Human bettor beats ChatGPT in NFL wagering results

The Week 7 Showdown

A human bettor recently faced off against ChatGPT to see which could better predict NFL outcomes. The competition focused on Week 7 of the season, where both participants were tasked with selecting five bets from a shared odds board. The rules required that no bet be priced worse than -120, ensuring a standardized level of risk for both the human and the AI.

The results showed a clear victory for human intuition over the machine. The human analyst finished the week with a 4-1 record, successfully picking up 3.11 units of profit. In contrast, ChatGPT struggled to find consistent winners, finishing with a 2-3 record and a loss of 1.33 units.

The human bettor noted that some of the winning picks were particularly close. Specifically, the 49ers-Falcons under and the Packers cover both cleared by only half a point. Despite the narrow margins, these results counted exactly the same as a larger victory, such as the Cincinnati Bengals winning by 22 points.

How the AI processed data

To give ChatGPT a fair chance, the human bettor used a highly specific prompt for the Week 8 selections. The AI was instructed to use GPT-6.1 Sol with the 'thinking effort' set to ultra. It was required to review every eligible spread, total, and moneyline based on the provided odds board captured on Saturday at 12:00 p.m. ET.

The prompt demanded that the AI use at least 10 different sources for each game to build a broad understanding of the situation. It also had to provide a detailed explanation for every selection it made. Furthermore, the AI was required to explain exactly why it passed on other eligible bets that fell within the price parameters.

Despite these rigorous instructions, the AI's performance in Week 7 was described as a 'dud.' It went 0-3 on its final three picks, including a miss on the Kansas City Chiefs. The human bettor even attempted to warn the AI about these potential losses in a pre-week analysis, but the model did not adjust its strategy.

Human vs. Machine logic

The human bettor's strategy often involved taking 'juice'—paying a higher price for a specific number—to secure a preferred outcome. For example, the human chose to take the Dallas Cowboys at -6½. This was based on the belief that the Cowboys' offense would break out for a big week while the Arizona Cardinals struggled to score in the red zone.

ChatGPT's logic for the same game was more cautious. While it agreed that the Cowboys had a red-zone advantage, it expressed concern over the Cowboys' defense allowing Arizona to stay close. Because of this vulnerability, the AI kept the Cowboys off its final five picks, opting instead for a different set of outcomes.

The two also differed on the Baltimore Ravens and Buffalo Bills game. The human bettor preferred to take Baltimore at +3½, viewing the spread as a way to get value against the Bills. ChatGPT, however, preferred the 'over' on the total points. This highlights a fundamental difference in how humans and AI weigh risk versus potential reward in sports.

The season standings

Looking at the broader season report card, the human bettor has maintained a lead over the AI. The human currently sits at 19-15-1 with a total of +3.20 units. This success is spread across different types of bets, with the human leading significantly in 'totals' bets, which are picks on the combined score of two teams.

ChatGPT has struggled more consistently, particularly on spreads and moneylines. The AI currently sits at 15-19-1 with a total of -5.36 units. Its performance on spreads and moneylines has been especially difficult, resulting in a -4.90 unit deficit in that specific category compared to the human's +0.54 units.

The human bettor plans to continue the competition into Week 8 with five new selections. The goal is to see if the human can maintain the winning streak or if the AI will find a way to correct its errors. The stakes are adjusted so that every winning bet produces exactly one unit of profit, making it easy to track the ongoing competition.

What comes next

The upcoming Week 8 slate will test several high-profile matchups, including the Dallas Cowboys, Baltimore Ravens, and Kansas City Chiefs. The human bettor is looking to follow up a successful week by finding value in the Chargers' performance against the Rams. This bet is described as a 'Plug Your Nose and Hit Place Bet' due to the high level of uncertainty involved.

The competition will also see both participants looking at the Indianapolis Colts and Jacksonville Jaguars. Both the human and the AI agreed on the 'under' for this game, citing the Jaguars' offensive struggles and the Colts' recent roster uncertainty. This rare moment of agreement suggests that some outcomes are more predictable to both humans and machines than others.

As the season progresses, the results will provide more data on how AI handles the nuances of sports betting. While the AI can process a massive amount of information from multiple sources, the human's ability to weigh 'sweats' and subjective team momentum remains a key factor. Readers can follow the results to see if the robot eventually closes the gap.