Salesforce

On the Fragility of Self-Improving Agents: Variance, Task Order, and Underspecification

自改进智能体的脆弱性：方差、任务顺序与规格不足

Qinyuan Ye, Yu Li, Yada Pruksachatkun, Jiaxin Zhang, Chien-Sheng Wu

机构 * Salesforce AI Research（Salesforce AI研究院） 纠错

AI总结 本研究通过多轮运行与打乱任务顺序的实验，揭示当前基于记忆的自改进智能体存在脆弱性，发现其性能受方差与任务顺序影响，还指出规格不足是相关诱因，呼吁采用更严格评估方案与人工监督机制。

Comments Code: https://github.com/SalesforceAIResearch/self-improve-fragility Data: https://huggingface.co/datasets/Salesforce/self-improve-fragility