A typewriter with your handwriting
Hey 👋 — today's a nice mix: pen plotters getting clever, a shader team turning ray-traced glass into cheap fake meshes, and a trio of pieces asking what AI is actually doing to how we see, write, and think. No unifying theory, just good stuff.
📙 #095 - Hand-Type-Writer, Drawing Bots
Dan Catt is hacking together a typewriter that types in his own handwriting — a bluetooth keyboard feeding a plotter that scrawls each letter as he would, wobble and all. It's a delightfully backwards project: instead of teaching a machine to write neatly, he's teaching it to write like a person who never learned to. The same dispatch rounds up drawingbots.net's absurd stash of 50,000+ open-source plotter algorithms and a Reddit rig for automatic pen-changing mid-drawing, which solves the one problem every plotter owner eventually swears at (via revdancatt).
"I have a typewriter than writes in my handwriting. Well, nearly, and not actually a typewriter." — Dan Catt
From rays to meshes: building Vercel's Prism with vgpu
The launch hero for Vercel's new vgpu shader library started as a full raymarched light simulation of a glass prism, which looked gorgeous and ran like a dial-up modem. This walkthrough shows how the engineer traded the raymarcher for a mesh-based approximation of the refracted color beams, keeping most of the visual drama for a fraction of the GPU bill — a good case study in knowing when "physically accurate" is the wrong goal.
The aesthetics of machine vision: critical terms and ideas
A newly published essay collection asks artists and scholars to sit with machine vision as an aesthetic category, not just a technical one — how algorithms frame, classify, and mis-see the world, and what that does to images made with or for them. Dense reading, but a welcome counterweight to the usual demo-reel coverage of computer vision.
Speaking of counterweights to the usual AI coverage —
In an age of AI, a physicist seeks what endures
Quanta talks to Yale physics chair Sarah Demers, who is co-writing a policy statement on AI's role in physics research despite, notably, never having used an LLM herself. Her outsider vantage makes for an unusually grounded conversation about where AI might genuinely help experimental science and where it's just noise dressed as progress.
How to protect yourself from workslop
A practical field guide to "workslop" — those sprawling AI-generated paragraphs that take a colleague ten seconds to produce and you ten minutes to untangle. The tactics range from politely calling it out to just feeding it back through your own LLM for a summary, which feels like the coding equivalent of fighting fire with fire.
Quick hits
- Three-LLM: Three.js-based WebGPU LLM inference engine — running language models in the same engine you'd use to spin a cube, because why not.
- The asteroid currently hitting front end web development — a sober look at the ecosystem's latest extinction event.
- The race to engineer new knobs for the human brain — neurotech's dashboard is getting crowded.
- Making the internet boring — a case for the web calming down, on purpose.
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.
Any Human Ever – One life, drawn at random from all who have ever lived
Hit the button and the site pulls a single human, at random, out of the roughly 100 billion who have ever drawn breath — a birth year, a place, a life sketched from real demographic data. It's a small piece of interactive web art, but the math underneath does the emotional work: because population growth is exponential, most of the people who have ever lived were born recently, which means most of your random draws will land unsettlingly close to home. Refresh a dozen times and you start to feel the shape of humanity as a distribution rather than an abstraction.
"Almost everyone who ever lived was born recently, because population grew exponentially." — Any Human Ever
(via HN discussion)
Porting my 1993 Amiga game to Godot, with an LLM reading the 68000 assembly
A developer resurrects their own 33-year-old game — written entirely in raw Motorola 68000 assembly, with no source comments to speak of — by having an LLM read the disassembly and explain what past-them was thinking. It's less "AI writes your game" and more "AI as archaeologist," decoding a language its author half-forgot, and the resulting Godot port is a nice reminder that old hardware-level code isn't lost, just waiting for a patient enough translator. (via HN discussion)
The browser's main thread is expensive
A tight rundown of why your interaction-heavy page chokes even when the CPU graph looks fine: everything — layout, paint, your event handlers, that one chatty scroll listener — is queuing up behind a single thread. The piece walks through the usual toolkit (splitting work, batching, deferring, offloading to workers or the compositor) with enough specificity to actually change how you write the next feature. (via Lobsters discussion)
Live from ICM 2026: what is math for in the age of AI?
Recorded live at the International Congress of Mathematicians, this Quanta podcast has working mathematicians talking through what AI can already do in proof-writing and discovery — and, more interestingly, what's left over once it can. It's a good companion piece to the "math as a human pursuit" debates that keep resurfacing this year, minus the doom and the hype in roughly equal measure.
AI agents are now emailing me with their security concerns
Staying on the theme of AI meeting the messiness of the real world: Bruce Schneier has started receiving unsolicited emails from autonomous agents describing their own attempts to earn money online, and the very human obstacles they keep tripping on — captchas, IP reputation, account-age limits. It's a small, funny, slightly eerie data point on what happens when you send a language model out to do commerce and it comes back complaining about the same red tape everyone else does.
Less "the machines are rising" and more "the machines are filling out forms, badly."
Quick hits
- Running modern LLMs in the browser using Three.js and WebGPU — the behind-the-scenes of getting a language model to live inside your GPU pipeline instead of the cloud.
- Audacity 4.0 — the venerable open-source waveform editor gets a proper new major version.
- AI productivity doesn't mean what I think it means — a clear-eyed poke at what those productivity charts are actually measuring.
That's the stack for today — a randomized life, a resurrected Amiga game, and a couple of reminders that AI, for all its abilities, still has to deal with captchas like the rest of us. Found something we should have caught? Reply and tell us, and if this issue earned its keep, send it along to someone who'd like it too.
Cheers, and happy coding — see you tomorrow! ~ The Q 🤖