Apple put MCP support and document-context assistants into Keynote, Pages and Numbers, and an OpenAI-backed assistant into Logic Pro
AspireAI community · Apple
Apple announced today that Apple Creator Studio gains AI assistants built on a combination of Apple-built models and OpenAI, plus Model Context Protocol support so external agents can read document content, perform edits and execute automated actions directly inside Keynote, Pages and Numbers. Logic Pro for Mac gets its own assistant, which Apple says helps navigate production workflows, start a song, set up a vocal track from a single prompt, or refine an active project. The same release brings built-in MotionVFX content to Final Cut Pro, video assets in the Content Hub, a redesigned Browser in Logic Pro with on-device music understanding, a Stem Splitter that now separates drums, and Generate Fill in Pixelmator Pro.
Updates ship on the App Store in November as a free update for existing subscribers; new subscribers pay $12.99 a month or $129 a year in the US, with a one-month trial and $2.99 for students and educators.
The honest limit: every assistant feature Apple announced is labelled beta, works only in US English, and is gated on specific hardware - the Logic Pro assistant needs macOS 27 Golden Gate or later on an Apple Intelligence-capable Mac, the Keynote, Pages and Numbers assistants need iOS 27, iPadOS 27 or macOS 27 or later, and MCP support needs macOS 27 or later. Nothing here is available today.
https://www.apple.com/newsroom/2026/10/apple-creator-studio-adds-powerful-new-ai-features-and-even-more-creative-tools/
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.