Learning AI #20

Prediction markets are the hot new thing. The two biggest are Polymarket and Kalshi. They are legalized online casinos that let you bet on most anything. AI can help you make good bets more often, but not where you think.

The obvious play is to ask AI who wins the next presidential election or the Super Bowl. Everybody does that.

The problem is that the big prediction markets already have thousands of knowledgable bettors known as “sharps” pushing those odds toward accuracy. By the time you place your bet, the market has become economically efficient, meaning buyers and sellers all have the same information, so it has “priced in” most of what AI would tell you.

The edge is in niche markets that have thin crowds. A ballot measure in a mid-sized city, or an FDA advisory committee vote on a drug most people have never heard of. These markets exist on the major platforms but draw little attention.

Here's how to use AI in a niche market.

Pick an event with a clear win-lose outcome and publicly available information. Say a minor league baseball game between the Hartford Yard Goats and the Jacksonville Jumbo Shrimp (real teams). Kalshi has the Goats listed at 55% odds of winning.

You spend 20 minutes with an AI, feeding it everything you can find about the past performance of the teams, win-loss records at night games on the road, and anything else the AI can find in statistical databases and on social media.

The AI synthesizes it and gives you a probability of 74% that Goats defeat the Jumbo Shrimp. You ask it to argue the other side. It finds one weak point: The Jumbo Shrimp have a proven major league player on their roster for a few days as he is recovering from an injury. You factor that in and settle on a 70% chance of the Goats winning.

The posted value on the prediction market says the probability of a Goats win is 55%. Your AI-powered research says 70%. That 15-point gap is the bet.

By buying “Goats” shares at $0.55 on Kalshi, you are getting a great deal since your research says they are worth $0.70. This is what experienced investors call “making money on the buy,” or buying at a price well below the true value.

When the bet resolves after the Goats defeat the Jumbo Shrimp, you collect $1.00 for that $0.55 bet. (With prediction markets, you do not have wait for the outcome. If the price in our example moves from $0.55 to $0.65, you can sell out at $0.65 and take the profit.)

Do this a few dozen times a month, which is made easier by using your AI assistant to do the research, and you have a nice side hustle.

Remember that thin markets are thin and tricky for a reason. Sometimes the crowd is lazy. Sometimes they know, or think they know, something you don't. And low liquidity means you can't bet large without moving the price yourself.

But the advantage is real. In a presidential election, you compete against professional traders with proprietary models. On a minor-league baseball game, you compete against three guys who heard about it on ESPN 8.

AI won't make you a guaranteed winner. It gives you faster, broader research than any individual can do alone, and forces you to think through probabilities more carefully than most retail bettors. You can spread your risk across dozens of bets in the same time it took research one pre-AI bet.

In a market full of lazy guessers, that's enough.

My book is off to a great start (see below), thanks to your support. It was the number one new release in its category last week, You can check out the Kindle and paperback versions on the book’s Amazon page. Kindle price still at $2.99 for a few more weeks.

The audiobook has it own page on Spotify and its free for Spotify premium users. I cloned my voice and had it do the narration. Sounds better than if I spent a week narrating it myself.

Things I think about

A typical hand-crocheted baby blanket uses more than two miles of yarn.

**********