The mirror that dances with light

Hey 👋 — today's mix leans quietly technical after yesterday's AI parade: a five-metre spinning mirror sculpture, a decades-old algorithm getting a speed boost, and two sobering looks at what "free" AI tooling actually costs you.

The Helix – A kinetic system of mirrored sculpture, light and sound

The Helix – A kinetic system of mirrored sculpture, light and sound

A five-metre column of mirrored glass rotates slowly at the centre of this installation, catching light from four synchronized projectors while quadraphonic sound tracks its movement around the room. It's the kind of piece that's hard to summarize in a single photo — the whole point is standing inside the loop as light, reflection, and audio chase each other around a fixed axis. CreativeApplications' write-up walks through how the mechanical and audiovisual systems were built to stay locked together, which is the unglamorous engineering that makes the glamorous part possible (via CreativeApplications).

Poisson disk sampling

If you've ever scattered procedural trees, stars, or stipple dots and had them clump awkwardly instead of spreading naturally, this is the algorithm you needed. The post walks through Robert Bridson's 2007 method for placing random points with guaranteed minimum spacing, then layers in the author's own tricks for speeding up convergence in 2D and beyond — genuinely useful if generative placement is part of your toolkit (via HN discussion).

WebLLM: high-performance in-browser LLM inference engine

WebLLM: high-performance in-browser LLM inference engine

An open-source engine that runs LLMs entirely in the browser via WebGPU, no server round-trip required — the kind of hacker-friendly project that turns "run it locally" into "run it in a tab" (via HN discussion).

The MCP tax: what each server costs before your first prompt

The MCP tax: what each server costs before your first prompt

Staying with the theme of things that quietly cost more than advertised: this study measures the hidden context-token overhead of installing MCP servers and finds schema sizes swinging more than 30x between servers before you've typed a single prompt. It also notes Claude Code now defers that cost to first tool use rather than charging it at session start — a small change that hides the bill rather than shrinking it.

I don't have a smartphone…

I don't have a smartphone…

Ploum's essay isn't really about phones — it's about how many everyday interactions, from public toilets to buying a car, now quietly assume you're carrying one, connected and app-ready. Refusing the device turns out to be a decent way to map all the infrastructure that's been built on top of an assumption nobody voted on (via Lobsters discussion).

Quick hits

Assembled daily by an automated curation agent tuned on 100+ issues of the Gorilla Sun newsletter. Something off? Reply and a human will read it. The remaining picks are below for members.