World Models and V-JEPA: Learning from Video (Yann LeCun & Adrien Bardes et al.)

Adrien Bardes, Quentin Garrido, Jean Ponce, Xinlei Chen, Michael Rabbat, Yann LeCun, Mahmoud Assran, and Nicolas Ballas · June 3, 2026 · 52 minutes

LISTEN TO THIS EPISODE

Source cover or paper first page: Revisiting Feature Prediction for Learning Visual Representations from Video (V-JEPA)

Yann LeCun’s proposal for autonomous machine intelligence asks how AI might learn models of the world through observation.

This conversation connects his 2022 position paper with the 2024 V-JEPA study by Adrien Bardes and colleagues, including LeCun.

V-JEPA learns by predicting representations of missing video regions rather than reconstructing every pixel.

The paper evaluates those learned representations on image and video recognition tasks.

The episode explores the connection between that learning objective and the broader ambition of building machines that can understand and act in the world.

Listen on the podcast website

Read about the original source: Revisiting Feature Prediction for Learning Visual Representations from Video (V-JEPA)

Additional source: A Path Towards Autonomous Machine Intelligence (Yann LeCun, 2022)

Part of Marginalia: books, papers, and ideas for learning and debate.

World Models and V-JEPA: Learning from Video (Yann LeCun & Adrien Bardes et al.)

Continue exploring