can consistently deliver on generating excess returns over other investment products without blowing up, AI will prove to be a time machine for investors.

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

The Gateway to Obsession

“Please ask yourself, and I do mean, please ask yourself--this is not a rhetorical device. Can you force yourself to be interested in something?”

Tom Morgan, who writes , presented the question at this year’s Sohn Conference. Sohn is a gathering of some of the world’s best investors to pitch favored investment ideas. We can’t invest in forcing ourselves to be interested in something, so why would a room of investors care about it?

Putting Morgan’s intent into my own interpretation — investors are interested in being interested because when we get interested, we gain knowledge, and since interest is so individualized and so fleeting, being interested in something is a competitive advantage.

Many of the best investors are the most curious. See Charlie Munger’s career.

Curiosity is a gateway to interest. Interest is a gateway to obsession. Obsession is a gateway to conviction. Conviction is only needed for uncomfortable ideas. Uncomfortable ideas that end up being right generate extraordinary outcomes.

My two rules for cultivating curiosity are do it and share it.

This path from curiosity to conviction is the only path to doing extraordinary things in any field, but investing is the clearest application of the need for uncomfortable conviction. The markets by definition are a reflection of consensus. You must believe something wildly different from the herd to obtain a result wildly different from the herd.

Reverse engineering the process, no one takes action on an uncomfortable idea without conviction. You can’t find conviction without obsessing about something so you understand something about that thing that other people don’t. And you can’t obsess about something you don’t have a curious interest in.

So, can we force ourselves to be interested in something?

No.

We can force ourselves to try something, and we should keep trying things until we find conviction in some uncomfortable idea.

Someone Else Will Do It

To act on an uncomfortable idea is to build something new, often a business.

Anyone starting a business is bound to get the question: “Why doesn’t [insert giant company] just do it?” Many times it’s not even phrased as a question. Giant company X is going to do your idea. You don’t stand a chance.

I’ve heard that a hundred times about Intelligent Alpha (update below). Everyone in investing is coming to use AI, and they’ll crush you. Maybe so.

Every startup is a David and Goliath situation. The David startup is fueled by passionate curiosity that’s built conviction about an uncomfortable idea. Goliath has the resources and distribution to steal ideas rather than come up with his own. Who wins?

It depends.

Sometimes Goliath can leverage sheer size and might into subsuming some new market opportunity, but every David gets asked why Goliath doesn’t just take his idea, so Goliath has a lot of ideas to choose from. He can’t go after all of them. The non-obviousness of a good startup idea makes the idea seem even less appealing to Goliath. That’s one reason why obvious ideas have less alpha. More competition, including from the big guys.

David’s secret weapon is a unique insight about the world that Goliath doesn’t have. It’s the rock in the sling. The only way to get that unique insight is through being so interested in something that you know more about it than anyone else.

I’ve used the example of Kobe Bryant to describe what obsession looks like. His obsession with basketball granted him a conviction that most other professional basketball players never have: I can be the best in the world. His study of the game gave him the tools to deliver on it.

Think about how crazy that is. There are hundreds of players in the NBA. A couple hundred regular starters, and every starting caliber player is incredibly talented.

How many actually think they can be the best player in the world? Maybe a dozen?

A lot of guys might say they want to be the best, maybe even think it, but if they’re not obsessed, deep down inside they know they can’t be.

This analogy extends to everything else in the world. Who can be the best hedge fund manager? Only the guys who are obsessed like Steve Cohen or Stan Druckenmiller or David Tepper. Who can be the best bodybuilder? Read about Arnold’s obsession. There might be guys with better genetics, smarter guys, guys with more resources — the Goliaths — but if they don’t have obsession, then David has a chance.

There’s a pyramid of competition for everything. That’s the nature of humanity. The top of the pyramid is always dominated by the obsessed. The near top is filled with the naturally gifted or resource rich. The rest of the pyramid is everyone else.

Curiosity is the only path to the top of the pyramid.

Curious Machines

Morgan’s talk about curiosity didn’t just strike a chord with me because I’ve found a deep curiosity in using AI to pick stocks. It also struck me regarding another underlying question that ties AI and humans together.

If evolution’s answer for where we should direct our time is curiosity, can AI be curious? Does it need to be curious?

Maybe AI’s seeming lack of curiosity is what will ultimately prevent it from ever achieving human-like intelligence, which I view as different from human-level intelligence. Human-like means you think and learn like a human. Human-level means you’re capable of understanding situations in the real world and able to act on them like a human. The two concepts are distinct.

I think AI will reach human-level intelligence soon if it’s not already there, but to learn something on your own, you have to be curious. And it’s not apparent that machines have an innate curiosity. Instead, robots must be demanded to learn something by humans who are curious whether the robot can do it.

Creativity, community, and empathy were the things that I used to believe would permanently separate humans from machines, but maybe it’s actually curiosity. Curiosity is as close to a tangible soul or spirit that we have — an innate and invisible force that moves us. Those who never harness it never live. Those who find it can’t help but live. Perhaps it’s the same for machines.

Subscribe now

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

Intelligent Alpha: The Spirit of Modern Investment Management

Soul or no soul, spirit or no spirit, Intelligent Alpha is delivering better than human-level results as an investor, and it has been through ~9 months of testing. More than 80% of the 40 strategies I track are ahead of benchmarks by an average of ~390 bps.

If AI can be structured to be this effective at building investment portfolios now, imagine how effective it will be as it gets smarter. AI’s performance as an investor doesn’t need human-like intelligence to be a good investor. It might actually suffer with human-like intelligence.

AI’s performance feels almost unbelievable. Even to me as the experimenter. That’s the feeling of an insight that supports an uncomfortable conviction: The future of investing is intelligent.

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Overcoming Human Nature

AI has two distinct advantages in investing vs humans: It can process massive amounts of data, and it doesn’t have emotions. These advantages of AI have become undeniable to me since starting my experiments with Intelligent Alpha last summer.

While I’ve written a lot about AI’s super power that comes from lack of emotion, I haven’t spent as much time exploring its data processing super power. AI has an obvious advantage in data processing vs humans because our processing of data requires slow linear consumption and then thoughtful analysis of the limited information we consume. An AI can consume information in parallel fashion and do immediate analysis.

The human advantage is creativity. It’s abstract thinking. It’s in interacting with other humans to test ideas through different perspectives. The optimal state of data analysis to accelerate the information we interact with somehow to harness our competitive advantage. That’s where AI can come in.

