https://blog.qife122.com/tags/%E7%A4%BE%E4%BC%9A%E8%A7%84%E8%8C%83%E5%AF%B9%E9%BD%90/ Recent content in 社会规范对齐 on 办公AI智能小助手 Hugo zh-cn qife Tue, 02 Sep 2025 01:14:18 +0800 https://blog.qife122.com/p/%E5%9F%BA%E4%BA%8E%E7%A4%BE%E4%BC%9A%E8%A7%84%E8%8C%83%E5%AF%B9%E9%BD%90%E9%9A%90%E5%BC%8F%E5%AF%B9%E8%AF%9D%E7%9A%84%E5%8C%85%E5%AE%B9%E6%80%A7%E6%84%8F%E8%A7%81%E8%A1%A8%E5%BE%81%E6%8A%80%E6%9C%AF/ Tue, 02 Sep 2025 01:14:18 +0800 https://blog.qife122.com/p/%E5%9F%BA%E4%BA%8E%E7%A4%BE%E4%BC%9A%E8%A7%84%E8%8C%83%E5%AF%B9%E9%BD%90%E9%9A%90%E5%BC%8F%E5%AF%B9%E8%AF%9D%E7%9A%84%E5%8C%85%E5%AE%B9%E6%80%A7%E6%84%8F%E8%A7%81%E8%A1%A8%E5%BE%81%E6%8A%80%E6%9C%AF/ <p>塑造包容性表征以拥抱多样性、确保公平参与和价值反映，是许多基于对话的模型的核心目标。然而，现有方法多依赖表面包容性，如提及用户人口统计特征或社会群体行为属性。这类方法忽视了对话中嵌入的细微、隐性的意见表达。此外，过度依赖显性线索可能加剧模型输出中的错位，并强化有害或刻板表征。</p>