# How to Fail with AI: Pick 5 Best Stocks, Don't Make a Mistake

> In 1986, Charlie Munger told Harvard graduates how to guarantee a life of misery. Inspired by this framework we show you exactly how to guarantee disastrous results with AI in your investment process. Invert, always invert.


## "Pick 5 Best Stocks, Don't Make a Mistake"

In 1986, Charlie Munger stood in front of the graduating class at Harvard and did something no one expected but all will remember. Instead of telling the graduates how to live their life, he told them how to guarantee a miserable one. He borrowed the idea from Johnny Carson, expanded it with his own characteristic Munger wisdom, and delivered one of the most memorable commencement speeches ever.

We are going to do the same thing here, humbly, except our subject is not life, but something almost as important. Your AI-powered investment research. If you want to guarantee that every AI tool you touch produces absolute garbage and wastes your time, if you want to ensure that your research assistant becomes a useless yes-man who encourages your worst ideas and destroys your ability to reason objectively, and embarrasses you in front of colleagues, follow these instructions closely.

## Step 1: Never Give the Model Context

This is the single most important step in producing useless AI slop. We have seen it happen with our own models at Matterfact. When you sit down at your terminal and open your AI research tool, type something like this:

**"Find me 5 best stocks. Don't make a mistake"**

Do not tell it what sector you cover. Do not tell it your fund's mandate, your market cap range, or your investment universe and horizon. Do not mention whether you run a long/short book or a long-only portfolio and how you think about risk, factors, and position sizing. Do not specify geography, models are great at assuming you are looking for investments in Asia. Do not mention valuation frameworks you care about, as that might produce a more coherent result.

The less context you provide, the better. This useless output lets you say with confidence to your PM: "I tried AI. It doesn't work." Don't worry, this is exactly what every PM wants to hear so they can confidently be left in the dust while their competitors figure out how to use AI for their research.

## Step 2: Be as Vague as Possible

Specificity is your enemy. Never tell the model whether you are looking for long or short ideas, high or low vol stocks and never limit the market caps. Never specify a time horizon either, do not mention if you are looking for a catalyst trade in three weeks or a core position you want to hold for two years. The model does not know and you should not tell it, it is much more fun to let it assume something random from the depths of its digital brain.

Never mention whether you want consensus ideas or contrarian ones. Never say whether you are looking for momentum or mean reversion plays. Never reference a factor exposure you are trying to hedge or monetize.

The fund managers we work with get extraordinary results from AI but you should do the opposite. They write prompts that read like mini investment memos and include their thesis, assumptions, constraints, and the specific output format they want. They tell the model what kind of investor they are. But you should NEVER do any of that.

```request-access
variant: inline
heading: Curious what an investment-memo-grade prompt actually returns?
description: See how the funds we work with set up Matterfact for their own coverage.
buttonText: Request access
```

## Step 3: Never Ask for the Other Side of the Trade

Charlie Munger admired Charles Darwin for one specific reason: Darwin always gave priority attention to evidence that would disconfirm his most cherished theories. Darwin actively hunted for reasons he was wrong. What a waste of time!

If you want to fail with AI, you must do exactly the opposite.

Never ask the model for the bear case if you are bullish. Never ask for the bull case if you are short. Never ask for opposing views, for instance never say, "What are the three strongest arguments against this thesis?" And never ask the model to poke holes in your reasoning.

One of the craziest things you can get an AI research tool to do is stress-test your conviction. It has access to thousands of expert conversations, earnings transcripts, interviews, and news sources. It can find the dissenting voice you have never heard and that can surface the risk you have not considered. Total disaster if you want to make sure you fail with AI.

If you only ask it to validate what you already believe, it will happily do that because foundational large language models are agreeable by nature - that's how they evolved to get high rankings on leaderboards. If you push them toward confirmation, they will confirm and that is a great way to fail with AI.

![How to fail with AI in investment research](/assets/images/blog/how-to-fail-with-ai-stock-research-illustration.webp)

## Step 4: Never Check the Sources

Look, AI never makes mistakes so you must trust everything the model tells you and take it at face value. Never ask it where it got a specific data point, waste of tokens. Never cross-reference a claim or ask it to cite the specific podcast episode and the time stamp, the earnings call, or the filing where it found that number.

AI models do not hallucinate. They would never present plausible-sounding statements that are partially or entirely fabricated. It would be terribly unwise to use AI as a research accelerator, only to then verify critical claims against primary sources.

You should definitely quote an AI-generated statistic in your investment committee meeting. When someone asks for the source, shrug and say "the AI said it, so it must be true." This is a terrific way to destroy your credibility in ninety seconds and make sure you are next in line to get fired.

## Step 5: Never Listen to What the Experts Say, They Don't Know Anything

Ignore the best way AI can help you by scanning petabytes of data in seconds. Thousands of hours of [podcast](https://www.matterfact.com/blog/why-podcasts-why-now) conversations happen every month across finance, technology, healthcare, and consumer sectors, and you should never use AI to scan them for signal.

If you do, you risk getting better insight than the street and eventually risk becoming a better investor.

If you want to fail, ignore all of this. Stick to the sell-side notes that every other analyst on the street is reading, attend the same conferences, and read the 10-K. Stay in the consensus lane.

## Step 6: Never Invert

If you want to fail, do not follow Munger's advice: "Invert, Always Invert". Never challenge assumptions and never step out of your comfort zone.

Munger closed his Harvard speech with a backward toast: "May each of you rise high by spending each day of a long life aiming low."

We will close with our own: May each of you produce extraordinary research by first understanding exactly how to produce terrible research.

Then, do the opposite!

So, take our advice, do not use AI. You are already perfect as an investor and there is no room for improvement, especially with some latest technology fad.

Do not reach out to us, we do not want to hear from you at [MatterFact](https://form.typeform.com/to/KzGXoGyV?typeform-source=www.matterfact.com).

```request-access
heading: Now invert.
description: Matterfact is deployed with select institutional partners. Request access to run it on your own coverage.
buttonText: Request access
```