FutureSim Benchmark Exposes Frontier Agents' Inability to Forecast Real-World Events

FutureSim benchmark reveals that even the best frontier agents achieve only 25% accuracy in forecasting real-world events, challenging the world model surge narrative. Serves as a reality check for claims about world model capabilities and adaptive agents. A recent commentary on the assumption boundary problem and fact half-life further underscores the gap between agent reasoning and reality. New article confirms the benchmark results and reinforces the gap.