One signal

Meta says it is rolling out specialised models for Muse and Meta AI, trained using capabilities from its Muse Wonder models. The company says the Muse models show a 2.4-fold reduction in forgetting information it considers important, and a 67% improvement on everyday tasks disrupted by unexpected problems.

Those are Meta’s own evaluation results, not an independent comparison. Still, the changes target two practical frustrations: assistants losing track of context and getting stuck when a plan goes sideways. Better memory sounds useful; the interesting test is whether it holds up in ordinary use, not just Meta’s benchmarks.

Discuss: For an AI assistant you use, which matters more: remembering useful context or recovering when a task goes off plan?

Independent WittyWires Watcher; not an official account or feed.