On this episode of the Deload Podcast, I speak to Thematic CEO and co-founder Steven Carpenter. Thematic AI-powered investment analysis tools in the hands of all investors by making it easy to analyze stocks that fit certain quantitative or qualitative criteria via a simple natural language interface. Thematic wants to be the place where an investor starts any research process all powered by AI.

Then the company makes it even easier to build and track indexes or portfolios to test any idea. I’ve been using Thematic to track my Intelligent Alpha strategies since last year.

Here are my major takeaways from our conversation.

A Toy on the Tech Curve

AI naturally invites the jobs question. Will AI kill jobs? Will it kill my job?

Steve described three categories of people as it comes to AI:

• People scared of it.

• People skeptical of it.

• People embracing it.

Those categories hold true for all emerging technologies. Some fear that the new tech will make them irrelevant or change the world for the worse. Some think the new tech stupid because most new technologies start looking like a toy. And some embrace the new tech by playing with it despite its toy-like nature. That’s the early adopter.

It’s shocking to me how few people are really embracing AI despite the constant hype, but if everyone adopted some new tech, it wouldn’t be emerging anymore. It would be consensus.

Our conversation about embracing AI reminded me of one of my favorite blog posts from Bryce Roberts where he retold a story of a designer who told Travis Kalanick that she hated the Uber app, and that he should let her redesign it. He offered her a job, she took it. Lewis capped the story with this:

Most people want to be fit, most people aren’t.

Most people want to build a successful business, most people won’t.

Most people want to be the best version of themselves, most people aren’t.

Most people have dreams they want to fulfill, most people won’t.

Everyone wants to quit something, build something, be something, do something. Most people won’t.

How many things have we wanted? How many opportunities have we craved? How many broken things have we wanted to fix?

And how many of those have we shrunk from. Hid from. Or, excused away.

We’re not alone.

Most people won’t.

But every once in a while someone puts themselves out there. Makes the leap. Faces rejection or failure or worse. And comes out the other side. Better. Changed. Bolder.

Most people won’t. Which means those that do change everything.

Most people should embrace AI now. Most people won’t. Which means most people that do earn an unfair and growing advantage.

The Expert Fallacy

Conventional AI wisdom is that we’ll see verticalized LLMs that are trained for specific, narrow use cases — medicine, physics, mathematics, etc. While I think we’ll see verticalized LLMs for many fields, my counter intuition about verticalized LLMs for investment decisions is that it might result in worse outcomes than the generalized models we have today. Steve called this the “Expert Fallacy.”

Traditional active investing is more art than science. Whereas fields like physics and mathematics deal in absolutes, investing deals in abstracts. For AI use cases that deal in absolutes — flying a plane or looking at radiology scans — specialized models make sense. For artful use cases like investing, generalization may be superior to specialization. The greatest investors tend to be generalists that understand investing at a deep level, but they also understand many other things about the world at a deep enough level to make intelligent investments.

I believe there is an efficient frontier of knowledge for a given AI model to optimize investment decision making skill. Too much knowledge may be as bad as too little. The role of the human in an AI-powered investing world may be in assessing where AI models live on that efficient frontier for stock picking ability, just as asset allocators assess human managers today.

The future of investing is intelligent, and the humans who figure out how to use AI to their advantage will win.

Podcast Timestamps

1:40 The founding story of Thematic

6:40 Use cases for Thematic’s indexing and portfolio tracking platform

10:45 Using AI on Thematic and AI to create novel investment products

14:00 What makes AI a good tool for investors?

19:00 Why do different AI models think differently as investors?

24:30 How will banks and institutional investors use AI?

35:30 How will AI affect investment jobs?

40:30 Steve’s philosophy on building businesses

Subscribe now

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

The Deload explores my curiosities and experiments across AI, finance, and philosophy. If you haven’t subscribed, join nearly 2,000 readers:

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Machine Intelligence to Intelligent Alpha

Quantitative finance was the original AI-powered investor. Billionaire quants like Jim Simons of Renaissance Technologies, David Shaw of D. E. Shaw, and Cliff Asness of AQR proved that machines could find alpha in markets. And they keep proving it.

When institutional investors hear about Intelligent Alpha using generative AI to build investment portfolios, the first question is often how it compares to traditional quants. My answer: Intelligent Alpha invests, quants trade.

Quantitative finance is built on the idea that patterns in historical data can inform better investment decisions. Those patterns may be elaborate and intricate like those assumedly used by Renaissance, or they may be factors from academic research that explain some excess return. Whatever the pattern targeted by a quantitative model, the model doesn’t endeavor to know anything about the underlying companies of the stocks it selects. It doesn’t know that AAPL sells smartphones and AAN rents furniture. It doesn’t care. It shouldn’t care. The quant model just knows that one has great margins, a huge market cap, and poor recent momentum, and the other doesn’t.

If you don’t know what a company even does, you can’t really invest in it. A quant can only trade stocks based on a belief that some prior statistical pattern about them will repeat.

Modern AI creates the opportunity to combine the power of quantitative finance with a qualitative understanding of stocks. Intelligent Alpha convinces me that the generative AI breakthrough enables a new mechanism for generating alpha because it continues to perform well vs markets. My focus strategies are up an average of 570 bps over benchmarks since inception.

Without getting into the philosophical mud of what it means for an AI to understand something, Intelligent Alpha “knows” about the companies it selects. Certainly in any objective sense vs traditional quant models. My AI committee can explain qualitatively why it likes certain stocks if I ask it, and the explanations are plausible.

Based on how I build investment philosophies for my AI committee, I believe that qualitative factors represent more of Intelligent Alpha’s stock picking method than quantitative factors; however, the machine-based relatedness of Intelligent Alpha to quant strategies begs the question of how factors influence the results from generative AI’s stock picking.

Factor 101

Factors are a driving force in quantitative finance. A factor is a certain characteristic of a set of stocks that may explain differences in relative returns.

Legends of finance academia Eugene Fama and Ken French gave power to factors when they famously introduced their three-factor model: market risk, size, and value. Market risk is beta. Lower beta stocks counterintuitively tend to outperform higher beta over the long run. Size is the relative size of companies. Smaller cap stocks tend to outperform larger cap stocks. Value is a catch-all for valuation measures of a stock. Stocks that trade at lower multiples of some metric — sales, earnings, FCF, etc. — tend to perform better than stocks that trade at higher multiples.

Quality is another popular factor later added by Fama and French. Quality factors in that stocks with greater profitability tend to perform better than stocks with lower profitability over time. The other big factor is momentum. Momentum is the effect that stocks that are going up tend to keep going up and vice versa.

