报告指出多AI系统存在风险, 呼吁采用新的测试方法来确保安全. Report identifies risks in multi-AI systems, calls for new testing methods to ensure safety.

渐进研究所的一份新报告突出强调了多个AI代理机构共同合作时的六大风险,包括性能不一致、通信中断和群体思维。 A new report by the Gradient Institute highlights six key risks when multiple AI agents work together, including inconsistent performance, communication breakdowns, and groupthink. 单一人工智能剂的传统测试方法不足,因此,报告建议采用控制模拟和监测试点方案来管理这些风险。 Traditional testing methods for single AI agents are inadequate, so the report recommends using controlled simulations and monitored pilot programs to manage these risks. 目的是随着企业越来越多地采用多种试剂系统,确保安全可靠的AI部署。 The aim is to ensure safe and trustworthy AI deployment as businesses increasingly adopt multi-agent systems.