Unverified50% confidenceFactExact time
Self-preference and style blind spots contaminate the downstream optimization loop: if the reward model favors a certain type of output, the policy model drifts in that direction rather than toward users' true preferences.
1
Sources
50%
Confidence
Long-term
Relevance
11/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSelf-scored RLHF replaces human annotation with the model's own scores, reducing costs, but the reward signal comes from the model itself, which may amplify existing biases and lead to a self-reinforcing loop.76% similarUnverifiedWhen external incentives (XP, leaderboards) are too strong, they may actually undermine users' intrinsic motivation for the habits themselves.74% similarUnverifiedThe "Inoculation Strategy" marketing tactic relies on two psychological mechanisms—anchoring effect and loss aversion—by first exposing the risks of low-price solutions and then presenting one's own pricing.74% similarUnverifiedCurrent large models generally exhibit a "pleasing tendency" (sycophancy), tending to affirm users' views rather than challenge them.74% similarUnverified用户在面对海量选择时倾向于选择已有社会证明的项目,形成强者愈强的累积优势正反馈循环73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1160923API
curl https://kongchang.com/api/v1/knowledge/claims/1160923MCP
get_claim(id=1160923)