普林斯顿与 Meta 提出 Self-Direction：把智能体纠错经验存成 LoRA 适配器复用
Nice new paper on continuous improvement of AI agents from Princeton and Meta
普林斯顿与 Meta 的新论文提出 Self-Direction 方法，让失败的智能体反思本次运行，并把教训训练进一个约为模型大小 0.02% 的 LoRA 适配器，之后由路由器在遇到类似失败时加载最接近的适配器。在 Qwen3.6-27B 上，GAIA 任务成功率从 38.8% 提升到 55.8%，优于上下文反思和 Skills.md 文件方案。在 185 个此前出现过的错误中，到第三次遇到时仍有 67.3% 的修复生效，而单纯重试为 42.9%。
推荐理由
AI 评分 58
论文把智能体的纠错经验固化成约 0.02% 模型大小的 LoRA 适配器并按需路由加载，给出了 GAIA 上的成功率对比。
主题
正文
Nice new paper on continuous improvement of AI agents from Princeton and Meta
Finds that agents repeat fewer mistakes when each fix is saved as a LoRA adapter and reused on similar tasks.
Agents can fix a mistake within a session, then often repeat it in the next one. Notes in context get forgotten, and full fine-tuning is costly and can disrupt other skills.
With Self-Direction, a failing agent reflects on its run and trains the lesson into a LoRA adapter about 0.02% of the model's size. A router later loads the closest adapter when the agent fails a similar task.
With Qwen3.6-27B, task success on GAIA rose from 38.8% to 55.8%, beating in-context reflections and a Skills.md file. Across 185 earlier mistakes, 67.3% of repairs still worked by the third encounter, against 42.9% for plain retrying.
If your agents keep making the same mistakes, save each fix as a small adapter they can reload, not only as prompt notes.