知识卡片
关键词屏蔽忽略社会语境
同一个词可出现在攻击、医学讨论、身份表达或反歧视文本中。按词整篇删除会同时漏掉变体攻击并误删合法内容,应在语境和目标群体上检查影响;屏蔽词越多不自动等于模型越公平。
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
- · 预训练数据
原书出处 · 1 处
- · 查看原书依据
Keyword-based methods simply do not capture this nuance. The question of whether it is more effective to handle these issues at the pre-training stage or further downstream is an open area of research. We will explore techniques that can be employed downstream in
参考来源
- 位置:《Designing Large Language Model Applications》第02章《预训练数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html)
- 结论依据:摘录中关于关键词屏蔽忽略社会语境的定义、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:Keyword-based methods simply do not capture this nuance. The question of whether it is more effective to handle these issues at the pre-training stage or further downstream is an open area of research. We will explore techniques that can be employed downstream in