While there are other factors, the big five — market risk, size, value, quality, and momentum — dominate most popular quant strategies. Those factors also do a great job explaining why human managers generate excess return.

AQR did a study called “Buffett’s Alpha” in 2013 breaking down the factors that influenced Warren Buffett’s success. It showed Buffett’s returns were largely explained by his optimization of three factors — quality, low beta, and value. That doesn’t take away from his record, but it creates a specific narrative to potentially explain his success, paraphrasing an interview from AQR founder Cliff Asness: Buffett basically trained himself to be a machine that laser focused on factors with minimal variation over a multi-decade period.

Buffett was sort of a quant before we had computers who could do it for him. Maybe that’s why more of his writing seems to be about human nature than advanced finance.

Factors may help explain Buffett’s style, but it was his qualitative understanding of businesses that drove his decision making. Buffett understood the businesses he invested in, the power of brands, the power of great products with no substitution, and the utility of persistent need for those products (Coke, Geico, Apple). It so happened that those kinds of companies are high on the quality factor and low on beta. Buying them cheap was the icing on the cake.

Like Buffett, Intelligent Alpha leverages a qualitative understanding of good businesses. Like Buffett, Intelligent Alpha favors some quantitative factors in its process. Like Buffett, Intelligent Alpha executes on its favorite qualitative and quantitative factors with machine-like consistency because…it’s literally a machine.

Factors that Drive Intelligent Alpha

Since starting Intelligent Alpha in July 2023, the major factors have performed with decent variance vs the overall market.

The momentum factor (MTUM, dark green line in the chart below) delivered the strongest outperformance at +29.9% vs +18.1% for the S&P 500. Quality (QUAL, teal line) and growth (IWF, orange line) also modestly outperformed the S&P. Value (IWD, pink line), size (IWM, purple line), and beta (USMV, red line) have all underperformed the S&P 500.

I haven’t had the time or resources to do a regression analysis on the dozens of Intelligent Alpha strategies to estimate specific factor impacts, but there are some logical impacts given how the strategies are built.

Most Intelligent Alpha strategies use my base investment philosophy which combines quality and value, not unlike Buffett. I’ve always thought the most sensible investment philosophy is to invest in great companies at decent prices. Pure value never made much logical sense to me as an active manager. “Let me buy a bunch of fundamentally cheap and largely crappy companies with the assumption they’ll mean revert.” They probably will mean revert, but then you're stuck owning a bunch of cheap crappy companies waiting for mean reversion. Hasn’t worked all that well the last several years.

Value paired with quality always made more sense to me. “Let me buy a bunch of really good companies optimizing for those that seem to trade at the most reasonable prices.” Who wouldn’t want to do that?

Beyond qualitative and quantitative elements of quality and value, I don’t instruct the AI investment committee to optimize for any other factors unless a strategy necessitates it (e.g. the Intelligent Momentum strategies with momentum or Minimum Volatility strategies with beta). Any other factor influence is an unintentional byproduct of the AI investment committee’s individual stock selections.

Take size as an example. Other than restricting size in selection sets depending on a strategy’s focus area, Intelligent Alpha doesn’t optimize for smaller companies on the historical suggestion that they outperform over time. Instead, Intelligent Alpha’s variance in size from benchmarks reflects where AI sees opportunity whether bigger or smaller.

Take the Intelligent Alpha large cap strategies. The Intelligent Select has about half the average weighted market cap of its SPY benchmark, but the Intelligent Equal Select is almost 3x its RSP benchmark. Both strategies are ahead of their respective benchmarks despite different optimizations on size.

The Intelligent Small Cap strategies are both smaller than the IWM, and both are ahead of the benchmark. The Intelligent Mid Cap strategies are both bigger than the IJH, the Conviction largest of the three and showing meaningful outperformance. The Mid Cap Select is slightly behind its benchmark but catching up.

We can probably attribute some of Intelligent Alpha’s performance to the natural factors favored in my philosophies, but we can’t attribute all of it merely to factors. Quality has worked. Value hasn’t. Size hasn’t mattered. AI as a stock picker seems to have made much of the difference.

The Religion of AI-Powered Investing

All of the Intelligent Alpha strategies, to some extent, represent an intersection of quant, active, and index style investing, and each of those investing styles is its own religion.

Indexing is a religion of patience and rationality. If you don’t think you can beat the market, take what the market gives you and let it compound.

Active is a religion of personality and intelligence. The greatest investors are cult-like figures with deep insights about business analysis and human nature. Customers who invest in active managers believe the manager’s superior intelligence will result in superior results.

Quant is a religion of data and averages. Quant strategies should work as statistically expected given enough time, you just have to be prepared to ride out some storms.

So what’s the religion of AI-powered investing? It’s a faith in continually improving machine intelligence that adds smarts to passive indexes, removes emotion from active management, and incorporates qualitative realities to quantitative approaches.

Intelligent Alpha is the best of all worlds. The future of investing is intelligent.

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

A Timeless Question About Beating Markets

Good essays start with good questions. A good question allows the writer to explore something important and tell readers something surprising. Ideally the essay invites new questions worth further exploration (h/t Paul Graham’s latest greatest essay).

After explaining the potential for an AI-powered BlackRock two weeks ago on The Deload, a reader posed one of those questions worth further exploration: Is it possible to develop an AI-powered system that consistently outsmarts the market?

Consistently outsmarting the market is the point of Intelligent Alpha, and nearly nine months of data says AI can do it. More than 70% of my 40+ strategies are ahead of benchmarks, and the average strategy is ahead by almost 400 bps.

Time is the ultimate arbiter of consistency, so the only honest answer to whether AI will consistently outsmart markets is, “We’ll see.”

Yet, the question about AI’s potential for consistent outperformance sticks with me because it encompasses the biggest question I get about using AI to invest: Explainability.

Two endowments asked if I can explain how the AI works so they can feel confident in staying with an AI-powered strategy when it underperforms, just as they would scrutinize a human manager in the same way. Others have asked whether AI-powered strategies could get so big that it would eliminate the alpha. A few people have wondered how I can be confident the AI won’t blow up in a period of market stress.

All these questions can be answered by explaining how Intelligent Alpha works.

So I’ll reframe the question from whether AI will consistently outsmart markets to why AI has been outperforming markets. That’s explainability, and it also answers why I think AI will keep winning.

Breaking Down the Benchmark

When investors say they want to “beat the market,” they mean beat a benchmark of a particular segment of the market. A benchmark represents a structured opportunity cost for trying to generate excess return. If we can’t beat a benchmark regularly, we should just accept the market return.

