Tyler Frink·Apr 25The Geometry of Multimodal FusionDeepMind’s Gemini Embedding 2 and the 3072-dimensional latent manifold — how modality-specific encoders enable native cross-modal fusion
Tyler Frink·Jan 12Inside-Sora’s-Architecture.How Modern Video Diffusion Models Learn Motion, Time, and Sound — and Why Coherence Still Breaks
Tyler Frink·Jul 14, 2025The Anatomy of JEPA: The Architecture Behind embedded Predictive Representation LearningHow Yann LeCun’s Vision for World Modeling Could Reshape Deep Learning
Tyler Frink·Jul 9, 2025The Anatomy of Veo 3:DeepMind’s Audiovisual Diffusion ModelHow 3D Convolution, Unified Latents, and Prompt Control Power the Future of Video Generation