The future of the metaverse appears to be taking a turn toward virtual reality. On June 5, Apple unveiled its own virtual-reality headset, a $3500 device that’s targeted at developers and expected to make inroads into the enterprise. While it may take several years for the device to be adopted widely, its significant investment and research hint at Apple’s desire to influence technology innovation in a major way.
This development could spell trouble for the social media giant Meta, which is attempting to leverage its huge user base to enter the virtual world of the metaverse, where people can interact with each other using their digital avatars. Despite the hype surrounding the metaverse, it remains difficult for companies like Meta to get their products in front of the right audience.
To address this challenge, Meta announced on Tuesday that it would provide researchers with access to components of a new “human-like” artificial intelligence model that it said can analyze and complete unfinished images more accurately than existing models. The model, dubbed I-JEPA, uses background knowledge about the world to fill in missing pieces of images, rather than looking only at nearby pixels like other generative AI models. This approach is designed to reduce errors commonly found in AI-generated images, such as hands that have extra fingers.
Developed by Meta’s top AI scientist, Yann LeCun, I-JEPA uses an abstraction technique that is similar to how humans recognize and understand objects. It also focuses on the relationships between objects and scenes, rather than just individual pixels. These features allow I-JEPA to create realistic images that are virtually indistinguishable from human-generated images.
To create an image, I-JEPA starts with a sketch that the user provides. Then it analyzes the sketch to identify what parts of the scene are most important, and then uses its knowledge of the world to fill in those areas, generating an image that is indistinguishable from the original sketch. I-JEPA is able to generate realistic and detailed images with only one pass, which is faster than previous algorithms that required multiple passes.
Meta is continuing to invest heavily in AI, and has a long history of publishing open-source research conducted by its dedicated internal research lab. The company’s CEO, Mark Zuckerberg, believes that sharing research models helps to drive innovation, assists in identifying safety concerns and cuts costs for Meta by allowing the company to leverage advancements made by others. The company has also started incorporating generative AI functionality into its consumer-oriented products, including ad tools that can generate image backgrounds and an Instagram product that can modify photos based on text prompts. Ultimately, these technologies can help Meta to attract more users to its products. Benzinga’s experts are on the hunt for hidden gems in this space.
