In this episode, Katherine Forrest and Scott Caravello break down three generative AI architectures—transformers, JEPA, and diffusion models—exploring what sets each apart and how they overlap. They also discuss Manifold-Constrained Hyper-Connections, a recent innovation aimed at improving how transformer layers communicate during training.

For the sources referenced in this episode, please see the links below:

DeepSeek AI: mHC: Manifold-Constrained Hyper Connections

##

Learn More About Paul, Weiss’s Artificial Intelligence practice: