Meta AI Gets Better Memory, Can Now Handle Tasks Without Giving Up
Meta is making changes to its AI assistants, aiming to make them more useful in everyday situations. The company is upgrading the models behind its Muse agent and Meta AI with two new capabilities borrowed from its advanced Muse Wonder model. The improvements are designed to help AI remember important details from previous conversations and find alternative ways to complete tasks when unexpected problems occur.
According to Meta, the specialised models will begin rolling out to Muse and Meta AI users from Friday. Mark Zuckerberg also announced the development on X, highlighting the improvements in memory and task completion. He said the models were “2.4x less likely to forget critical info” and “67 per cent better at self-correcting and completing everyday tasks”.
Users who want to handle more complex requests can continue using Muse Wonder itself.
Meta AI gets improved memory
A big new feature is Active Memory. Active Memory lets the AI assistant keep information from earlier chats instead of starting every conversation, from scratch.
Meta says the improved models can tell which details are term and which are important enough to keep in mind for later talks. This might be very helpful when people use AI for jobs that last days.
For example, someone planning a holiday may share their travel budget, preferred destinations and other requirements during an initial conversation. With Active Memory, the assistant can use those details later without asking the user to provide the same information again.
Meta claims its upgraded Muse models are 2.4 times less likely to forget important information. During the company’s internal testing, the models successfully used relevant context in 90 per cent of tasks, compared with 76 per cent earlier.
AI assistants can recover from problems
The second addition, called Active Focus, aims to improve how AI assistants respond when something does not go according to plan.
Until now, an AI assistant could struggle when a website stopped responding, a booking process failed or an unexpected issue interrupted a task. With Active Focus, the upgraded models are trained to look for alternative solutions instead of abandoning the process.
Meta reports a 67 per cent improvement in completing everyday tasks when unexpected difficulties arise. The models can also determine when they should continue working independently and when they need assistance from users. They can adjust their approach when users provide corrections during a task.
Meta’s strategy behind specialised AI models
The upgrade is also part of Meta’s broader approach to developing AI assistants. Rather than making its most advanced and expensive model available for every everyday request, the company is using Muse Wonder to train specialised models for specific purposes.
These models are designed to focus on practical abilities such as remembering information, following instructions and completing tasks.
Muse is intended to function as an AI agent while Meta AI can be found on many platforms such, as WhatsApp, Instagram, Facebook and Messenger.
Meta says its aim is to bring “even more capable superintelligence” to more people. Whether these improvements translate into a noticeably better everyday experience will become clearer as more users get access to the upgraded assistants.