生數科技完成近20億元B輪融資，以通用世界模型定義下一代數位與物理世界生產力底座

🔑 增強重點摘要

• •The funding round was led by prominent investors including Alibaba, Baidu, and Zhipu AI, signaling strong strategic backing from China's major AI ecosystem players.
• •Shengshu Technology is best known for its Vidu model, a video generation AI capable of producing high-consistency, 16-second long videos from text prompts.
• •The capital injection is specifically earmarked for scaling compute infrastructure and accelerating the development of multimodal world models that move beyond simple video generation into interactive simulation.

🛠️ 技術深入

• •Utilizes a U-ViT (Unified Vision Transformer) architecture, which treats visual data as tokens, allowing for more efficient scaling compared to traditional U-Net based diffusion models.
• •Employs a proprietary 'Diffusion Transformer' approach that integrates spatial-temporal attention mechanisms to maintain object permanence across long-duration video sequences.
• •Focuses on 'World Model' training objectives, where the model is trained to predict future frames based on physical laws and causal relationships rather than just pixel-level interpolation.

🔮 前景展望基於引用來源的 AI 分析

⏳ 時間線