Jinsong Wu
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Beyond Causal Masks: Auditing and Localizing Hidden Future-Token Leakage in Adaptive Sequence Models
According to the abstract, a preprint shows how an adaptive routing decoder looked far better in validation, with loss of 1.4259 against 3.5958, because of hidden future-token leakage despite a correct causal mask. Two repaired models removed the apparent advantage.