Unverified50% confidenceBenchmarkExact time
Muse Glimmer在Artificial Analysis第二版测试中幻觉率高达82%，知识工作和幻觉是其最大弱点
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/19/2026
First Seen
Valid until: 11/17/2026
Sources
Related Claims
Unverified在GPQA V2版本测试中，Glimmer的幻觉率高达82%72% similarVerifiedResearch shows that even the most advanced models have hallucination rates in factual question-answering tasks that still fluctuate between 5% and 20%.66% 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65% similarUnverifiedIn scenarios where AI-generated works account for a low proportion (e.g., only 5%), even with a detection accuracy of 90%, more than two-thirds of the flagged works are actually human-created—this is the base rate fallacy in Bayesian reasoning.65% 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62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/772087API
curl https://kongchang.com/api/v1/knowledge/claims/772087MCP
get_claim(id=772087)