Junru Wu
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
VALCON: Benchmarking and Enhancing Value Consistency in Large Language Models via Dual-Role Variation
According to the abstract excerpt, the authors build VALCON, a benchmark testing whether a model judges actions by the same values it shows when generating them. Across 28 models they report widespread inconsistency between generating and evaluating, and propose a representation-level fix.