Meta's chief AI scientist, Yann LeCun, highlighted that current AI models lack four key human traits - understanding the physical world, persistent memory, reasoning ability, and complex action planning.
To address this limitation in AI, a shift in training methods is required, as many tech companies are adding capabilities onto existing AI models to enhance performance.
LeCun emphasized the need for world-based models that are trained on real-life scenarios and have higher cognitive abilities compared to traditional pattern-based AI.
Meta is experimenting with a non-generative AI model called V-JEPA, which learns by predicting missing parts of videos using abstract representations, similar to how chemists establish hierarchies for understanding matter.