LeCun clarifies JEPA: the real split is joint embedding vs generative, not the predictor
jbhuang0604 · x · 2026-11-10
Responding to ongoing debate, Yann LeCun explains that the key distinction isn't between JEPA variants with or without a predictor, but between joint-embedding and generative architectures. Generative models predict y from x with a divergence loss, while joint-embedding architectures compare x and y in embedding space; whether a predictor is needed depends on the task. jui-huang (jbhuang0604) distills the argument into a single diagram.
More from Research
- Hazard-free Boolean circuits can be exponentially more expensive — and NP-hard to build — thomasahle · 2026-11-10
- Constructing hazard-free Boolean circuits proven NP-hard — thomasahle · 2026-11-10
- Unitree Go2 robot asks clarifying questions via Gemini and Bayesian active learning — DynamicWebPaige · 2026-11-10
- Visual Token Routing cuts computer-use agent inference cost up to 3.3x with 2.5-5.8x fewer tokens — wellecks · 2026-11-10
- SimFoundry auto-builds interactive 3D sim scenes from video for robot policy training — DynamicWebPaige · 2026-11-10
- CMU paper: one unmodified computer-use agent wins prizes across four embodied AI contests — wellecks · 2026-11-10