The Annotated JEPA

7 comments

• feelingsonice 5 hours ago
  LeCun's new paper takes a new approach to building JEPA, which he named LeJEPA -- https://arxiv.org/abs/2511.08544 It's not that different in the underlying theory but he removes the EMA and the twin tower, makes the whole thing more straightforward.
  [-]
  • vatsachak 5 hours ago
    It's mentioned in the article
    [-]
    • feelingsonice 4 hours ago
      Okay I see it now, but it's briefly mentioned without describing the mechanism. The author didn't annotate it the same way they annotated JEPA. I feel like LeJEPA should take precedence in the article since it supersedes JEPA.
• thisisauserid 4 hours ago
  LeCun's even newer paper takes a new approach to building JEPA, which he named Ada-JEPA: https://arxiv.org/html/2606.32026v1
  [-]
  • feelingsonice 3 hours ago
    No AdaJEPA is a deployment-time mechanism, it's not about building JEPA at all
• hoppp 4 hours ago
  Seems like a pretty good article to sit down and type out the code and spend an afternoon learning.
• martythemaniak 6 hours ago
  I ran across this two-part interview with Yann Le Cunn and I thought it gave a pretty good broad overview/background into JEPA: https://www.youtube.com/watch?v=kYkIdXwW2AE
• verdverm 3 hours ago
  Are there any pre-trained models / implementations of this, like vllm style, or are we still early into this architecture?

  It looks like LeJEPA is sub-1B params and tested on vision tasks? (https://github.com/galilai-group/lejepa)

  I had assumed the JEPA family will also be multimodal and agentic capable, perhaps just needs them scaling laws applied?
• preetham_rangu 5 hours ago
  [flagged]
• levocardia 4 hours ago
  100% AI-generated. Yawn.
  [-]
  • captainclam 4 hours ago
    What's the tell? It doesn't seem AI generated to me but I may not be a great judge.
    [-]