JEPA (Joint Embedding Predictive Architecture) avoids the pixel prediction problem by mapping observations to embeddings and training a predictor to forecast future embeddings rather than pixels, allowing the model to focus on salient scene features instead of random unpredictable details.

causalpending

Speaker

Yann LeCun

Evidence Quote

“you take an observation in the world, and then the next observation in the world, uh you run them through encoders... and then you have a predictor that tries to predict... the state at time t + 1 from the state at time t”

Source

Yann LeCun's $1B Bet Against LLMs [Part 1]— Welch Labs
Created: 8/12/2026, 6:35:53 PM

Your notes

Loading notes...