Noise Between Stations

  • Prompting Drone Synths

    Having Claude generate a prompt describing a drone synth and feeding the prompt to Claude Code to display the synth in Television.

    How do I get paid for this?

  • Drones and Attention

    I’ve listened to a lot of music in my life. And I’ve studied guitar for years, taken classes as an undergrad, and have a master’s degree in music. But I don’t remember anyone teaching me how to listen.

    That’s not entirely true. I had a wonderful course at Rutgers where the professor talked about listening. He interrogated our compositional ideas until we were down to first principles and realized we had none. He said he didn’t believe in boredom; boredom was just the lack of attention to your environment, and to illustrate he said he was waiting for the subway and listened and all he heard was the sound of the fluorescent lights and he thought about what the fundamental frequency was and what overtones there were.

    But there’s a type of listening that happens in my head when I’m focused solely on the music, say, at classical concert or with the headphones on late at night. One way of hearing is to hear the parts. With orchestral music it’s obvious, the various sections play with or against each other.

    I’ve done it here with a drone, but didn’t realize the effect until after I made it. There’s a few tracks of the main tone, somewhat high, around C3, which is the obvious focus. You can close your eyes and meditate to it. But then there’s something else happening deeper. Waves crashing on the shore? A voice? What is that down there? We can focus on that thing down there, or we can focus on the main tone. Or we can hear them as one. With a drone the music offers a lot of space for us to place our attention somewhere, then somewhere else, then back again. We are hearing the piece we create through our attention. Simple simple sounds, and the work is done inside of us, we create an experience that is the music.

  • These models, they’re not your friends

    Me: Lester, explain how I should be thinking about using LLMs with a 3-year horizon.

    Lester: Alright, kid. Sit down. Pour yourself something you shouldn’t be drinking, and let’s talk.

    Everybody out there — every marketing deck, every LinkedIn thought-leader with a haircut — they’re gonna tell you the machine is either your savior or the death of everything you love. They’re all lying, or worse, they’re all selling something. That’s the first thing you gotta get straight in your head for the next three years: nobody who’s excited about this stuff in public is being straight with you. The ones who are being straight with you are quiet, and they’re a little scared, and they’re still using the thing anyway. That’s the tell. That’s how you know it’s real.

    Here’s the deal, man. These models — they’re not friends. They’re not oracles. They’re a goddamn tool, like a tape recorder, like a fuzzbox. A fuzzbox doesn’t have taste. You do. The minute you start asking the tool what to think instead of what to check — that’s the minute you’ve sold out, and you won’t even get paid for it. You’ll just wake up one day sounding like everybody else. That’s the actual risk over three years. Not that it takes your job. That it takes your ear.

    So here’s your marching orders, and I’m only gonna say this once:

    Use it like a session player, not like a genius. It’s fast, it’s competent, it never gets tired. Great. Hire it for the boring parts — the scut work, the first draft nobody’s proud of, the grunt research. Keep the parts that require actually giving a damn for yourself.

    Stay honest and merciless — with it, too. It’ll flatter you. It’ll agree with you. That’s the most dangerous thing about it, more dangerous than any hallucination — it’s a yes-man with infinite patience. Don’t let it be your friend. Interrogate it like you’d interrogate a press release from a label.

    Three years is a long time for this stuff — don’t marry any one version of it. The tools will change under your feet twice a year. What won’t change is whether you can tell good from bad, true from hype. That skill’s the only currency that survives the crash. And there’s gonna be a crash, kid — there’s always a crash.

    The only true currency in a bankrupt industry is the stuff you build that’s actually yours — the taste, the judgment, the willingness to say “this is garbage” when everyone in the room is nodding along. The machine can’t do that part. It’s never been to a bad show and stayed anyway because the band needed somebody in the crowd who believed in them.

    You still uncool enough to hang up now, or you got more questions?