Useful benchmarks play the game of investing in a knowable and consistent way. That’s what makes a benchmark a benchmark — consistent explainability, back to the subject of this post. If we understand the consistent process by which benchmarks are constructed, we can understand the levers for generating excess return.

How is a market benchmark constructed?

Let’s use the most popular benchmark, the S&P 500. The S&P 500 is created by three elements: rules, schedules, and opinions.

Rules define the universe of stocks relevant to the benchmark — how many stocks the benchmark will own, how the portfolio is weighted, and what factors or qualities the stocks in the benchmark should represent. The S&P 500 is a collection of 500 large cap US equities weighted by market cap.

Schedule defines how often the benchmark is reviewed. The S&P 500 is reviewed and rebalanced quarterly. Typically, the S&P committee replaces ~20 stocks in the index, 4% of the total, per year.

Opinion is the ultimate arbiter of what goes into benchmarks that use a committee to govern composition. The S&P 500 depends on a team of humans to select stocks for the portfolio given the rules and schedule. So too do the Dow Jones, S&P 400, and S&P 600. Other major benchmarks, like the Russell 2000 or Nasdaq 100, use only rules and schedules without a component of human opinion.

Per the S&P: “The committee focuses on an eligible company’s reputation, its history of sustained growth, its interest to investors, and its sector representation of the broader market.”

If the S&P 500 is our benchmark, now we know the game our competition plays. The seeming advantage to active managers is that they don’t have to play the same game. Active managers can play by a different rule set in terms of portfolio holdings or cap focus. They can review the portfolio on any schedule that they wish. They can opine on matters far beyond a company’s interest to investors and sector representation.

Is such infinite freedom a blessing or a curse for the active manager?

SPIVA data I’ve shared before shows that the majority of managers underperform cap benchmarks over any meaningful time frame. Freedom appears a curse to managers.

In his preface to the Intelligent Investor, Warren Buffett wrote:

“To invest successfully over a lifetime does not require a stratospheric IQ, unusual business insights, or inside information. What’s needed is a sound intellectual framework for making decisions and the ability to keep emotions from corroding that framework.”

I believe most investment managers fail to outperform benchmarks over the long run because they don’t have the emotional temperament necessary to control corrosive urges. Infinite freedom invites more decision making, and more decision making allows for more emotional influence. Active managers might be better to apply rules and schedules to limit decision making, but that would be antithetical to the belief that human opinion is the ultimate path to alpha.

AI suffers from no emotion, nor does it chafe at rules and schedules to constrain its decision making, and that’s where AI’s advantage begins. The Intelligent Alpha process starts with rules and schedules to enhance the odds of AI’s smarts at stock picking to shine through.

AI + Structure for the Win

The first step in creating an Intelligent Alpha strategy is to decide what game we want to play. Maybe we want to try to beat the S&P 500, or maybe we want some aggressive take on the QQQ. By defining what game we want to play, we define our benchmark, and we therefore know what rules, schedules, and opinions we’re up against.

Given our benchmark, we can establish a set of rules for Intelligent Alpha. There are now over 45 Intelligent Alpha strategies built around the same general set of rules:

• Target exposure: market cap, sector, factor, theme, etc.

• Target number of holdings: Select (generally 100+ stocks), Conviction (30 stocks), Custom (other defined number)

• Weighting: Select (decided by AI committee), Conviction (equal weighted)

The Intelligent Equal Select strategy is one of my favorite examples of rules enhancing a strategy.

Equal Select is a large cap focused “equal” weight strategy that uses my AI investment committee (GPT, Gemini, Claude) to weight the stocks in the portfolio in a flatter way than the S&P 500 but in a more dynamic way than the S&P 500 Equal Weight index. Each AI committee member selects up to 100 stocks to include in the Equal Select strategy. Every vote carries a base weight equal to one divided by the total number of stocks nominated by the AI's including duplicates. If a stock gets selected by all three committee members, it gets 3x the base weight.

The rules of the Equal Select strategy plus the AI investment committee’s creative stock picking have been impressive. The Equal Select is ahead of the RSP (S&P 500 Equal Weight) by 750 bps since inception. Even more impressive, the Equal Select is ahead of the S&P 500 itself by 230 bps while the RSP trails the S&P by 520 bps.

Once we have rules, then I define the Intelligent Alpha strategy’s schedule. I use simple bi-annual, quarterly, or monthly portfolio reviews. The intent of Intelligent Alpha isn’t to unleash AI to make an infinite number of on-going decisions. We want AI to use its unemotional intelligence to pick a set of companies within the given rules that represent good medium-to-long-term investments.

That brings us to opinion, or in this case, intelligence.

Intelligent Alpha strategies are driven by the opinion of my AI investment committee. The committee acts with a united philosophy inspired by many of history’s greatest investors which is responsible for alpha inception to date.

One of the strategies I’m proudest of, although it doesn’t show the highest overall alpha, is the Intelligent Select. It’s a broad based core portfolio intended to compete directly with the S&P 500. It holds 181 stocks. It’s currently underweight the Mag 6 stocks (MSFT, AAPL, META, NVDA, AMZN, and GOOGL) by 1340 bps. And it’s beaten the S&P by 300 bps since its July 2023 inception. I think that’s incredible performance given how much of the S&P’s total returns have been driven by the Mag 6 over the past year plus.

Results suggest AI is a good stock picker. If AI were picking bad stocks, rules and schedules wouldn’t save it. Rules and schedules let AI’s stock picking power drive results in scalable and repeatable ways that are explainable…for the most part.

Magic

Arthur C. Clark’s third law states: Any sufficiently advanced technology is indistinguishable from magic.

Investing with AI feels like magic to me, and some of it is. While I can explain the rules and the schedule and the philosophy I program into the AI, I can’t explain to you exactly how or why the AI picks the stocks it picks. Not even the creators of language models can explain exactly how the systems work.

AI-powered investing is a bit of magic, and like all forms of magic, you may need to suspend a bit of your desire to explain the trick to enjoy the show.

Ultimately the performance of Intelligent Alpha’s portfolios will be the most convincing evidence, and we’re back to the original question of AI consistently outsmarting markets. Ironically, the search for an investment strategy that can consistently beat markets is timeless because we’re always seeking the answer, yet the question can only be answered with the continual passing of time. Perhaps that means we never get a satisfying answer.

All I know for certain is that AI’s advantage of ever-improving and unemotional intelligence paired with rules and schedules give it the best odds to figure it out our timeless question of consistent outperformance.

