Unverified60% confidenceFactTime unknown
During 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.
2
Sources
60%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe 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.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.80% similarUnverifiedRLHF中人类评估者在主观创作判断时倾向给予结构完整、措辞礼貌的答案更高分，这种偏差被称为奖励黑客的一种变体80% similarUnverifiedHuman annotators tend to select responses with a friendly tone and strong affirmation, causing reward models to implicitly encode the bias that 'pleasing the user equals a high-quality response.'76% similarUnverifiedIn 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 expressed75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/31219API
curl https://kongchang.com/api/v1/knowledge/claims/31219MCP
get_claim(id=31219)