Meta Platforms announced on Tuesday that it would give researchers access to components of its new artificial intelligence (AI) model, I-JEPA, which can complete unfinished images with higher accuracy than previous models. The AI model utilizes world knowledge to fill image gaps, leading to fewer errors often seen in AI-generated images, such as extra fingers on a hand.
Meta, the parent company of Facebook and Instagram, frequently publishes open-source AI research through its internal research lab. CEO Mark Zuckerberg believes sharing models developed by Meta’s researchers aid innovation, identify safety gaps, and reduce costs.
Despite industry warnings about the potential hazards of AI technology, Meta’s executives have refrained from signing a statement equating its risks to pandemics and wars. Yann LeCun, Meta’s top AI scientist and one of the “godfathers of AI,” advocates for including safety checks in AI systems and criticizes “AI doomers.”
Meta is also beginning to integrate generative AI features into its consumer products, such as ad tools capable of creating image backgrounds and an Instagram feature that modifies user photos based on text prompts.