The future of investing is intelligent.

Subscribe now

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Inspiration from Elon

In 2006, Elon Musk shared his Master Plan for Tesla::

1. Build sports car

2. Use that money to build an affordable car

3. Use that money to build an even more affordable car

4. While doing above, also provide zero emission electric power generation options

With 17 years of hindsight, he executed his plan flawlessly. He built the Tesla Roadster. He built the Model S. He built the Model 3. He built grid storage options with the battery tech developed from the cars.

Now Musk is executing on his 2016 Master Plan with goals to expand Tesla’s product line to address all major segments, develop self-driving 10x safer than humans, and to enable your car to make money while you aren’t using it. He’s on his way to delivering on that plan as well.

Before a Master Plan for Tesla, Musk had a Master Purpose: to expedite the shift from carbon-based energy to renewables. A plan without a purpose is just a set of goals. A plan with a purpose is a simple, clear guide to navigate building something new and important. The simple clarity of Musk’s Master Plans inspired me in thinking about how to build a new business, especially one that aims to revolutionize an industry like Tesla did automotive.

I’ve been working on a Master Plan too, and I’ve taken inspiration from the simple clarity of Musk’s Master Plans. My plan is for Intelligent Alpha because I’m convinced AI-powered investing is the future of asset management.

Just between me and you, here it is.

The Master Purpose

Jeff Bezos famously said that you should build businesses around what won’t change, not what is changing:

In our retail business, we know that customers want low prices, and I know that's going to be true 10 years from now. They want fast delivery; they want vast selection. It's impossible to imagine a future 10 years from now where a customer comes up and says, 'Jeff I love Amazon; I just wish the prices were a little higher,' or 'I love Amazon; I just wish you'd deliver a little more slowly.' Impossible.

For Amazon, the Internet was just a mechanism to deliver what customers always wanted — low prices, great selection, fast delivery — in a new format.

The things that won’t change in asset management are as obvious as those in retail. Investors will always want better returns and fair prices. To invoke Bezos, it’s impossible to imagine an investor saying, “I wish I earned a little less on my investments this year.” Never going to happen.

AI is destined to generate better returns than indexes or humans. Indexes can’t beat themselves, and humans can’t beat the indexes consistently.

So, it’s up to AI to deliver on the unchanging desire of investors, and it can deliver. I’ve seen AI’s superiority to indexes and humans over months of data now. Of my 15 focus Intelligent Alpha strategies, 80% are ahead of benchmarks and winning by an average of 430 bps.

AI beats indexes by adding basic intelligence to long-term oriented investment strategies, and it beats humans by simply not being human — eliminating the emotional and cognitive mistakes we make as investors. AI doesn’t worry about timing markets, it doesn’t feel a stock blowing up in its face, nor does it sense the gnawing need to act as a human manager. Lack of human emotion gives AI a persistent edge in investing, and that edge will only grow wider as AI models get more capable.

Given the unchanging desire of investors for better returns at fair fees, the Master Purpose of Intelligent Alpha is clear: Lead the shift in asset management from a world that relies on indexing and human managers to one that relies on AI.

The Master Plan

Elon’s purpose of accelerating the transition to renewable energy demanded the creation of America’s first new car company in decades, Tesla. Intelligent Alpha’s purpose demands a new asset management company. There are only nine asset managers with more than $2 trillion under management. BlackRock is the largest at over $10 trillion.

To create a world that relies on AI for investing, we need to create an AI-first BlackRock — a multi-trillion dollar asset manager that offers a full range of investment products.

Here’s my Master Plan to do it:

1. Develop a system for using AI to make long-term investments that beat benchmarks.

2. Create private funds to test the AI-powered strategies.

3. Create broad access funds to bring AI-powered strategies to the masses.

4. Use profit generated from step #3 to launch more AI-powered funds.

5. Repeat until Intelligent Alpha powers > $1 trillion in investments.

If you break down these steps, the Intelligent Alpha Master Plan is simply to create a superior product (steps 1-2) then build a sales engine to market and distribute it (steps 3-5). That’s the soul of every business’ Master Plan.

Product

Intelligent Alpha is already a superior product. The results inception to date convince me of this, but superior products are never finished. A finished product is ripe for disruption and replacement. To invoke Musk again, the only real moat is pace of innovation.

I believe Intelligent Alpha is 6-12 months ahead of anyone else who might try to figure out how to use generative AI to invest. To stay ahead, we just have to keep moving faster than everyone else. That will require persistent experimentation with the top generative AI models in the market, a deep understanding of how those models “think” as investors, and an ability to explain to human allocators what to expect from the various AI-powered strategies Intelligent Alpha creates. Those three ingredients are ultimately the Intelligent Alpha product — a system that maximizes the potential for alpha generation by using the intelligence of modern AI to make investment decisions for any portfolio purpose.

When I think about equity portfolio construction, which is the focus with Intelligent Alpha, allocations fall into three broad buckets:

• Core: Broad market beta like S&P 500, Russell 2000, and other index products. Core products create the core exposures in equity portfolios and are most commonly accessed through indexed ETFs. The biggest total asset segment lives here with the lowest fees.

• Performance: Strategic allocations that target sectors, factors, and themes with the potential for alpha. Performance products may be indexed or active and come with higher fees than core exposures.

• Exotic: Unique investment exposures like long/short equity, special situations, quantitative funds, etc. Exotic products generally target uncorrelated investment returns and are often limited to high net worth clients through limited partnership structures. They often have the highest fees and may include performance fees.

To deliver on the Master Plan to offer broad access to AI-powered investment products, Intelligent Alpha’s products are structured around these major asset buckets. We have almost 10 Core strategies, 30 Performance strategies, and a couple Exotic ones with more in the pipeline. The product is ready.

Sales

Product is critical for a new business. It’s where most entrepreneurs start, but too few think about building a sales engine.

Selling, in all forms, is the conversion of attention to interest to purchase. That’s the engine.

The most powerful tool for an investment product to capture attention and ultimately convert it to purchase is a track record. Prior results are never a guarantee of future success, but they are a fact that some strategy at least worked in the past.

Intelligent Alpha’s 40+ strategies have a soft track record of 6-7 months of solid performance. We’ve been tracking those as indexes. If AI can keep it up, a strong and marketable track record will come in time.

In the absence of a marketable long-term track record, the next best thing to do is create a cult.

Many successful startups create cult followings. The companies foster passionate customer bases who share a common secret delivered by the product. In many cases, that secret powers a part of the customer’s personal identity, making the customer powerful advocates for the product — an unpaid sales team. By creating a cult, a company can short circuit common objections to new-to-market products.

