On October 30th, the AI research account BallotBench published a comparative experiment, comparing the win probability predictions of nine AI models for the election in Maine's 2nd Congressional District with two prediction markets, Kalshi and Polymarket. The results showed that the nine models predicted an average win probability of approximately 50% for Democratic candidate Matthew Dunlap, with each model predicting a range of 44%–54%, and five models giving a win probability exceeding 50%. In contrast, Kalshi gave Dunlap a 44% win probability, and Polymarket gave him 46%, with both platforms giving Republican candidate Paul LePage a higher win probability. The election in this district is scheduled for November 3rd, and this comparison provides an observational sample for evaluating the differences in election prediction between large models and prediction markets formed by money trading.
