SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture

要約

Recent large vision-language models (VLMs) remain fundamentally constrained by a persistent dichotomy: understanding and generation are treated as distinct problems, leading to fragmented architectures, cascaded pipelines, and misaligned representation spaces. We argue that this divide is not merely…