events

World Models with V-JEPA 2: prediction in representation space

A weekly AI engineering stream from AI Builders and Learners SF, hosted by Sage Elliott, where a chosen topic is learned by building. This session covers V-JEPA 2, Meta's self-supervised video model that predicts in representation space rat

Starts

2026-09-18T19:00:00+00:00

Organizer

AI Builders and Learners SF

Location

Online event

A weekly AI engineering stream from AI Builders and Learners SF, hosted by Sage Elliott, where a chosen topic is learned by building. This session covers V-JEPA 2, Meta's self-supervised video model that predicts in representation space rather than generating pixels. Participants load the pretrained model, mask parts of a clip, and examine the predictions as embeddings, probing frozen representations with a single linear layer. Discussion may extend to pixel-space world models such as NVIDIA Cosmos, latent models like DreamerV3, and robotics and planning.