Meta Platforms, the company that owns Facebook and Instagram, has announced that it will be giving researchers access to components of its new human AI model. This model will allow you to analyze and complete unfinished images much more efficiently than current models.
The revolutionary approach of the I-JEPA model
The I-JEPA model uses the accumulated knowledge about our world to fill in the gaps in the images. Instead of relying only on neighboring pixels, as modern generative models do, I-JEPA applies human-like reasoning. This helps avoid errors that often occur in AI-generated images, such as images with the wrong number of fingers on the hands.
Publication of research results and collaboration
Meta actively publishes the results of its own open source artificial intelligence research. The company seeks to collaborate with researchers by providing them with the opportunity to use the models they have developed. The head of the company, Mark Zuckerberg, argues that such a partnership will help stimulate innovation, reduce costs and identify possible security problems.
Response to industry warnings
Meta management has effectively dismissed warnings from other leading companies about the dangers associated with the development of artificial intelligence. They refused to sign a joint statement backed by Open AI, DeepMind, Microsoft and Google that compares the risks of AI development to a pandemic and nuclear war, notes NIX Solutions. Yann LeCun, an AI expert at Meta, has spoken out against a lot of panicky predictions, arguing that it’s enough to implement security algorithms in AI systems.
Integration of generative AI functions
Meta is also starting to implement generative AI features in custom products, including advertising tools that can create backgrounds for images or products for Instagram. These tools allow you to edit user photos based on text queries.