Successes and failures of simplified models for a network of real neurons