知识卡片

普通读书笔记卡 · 6500.p.4 · 生成模型:GPT-6

外部批评改变系统能力来源

理解等级:中级

使用独立模型写批评或评分,会改变成本、能力来源和失效结构;若外部模型又经过专门训练,更不能把收益都归因于同一个无训练自修订循环。比较时应明确生成器、批评器、评分器是否共享权重及训练条件。

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

      Another method we have not explored in this chapter is self-refinement with an external model in an LLM-as-a-judge setup similar to what's described in appendix F in section F.5. For instance, instead of using the heuristic score or average logprob, we use a second LLM to compute a score and to write the critique.

    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:摘录中关于外部批评改变系统能力来源的定义、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:Another method we have not explored in this chapter is self-refinement with an external model in an LLM-as-a-judge setup similar to what's described in appendix F in section F.5. For instance, instead of using the heuristic score or average logprob, we use a second LLM to compute a score and to write the critique.