According to the official blog, the human-like model excels in analysing and accurately completing unfinished images, surpassing the capabilities of current models.
The I-JEPA model utilises contextual understanding of the world to fill in gaps in images, instead of relying solely on nearby pixels like traditional generative AI models.
Helmed by Meta’s Chief AI Scientist Yann LeCun, the AI model which ‘learns by creating an internal model of the outside world, and compares it with the abstract representations of images’.
Meta is open-sourcing I-JEPA's checkpoints and training code, as the company believes that sharing its models with others will aid the firm in "spurring innovation, spotting safety gaps, and lowering costs".
Cyclone Biparjoy: IMD Issues Red Alert A Day Ahead Of The Cyclone’s Landfall In Gujarat