AssocVTC: Association-trained Visual Token Compression — train an image encoder to pack a whole scene into few decisive visual tokens for MLLM/embodied-AI context budget. (formerly SR-Diffusion-v3)

鲸鱼也是鱼(a whale is also a fish)

任何简单系统，只要规模足够大且内部存在足够丰富的相互作用，就有可能涌现出智能。——把这句话拆成六条可证伪判据，并逐条用实验检验。 Any sufficiently large simple system with sufficiently rich internal interactions may give rise to intelligence.

Local Graph Neural Network for Sentence Embeddings — PyG GCNConv + learnable adjacency + InfoNCE

DINOv2-giant (frozen) + Qwen3.5-4B VLM for Construction Site image captioning