知识卡片
识别假设不能由流畅答案替代
LLM给出因果概率数值公式时,应先回到目标定义,再查其识别假设。自然观测频率的比值不能一般替代跨世界概率;输出再流畅、代码再整齐,也无法填补未说明的混杂、单调性或支持条件。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 因果与大语言模型
原书出处 · 1 处
- · 查看原书依据
answer is wrong. This is what is commonly called a “hallucination,” a convincing lie. This example is especially convincing—that formula is true only when the relationship between the treatment and outcome is monotonic, when there are no confounders, and when P(Treatment) = P(Outcome). Those assumptions are so strong that we can’t forgive ChatGPT for not mentioning them. Other generations from the model were even more
参考来源
- 位置:《Causal AI》第13章《因果与大语言模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-13.xhtml)
- 结论依据:摘录中关于识别假设不能由流畅答案替代的定义、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:answer is wrong. This is what is commonly called a “hallucination,” a convincing lie. This example is especially convincing—that formula is true only when the relationship between the treatment and outcome is monotonic, when there are no confounders, and when P(Treatment) = P(Outcome). Those assumptions are so strong that we can’t forgive ChatGPT for not mentioning them. Other generations from the model were even more