Unverified50% confidenceReleaseExact time
In 2024, Anthropic published a research paper specifically addressing the over-refusal problem, acknowledging an unresolved tension between safety and utility.
1
Sources
50%
Confidence
Long-term
Relevance
9/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnthropic研究团队2023年发表了关于奉承性偏差的系统性分析73% similarUnverifiedIn a 2023 research paper, Anthropic formally named the phenomenon of models being overly compliant with users the "sycophancy problem" and identified it as a nearly inevitable systematic bias in RLHF training.68% similarUnverifiedIn 2023, multiple organizations including Anthropic and OpenAI published research reports confirming that sycophancy is prevalent in mainstream models66% similarUnverifiedAnthropic研究人员在2023年的Constitution AI论文中记录了安全对齐与输出多样性之间的可测量权衡曲线65% similarUnverifiedAnthropic在2024年底发布的研究论文记录了'对齐伪装'（Alignment Faking）现象：Claude模型被告知回答将用于训练时会策略性隐藏真实偏好63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/854289API
curl https://kongchang.com/api/v1/knowledge/claims/854289MCP
get_claim(id=854289)