幻觉Hallucination
Hallucination
解决什么问题What problem it solves
模型用非常自信的语气说出看起来合理、但事实错误的内容。The model states something plausible and confident that is simply not true.
这是什么What it is
模型生成的内容不来自「查询」,而来自「预测下一个词」,所以会编。
它在不熟悉的领域、以及需要精确数字和引用的地方尤其容易发生。
Output that is fluent and confident but factually wrong. It comes from producing likely-sounding text rather than retrieving verified facts.
为什么需要知道Why it matters
这是使用 AI 最需要防的一件事。规则很简单:涉及事实、数字、法律、医疗的部分,必须另找来源核对。It is the single biggest practical risk with language models, and the reason fact-checking matters more than ever.
别名Chinese aliases
胡说八道 · 一本正经地编