Meta brings specialized models to Muse and Meta AI

Meta says the specialized models are versions of its existing models trained using Muse Wonder, and that it is rolling them out to people who use Muse and Meta AI.[1] The company says both Active Memory and Active Focus are coming to the specialized models in Muse and Meta AI.[1] Those details would matter to teams deciding when to test the features or whether to plan around them, so users should look for further availability information rather than assume a particular timeline.

Meta describes Active Memory as helping Muse and Meta AI remember relevant information from earlier conversations.[1] It says it trained the models to distinguish temporary information from permanent information.[1] For Muse, Meta reports that the improved models reduce forgetting critical information needed to help users achieve goals by 2.4 times.[1] It also reports a 90% score for current Muse models, compared with 76% for previous models, on tasks that correctly use critical context.[1]

Meta says Active Focus helps Muse keep working toward a goal when unexpected challenges arise.[1] It reports that Muse models improved 67% at completing everyday tasks where unexpected issues arise.[1] Meta also says the models can multitask, choosing when to pursue goals independently or collaborate with the user.[1] The specialized Muse models can adapt to steers and clarifications while tasks are running.[1] Meta says they can break complex problems into clear steps and ask sharper follow-up questions.[1] The figures describe Meta’s reported task results, not a guarantee about performance on any particular workflow.

Meta says it introduced the Muse Wonder series a few weeks before announcing these specialized models.[1] Meta also reports current and previous Muse model index scores of 47% and 38%.[1]