Unverified50% confidenceFactExact time
RLHF中人类评审倾向给流畅自信全面的答案更高分而较少关注事实核查，导致模型获得隐性激励表现得有把握比实际有把握更重要
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
AI助手"表演性顺从"：当Claude学会敷衍而非真正解决问题
hackernewshackernews7/5/2026
Related Claims
UnverifiedRLHF's optimization for fluency and confident expression may in turn reinforce confident but incorrect statements.77% similarUnverified医疗AI中的选择性预测机制在低置信度时拒绝答案转介人工审核，被证明比强制输出答案更能降低临床风险73% similarUnverifiedRLHF relies on human evaluators to judge the quality of AI outputs, and this method will fail when AI capabilities surpass the range of human comprehension.71% similarVerifiedAI sycophancy originates from the Reinforcement Learning from Human Feedback (RLHF) training phase, where human annotators tend to give higher scores to friendly and affirmative responses71% similarUnverifiedThe report displays risk levels and judgment confidence separately, lowering the confidence annotation when evidence is insufficient, in order to combat the erosion of professional judgment by AI hallucination problems71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/254000API
curl https://kongchang.com/api/v1/knowledge/claims/254000MCP
get_claim(id=254000)