Trust Your Guide Only When Certain: Uncertainty-Aware Sparse Alignment at Inference Time
This work proposes TUSA (Trust-based Uncertainty Sparse Alignment), a dynamic arbitration process that reframes alignment as a dynamic arbitration process, introducing an uncertainty-aware arbiter that authorizes intervention only when two conditions are met: the supervisor is confident and the token is semantically salient.
Zeen Zhu, Zhuo Li, Wei-Yang Guo et al.
· 0 citations