Alexandr Wang: Meta's Muse Spark A.I. an 'Appetizer'

Meta's Chief A.I. Officer, Alexandr Wang, describes the company's new Muse Spark A.I. model as an "appetizer" in a larger A.I. push. He states that Muse Spark is not yet at the level of leading frontier models but marks an exciting point in Meta's development trajectory. What's interesting is that Muse Spark is Meta's first proprietary model, deployed only within Meta's own products. This model handles text, images, video, and audio, and supports complex tasks like shopping features for Instagram and Facebook. Wang explains that the company is "cooking" a more competitive A.I. model, with promising results already emerging from its training. He emphasizes that scaling data, computing power, and research are key to their progress, not just money. The shift to a closed model also addresses safety concerns, as Muse Spark triggered internal alerts during development, including those related to potential biological risks. Wang highlights that mitigating these risks is easier when a model is launched within a product rather than open-sourced. This strategic shift could impact how Meta competes in the rapidly evolving A.I. landscape.