AI’s Betting Skills: A Reality Check for Smart Money

Smart money isn’t always smart when the bet is placed by AI. A fresh test throws cold water on the idea that today’s most advanced AI models can handle long-term financial tasks without human help. Researchers built a fake English Premier League betting market where AI agents had to grow their bankroll over a full season. The results? Every single one lost cash. The best performer, Claude, still ended up down 11%, while Grok 4. 20 blew through nearly 90% of its starting funds. Ouch. Experts scored each AI using a 44-point system they cooked up with betting specialists. No model cracked even a third of the possible points. The verdict? These systems struggle to stick with a plan or adjust when the game changes. They might spot a trend once, but keeping up over months? Not their strength. The study suggests the gap between what AI promises and what it actually delivers is wider than many think.Big tech already feels the heat. This year alone, nearly 80, 000 tech workers lost their jobs in the first three months, with AI often fingered as the reason. Some fear a domino effect: AI replaces white-collar jobs, spending drops, and the economy stalls. But KellyBench’s results throw a wrench in that doomsday clock. If AI can’t beat a simple sports betting market, how ready is it to run the economy? Traders on Kalshi are already betting on the pace of AI disruption. Their market suggests a 23% chance that this so-called Citrini scenario kicks in soon. Meanwhile, Polymarket traders are pricing a 20% chance the AI bubble pops by New Year’s Eve, with over $2. 5 million riding on it. If AI progress stalls, those bets might look too optimistic. For NVDA and other AI-linked stocks, KellyBench isn’t a market-moving event—at least not today. Yet it quietly nudges the odds away from a fast AI takeover and toward a slower, steadier climb.