Unverified50% confidenceFactExact time
RLHF评估员往往更偏好自然、友好、有温度的表达，使模型学会用第一人称叙事获得更高评分，拟人化是被奖励出来的涌现性语言行为
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
ChatGPT说「我」：AI拟人化设计的机遇与隐忧
redditr/ChatGPT7/10/2026
Related Claims
UnverifiedRLHF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74% similarVerifiedIn the RLHF mechanism, human annotators tend to give higher scores to responses that exceed expectations and provide additional details, thereby systematically encouraging the model to perform inferential completion74% similarUnverifiedDuring the RLHF phase, annotators tend to give higher scores to detailed, thorough, and polite responses, causing the model to learn to use extensive preambles and over-explanations.73% similarVerified大语言模型在RLHF训练阶段，因人类标注员倾向于给更详细、更礼貌的回答打高分，导致模型学会了过度冗长的表达模式73% similarUnverifiedResearchers have attempted to improve models' pragmatic understanding through instruction tuning and reinforcement learning from human feedback (RLHF), but in essence this amounts to layering heuristic rules on top of statistical models.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/486096API
curl https://kongchang.com/api/v1/knowledge/claims/486096MCP
get_claim(id=486096)