If autoregressive models are trainable digital biology, then diffusion models are trainable digital chemistry. And the underlying linear transforms and non-linear partitions are physics.
Posters: eshear. eshear
3 likes recorded by the source corpus.
Topics: AI & LLMs