Northwestern与Google论文:Harness2实现Agent自优化
New Northwestern and Google paper shows that agent harnesses can improve themsel…
Agent工程领域的硬核突破,Harness2展示了无需人工标注即可自动优化Agent配置的方法,对做Agent开发的同学极具参考价值,建议收藏研读其验证逻辑。
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.
西北大学和谷歌的新论文表明,智能体工具包(agent harnesses)可以通过测试运行修改并保留有效部分,在没有答案键的情况下在新任务上实现自我改进。
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.
智能体的工具包是包裹在模型周围的提示词、工具、技能和记忆。手动调整它速度缓慢,而新的现实世界任务没有可用于调整的答案键。
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.
Harness2 运行智能体,由编辑器模型提出工具包的修复方案,然后重新运行任务以查看哪些修复改善了智能体的行为。每个任务的经验教训会被存入一个库中,用于指导下一次任务的修改。
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.
单次改进步骤使 GLM-5 的 WorkBuddy-Bench 得分提高了 11.5 分,在其原始工具包上超越了 Gemini 3.5 Flash。若不在这些测试运行的基础上进行编辑,编码分数反而从 73.4 下降至 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.
将智能体的工具包视为针对每个任务需要修订的对象,并仅保留那些在实际运行中能改变其行为的修改。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力