Meta has introduced new AI models, Muse Wonder, for personal AI agents.

Meta has introduced the Muse Wonder series of AI models, which includes Muse Wonder and Muse Wonder Full. According to the company, these are its most powerful models to date. They are designed to work with personal and professional AI agents that help perform complex tasks in work and daily life.

The new models feature two key capabilities: Active Memory and Active Focus. Active Memory allows the model to retain important information about the user and consider previous conversations when performing new tasks. Active Focus enables the model to work longer on complex requests, try different solution approaches, and adapt to new information. When necessary, the model can ask clarifying questions and coordinate important actions with the user.

Muse Wonder and Muse Wonder Full can also handle professional information in the fields of law, finance, and health, analyze documents, and write code. Meta evaluated their capabilities using tests in programming, memory, and problem-solving. The company also noted that the new models perform better on unfamiliar tasks that were not a primary part of their fine-tuning.

Meta paid special attention to safety. Since personal AI agents may access messages, calendars, and personal documents, the models were trained to handle confidential information carefully and resist malicious instructions. According to the company, the new models have improved mechanisms to prevent abuse in cybersecurity and other high-risk areas.

Muse Wonder is already available via the Meta Model API and Muse Code. The more powerful Muse Wonder Full version is available to Muse Code subscribers. Soon, Muse Wonder will begin operating within the Muse service, initially becoming available to subscribers. Meta also plans to use the new model to create specialized AI tools and automate tasks requiring more complex information processing.

The development of Muse Wonder is part of Meta’s strategy to build a personal superintelligence—an AI capable of understanding an individual’s needs and helping them achieve their goals. The company will continue to advance this direction and plans to publish additional results from its model safety evaluations.

Source: meta.ai