interpretability
critique
bearish
Existing multimodal knowledge editors rarely control the semantic boundary of each edit, leading to a scope gap where instance-level success doesn't guarantee proper generalization
Computation and Language27 Jul 2026