Unverified50% confidenceFactExact time
OpenAI、Anthropic、Google等公司持续在有益（Helpful）、无害（Harmless）、诚实（Honest）三者之间寻找平衡点
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
AI越骂越聪明？拆解大模型的行为逻辑与提问技巧
redditr/GoogleGeminiAI7/10/2026
Related Claims
Cite This Claim
Stable URI
https://kongchang.com/claim/488249API
curl https://kongchang.com/api/v1/knowledge/claims/488249MCP
get_claim(id=488249)