大语言模型的外在幻觉：成因与检测

AI 摘要

Lil'Log 博客文章聚焦大语言模型的外在幻觉问题，将其定义为输出与预训练数据或世界知识不符的捏造内容。文章分析了幻觉的成因，包括预训练数据问题及微调引入新知识的影响，并介绍了幻觉检测方法，如 FactualityPrompt 基准和 FActScore 指标。

正文节选

Hallucination in large language models usually refers to the model generating unfaithful, fabricated, inconsistent, or nonsensical content. As a term, hallucination has been somewhat generalized to cases when the model makes mistakes. Here, I would like to narrow down the problem of hallucination to cases where the model output is fabricated and not grounded by either the provided context or world knowledge. There are two types of hallucination: - In-context hallucination: The model output shoul

发布时间：2024-07-07 08:00

抓取时间：2026-08-02 00:26

来源机构：Lilian Weng