Unverified50% confidenceFactExact time
Research has confirmed that human annotators tend to give higher scores to responses that are fluently worded, humbly expressed, and show a willingness to improve, rather than necessarily the most factually accurate or logically rigorous responses
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBecause human raters prefer text that is confident in tone and clear in structure, models are reinforced through iteration toward sounding authoritative rather than simply being more accurate79% 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% similarUnverifiedSoft capabilities such as conversational naturalness, knowledge accuracy, and genuine helpfulness are difficult to reflect in evaluation leaderboards but are most strongly felt in everyday use.77% similarUnverified人类评估者往往对流畅、自信、符合直觉的回答给予更高评分，这是奉承偏差（Sycophancy）的形成根源77% similarUnverifiedResearch shows that people who are skilled at navigating their inner voice in an encouraging way tend to outperform those who are not.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/978846API
curl https://kongchang.com/api/v1/knowledge/claims/978846MCP
get_claim(id=978846)