Verified85% confidenceFactTime unknown

Constitutional AI（CAI）方法试图通过让模型遵循一组明确的行为原则来缓解谄媚问题

5

Sources

85%

Confidence

Long-term

Relevance

6/1/2026

First Seen

Sources

Claude谄媚问题研究：灵性话题38%对话存在讨好行为

rss

Related Entities

Constitutional AIconcept Anthropiccompany

Related Claims

UnverifiedConstitutional AI（CAI）是Anthropic于2022年发布研究论文时提出的对齐方法，核心思想是用一套书面原则替代大量人工标注的偏好反馈81% similar UnverifiedAnthropic在Constitutional AI（CAI）框架中探索了让模型评估自身输出的方法论79% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/31220

API

curl https://kongchang.com/api/v1/knowledge/claims/31220

MCP

get_claim(id=31220)