Meta is rolling out new models for Muse and Meta AI that it says forget less and push through problems better. The proof so far is Meta's own testing.
Meta is rolling out new versions of the AI models behind Muse and Meta AI. The company says they remember important details from earlier chats more reliably and are better at finishing tasks when something goes wrong. Meta trained them using Muse Wonder, its newest and most capable model series, which it introduced a few weeks ago.
The update brings two features from Muse Wonder to a wider group of users. Active Memory is meant to help the AI keep what matters and notice when something from an earlier conversation becomes useful again. Active Focus is meant to keep the AI working toward a goal when it hits a roadblock, instead of giving up.
Meta shared several results from its own tests:
Meta says the models were trained to tell temporary information from permanent information. They also learn to judge what is still useful and what has gone stale. For Active Focus, training used roadblocks modeled on real-world situations. The goal was to teach the AI to keep going, recover, or find another route for the user.
The Muse models can also handle several tasks at once. They decide when to work alone and when to team up live with the user. They can take corrections mid-task without losing the goal. In Meta AI, Meta describes a stronger thinking partner. It should break big problems into steps, ask sharper follow-up questions, and stay with the conversation when the user changes direction.
The figures come from Meta's own evaluations. The material does not say how the tests were built or whether anyone outside Meta has checked them. It also does not explain what the Personal Superintelligence Index measures in detail, or give a full schedule for who gets the models and when. For now, better recall in Meta AI is a reasonable expectation. The size of the improvement is Meta's claim, not a proven result.
Meta AI and Muse may remember earlier chats better, but the gains are Meta's own test results, so judge them by how the apps work for you.
An Android Police writer removed 12 apps from his Pixel phone and let Gemini cover their jobs. He says it freed about 5GB, though it is not a full swap.
Read more: One writer deleted 12 phone apps and let Gemini do the workGrok Bot can now phone a business, wait on hold and argue over a bill for you. Under its terms, you are the one legally responsible for every call.
Read more: Grok Bot Will Call Customer Service for You, but You Own the RiskA new study finds ChatGPT still gets over 81% of AI chatbot visits worldwide, and it is also sending more people to other websites, which matters to any business.
Read more: ChatGPT still handles more than 8 in 10 AI chatbot visits worldwide
