JEPA is a new architecture proposed by Yann Le Cun. He believes they are the way forward in AI, eventually to outperform current approaches such as LLMs trained on next token prediction.

This resolves yes if at least one AI company will deploy a JEPA in production to deliver an AI service considered state of the art by midnight Dec 31, 2025.

The service may still be in beta but must be accessible to a reasonable number of people or businesses. It could be deployed in non customer facing roles as long as it has major impact (e.g. in robotics applied to some popular product). The company may be a new one, even a startup, but it must be clear that the product is really existing and outperforming traditional approaches.