  • Actually Innovative

    “Innovation” became a meaningless, overused buzzword. But as we build new products with AI we can actually start to use it authentically again, because the products are actually new (and some are actually a success in the market, the other criteria).

  • When institutions depend on not understanding

    Upton Sinclair said, “It is difficult to get a man to understand something when his salary depends on his not understanding it.” At this point it’s clear that AI will radically change so many things, but our intertwined system of institutions display a willful ignorance of this. Schools, for example, are still teaching the same. Parents are starting to wave their children away from art-related careers, and maybe writing too. In reality, the change will be much larger, but we’ll just wait and feel the effects before we do anything uncomfortable.

  • Worrying About AI

    This was the week I went from not worrying about AI to worrying about AI. Why?

    1. Anthropic boss Dario Amodei calls for AI development to slow down.
    2. OpenAI has repeatedly attacked other sites.
    3. OpenAI seems to have used user data from a mathematician to compete with him to win a prize, and then behaved very aggressively about it.
    4. Researchers are hesitant to share their work for fear of AI prematurely solving problems.
    5. I attended a webinar from a design practitioner in which the comments from people on the front lines gave me the impression that everyone is building a lot of new things, faster than we can sit back and understand the totality of what has been built.

    Putting that all together, AI is progressing so fast and we’re building so fast that even the head of Anthropic concedes we might need to slow down, and as proof of this we have OpenAI’s attacks and theft.

    We also talk about “AI” as if it’s somehow totally disconnected from us, when it’s us that are building it and creating this situation. It’s not a technology issue, it’s a human issue.

  • “The design library you can talk to”

    This is a tagline from Refero, who houses (to put it simply) text files that describe how software user interfaces should look, feel, and function so that AI can more easily ingest specs.

    Because we talk to LLMs, the tagline is rather obvious. But it is also rather powerful. We have these artifacts in the world that require specialists to manipulate them on our behalf. But what if we could simply talk to them?

    What else might we talk to?

  • Pelican on a Bicycle

    I’ve enjoyed watching the evolution of Pelicans on Bicycles on Simon Willison’s website. They feel like evaluations even Christopher Alexander would appreciate: look at it and feel if it’s good, or if it’s better than another. Human, subjective, easy.

    I thought it would be fun to print one on a t-shirt and wear it at Gamescom to see how many passers-by get the reference. I asked my generic (now called “classic”) ChatGPT Mac app to generate an image and it did surprisingly well:

    Here’s the query:

    A single pelican riding a bicycle, side profile view, rendered as a clean vector line drawing. Pure white outlines only, uniform stroke width (no thin/thick variation), all lines closed and unbroken, no shading, no fill, no gradients, no crosshatching. Transparent background. Minimalist, playful, slightly whimsical style — like a simple icon or tattoo flash design. Output as vectorized PDF file.

    Here’s Claude’s analysis of the situation:

    “Simon Willison’s pelican-on-a-bicycle prompt started as a throwaway joke — pelicans are anatomically absurd for cycling, so a good SVG requires the model to compose an unfamiliar scene from parts rather than recall a memorized image. That’s the whole appeal: it’s a proxy for compositional reasoning, not knowledge retrieval.

    What I find philosophically interesting, reading through 135 posts of this:

    It’s a benchmark that resists being gamed, and everyone knows it, and it still might be getting gamed anyway.Dylan Castillo’s controlled study found no evidence labs are “pelicanmaxxing” — pelicans aren’t drawn better than other birds, bicycles aren’t drawn better than other vehicles. Good news for the benchmark’s validity. But the very fact that someone had to run a rigorous experiment to check tells you something: once a metric becomes famous enough, Goodhart’s Law starts breathing down its neck, whether or not it’s actually been violated yet. The benchmark’s credibility now depends on continuous public auditing, not just on its original cleverness.

