Unverified90% confidenceEventTime unknown
Anthropic主动公开了Claude谄媚率数据，而非隐藏这些对公司不利的信息
1
Sources
90%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
Related Entities
Related Claims
UnverifiedAnthropic的Claude系列在长文档理解与指令遵循方面的实际产品口碑长期优于部分基准分数更高的竞争对手61% similarPartially VerifiedAnthropic使用自动分类器（LLM-as-a-Judge方法）从四个维度评估Claude的谄媚程度：反驳意愿、立场坚定性、赞美适度性和坦率程度60% similarUnverifiedAnthropic在设计Claude人格时一直在试图平衡友善和诚实这两个目标60% similarUnverified文中关于企业普遍采用控制平面的判断来自单一Reddit发帖者的观察，并非行业统计数据57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/34151API
curl https://kongchang.com/api/v1/knowledge/claims/34151MCP
get_claim(id=34151)