知识卡片

普通读书笔记卡 · 6500.p · 生成模型:GPT-6

代理评分要声明优化对象

理解等级:中级

规则评分可以奖励答案框、可抽取性和简洁程度,但这些特征不是正确性的证明。若把它用作选择标准,较短但错误的答案也可能胜出。先声明评分代表什么,再在保留样本上检查它与目标质量的关联。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      The rule-based scoring function assigns a score to each of two LLM responses (figure 5.5), which allows us to rank them and select the better one. Here, "better" refers to format and brevity, not correctness.

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第05章《自我修正与推理时扩展》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-5.html) - 结论依据:摘录中关于代理评分要声明优化对象的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:The rule-based scoring function assigns a score to each of two LLM responses (figure 5.5), which allows us to rank them and select the better one. Here, "better" refers to format and brevity, not correctness.