Datasets, Borges, and one brave t-shirt
Hey 👋 — today's mix leans thoughtful: a decade-long art practice built on hand-curated datasets, a fresh skirmish over who owns the internet's training data, and one absolute unit of a language model squeezed onto a laptop that has no business running it. Plus a t-shirt that turned out to be a shell script. Let's get into it.
An Interview with Anna Ridler
Right Click Save sits down with generative artist Anna Ridler, who's spent ten years building her own datasets by hand rather than scraping the web for them — treating the labor of classification itself as the artwork, not the plumbing behind it. The conversation touches her recent Oxford commission, which pairs 17th-century taxonomies with Borges's famous skewering of any system that claims to organize all knowledge. It's a sharp reminder that "the dataset" isn't a neutral ingredient; someone chose it, and that choice carries meaning.
"The process, the labor, the choice of what to include and what to exclude has always been the subject, not the method." — Anna Ridler
Building an interactive wave propagation cube grid with Three.js
Franky Hung's Codrops tutorial builds a grid of cubes that ripples outward from your cursor like a pond that's been to physics class — mouse-trail interaction, wave math, post-processing, and performance tuning, all with source included. Good one to bookmark if you've been meaning to graduate from "spinning cube demo" to something with actual choreography.
Show HN: Getting GLM 5.2 running on my slow computer
Colibri is a pure-C, zero-dependency engine that streams experts off disk to get a 744-billion-parameter mixture-of-experts model, GLM-5.2, limping along on a machine with just 25GB of RAM. It's the software equivalent of towing a cruise ship with a bicycle, and it works, which is the fun part (via HN discussion).
Publishers prepare to opt-out of Google search over AI training
Staying on the theme of who gets to use what: Adweek reports that major publishers, emboldened by Cloudflare's new default bot-blocking, are weighing whether to cut Google's crawler off entirely rather than let it index their pages and feed AI training on the same pass. It's a notable shift from the old bargain — searchability in exchange for scraping — and suggests the next round of these fights won't be settled quietly.
The dataset question from Ridler's interview, playing out at web scale.
Obfuscated bash script by Akamai found on a Uniqlo t-shirt
Someone spotted a base64-encoded, self-evaluating bash script printed on an Akamai-designed Uniqlo shirt and, naturally, decoded it — turning up a harmless Easter-egg message rather than anything sinister. Proof that "curiosity killed the cat" was never really about cats (via Lobsters discussion).
Quick hits
- AI content is everywhere on social media, especially LinkedIn — a look at just how much of your feed isn't human, starting with the platform most primed for it.
- A bug which only affected left-handed users — a debugging tale that'll make you double-check your assumptions about "everyone."
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.