Competition in the field of artificial intelligence is accelerating toward deep interaction and the long-term ability to execute tasks. On November 6, Meta officially released a new research announcement, announcing that it is rolling out a series of specialized models trained on the cutting-edge large model Muse Wonder to the broad user bases of Muse and Meta AI, fully empowering everyday conversations and complex task handling.
A Leap in Core Capabilities: Personal Superintelligence Index Rises Sharply
The specialized models introduced this time inherit the industry-leading performance demonstrated by Muse Wonder in handling complex tasks in work and daily life.
In the latest test of the authoritative evaluation system “Personal Superintelligence Index,” the score of the new-generation Muse model achieved a significant leap, rising sharply from 38% in the previous generation to 47%, demonstrating stronger overall intelligence.
Two Core Capabilities Arrive: Active Memory and Active Focus
To address the industry pain points where traditional AI easily forgets key information and is prone to interruption when facing unexpected situations, Meta has introduced two major killer features in the new model: Active Memory and Active Focus.
In terms of Active Memory, the model has the ability to perform deep reasoning across time, accurately discern the timeliness of information, distinguish what is temporary from what is long-term, and thereby remember important details over conversations lasting months with users. Official evaluation data shows that in the Active Memory evaluation, the current model correctly uses key context in as many as 90% of tasks, a qualitative leap compared with 76% in the previous generation; the error rate for forgetting key information has also been greatly reduced by a factor of 2.4.
In terms of Active Focus, the model has undergone specialized training for various typical task bottlenecks in the real world. Even when encountering unexpected challenges while performing everyday tasks, the model can try harder, recover from setbacks, or proactively find alternative solutions for users. Relevant data shows that when handling everyday tasks prone to unexpected problems, the new model’s performance improved by as much as 67%.
Behavioral Pattern Optimization and Full Ecosystem Rollout
In addition to the significant enhancement of underlying capabilities, Meta has also fine-tuned the specialized models for the different application scenarios of its two flagship products.
In Muse, the model has learned how to become an excellent multitasker, clearly discerning when it should independently advance the user’s goals and when it should collaborate with the user in real time, while perfectly adapting to user guidance and clarification during execution to avoid confusion. In Meta AI, the same model upgrade transforms it into a better thinking partner, able to break down complex problems into clear steps, ask more precise follow-up questions, and seamlessly keep pace when the user changes the direction of the conversation midway.
By bringing these cutting-edge technologies down to its mainstream product matrix, Meta is enabling more ordinary users to truly experience superintelligence with long-term memory and autonomous task-focusing capabilities.
