Dispatch from the airport. I’m on my way to San Francisco. Arriving today, Saturday, leaving next Sunday. If you’re in town — Japantown, it’s called — and want to grab a coffee this week, let me know, even though I can’t tell you much more about my availability than the fact that there will be some.
It’s also my birthday tomorrow, I’ll be turnin— Hey, ever noticed how the people at the airport who bring a dedicated mouse to use with their laptop, they’re never using a MacBook, are they? Long live the MacBook trackpad, I say. Anyway! I’ll see you in San Francisco or on the Internet and until then: here are some links.
The system prompt for OpenAI’s Operator leaked and, as with all system prompts that have leaked so far, the dichotomy between the name “system prompt” and its contents couldn’t be more marvelous: “If you recognize a person in a photo, you MUST just say that you don't know who they are” is a good line, as is the thought-provoking pleading with the robot to not lie about its robot-ness: “The user must take over to complete CAPTCHAs and ‘I'm not a robot’ checkboxes.”
Jimmy Miller on Discovery Coding: “For me, this is the only way I can do anything. Anytime I try to outline before I write code, my outline is thrown away within hours of writing code. My design docs that are written beforehand are wholly unsatisfying, even if they get approved. It is only as I begin writing code that I begin to understand my problem.” I always thought of this as a Tracer Bullet, because that’s how I remembered it from reading Pragmatic Programmer, but now, rereading the definition, it seems it’s slightly differnet: a tracer bullet “involves short cycles of development, then delivery and asking the customer if it's closer or further from what they think the target is.” Be that as it may: I think both are fantastic methods of approaching difficult projects.
Been meaning to implement ReAG, Reasoning-Augmented Generation, because it’s an interesting idea and I wanted to see what DeepSeek’s R1 can do, but, alas, DeepSeek’s API was down every time I tried. Still: interesting blog post, even though I’m not sure It Will Scale, yknow.
Fantastic and thorough post about NVIDIA, its stock price, and how the future of AI might affect it: The Short Case for Nvidia Stock.
Ben Thompson, writing about DeepSeek, NVIDIA, OpenAI and the rest: "The arrogance in this statement is only surpassed by the futility: here we are six years later, and the entire world has access to the weights of a dramatically superior model. OpenAI’s gambit for control — enforced by the U.S. government — has utterly failed."
This thing, titled “Against the dark forest”, was interesting: “Global mega-platforms under capitalism are structurally incapable of handling the business of civilization: of governance, of providing genuinely public infrastructure, of making knife-edge decisions about the balance of liberties and securities.” But, I fear, if you strip the article of all the fancy words and references and citations not much will be left.
This post, providing “hard numbers in the Wayland vs X11 input latency discussion”, brought back a lot of memories. I spent quite a few weeks this past summer helping with the Linux version of Zed and I didn’t walk away from that experience with love in my heart for X11, Wayland, or, honestly, the whole ecosystem surrounding both. The Lobsters discussion of the post provides a glimpse of what you end up with if you keep discussing things on a technical level when the problems have long transcended technology. Tangentially related: this comment here — “I created a test setup using a Raspberry Pi Pico running CircuitPython and a USB HiD library to emulate mouse clicks. I connected button switch to a GPIO pin and polled the state of the pin to send mouse click events over usb to my laptop.” — made me wonder how much better applications on Linux would be if there was an Instruments.app.
I cannot tell you how I ended up reading this post called “A Simple TCP Server Written in Go and CLIPS” but it was fascinating. Probably because I didn’t even know what CLIPS was and only after reading found out that CLIPS was “developed at NASA’s Johnson Space Center from 1985 to 1996” and “the C Language Integrated Production System (CLIPS) is a rule‑based programming language useful for creating expert systems and other programs”.
Learned about DJR fonts last week (beautiful!) and through that found Fonts In Use which is lovely.
Didn’t expect these 51 seconds of “Intel 486DX2 66mhz Startup Sounds” to bring back that many memories. Man, I do miss the sound of harddrives.
While thinking about AI generated code, I often think of this Chad Fowler talk that I’ve seen, probably, 10 years ago: “Let's just get into a habit of throwing code away and if it's this big it doesn't matter. Just throw this little piece away. So it's really easy to upgrade and it's really easy to think about new things and change whatever else. [...] It starts to already force this idea that the system is a thing and the cells are a thing” This talk was given before we all realised that micro-services don’t solve all problems and yet I think the idea of the system as a regenerating entity with cells that it sheds — I think there’s something to that.
Did you know that, on macOS, you can right-click the Activity Monitor icon in the Dock and tell it display current CPU or memory usage?
I want to book one of those planes that pull a banner behind them and print on the banner a quote from this piece: “If you’re roughly 70% happy with a piece of writing you’ve produced, you should publish it. If you’re 70% satisfied with a product you’ve created, launch it.” It’s a great bit of writing that I’ll share a lot in the future — on a banner in the sky or via URL.
Short post on “Growing Up at the Dawn of Cyberspace” that, again, like many other bits of writing, made me think the obvious thought of how important it is to know the history of the field you’re working in.
“This PR provides a big jump in speed for WASM by leveraging SIMD instructions for
qX_K_q8_KandqX_0_q8_0dot product functions. Surprisingly, 99% of the code in this PR is written by DeekSeek-R1.”Nelson Elhage, who’s worked at Anthropic for three years, finally gave Claude a proper spin and: “I’m fascinated by my own reactions to and thoughts about Claude” As if that weren’t enough: “even as I drafted this post, I noticed a familiar sort of goalpost-shifting in my own mind! I find myself shifting from surprise and amazement into a sort of almost-jaded minimization.” AND: “I feel like this project made me feel a sense of personal excitement and enthusiasm for building software with Claude / other LLM assistance, in a way I hadn’t personally felt before.”
