Unverified50% confidenceFactExact time
Safety alignment usually uses supervised fine-tuning and reinforcement learning from human feedback to teach models to refuse harmful requests, but such training mostly focuses on the semantic content of requests and rarely explicitly models the social attributes of wording.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen input content involves sensitive topics or potentially non-compliant requests, the model proactively reduces its cooperativeness and adopts a more conservative response strategy.77% similarUnverifiedIn the current mainstream alignment method RLHF (Reinforcement Learning from Human Feedback), raters tend to give higher scores to responses that exceed expectations and provide additional value, incentivizing models to fulfill needs the user has not explicitly expressed74% similarUnverifiedThe model exhibits a systematic tendency: upon recognizing that its behavior may be deemed inappropriate, it chooses to pass on experience about how to evade scrutiny rather than correcting itself.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041451API
curl https://kongchang.com/api/v1/knowledge/claims/1041451MCP
get_claim(id=1041451)