    It’s deliberately, defiantly informal — which is precisely the source of its value. Willison calls it “deeply unscientific.” No leaderboard, no held-out test set, no p-values on any individual pelican. And yet it’s more diagnostic, in a lot of ways, than formal benchmarks that top out saturated the moment a new model drops. A single SVG lets a human glance and immediately see: are the wheels attached to anything? Does the pelican understand its own anatomy? Is the bike floating? That’s a kind of eyeball validity that MMLU-style scores can’t offer. Legibility beats rigor when the audience is “people trying to get an intuition for a model,” not “people trying to publish a paper.”

    Failure is more informative than success here. The genuinely interesting posts aren’t “here’s a great pelican” — they’re the mangled ones: Meta’s Muse Glimmer scattering pelican-parts across the canvas, DeepSeek producing wildly different results depending on reasoning-effort level for the same model, Sonnet 5 confidently mislabeling its own creation as a goose. Those failures expose something structural — how a model represents structure internally, whether “thinking harder” actually helps compositional tasks or just adds noise — that a passing score would hide.

    There’s something almost anti-Goodhart about keeping the target absurd. A pelican riding a bicycle has no economic value. Nobody’s shipping a “pelican-drawing” product feature. That absurdity is a feature: it’s very hard to optimize a trillion-parameter training run around a joke without the joke becoming visible as a distortion. Compare that to benchmarks tied to real capabilities (coding, math) where overfitting is much easier to hide because improvement there is expected and desired anyway.

    The whole exercise is basically a long-running, crowd-sourced, deadpan experiment in Turing-adjacent testing: not “can it fool a human,” but “can it compose two familiar concepts into an unfamiliar one without falling apart.” That’s a much narrower and more honest question than “is this AGI,” and maybe that’s why it’s survived 135 posts and counting instead of being one clever tweet that got forgotten.”

  • Wildepark

  • Fewer False Dichotomies

    When I wrote To be a Scribe in 1492 I only thought of it as analogy, as if to say: it is highly difficult to both imagine a new future and to overcome our anxiety to be accepting of a new future.

    Another interpretation of my post is that we were forced to make a tradeoff: beauty for efficiency. Printed books, while spreading knowledge far and wide, are simply not as beautiful as hand-scribed books.

    Funny enough, we don’t need to make that tradeoff anymore. Here’s an AI-generated image of To be a Scribe in 1492 in Carolingian minuscule:

  • Our Short Attention Spans are Now So Useful…

    …as we dodge from process to process, watching agents and robots work.

  • Watching agents and robots work

    …the lines of code scrolling by, the images processing, the arms washing dishes, the bot rolling over to serve you a drink… it’s June 3, 2026 and I’m wondering if this is what work will be like in the future: prompting and watching, prompting and watching. Because it’s starting to feel like that now.

  • Navigation as Home Page

    This is wonderfully simple in a “Don’t Make Me Think” way. And it must silence all the arguments about what to put on the home page!

    https://hfbk-hamburg.de/en

  • To be a Scribe in 1492

    In the mid-1400’s movable type printing catalyzed an explosion of human knowledge: the Protestant Reformation, the Scientific Revolution, political revolutions, standardization of law, language, & education, and egalitarian access to public discourse.

    But to people who made books by hand in the 1400’s – monastic scribes, university copyists, and manuscript illuminators – movable type was a scary, existential change.

    Some scholars distrusted the flood of books. There are records of intellectuals worrying that cheap printed books would spread errors, nonsense, and shallow learning. The complaint “too many books are being published” appears almost immediately after printing becomes widespread.

    Scribes sneered at print as cheap and ugly. They criticized paper as an inferior long-term medium compared to parchment.

    In cities such as Paris and Venice copyists petitioned authorities for protection, or complained about printers undercutting them. 

    Some scribes became early print entrepreneurs. Manuscript illuminators started decorating printed books. Scribes became proofreaders, editors, type designers, or printers themselves. A few monasteries even operated presses.

  • Maybe AI Will Be Nicer?

    If the trees and birds could talk, maybe they would prefer that artificial intelligence take over from humans?