AI researchers rediscover the "eval to production gap"

Researchers find that systems optimized for static metrics fail in dynamic environments, bridging the gap between quantitative finance’s "overfitting" and AI’s "eval to production gap."