FAiE：用失败记录训练智能体，不再手动改提示词
精选理由
这条讲的是 FAiE 的思路：把智能体失败的轨迹拿来训练模型，而不是反复改提示词，看着挺聪明的路子。
FAiE 的做法是把智能体的每次失败轨迹收集起来，直接用于训练模型，而不是靠人工修补提示词。它还会收集用户偏好数据来对模型做再训练。目标是让智能体从自己的错误中持续学习。
原文 · Ate-a-Pi
I hope this works. It looks very clever:
Basically, you get an agent that learns from mistakes. FAiE takes every failure trace and, instead of patching prompts, uses it to train the model to avoid those mistakes.
They also use user preferences to retrain their model.
