AI 热点

Northwestern 与 Google 论文提出 Harness2,让智能体外壳无需答案键自我改进

Rohan Paul更新于 11月15日 09:00

New Northwestern and Google paper shows that agent harnesses can improve themselves on new tasks without answer keys, by test-running edits and keeping what hel

AI 导读

Northwestern 与 Google 的新论文提出 Harness2,让智能体外壳在没有答案键的新任务上自我改进:先运行智能体,由编辑模型提出外壳修改,再重跑任务看哪些修改真正改善了行为,并把每个任务的经验存入库中指导下一次修改。一次改进步骤让 GLM-5 的 WorkBuddy-Bench 得分提升 11.5 分,超过使用原始外壳的 Gemini 3.5 Flash;而不做这些测试运行直接改外壳,编码得分从 73.4 降到 63.2。

推荐理由

AI 评分 58

原文给出无答案键场景下自动改写智能体外壳的具体流程,以及一次改进带来的分数变化和反例。

主题

标签

正文

New Northwestern and Google paper shows that agent harnesses can improve themselves on new tasks without answer keys, by test-running edits and keeping what helps.

An agent's harness is the prompts, tools, skills, and memory wrapped around the model. Tuning it by hand is slow, and new real-world tasks come with no answer key to tune against.

Harness2 runs the agent, has an editor model propose harness fixes, then reruns the task to see which fixes improved the agent's behavior. Lessons from each task go into a library that guides edits on the next one.

A single improvement step raised GLM-5's WorkBuddy-Bench score by 11.5 points, past Gemini 3.5 Flash on its stock harness. Editing the harness without those test runs instead dropped coding scores from 73.4 to 63.2.

Treat your agent's harness as something to revise per task, and keep only the edits that change its behavior on real runs.

来源

Rohan Paul