Ronald Mannak

I asked Fable 5.2 to "go wild" making a game, and all I got was this Pong clone. And then I played it ... 👇

Over the last 30 days, Sosumi.ai has served tens of millions of requests to coding agents looking up Apple docs, from about 6,900 (nice) unique IPs/day. But does it actually help? I built an eval to find out: 41 Swift tasks, mostly new APIs in the OS 27 SDKs, checked by the compiler and per-task rules. Without docs, Opus, GPT-6.1 Sol, and Sonnet pass about a third of them. With Sosumi, they pass 98%, 97%, and 94%. Compared to built-in web search, Claude models score 12 to 30 points higher with Sosumi, at about half the cost or less. GPT does slightly better with web search (99%), but at 3× the cost. Running the eval also turned up a handful of fixes, now live (one framework page was 3 MB 😬). The eval is open source: github.com/NSHipster/sosumi.…

Qwen4 is coming. Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities. Built for real-world tasks — from complex coding and knowledge work to front-end builds and professional software workflows. Qwen4-Max is now in preview on Alibaba’s Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio. Put it to the test on a real problem. Show us what you build. 👀 Token Plan: qwencloud.com/pricing/token-…

Apple is working on a new app called Harbor Harbor turns your Mac into a private AI inference server. Run open-weight models locally on Apple silicon—your data never leaves your hardware, and nothing reaches the cloud. Connect multiple Mac computers over Thunderbolt 5 to pool unified memory, enabling large models and higher concurrency that a single computer can handle alone. Set up in minutes, not hours Harbor automatically detects your devices over Thunderbolt 5 to form a cluster. Plug in, start, and you’re running—no manual configuration, no command line, no infrastructure expertise required. Download and run models directly Browse, download, and load open-weight models from within the app. Harbor handles model management so you can focus on the work, not the setup. Cluster for scale Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource. Run models that exceed any single computer’s capacity, or serve more concurrent requests across your cluster. Harbor is currently in a private preview under the Claris brand.

Even the simplest weight edit, a deletion, can teach an LLM new skills! Meet Deletion Tuning (#NeurIPS2026): just zeroing <0.1% of weights matches full fine-tuning's accuracy at scale. It's thousands of times smaller and avoids forgetting, a step toward continual learning. 🧵👇

Skimmer is the first step in Siri AI’s on-device processing. It uses regular expressions to match requests to tool flows. On macOS, you can browse the rules by locale: /System/Library/PrivateFrameworks/IntelligenceFlowPlannerSupport.framework/Versions/A/Resources/enigma_patterns/<locale>/Enigma.json

This is why Elon Musk started making chips. 🤣

JUST IN: Nvidia $NVDA crashes 6.5% at open after CEO Jensen Huang admits to having an affair with an alleged Chinese spy.

Hi friends! I’ll be giving a talk at tomorrow’s AiOS meetup about @bitrig‘s upcoming build and simulator system. There are some cool details in here about how to design AI-first tooling, so if you’re interested at all in that space, or Apple platforms development, come through! luma.com/wy0z8tek

Big news for virtualization: macOS 27.3 lets Core AI in Mac VMs use the host’s GPU and Neural Engine. In some tests, the new paravirtualized paths can run Core AI tasks up to 7× faster¹ than the guest CPU, achieving 50 to nearly 100% of the host’s native speed. The new Core AI accelerator currently requires private APIs and a restricted entitlement. Here's how to set it up: • Install macOS 27.3 on the host and guest. • Build a VM host app with com\.apple.security.virtualization and com\.apple.private.virtualization entitlements. You’ll need to relax AMFI for the second entitlement. • On VZVirtualMachineConfiguration, add the accelerators using _setAcceleratorDevices with: • _VZMacCoreAIDeviceConfiguration (GPU, adds PVAIDevice) • _VZMacNeuralEngineDeviceConfiguration (existing ANE accelerator, adds Neural Engine for Core AI) On the guest, enable the feature flag: sudo mkdir -p /Library/Preferences/FeatureFlags/Domain && \ sudo defaults write /Library/Preferences/FeatureFlags/Domain/ParavirtualizedCoreAI enable_pv_delegate -dict Enabled -bool true and restart the guest. You can now call Core AI on the guest. If set up properly, ComputeUnitKind.availableKinds should list gpu and neuralEngine. —— There's a second, currently disabled host-side feature flag (ParavirtualizedCoreAIHost/enable_save_restore). When enabled, the Core AI accelerator advertises support for saving and restoring VM state. —— 1. Tested on a Mac with an M4 Max chip using public models (YOLO, CLIP, Depth Anything) and matmul/convolution benchmarks. Crossing the VM boundary adds ~1 ms per inference call. Qwen3-0.6B in Apple’s llm-benchmark ran at about 10% of host speed. Core AI’s paravirtualized path repeatedly sets up host access to large shared KV-cache buffers for each generated token.

Longtime readers know I'm a fan of Space Disco. Perhaps the only fan. So I make no apologies for introducing you to Spanish cosmic disco-funk troubadours Zoom, and their minor 1978 hit Space Rescue! This track is a full cardiovascular workout, accompanied by some excellent spaceships that wouldn't be out of place on Panther editions of Isaac Asimov novels. It barely troubled the charts, but the full version ("Rescate Espacial") is available on their debut album Zoom from Polydor Records. Don't thank me, it's what I do. Please don't try this dance on a full stomach...