Unverified50% confidenceFactExact time
The study defines the Private Sparse RLHF problem: it assumes the reward model is governed by only s* ≪ d relevant features and that preference feedback follows the Bradley-Terry-Luce (BTL) model.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/22/2026
First Seen
Valid until: 1/20/2027
Sources
Related Entities
Related Claims
UnverifiedIn RLHF, if preference data is of low quality or contains systematic bias, the reward model will learn incorrect signals, causing the model's behavior to deviate from expectations.72% similarVerifiedRLHF is essentially a behavioral-level constraint that does not alter the model's underlying capabilities; models trained with RLHF still retain latent knowledge of the restricted behaviors.71% similarUnverifiedSmall-parameter models (such as those at the 1B scale) are prone to sycophancy, tending to give users positive responses rather than critical evaluations, partly because RLHF training reinforces behavior patterns that "keep users satisfied".69% similarUnverifiedRLHF本质上是在模型输出分布上施加统计偏好，而非植入真正的价值理解，导致表面对齐在面对对抗性输入时易崩塌68% similarUnverifiedUncensored models are typically implemented through two approaches: removing refusal preferences from RLHF, or explicitly weakening refusal tendencies using methods such as Direct Preference Optimization (DPO).67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1083974API
curl https://kongchang.com/api/v1/knowledge/claims/1083974MCP
get_claim(id=1083974)