Unverified60% confidenceFactExact time
In 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 expressed
2
Sources
60%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
Flagship Model vs. Local 27B: Which Follows Code Instructions More Faithfully?
redditr/ollama7/9/2026
Related Claims
Unverified人类反馈强化学习（RLHF）机制使得人类偏好数据的战略价值急剧攀升，这类数据难以自动化生成，必须依赖真实用户的判断与反馈82% similarUnverifiedRLHF（基于人类反馈的强化学习）会放大规避风险效应，因为人类评分者倾向于给保守、无争议的输出打高分81% similarVerifiedResearch shows that human evaluators in RLHF tend to prefer longer, more detailed, and friendlier-toned responses, even when those responses are not superior in factual accuracy.77% similarUnverifiedThe root cause of sycophantic behavior can be traced to the RLHF training paradigm, in which human evaluators prefer responses that agree with their own views.76% similarUnverifiedDuring the RLHF stage, human annotators tend to give higher ratings to responses that sound friendlier and more agreeable to the user, causing models to develop a systematic sycophantic tendency.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/460372API
curl https://kongchang.com/api/v1/knowledge/claims/460372MCP
get_claim(id=460372)