Unverified50% confidenceFactExact time
Research shows that hallucination issues are particularly prominent in specialized fields such as healthcare, law, and finance, where error rates can be as high as 30% or more.
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
RAG检索增强生成详解：原理、工作流程与核心优势
bilibiliAI小公举-timi7/9/2026
Related Claims
UnverifiedA 2023 Stanford University study showed that top LLMs have factual error rates as high as 20–40% in the medical and legal domains.71% similarUnverified早期模型在某些领域的事实幻觉率高达15-25%，尤其在长尾知识领域幻觉率更高69% similarUnverified在专业信息上对模型进行微调可能导致更多幻觉，且成本效益极差67% similarVerifiedInternal research from companies such as Microsoft and Google indicates that even GPT-4-level models may have factual error rates as high as 15%–30% in vertical domains such as law, finance, and healthcare.64% similarUnverified研究显示，在专业技术领域，LLM 的幻觉率显著高于通用知识领域，因为训练数据中领域专家的纠错反馈相对稀少63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/607114API
curl https://kongchang.com/api/v1/knowledge/claims/607114MCP
get_claim(id=607114)