跳到正文
𝕏 Rohan Paul(x.com)·· 8 天前品質評分52
重點導讀

AI 助理遇到新任務時,常沒有標準答案可用來調整週邊設定。

  • Northwestern 與 Google 的研究提出 Harness2:先讓 AI 助理執行任務,再由另一個模型提出提示詞、工具、技能或記憶設定的修改,重新執行後只留下有幫助的改動。
  • 每次任務學到的經驗也會存進資料庫,供之後調整參考。研究中,一次調整讓 GLM-5 在 WorkBuddy-Bench 的分數增加 11.5 分,超過使用原始設定的 Gemini 3.5 Flash。
  • 若不先實際測試修改,程式表現則從 73.4 分降到 63.2 分。

AI 伴隨問答

自備 Key 深度分析

針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:

Rohan Paul
Rohan Paul@rohanpaul_ai
·8 天前
𝕏

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.

推文來源 · X (Twitter)在 X 參與討論

讀者討論

理性交流 · 即時討論