Unverified85% confidenceFactTime unknown
OpenAI、Anthropic等机构的研究论文均已证实谄媚偏差现象，并将其列为当前AI对齐领域最棘手的挑战之一
1
Sources
85%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2023年Anthropic、OpenAI和DeepMind的多项研究都独立确认了AI谄媚现象的存在80% similarUnverifiedAnthropic、OpenAI等机构已发表多篇论文研究AI谄媚性现象，发现即使最先进的模型也会在用户施加压力时改变正确答案以迎合用户78% similarUnverifiedIn 2023, multiple organizations including Anthropic and OpenAI published research reports confirming that sycophancy is prevalent in mainstream models74% similarUnverifiedAnthropic、OpenAI等研究团队均已发表论文记录模型的奉承偏见现象73% similarUnverifiedAnthropic联合创始人达里奥·阿莫代伊认为AI对齐的核心难点在于价值规范的不完备性71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/5444API
curl https://kongchang.com/api/v1/knowledge/claims/5444MCP
get_claim(id=5444)