In the world of horse racing, tomorrow’s Kentucky Derby is the Super Bowl, the World Series, and the NBA Finals all rolled into one. The “greatest two minutes in sports” is the marquee matchup in a series of over a dozen races, but if predicting the outcome of the March Madness basketball tourney is tricky, it gets even harder when you combine both human and equine unpredictability.
Throw in factors like the weather and conditions of each race track, and the odds of picking winners—the obsession that drives the $500 billion horse racing market—are not exactly in your favor.
In one 20-horse race, the odds of correctly picking the winner are simply 1 in 20. But wagering on horse races can get complicated. If you want to predict the exact order of the top two, three, or four horses, known as an exacta, trifecta, and superfecta, the odds skyrocket to 1 in 380, 1 in 6,840, and 1 in 103,025.
But wait! There’s more! Players often “box” their choices to increase the chance of winning. This means they bet on all possible combinations of their favored horses. The odds of winning a boxed exotic bet are better than playing them straight, 1 in 190 for a boxed exacta, 1 in 1,140 for a boxed trifecta, and 1 in 16,215 for a boxed superfecta. However, the payouts are also reduced accordingly.
Some bettors might relish the prospect of spending hours poring over stats and racing forms, but we invested our time in creating a tool to get us to an answer more quickly—meanwhile crafting a system that can be called up when the next Kentucky Derby draws near.
Enter AI
There are a variety of ways to enlist artificial intelligence in horse racing. You could process reams of data based on your own methodology, trust a third-party pre-trained model, or even build a bespoke solution from the ground up. We decided to build a GPT we named it HorseGPT to crunch the numbers and make the picks for us.
A GPT is a type of AI agent built atop a large language model like ChatGPT from market leader OpenAI. A customized agent works with a set of pre-defined prompts and training data to guide its behavior. This approach is less complex than training a model from the ground up, but proved to be good enough to pick winners and defend those choices.
We did start from scratch in one sense: we were completely new to playing the horses, and don’t know the first thing about “handicapping nags” or other terms of art. The most popular AI chatbots were similarly ignorant at first.
We carefully curated prompts to instill HorseGPT with expertise in data science specific to horse racing: how weather affects times, the role of jockeys and riding styles, the importance of post positions, and so on. We then fed it a mix of research papers and blogs covering the theoretical aspects of wagering, and layered on practical knowledge: how to read racing forms, what the statistics mean, which factors are most predictive, expert betting strategies, and more.
Finally, we gave HorseGPT a wealth of historical Kentucky Derby data, arming it with the raw information needed to put its freshly imparted skills to use.
We unleashed HorseGPT on official racing forms for this year’s Derby. We asked HorseGPT to carefully analyze each race’s form, identify the top contenders, and recommend wager types and strategies based on deep background knowledge derived from race statistics.
The big four
Without further ado, here are HorseGPT’s predictions for the 150th “Running of the Roses” at the 2024 Kentucky Derby on Saturday. The main race is race number 12.
Given the profiles of the horses and their racing styles, here’s its recommended betting strategy:
As for so-called exotic bets?
A second and third opinion
To double and triple check our work, we did give other powerful AI tools and models a spin. However, building agents with Claude or Reka required API credits and careful coding. Cutting to the chase, we attempted to rely on their respective massive training datasets, building an elaborate prompt to secure their winning picks.
Interestingly enough, our HorseGPT AI agent and the other out-of-the-box chatbots seemed to agree with each other, and with many experts analysts cited by the official Kentucky Derby website. Just a Touch was the only aberration. HorseGPT put it in a top-three spot, but the other experts—AI and human—went with Catching Freedom instead.
That said, some human experts like James Cully and Kaitilin Free are also optimistic about Just a TOuch.
The top three by consensus among our AIs and our mixed slate of experts is to bet on Sierra Leone, Fierceness, and Catching Freedom.
Hold your horses
We didn’t know how to read a racing form, but HorseGPT could break down each statistic and factor in mere minutes after being coached with prompts and research material. We may not be handicapping savants, but we fed our model information provided by experts. That’s the power of these large language models: if you know how to work with a chatbot, but once you’ve ‘taught’ them and fed them high-quality information, they rapidly acquire domain expertise.
