AI models can trade real money, but the market design decides if they win

Researchers put AI models to work trading real money on live prediction markets. It turns out the models' success depends more on the market's rules than on the AI's intelligence.

Everyone assumed that if an AI could predict an event, it could profit from that prediction. This paper shows that the specific design of the market, like how trades are executed or settled, is a bigger factor in an AI's financial performance than its raw predictive power. This means companies building AI for financial trading now have to optimize for the market's structure, not just the model's smarts.