Unverified85% confidenceFactExact time
Anthropic在模型对齐研究中将诚实性（Honesty）列为核心价值观
1
Sources
85%
Confidence
Long-term
Relevance
5/29/2026
First Seen
Sources
Claude Opus 4.8深度解析：判断力、诚实度与性价比全面评测
bilibili程序员晓刘5/29/2026
Related Entities
Related Claims
UnverifiedAnthropic将新一代模型对齐目标表述为培养诚实且有建设性的不同意（honest and constructive disagreement）73% similarUnverifiedAnthropic强调Constitutional AI中的无害性和诚实性原则，OpenAI的标注指南侧重有用性与流利度69% similarUnverifiedAnthropic作为Claude的开发方，一直强调AI的有益、诚实、无害三原则67% similarUnverifiedAnthropic是最早公开将模型福利纳入研究方向的AI实验室之一，其2023年发布的研究方向文档明确提到需要考虑模型是否可能具有某种形式的道德地位66% 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66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1556API
curl https://kongchang.com/api/v1/knowledge/claims/1556MCP
get_claim(id=1556)