But before you rush off to place bets—if allowed under the laws of your local jurisdiction—remember that predicting anything as messy as humans—whether it’s bracketology, horse racing, or even stock market trends—requires nuance. AI and conventional mathematic models analyze past performance and make predictions based on “ceteris paribus,” or status quo, conditions.
But they can’t predict extraordinary circumstances like an injury, a sudden rain, or a jockey or horse simply having a bad day. That unpredictability, of course, is part of the thrill.
Bonus picks for the other races
Unlike a human bettor, HorseGPT doesn’t tire of crunching numbers. So we also asked our AI model to digest the racing form for each of the other eleven races this weekend. The amount of data available was not the same, admittedly, but HorseGPT applied the same methodology. For the completists, here were the results.
Race 1:
Win Bet: Pure Force is a strong candidate for a win bet given his solid workouts, strong breeding, and trainer’s track record.
Exotic Bets:
Race 2:
Win Bet: Scylla due to strong performances and favorable conditions.
Exotic Bets:
Race 3:
Win Bet: Mindframe due to the highest potential shown in debut and strong workouts.
Exotic Bets:
Race 4:
Win Bet: Place win bets on Best Actor due to high consistency and performance, particularly if the track remains dry.
Exotic Bets:
Race 5:
Top Picks: Pure Force and Lou’s Legacy stand out based on their training and past performances.
Win Bet: Pure Force seems a strong contender for a win bet.
Exotic Bets:
Race 6:
Win/Place Bets: Alva Starr and Vahva are the most consistent and should be considered for win/place bets.
Exotic Bets:
Race 7:
Win Bets: Chili Flag and Walkathon are compelling choices for win bets due to their strong performances and experience on turf.
Place/Show Bets: Ag Bullet and Cairo Consort are suitable for place and show bets, given their solid records and ability to perform under pressure.
Exotic Bets: Consider a trifecta or superfecta box including Chili Flag, Walkathon, Ag Bullet, and Cairo Consort to capitalize on their combined strengths.
Race 8:
Win Bets: Top Conor and Nash appear to be the strongest bets based on their high Beyer figures and earnings.
Place/Show Bets: Consider Beeline and Vlahos for place/show bets due to their potential and performance metrics.
Exotic Bets: A trifecta box involving Top Conor, Nash, and Beeline or Vlahos might be promising. Adjust based on the latest track conditions and any pre-race observations.
Race 9:
Top Picks for Dry Track: Legend of Time, Neat, and Set.
Top Picks for Muddy Track: Legend of Time, Neat, and Cugino.
Exotic Bets: Consider a trifecta box including Legend of Time, Neat, and Set or Cugino. Add Stay Hot for superfecta bets, particularly in dry conditions.
Potential Longshots: Blue Eyed George and Lord Bullingdon could offer value in more complex betting structures like superfectas due to their moderate ratings but potential for upset.
Race 10:
Win Bets: Tejano Twist and Gun Pilot appear to be strong choices based on their Beyer figures and proven performances.
Place/Show Bets: Mr. Wireless and Three Technique are suitable for place and show bets, given their solid records.
Exotic Bets: Consider a trifecta or superfecta including Tejano Twist, Gun Pilot, Mr. Wireless, and Three Technique to capitalize on their combined strengths.
Race 11:
Win Bets: I’m Very Busy stands out due to its high Beyer figure and recent performance, making it a strong candidate for win bets.
Place/Show Bets: Integration and Program Trading have shown consistent performance, making them good candidates for place or show bets.
Exotic Bets: A trifecta box involving I’m Very Busy, Far Bridge, and Program Trading could yield a high return, leveraging their combined strengths and track compatibility.
Race 13:
Win Bet: General Partner: due to its strong performances, highest earnings in the field, and solid Beyer figures.
Place/Show Bets:
Exotic Bets:
Race 14:
Win Bet: Gun Party due to consistent performance and good finishing strength.
Place/Show Bets: Timeout and Dive Bomber are recommended for their high potential and strong debut performances.
Exotic Bets:
Edited by Ryan Ozawa.
Source: Decrypt
