Definition
Plain language
A large ready-made vision system that already understands objects, motion, and layout in ordinary images.
As stated in the literature
A self-supervised pretrained vision backbone whose frozen features provide a smooth, semantically structured latent space reused here as the world model's frame representation.
Also called: DINOv2, DINO
Why it matters: Reusing a ready-made vision system saves enormous training effort and gives other models a reliable understanding of images to build on.
For example, a new system can skip learning to recognize objects from scratch and instead reuse DINOv3's existing sense of what's in a picture.
Heard on the show
“The teacher here is an exponential moving average of the student — a slowly updating running average of its own recent weights, borrowed from BYOL and DINO.”Episode 242 — Making a Vision Model Better by Showing It Blurry Images