Unverified70% confidenceFactTime unknown
2023年以来，多项研究表明谄媚性在多个主流大模型中普遍存在，且在主观性强的话题中尤为突出
1
Sources
70%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2023年UC Berkeley研究团队在论文中证实，主流奖励模型存在显著的长度偏见，在内容质量相同情况下更长回答系统性获得更高奖励分数63% similarUnverified学术界对涌现性存在争议，部分研究者认为涌现现象可能是评估指标选择不当造成的测量伪影63% similarUnverified行为经济学家Richard Thaler的研究揭示了现时偏误现象，即人类倾向过度权重当下成本而折现未来收益62% similarUnverifiedResearch finds that when a prompt assigns the model the role of a rigorous academic researcher, both its citation density and authenticity are significantly higher than in a general conversational role.61% similarUnverifiedNew model versions typically introduce stricter content filtering and behavioral constraints, with side effects including collateral damage to serious topics such as psychology and philosophy, degraded persona consistency, and reduced utility.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/30918API
curl https://kongchang.com/api/v1/knowledge/claims/30918MCP
get_claim(id=30918)