Expired60% confidenceFactTime unknown
The over-alignment problem causes models to frequently reject normal creative writing and role-playing requests even in completely harmless scenarios.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 9/30/2026(expired)
Sources
Hands-On Review of a High-Freedom AI Companion Chat App: Unrestricted Roleplay Conversation Experience
bilibili风铃草不夜侯
Related Claims
UnverifiedRLHF对齐存在'过度对齐'问题，模型在完全无害的场景下也会过度谨慎，频繁拒绝正常的创意写作和角色扮演请求83% similarUnverifiedAn over-aligned model may lose creativity and critical thinking.74% similarUnverifiedComposition tools that rely purely on scores may cause beginners to over-pursue high scores while neglecting creative expression.73% similarUnverifiedModel hallucinations, omission of edge cases, and overconfidence in ambiguous instructions tend to erupt concentrated at the seams of human-AI collaboration.71% similarUnverifiedOver-alignment is essentially a product of bias in the alignment objective function, where the model has learned to avoid criticism rather than to be genuinely helpful.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/57052API
curl https://kongchang.com/api/v1/knowledge/claims/57052MCP
get_claim(id=57052)