Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
The Rise of AI Slop Code: How Teams Can Tackle the "Vibe Coding" Crisis
redditr/cursor7/11/2026
Related Claims
Unverified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71% similarUnverified有研究表明当输入文本具有明确结构化标记时，主流LLM的信息提取准确率可提升20%以上，幻觉发生率相应降低69% similarUnverifiedStanford research shows that in complex debugging scenarios, the probability of LLM suggestions causing beginners to form incorrect mental models is approximately 23%, far higher than that of human tutors.69% 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% similarUnverifiedMultiple benchmark tests in 2023 (such as the HaHackathon dataset) showed that top LLMs still have significantly lower accuracy than ordinary human annotators on sarcasm recognition tasks67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/490316API
curl https://kongchang.com/api/v1/knowledge/claims/490316MCP
get_claim(id=490316)