ChatGPT can write a to-the-word-count high school history essay or doctoral thesis. It can help realtors target potential buyers and hone search engine optimization. But AI sports betting is a thing?
Yes, ChatGPT is a tool for the thorough, a crutch for the corner-cutters, and maybe the next step toward future tech dystopia, depending on the observer.
It couldn’t have seen the Jontay Porter scandal coming. It cannot win you your Super Bowl bets. (Yet. But that will get a lot closer to reality this year, a Microsoft executive believes.).
I asked. And it told me why. As a language model, the current belle of the artificial intelligence ball is just not designed for the task.
But it can help, able to aid skilled coders in building their own AI sports betting bot.
So says Siraj Raval, a self-styled YouTube educator and AI specialist with more than 746,000 subscribers.
Why does he make such a claim? Because he says he did it, laying out the steps in a YouTube video posted earlier this year for a bot called WagerGPT.
And he keeps doing it, he claims. Raval built another bot he calls OddsGPT to win an initial NHL bet on the Seattle Kraken. And now WagerGPT is generating wins on the race track, he says.
Granted, this is the first step in a process that requires much more labor, and honestly, vetting. Raval set out to concoct an arbitrage model that morphed into a straight win/loss prediction for nightly NBA games. Marshaling — tech terms! — sentiment analysis from Twitter and PyTorch for deep learning, Raval said, he turned $2,000 into “about $7,000” on two bets on a grey market sports betting site using the Polygon cryptocurrency.
And then he moved on to the next topic. It’s up to actual sports bettors to follow up on in his work, he told Gaming Today, with the coding freely available on GitHub.
“I made a few bets that it didn’t work, three test bets that were smaller when I was building it at first. Then I let it train on much more data and then I just went ahead and made a bet, that huge bet I made, and it worked,” he said. “But have I run it since then? No, I haven’t. And that’s because every week I just kind of try to think of a different project. I’m an ideation channel. I’m not a professional sports bettor. I’m sure I could just continually do that every week, but I’m more interested in growing the YouTube channel and getting people excited about using AI for many, many different things.”
Though Nick Cockerill, vice president of product and sports data at Stats Perform, told Sports Business Journal that the adoption of AI in sports betting is “currently a fairly slow burn,” there are companies like Quarter4 offering betting guidance using a neural network and deep learning system affectionately known to the founders as “The Brain.”
The house will undoubtedly brandish brains, too.
Artificial intelligence, by design, is tasked with finding patterns quickly, allowing financial traders to make decisions without hours of exhaustive research. Or Google to plaster your feed with advertisements germane to your search history, defensive coordinators to ascertain how an opponent defends blitzes. It, therefore, stands to reason that finding patterns in performance in either teams or individuals without spending hundreds of hours could be a lucrative and cost-effective way of applying a bankroll.
How Siraj Raval Built WagerGPT Utilizing ChatGPT
Raval told Gaming Today his main focus is education and empowerment. The topics of his weekly videos, a mix of egghead DIY, occasional dance bits, and joyous musings on math are often suggested by commenters, which was the case with the WagerGPT episode.
“I am not regularly a sports bettor. In fact, I don’t really watch sports that often. I’m more just like into the math part.”
The process of creating WagerGPT is much the same as building a bot to predict financial results, Raval said, based on a neural network.
“All of them have different configurations, but at the base, they’re all the same idea,” he said.
ChatGPT was at times a timid collaborator. Raval relied on the system to supply certain starting points for complex coding operations but was occasionally rebuffed about the legality of the enterprise in various locales. Generally, however, it would surrender for the needed information with a follow-up question. From there, Raval relied on the existing work of other coders to find useable sources of updated odds — he focused on the NBA — and ways to facilitate communication between the systems.
“That was hilarious. I really didn’t expect that,” Raval said of ChatGPT’s reticence. “It’s like prompt engineering is a new skill. It’s this new job category that’s emerged and it’s actually many different jobs. It’s not just one, because whether you’re a chef or a sports bettor or an athlete, you need to be able to converse with your AI and tell it the optimal prompt, what to do.
“I guess you could call this jailbreaking, kind of like what I used to do for the iPhone back in the day, where you can load different software. It’s kind of like you’re jailbreaking ChatGPT to do what you want.”
WagerGPT is Available for Free, Allowing Bettors To Vet AI Sports Betting
Raval made his bets on a decentralized grey market platform in part because it didn’t geo-block him, and because it dealt in the Polygon cryptocurrency for which he has an affinity and a previous business connection. The entire process, he estimated, took between 60-to-80 hours, including producing the video.
But what level of coding expertise is required to allow would-be AI-powered sports bettors to prove or disprove Raval’s work? And maybe make some money.
“Well, hopefully, zero now that the video is out there because my code is now available,” he said. “The only coding expertise is hitting run on that notebook, which runs in your web browser, whether it’s Chrome or whatever. You would go to a website, you open up that code — it’s called a Google Colab — it’s got a bunch of code in there, but all you have to do is hit play, and it’s going to run the model on Google’s server. It’s a neural network, and it’s using all the data that it’s been trained on up to today using FanDuel or whatever API, and then it’s going to output the expected value for each of the teams, specifically for the NBA, which is what I tailored it for. And then you can make your bets based off that.”
Various computer programmers and developers contacted by Gaming Today said that Raval’s coding process is sound, but said validation would require vastly more runs.
If a bettor is interested in another sport, Raval admitted, the process becomes much more labor-intensive.
“If you want to do a different sport, you’re going to need a different set of training data entirely,” he said. “There’s not a no-code way to do that yet, unfortunately.
Like WagerGPT, ChatGPT is free – mostly. But it’s no help at all for the lazy resourceful sports bettor seeking AI guidance on sports.
That could be changing quickly, though.
“I would imagine this entire process in like a year or two is going to be a conversation with an AI,” Raval said. “We’re going to see a lot of these competing AI. ChatGPT is just the first, and I’m sure somebody’s going to make the sports betting AI that you talk to, and they’re probably going to charge for it as well.”
About the Author
