The New Shelton Wet/Dry

Introducing Omarchy M

A 15-person team is building Omarchy M to put Linux on Apple Silicon Macs, with a first release targeting full M1 and M2 compatibility, including Pro and Max variants. Marcelo Alcantara’s native macOS installer sets up dual boot without a USB stick or terminal ritual, while Eduardo Martinez’s Try Omarchy runs the desktop as a hardware-accelerated Mac app without partitioning anything. The release targets GPU support, USB-C monitors, Touch ID, disk encryption, and Apple’s MLX machine learning framework running through Vulkan; parallel efforts tackle M3–M5 hardware and older Intel Macs. The pitch is straightforward: keep Apple’s hardware, lose its operating system, and give machines beyond Cupertino’s software ambitions something useful to do.

https://omarchy.org/news/2026/09/introducing-omarchy-m/

A Topological Picture Book, Rendered

An interactive renderer turns topological surfaces into hand-hatched plates inspired by Francis, Apéry, and Hilbert–Cohn-Vossen, with draggable parametric and implicit models including the torus, Boy’s surface, and Morin surface. Its GPU pipeline chains silhouettes from n·v = 0, triangulates zero sets with marching tetrahedra, traces principal-curvature streamlines, and handles occlusion with dashed hidden lines and paper-colored haloes. Broad-nib strokes add taper, wobble, overshoot, ink pooling, and fibrous bleed, while labels remain pinned to the rotating surface and dim when obscured. It matters because the page makes decades of non-photorealistic rendering research feel like a particularly eccentric, unusually legible mathematical sketchbook.

https://e-infinity.space/picture-book/

On-device intelligence for every product

Desert Ant Labs launches 18 specialized models for audio, vision, and text through one Swift, Kotlin, and JavaScript SDK, with free use up to 100,000 monthly active devices. Its 9MB Clear model enhances five minutes of audio in one second, Voz transcribes 10 minutes in two seconds, and 2MB Tongue identifies 84 languages from three words with better accuracy than a 293MB detector. The European lab grew from Detail’s effort to replace costly cloud APIs, including Claude Sonnet, with fast local inference that keeps personal data on the device. It makes a pointed case that much of AI’s repetitive work needs cheap little brains, not another frontier model billing by the token.

https://desertant.com/blog/introducing-desert-ant-labs/

Watched