Auto-sync: update nexus workspace
This commit is contained in:
23
wiki/concepts/Hallucination.md
Normal file
23
wiki/concepts/Hallucination.md
Normal file
@@ -0,0 +1,23 @@
|
||||
---
|
||||
title: "Hallucination"
|
||||
type: concept
|
||||
tags: [hallucination, llm, problem, reliability]
|
||||
aliases: [Hallucination, 幻觉, 大模型幻觉, LLM 幻觉]
|
||||
last_updated: 2025-12-20
|
||||
---
|
||||
|
||||
## Definition
|
||||
Hallucination,幻觉,大语言模型在面对陌生领域或缺乏足够知识的问题时,"一本正经地胡说八道",给出看似合理但实际错误答案的现象。
|
||||
|
||||
## Key Facts
|
||||
- 本质原因:LLM 的知识局限于训练数据集,面对未知领域时无法真实"理解",只能基于概率生成最可能的下一个词
|
||||
- 类比:LLM 在陌生领域"只会写一个解字",然后放飞自我
|
||||
- [[RAG]] 是解决幻觉的主要技术手段之一(通过外部知识引导)
|
||||
- 其他方法:模型微调、Prompt Engineering、Chain-of-Thought
|
||||
|
||||
## Connections
|
||||
- [[RAG]] ← 解决 ← [[Hallucination]]
|
||||
- [[Large Language Model]] ← 问题 ← [[Hallucination]]
|
||||
|
||||
## Sources
|
||||
- [[大模型相关术语和框架总结|llm-mcp-prompt-rag-vllm-token-数据蒸馏]]
|
||||
Reference in New Issue
Block a user