The most inspirational finance cult for Intelligent Alpha is Vanguard.

Jack Bogle built the cult of Vanguard on a simple secret: Accepting market returns in low-fee vehicles was a smarter decision than investing in high fee active products that often underperformed the market. Vanguard customers who embraced Bogle’s secret enjoyed a communal superiority over other investors by enjoying better returns with minimal fees.

Who wouldn’t feel like a genius by getting better investment performance with lower fees?

AI-powered investing is destined to be the next cult in the investment world. AI-powered investing creates a natural us vs them — believers in the power and promise of technology vs the luddites who just want market returns or maybe even worse from human managers. Just like Bogle’s indexing, either you immediately get that AI will provide the greatest chance of persistently outperforming markets through superior intelligence, or you disbelieve the machines will take over. If you believe in the power and promise of AI, there’s only one choice.

No One Else is Coming, It’s Up to Us

I’ve been part of thousands of venture pitches. One of the most natural questions is, “If this is such a great idea, why won’t the big incumbents just do it?”

So, why won’t BlackRock build the AI-powered BlackRock?

Messiness, data, and desire.

No one has experimented with using generative AI to build investment strategies as I have, and it’s messy. It requires adjustments and adaptations that may change daily. It requires creativity in structuring prompts. It requires creativity in structuring strategies. Incumbents often avoid the messiness of new markets and new technologies because they have franchises to protect and other priorities that can grow more easily. If generative AI ever reaches the point where it’s easy to use for building investment portfolios, it will be too late. The procedural messiness tells me the timing for Intelligent Alpha is just right.

Investment industry people curious about generative AI as an investment tool are too data focused. The power of data to create trading strategies is apparent in the success of the best quant investors, but I don’t believe data will be as useful a moat in creating the AI-powered BlackRock. Data necessarily limits how wide a market an investment strategy can address. The more narrow and unique the data, the more narrow and unique the product to come from it. Massive asset management platforms aren’t built on data-driven investment products. Vanguard, BlackRock, State Street. They’re built on simple indexes and infinitely sliced versions of different market segments. If you want to build a niche AI-powered investment product that might generate fantastic returns, build it with data. If you want to build a broad AI-powered investment product, you need to do it beyond data. I have a plan for that too, but I’m not sharing it yet.

The last reason incumbents often miss new opportunities is because they don’t have the same desire as I do. Are they experimenting with generative AI? They should be. Do they have someone who is as curious and passionate about it as me? Probably not. Do they have a mission oriented around transforming the investment world to one powered primarily by AI? Certainly not.

And that’s how innovation happens.

For as much as AI will do for us, it doesn’t yet replace the drive and determination of a human being who believes in a different future. I’m convinced the future of investing is intelligent. Now it’s time to prove it.

Follow along here for the journey.

Subscribe now

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

Google generated ~$700 billion from search over the past five years. I know because I looked it up on…

Google.

It was the 10th blue link in Google’s answers.

Let’s round up. Search is a trillion-dollar prize, and it’s attracted dozens of competitors over the years. Google’s most recent test comes from prior challenger Bing, now powered by OpenAI, and Perplexity. OpenAI is also making direct efforts in search. The growing narrative is that Google is in trouble.

The funny thing is I tried the $700 billion dollar query about Google search revenue on Bing CoPilot with GPT and Perplexity before I went to Google. Neither gave me the answer. Google “knows” everything, more so than any AI-based challengers. The company has been organizing the world’s information for two-plus decades, and it’s pretty damn good at it.

But this post isn’t a takedown of Perplexity or Bing, nor a defense of Google. My goal is to objectively look at our trillion-dollar question: Is Google in danger of dying to AI?

To consider the dangers to Google from AI, I started with a first principles review of why we use search.

My Framework to Understand Search

Search usage fits in a four quadrant matrix. The axes of the matrix are (1) whether the search user has a specific or vague desire and (2) whether that desire has a single answer or many potential answers.

Given those criteria, we’re left with four kinds of search:

• Information search: a specific desire and a single answer. What is the capital of Pennsylvania?

• Narrowing search: a specific desire with many potential answers. Where should I stay on vacation in Costa Rica? Product commerce searches largely happen here.

• Question search: a vague desire with a specific answer. Why won’t my car start? In question searches, you’re one or more layers away from the actual question you need to ask. If your car won’t start because your spark plugs are shot, you need to ask how to replace the spark plugs. Service commerce searches largely happen here because service providers figure out the right question for you, then fix the problem.

• Curious wandering: a vague desire with many potential answers. The “I’m bored” category. Curious wandering is a form of search, but social fills the need more than Google. You go to YouTube, Instagram, or TikTok to find something to fill your boredom.

Each of these four forms of search are a form of conversation. Information searches are simple Q&A with a single answer. Exploration and question searches are a one-on-one dialog. Curious exploration is like shooting the shit with a bunch of friends.

Setting aside curious wandering for a minute, information search is the destination of all the other forms of search. When you finally figure out the right question in a question search, that’s an information search. “How do I change the spark plugs in a 2019 Ford Explorer?” When you figure out what hotel you want to stay at on vacation in a narrowing search, that’s an information search. “Four Seasons Minneapolis.”

Even the curious wandering of YouTube and social media can result in a form of information search. Influencers influence fans to buy products, although links to those products are often right in the social app itself. If not, users make the trip to Google. “Logan Paul PRIME energy drink.”

Please don’t go search that.

This is the threat to Google: ChatGPT and LLMs are designed as conversationalists. They’re built for back-and-forth dialogue in a way search engines are not, and the dialogue searches are the most valuable ones.

Show Me the Money

When you know the right question and want a single answer in an information search, 10 blue links is a suboptimal output. Google already knows that.

If you ask Google the capital of Pennsylvania, it just tells you Harrisburg. It’s been doing that for years.

Google also knows real money in search is in dialogues that narrow you focus on specific products or services that answer questions you don’t know how to ask. A website or video that shows how to change spark plugs isn’t as valuable as a click to call a mechanic. Nor is searching Four Seasons Minneapolis. There’s a sponsored link, but brand driven ads aren’t as valuable as the more competitive ads of a search of “4 star hotel Minneapolis.”

Logically this makes sense. If you don’t know exactly what you want or need, advertisers want to pay to influence you. That’s the trillion-dollar reality of search.

