Unverified50% confidenceOpinionExact time
Because 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 accurate
1
Sources
50%
Confidence
Long-term
Relevance
9/22/2026
First Seen
Sources
Related Entities
Related Claims
Unverified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78% similarUnverifiedUsing the same model for both generation and self-evaluation introduces self-reinforcement bias; introducing an independent evaluator provides a more objective quality assessment.74% similarUnverifiedResearch shows that LLM judges tend to favor outputs that are confident in tone, longer in length, or stylistically similar to themselves, and are more lenient toward jailbreak-style phrasing.74% similarUnverifiedResearch 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.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/940222API
curl https://kongchang.com/api/v1/knowledge/claims/940222MCP
get_claim(id=940222)