Yong-An Guo

Structural semantic organization distillation for cross-task transfer from prompt-conditioned foundation models to CNN detectors

It is argued, and empirically observe, that value-level alignment saturates in this setting: whether through concept-guided response matching or a learnable projection, the supervision target remains segmentation-shaped, so bridging activations does not resolve the underlying mismatch.

Minimal AIF-inspired response control for evidence-aware and dependable LLM behavior