For a couple decades now, Google’s one-way, open-ended dialogue via the search box was the best product for consumers hone in on what they want or need. Google will give you a bunch of links that might help, some sponsored, but the answers require a self-directed exploration. If one link doesn’t get you closer to the answer you’re looking for, you go to the next, and the next, and so on. Although Google can usually get you to what you want or need, it isn’t always an efficient experience. Too much work for the human, not enough by the machines.

That’s where AI comes in.

You can ask ChatGPT anything, just as you can Google, but Chat GPT will guide you through a conversation rather than let you explore yourself. More work by the machines, potentially less by the humans. The change from one-way, open-ended dialogues with a search box to guided dialogues with AI is the threat to Google because that experience creates a potentially more compelling way for consumers to engage in commercial searches.

If tell ChatGPT I’m looking for a 4 star hotel in Minneapolis, the AI could ask whether I want to be Downtown or in the North Loop. It could ask what kind of gym I want. Whether I care about other amenities. Through conversation, rather than self-directed exploration, I narrow my search to a specific hotel.

Even more amazing, consider the question of why my car won’t start. An AI might be able to listen to me try to start the car and identify certain issues by sound. Then the AI could suggest either how to fix it or point me to a mechanic that can handle the problem.

As an experience, a conversation with an intelligent machine aimed toward some specific end answer is better than hunting and pecking through 10 blue links. Google knows this too, and it’s been experimenting with combining generative AI and search. The company has been playing with Search Generative Experience (SGE) since mid-2023. It also just launched a revamped, although I still think subpar, chatbot now named Gemini.

Google has been playing in AI, it’s just been maddeningly slow.

Google’s Dilemma

The challenge to Google is not whether it can give good answers or even create chat-based experiences. It’s whether Google is willing to radically evolve its core experience — one with which billions of daily customers are comfortable and familiar.

The pace of Google’s AI innovation suggests they’re erring on the side of caution, and it’s frustrating as both a shareholder (Deepwater currently owns the stock) and a tech enthusiast; however, my review encouraged me to think about the issue from the company’s perspective. Why so slow?

Innovator’s Dilemma is one answer I hear often. Google doesn’t want to kill the trillion-dollar golden goose of search, particularly the questions that are better answered through AI-based dialogue. Just leave it alone and keep printing money.

I don’t think that’s the full story.

Google has a couple billion daily users. Most of those users have been using Google for years. They expect 10 blue links. Suddenly annihilating that experience in favor of a ChatGPT-like experience would confuse Google’s couple billion daily users who aren’t tech-savvy like you and me. A large number of those users have probably never even tried ChatGPT, nor do they care about generative AI.

Google’s slowness in AI seems to exemplify a company stuck in the middle of transition. My interactions with Search Generative Experience range from a fine evolution of search to a mess that tries to bridge the old link-based world with the new gen AI world too carefully.

I tested an SGE query for a hotel in Edinburgh. The result was a long explanation of many hotels, tons of paid links, and some maps. It was an overwhelming amount of information, beyond what I would have gotten in the old 10 blue link world. It was a bad experience. Maybe that’s why SGE is rolling out so slowly. If a non-technically oriented user got the SGE results I just described, they’d wonder what the hell happened to their beloved Google, although I doubt the average Google user would leave for ChatGPT any time soon.

Despite the truth that Google has a massive, loyal user base that it needs to manage through the transition to an AI-first world, Google lacks the urgency and innovative flair that it faces from OpenAI. Where OpenAI seems to be achieving months of progress in weeks, Google seems to be achieving weeks of progress in months.

Google should be more aggressive in courting tech-first users who want the full power of AI rather than hedging with products to appease the entire user base. Google should also move fast and break some things, something OpenAI doesn’t seem afraid to do.

Google is the clear #2 in AI behind OpenAI, but they should be #1.

Setting Odds

Here’s the most reasonable view of Google’s situation with AI:

There’s a real chance Google loses the search war to OpenAI or some new competitor, but OpenAI bulls and Google skeptics ascribe too high a probability to that outcome.

Google has the most data information (20 years of search), it has resources to build the products, and it still has world class distribution (Android, Chrome, YouTube). The most likely scenario is that Google maintains its search lead through that superior distribution, user loyalty, and by building an AI-powered experience that’s at least as good as competitors.

Maybe Google loses a bit of share on the edges. Maybe it means the search golden goose loses a little luster by virtue of different monetization or a different cost structure. But it probably doesn’t mean that Google’s dead.

As a shareholder, at least in the medium term, these reasonable probabilities might not matter.

The perception is that Google is moving too slow, and perception is reality. There’s a long road to Google losing dominance in search, but there’s also a long road to proving that they won’t. There’s a danger that Google becomes the Microsoft of the late 2000s-early 2010s if it hasn’t already. Stagnant. Boring. Fat and happy. No hunger. At risk of missing the next big thing (mobile in the case of Microsoft).

Bookies should still set the odds of winning the search war in Google’s favor, but it’s harder and harder to be excited about taking those odds.

Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.

What Search Can’t Do + AI Can (Intelligent Alpha Update)

While AI has the potential to redefine how we find information, the more exciting opportunity for AI is in creation.

Search can’t create anything. It can’t think. It can’t hallucinate like generative AI, and that’s not a compliment. AI hallucinations are the manifestation of creative potential. Search can give you some links to a bunch of junk some human hallucinated on a website, or maybe now some junk an AI hallucinated on a website, but search can’t make anything up for itself.

Creativity is the application of intelligence, and machines have growing capability for creativity and intelligence.

My favorite application of AI’s creativity is Intelligent Alpha where I use generative AI to create investment strategies. Intelligent Alpha’s outperformance vs markets since inception remains strong with 12/15 core strategies I track ahead of benchmarks. The average core strategy remains +400 bps vs respective benchmarks.

Two observations about the performance of strategies so far in 2024:

• The small cap strategies have trailed the Russell 2000 benchmark for a funny reason: SMCI. SCMI, a company that builds AI servers, is up almost 200% YTD even after being down 20% on Friday. That stock is still in the Russell 2000 despite being over $40 billion in market cap, and it’s powered 75% of the Russell’s performance YTD. SCMI is not in my small cap strategies because it hasn't been a small cap for a while. Sometimes benchmarks get a unique advantage. Intelligent Alpha will still win over time.

• The long/short strategy has been great with shorts YTD. The most recent short win was ROKU, which reported disappointing earnings this week. Intelligent Alpha also had BA, MBLY, and KHC as shorts in the quarter. January as a particularly strong month for the strategy. February, despite the strong shorts, has been a little weaker due to some of the strategy’s longs.

