On-Device AI Across iOS 27: Spotlight and Media
iOS 27 threads the on-device model through the system: SpotlightSearchTool makes Core Spotlight LLM-grounded, and AVFoundation generates subtitles on device.
AI & TechnologyThoughts on design, development, AI infrastructure, and building products.
iOS 27 threads the on-device model through the system: SpotlightSearchTool makes Core Spotlight LLM-grounded, and AVFoundation generates subtitles on device.
AI & TechnologyiOS 27 reworks SwiftUI lists, documents, toolbars, and errors: drag-to-reorder, a readable/writable document model, toolbar overflow, item-based alerts.
AI & TechnologyiOS 27 adds GenerationOptions.ToolCallingMode to steer how the on-device model uses tools, plus built-in Vision tools: OCRTool and BarcodeReaderTool.
AI & TechnologyWWDC26 made Metal a first-class ML compute surface: Metal tensors for custom kernels, the M5 neural accelerator, real-time neural rendering, and MLX Swift.
AI & TechnologyvisionOS 27 adds Spatial Preview from the Mac, Foveated Streaming from a PC, immersive Safari environments, metric-space object tracking, and RealityKit.
AI & TechnologyiOS 27 cuts camera launch in half with Deferred Start, adds Pro Video Storage for ProRes writes, and brings the Center Stage square front camera to your app.
AI & TechnologyiOS 27 adds a server-scale Foundation Model on Private Cloud Compute with on-device privacy, plus a protocol to plug in your own LLM provider.
AI & TechnologySwift 6.3 and 6.4 from WWDC26: anyAppleOS availability, module selectors, borrow/mutate accessors, the Iterable protocol, Swift Testing interop, and MLX.
AI & TechnologyiOS 27 accessibility for reading apps and custom controls: text navigation linking, causesPageTurn, UITextInput, the adjustable trait, and direct touch.
AI & TechnologyCore AI is iOS 27's low-level model-execution framework: asset vs model, NDArray tensors, compute-unit targeting, and inference functions on Apple silicon.
AI & TechnologyiOS 27 gives SwiftData first-class change observation with ResultsObserver, persistent-history observation with HistoryObserver, and codable attribute storage.
AI & TechnologyAI agents have two untrusted inputs: code the model writes and tool output it reads. One now has a real WASM sandbox; the other, MCP tool poisoning, does not.
AI & TechnologyTechnical writing published at Introl
Comprehensive hardware recommendations and cost analysis for running large language models locally.
GPU selection guide comparing NVIDIA's latest datacenter accelerators for different AI workloads.
Deep technical dive into Google's Tensor Processing Unit evolution from TPUv1 to TPUv5.
Resource sharing strategies for GPU clusters in containerized environments.
Guide to building and managing distributed AI computing with Ray framework.
Analysis of open source LLM economics and DeepSeek's competitive positioning.
Future datacenter power requirements and NVIDIA's next-generation GPU roadmap.
Small modular reactor solutions for powering next-generation AI infrastructure.
Technical analysis of DeepSeek's Multi-Head Compression architecture innovations.