Meta applications are fully integrated with the Muse Spark large-scale model, enabling multiple agents to work in parallel with a single command.

According to AIMPACT, on May 13th (UTC+8), Beating reported that Meta announced the full rollout of its latest large-scale model, Muse Spark, developed by Meta Superintelligence Labs, to its core products, including WhatsApp, Instagram, Facebook, Messenger, and Ray-Ban Meta and Oakley Meta smart glasses. Billions of users will directly experience the next generation of Meta AI, featuring visual perception and multi-agent parallel processing capabilities. The biggest underlying change in this update is the introduction of a parallel processing mechanism for subagents. Previously, AI could only answer questions linearly; now, when a user makes a complex request, Meta AI can simultaneously activate multiple agents in the background to act independently. For example, when a user requests to plan a family trip, the system will have one agent schedule the itinerary, another compare destinations, and a third search for activities suitable for children. Finally, the results from all three will be synchronized, significantly reducing waiting time for complex tasks. At the interaction level, Muse Spark completes Meta's real-time multimodal perception capabilities. Leveraging smart glasses or smartphone cameras, users can ask AI questions using natural voice while observing the real world, interrupting or switching languages at any time during the conversation. The model can directly read environmental information, such as scanning airport shelves to find snacks with the highest protein content, or helping users interpret medical charts. Furthermore, the model supports visual code generation; users only need to provide a single-sentence prompt to directly generate customized web pages or mini-games. To fully leverage its super app matrix, Meta has also deeply embedded AI into social and e-commerce scenarios. Users can bring up a sidebar in group chats, allowing AI to provide private answers based on the chat context; the shopping mode enables price comparison and map aggregation across Facebook Marketplace and the entire internet. Compared to other large model vendors relying on independent apps, Meta is using its massive software and hardware traffic pool to transform large models into ubiquitous infrastructure. (Source: ME)