Verified75% confidenceFactExact time
A 2023 Stanford University study showed that top LLMs have factual error rates as high as 20–40% in the medical and legal domains.
3
Sources
75%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
The Sprout Experiment: Exploring Deterministic Symbolic Reasoning AI Without Neural Networks
redditr/learnmachinelearning7/10/2026
Related Claims
Unverified2023年的分析显示主流LLM在生成学术引用时的错误率从30%到超过60%不等77% similarUnverified斯坦福大学2023年的研究表明，主流AI检测工具误判率可高达20%以上74% 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.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/482347API
curl https://kongchang.com/api/v1/knowledge/claims/482347MCP
get_claim(id=482347)