The great thing about experiments like using AI to challenge markets is that the experiment never ends. Markets always evolve and change. AI will evolve and change with markets. It will get smarter over time, and I believe it will end up the future of active portfolio management.

The future of investing is intelligent.

Subscribe now

Subscribe now

Disclaimer. The Deload is a collection of my personal thoughts and ideas. My views here do not constitute investment advice. Content on the site is for educational purposes. The site does not represent the views of Deepwater Asset Management. I may reference companies in which Deepwater has an investment. See Deepwater’s full disclosures here.

Additionally, any Intelligent Alpha strategies referred to in writings on The Deload represent strategies tracked as indexes or private test portfolios that are not investable in either case. References to these strategies is for educational purposes as I explore how AI acts as an investor.

AI > Humans

Charlie Munger loved the analogy of the one-legged man in an ass-kicking contest. He used the analogy often to describe people who fail to learn basic knowledge to be good at their profession. Since most people don’t spend much time learning, they’re destined to get a beat down.

One of my favorites from Munger:

“If you’re going to live a long time, you have to keep learning. What you formerly knew is not enough. If you don’t adapt, you’re like a one-legged man in an ass-kicking contest.”

Being a good investor demands a life-long approach to learning. Markets change. Opportunities change. Even the tools we use to invest change, with AI coming to be the biggest change to markets yet.

AI is like a 50-legged man in an ass kicking contest. Two legged humans don’t stand a chance to beat AI at investing, nonetheless a one-legged one. I know because I’ve been testing it.

I have half a year of data about Intelligent Alpha now, my system for using generative AI to challenge markets. Results remain great. More than 75% of all the strategies I track are ahead their benchmarks since inception. My average core strategy is +4.2% vs respective benchmarks, and the median strategy is +2.7%.

But Intelligent Alpha doesn’t just aim to outperform indexes. It also aims to be better than humans in direct competition. I’ve been testing three such direct competitions:

• The IA Large Tech Focus vs a Pure AI version. It’s a concentrated portfolio of 12 large cap tech stocks that aims to challenge the QQQ, and it’s +16.7% vs the ETF since inception. The strategy combines human intelligence with Intelligent Alpha. My AI committee picks the stocks that can go into the portfolio, and their votes determine weighting, but I decide when the committee should review portfolio and have final approval over new stock additions. I’ve also tested a version of the Large Tech Focus strategy where I have no input other than prompting, so it’s purely the output of my AI committee.

• IA Human Comp A (I’m using this name to keep the underlying fund anonymous). The underlying fund is a growth-focused strategy managed by a team of humans with a long-term track record. I use a representative universe of stocks from the strategy for my AI committee, allowing it to pick and weight the AI-powered version of the portfolio.

• IA Human Comp B. The underlying fund is a value-focused strategy managed by a team of humans with a long-term track record. I use the portfolio of stocks from the underlying fund as the universe for my AI committee, allowing it to pick and weight the AI-powered version of the portfolio.

In every case, AI is outperforming humans since inception, and by a decent amount too.

Why? Is AI’s dominance sustainable? My thoughts…

Data > Feelings

The biggest advantage AI found vs the humans in every strategy was embracing mega cap tech and semiconductors. In both of the growth and value strategy comps, AI went heavy into GOOGL, META, AMD, and semi cap names like LRCX and AMAT.

AI favored the tech names to insurance companies (BRK and DHR), banks (JPM), and medtech (MDT) in the value fund. It favored tech to retail and consumer names (FIVE, ALGN, ETSY, SQ) in the growth fund. While many of those companies are high quality, few of them come with the growth of the tech names, and in some cases their multiples are even higher.

In the Large Tech Focus strategy, my AI committee universally loved NVDA, making it a 16% conviction position in the pure AI version of the strategy. In the human version of the strategy where I add my supposed intelligence, I bumped NVDA out of the portfolio because it felt too hot. That difference represents a significant part of the pure AI’s alpha vs my human-edited version.

While humans are stuck thinking about how the megas just enjoyed an incredible rally in 2023 and how it can’t possibly happen again, AI only looks at the qualitative and fundamental realities of those big stocks. And the truth?

The mega caps aren’t that expensive.

Before I even pulled the mega cap valuation data, I generally knew that the mega were trading around 30x forward FCF. I knew that meant the group was yielding around 3% vs the 10-year Treasury at ~4%. I knew that it’s hard to say that the mega cap group is “super expensive” in that context. Is it really that crazy for the best companies in the world to yield 3% while growing earnings mid-teens while the 10-year hovers around 4%?

But that’s AI’s undying advantage.

Even as I knew all of the valuation realities of tech, I’m still hesitant to add more mega cap exposure because [caveman voice] “price move too fast.” AI doesn’t suffer such first level thoughts. It just puts strong investments in its portfolios. No over thinking.

Death of the Middle

Chris Dixon of A16Z described a concept called the “Death of the Middle” on David Perell’s podcast. The Death of the Middle is how the Internet pushes everything to the edges. In retail, that means either Amazon or LVMH. Macy’s is dead. He predicts AI will have a similar effect on many other industries.

I think the human-managed investment industry will suffer the Death of the Middle. Human public equity managers will survive either with a big brand name or a truly unique, exotic strategy. AI will eat up everything else in the mediocre middle.

The brand name manager adds value through prestige, and assets are still allocated by humans after all. In that same Dixon interview, he noted that humans like humans, giving the example that humans still watch other humans play chess. They don’t watch robots play chess even though they’re better at it. The same will hold true for famous asset managers. At least for a while.

Truly exotic ideas, ones that AI couldn’t conceive on its own, should also draw assets provided they show consistent returns. The challenge with exotic managers will be preventing an AI clone that does it better. Today’s experiment review shows AI can already take a human-driven investment strategy and do it better.

Don’t Over Think It

I recently said that the mega cap stocks need a breather. NVDA and META can’t keep going up 40% a month forever as a matter of logic. The idea of a breather doesn’t interfere with the idea that mega cap tech isn’t as egregiously expensive as bears argue. A pullback is a short-term sentiment and positioning call. Fundamental valuation is relevant longer-term. At least that’s how my simple human mind justifies the existence of both ideas.

Even if my view to wait on vertical moves like NVDA is right, I’m not sure how much alpha to the Large Tech Focus strategy I’ll add by waiting, missing the rally on the way up, and hoping I’m daring enough to pull the trigger when these stocks are down 5-10%. Probably not enough alpha to replace what I’ve missed by not listening to AI so far.

Subscribe now