Meta reveals Muse Wonder: New flagship AI model with active memory
New AI from Meta navigates unfamiliar tasks (collage: RBC-Ukraine)
Meta has unveiled its new flagship AI model, Muse Wonder, which the tech giant calls the most powerful in its portfolio. The series includes a base version and a scaled-up Muse Wonder Full—both designed for personal and professional AI agents, according to Meta.
Key features: flexible memory and ability to maintain focus
The development of the new lineup resulted from a year-long review of Meta’s AI strategy. Both models were trained on the Prometheus data center with a capacity of one gigawatt, which, according to the company, helped enhance their capabilities.
Developers highlight two key technologies that make the algorithms autonomous:
- Active Memory: The artificial intelligence remembers important details about the user from previous conversations and independently recalls them exactly when needed for a new task.
- Active Focus: The algorithm is capable of working independently on large projects. If an obstacle arises, it seeks alternative paths, and before making important decisions, it clarifies details with a human.
Thanks to this, the artificial intelligence easily navigates complex topics - from analyzing legal contracts and financial reports to decoding medical tests.
(Editor's note: The use of AI in interpreting medical indicators does not replace professional medical consultation!)
Full-scale coding and versatility
Meta placed special emphasis on software creation.
The new models are capable of supporting code development from start to finish: they remember the overall project structure, make corrections in large datasets, and test the program for errors themselves.
Furthermore, tests have shown that the new algorithms handle tasks well that they did not encounter during training. This makes them flexible assistants in real life, where unpredictable situations arise daily.
Privacy and data protection
Since the personal AI assistant has access to personal messages, calendars, and work documents, developers have paid maximum attention to information security. The models are resistant to manipulative instructions and reliably protect personal data from leaks.
The base Muse Wonder model is already available to developers via the Meta Model API, while the Muse Wonder Full version is open to Muse Code subscribers.
In the near future, the new technologies will also appear in the publicly accessible Muse chatbot on a global level.
More interesting:
