1 Matching Annotations
  1. Sep 2026
    1. Astra is our most aligned model, with substantial improvements in understanding user intent and model behavior—you can delegate tasks with greater confidence in Astra's judgment.

      【非共识】这一关于模型对齐的声明是主观的,没有提供客观衡量标准。'对齐'是一个复杂且多方面的概念,仅凭OpenAI的自我评估不足以证明这一点。需要第三方验证和更具体的评估方法来支持这一说法。