Meta Introduces Specialized Muse and Meta AI Models Trained With Muse Wonder

Meta has introduced specialized versions of its existing models, trained using the Muse Wonder series, and is rolling them out to people who use Muse and Meta AI. The company says the update brings key frontier capabilities from Muse Wonder to more users, including improvements measured on the Personal Superintelligence Index, Active Memory, and Active Focus. The move follows the recent introduction of Muse Wonder as Meta's most capable models to date.

Performance on the Personal Superintelligence Index

Meta reports that its current Muse model scores 47 percent on the Personal Superintelligence Index, compared with 38 percent for the previous Muse model. The company presents this as part of its effort to bring capabilities from Muse Wonder to the models behind Muse and Meta AI.

Active Memory and Active Focus

Active Memory enables Muse and Meta AI to remember what matters and recognize when earlier information becomes relevant, making it easier to build on past conversations. Meta says its models have been trained to recognize temporary versus permanent information, reason across time, and distinguish relevant details from stale ones. In Muse, improved models show a 2.4x reduction in forgetting critical information needed to help users achieve goals, and the Active Memory evaluation shows 90 percent of tasks correctly using critical context, up from 76 percent.

Active Focus enables Muse to continue working toward a goal even when unexpected challenges arise. Meta says it trained models on real world bottlenecks so they can try harder, recover, or find alternatives on a user's behalf. In Muse, models show a 67 percent improvement in completing everyday tasks where unexpected issues arise.

Specialized Model Behaviors

Meta says its specialized models also leverage Muse Wonder's performance on behaviors especially important to Muse. They are described as excellent multi-taskers that know when to push on goals independently or collaborate live with a user. They can also better adapt to steers and clarifications in the middle of running tasks without getting confused.

In Meta AI, the same improvements are intended to make the system a better thinking partner and collaborator. It can break complex problems into clear steps, ask sharper follow-up questions, and keep up when a user changes direction mid-conversation. Meta says the goal is to give more people access to even more capable superintelligence.

Sources: Meta-Ai-Research · The New Stack