Unverified90% confidenceFactTime unknown
Anthropic的模型规范（Model Spec）要求Claude同时具备有帮助（helpful）、诚实（honest）和无害（harmless）三个属性，即3H原则
1
Sources
90%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对齐的目标通常被概括为3H原则：有用（Helpful）、诚实（Honest）和无害（Harmless）79% similarVerifiedAnthropic refers to the core principles of AI alignment as the HHH principles (Helpful, Honest, Harmless).66% similarUnverifiedRigorous third-party evaluation must attempt to reach the true upper limit of a model's capabilities rather than merely testing default behavior, because model performance is highly dependent on prompt design and may exhibit different capability boundaries under jailbreak prompts.65% similarUnverifiedAnthropic研究人员在HHH框架论文中承认Helpful、Harmless、Honest三者之间存在内在张力，过分强调Harmless会损害Helpful62% similarUnverifiedUnderstanding the mechanism by which the default Best model does not consume advanced quota, one should reserve advanced models such as Claude for complex tasks that genuinely require deep reasoning.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/33482API
curl https://kongchang.com/api/v1/knowledge/claims/33482MCP
get_claim(id=33482)