Economists run a secret prediction game each year. When ChatGPT took part, here's what happened.

In an intriguing annual forecasting contest, economist David Seif invited ChatGPT to compete against a group of 160 economists, investors, and tech leaders. This game, now in its seventh year, challenges participants to predict around 30 significant events across various fields, including politics, business, and pop culture. As the 2025 contest began, predictions ranged from whether Taylor Swift and Travis Kelce would announce their engagement by April 1, to whether Bulgaria would adopt the euro as its currency by July 1. Sam Leffell, a hedge fund director, decided to include ChatGPT in the competition after testing its predictions during his own preparation. He shared the complex rules with the AI, allowing it to generate probability estimates for each question. The scoring system is unique: participants earn points based on the accuracy of their predictions, with a lower score indicating better performance. If a contestant assigns a high probability to an event and it occurs, they score fewer points; conversely, an incorrect high probability leads to a significantly higher score, which can be detrimental. When the results of the contest came in, Seif emerged victorious for the first time. ChatGPT, however, placed 80th among the competitors, raising questions about its predictive capabilities. While some viewed this as evidence that the AI was no better than the average human predictor, others suggested it demonstrated a remarkable level of competency. Seif noted that if participants had simply assigned a 50% probability to each event, they would have scored 75,000 points. ChatGPT's score of 82,925 indicated it ultimately fell short of adding value, largely due to its struggles with events lacking substantial data. Despite its mixed performance, Leffell highlighted ChatGPT's efficiency. The AI processed complex rules and produced predictions in minutes, whereas human participants typically invested hours researching and calculating probabilities. Leffell argued that, in scenarios involving larger datasets, AI tools like ChatGPT could become invaluable for making rapid assessments across numerous events. As the landscape of prediction evolves, the implications of AI's involvement in forecasting challenges remain a topic of interest. Can machines consistently outperform humans by leveraging data and speed? This contest may provide insights into the future of AI in predictive analytics.