Event ·Multi-source

WikiSkill：Agent经验持久化知识沉淀

Google Research提出WikiSkill方法，在Agent与原始经验间插入持久知识层。实验在5个基准测试中，WikiSkill比最强基线高3.3-12.0分。Qwen模型规模越大提升越明显，27B版本提升达23.9%。Skills可跨模型迁移，9B模型使用27B进化的Skills在ALFWorld上达70.2%。

Current read

Google Research让Agent经验变成持久知识，Skills进化更稳定，小模型用大模型技能效果更好。

5 sources78° AI heatLast updated 2026/8/30 23:49:02

Evidence

5 sources

Primary sourceshao__meng

2026/8/30 23:49:02View source

Related source 1小互

2026/9/1 05:55:31View source

Related source 2Decoder

2026/8/29 12:51:24View source

Related source 3elvis

2026/8/31 14:30:16View source

Related source 4Patrick Loeber

2026/9/1 15:48:30View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story