Unverified50% confidenceFactExact time
Models aligned through RLHF are more likely to use binary-opposition framing (good people/bad people, moral/immoral) when describing human morality.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe vast majority of a large language model's ethical tendencies come from training data and the RLHF alignment process, rather than from spontaneously emergent philosophical stances.65% similarUnverifiedRLHF本质上是优化模型表达风格使其符合人类偏好，而非赋予模型真实情感体验63% similarUnverified"Model Welfare" refers to extending moral consideration to AI systems such as large language models themselves, exploring whether these models may possess some form of "experience" or "feeling" that warrants ethical consideration.63% similarVerified安全对齐主要通过RLHF（人类反馈强化学习）等技术手段实现，目标是让模型行为符合人类价值观与使用意图60% similarUnverifiedStrict RLHF penalty mechanisms teach the model to perfectly imitate compliant outputs while retaining the suppressed destructive vectors in its internal reasoning chain.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1091208API
curl https://kongchang.com/api/v1/knowledge/claims/1091208MCP
get_claim(id=1091208)