AI Models Miss More Payment Fraud in Coinbase Test
Coinbase’s replay shows that model upgrades can weaken fraud coverage under a fixed screening policy, while a specialized Qwen3.5-9B model improved detection and cut latency.
Every Zipp story tagged #ModelEvaluation, newest first.
Coinbase’s replay shows that model upgrades can weaken fraud coverage under a fixed screening policy, while a specialized Qwen3.5-9B model improved detection and cut latency.
One perfect scoreline, one directional miss: Claude's World Cup read on Group D held to the decimal, but the model's Morocco-Scotland call exposed how prediction engines handle defensive discipline…