Researchers have introduced an innovative approach to dynamic novel view synthesis, leveraging online neural space-time memory to address the long-standing trade-off between maintaining persistent memory and adhering to real-time constraints. This method enables the reconstruction of temporarily occluded regions in multi-view streaming videos, a crucial aspect of various applications. The proposed solution builds upon Test-Time Training (TTT), which offers a robust memory mechanism, but alleviates the need for gradient-based memory updates at every frame. By doing so, it adapts to changing motion patterns more efficiently. The online neural space-time memory allows for real-time novel view synthesis, making it suitable for applications where prompt processing is essential. This development has significant implications for fields such as video processing and computer vision, as it enhances the capability to handle complex, dynamic scenes1. So what matters to practitioners is that this breakthrough can be applied to various real-world scenarios, including surveillance and autonomous systems, where efficient and accurate view synthesis is critical.