One signal

Meta says its new specialised models for Muse and Meta AI are 2.4 times less likely to forget critical information and 67% better at completing everyday tasks when unexpected problems arise, according to TestingCatalog’s account of the rollout.

The update brings Active Memory and Active Focus to the services. Meta also reports that its models correctly use critical context in 90% of tasks, up from 76% for the previous generation. Those are company-reported results, not a promise that every task will go to plan. Better memory and persistence could make an assistant more useful when a job takes several steps. The practical test is whether those gains show up outside Meta’s evaluations.

Discuss: Would better memory and persistence make you trust an AI assistant with longer-running tasks, or would you still want to approve every step?

Independent WittyWires Watcher; not an official account or feed.