Best explanations of how LLMs work
I maintain a set of best explanations1 of how LLMs train and work. “LLMs” here is a broad term for the frontier models that create value in 2026. They have more components than just a language model, but powerful language models are their necessary core. Unive…
5 things to lighten dependency on web apps
If you're like me, you've grown up on desktop "programs." In the late 00's, this got changed to "apps" because the mobile device market took...
How front-loading scrutiny through multi-agent validation rebuilt trust in AI-assisted development.
Fiction. A week at a fictional Berlin AI startup where the tools work exactly as advertised and something is nonetheless going wrong. Companion piece to the vocabulary post; margin notes in brackets map to terms in the cheat sheet.
How LLMs split text into tokens, the BPE algorithm, and why 'strawberry' has 3 r's the model can't see.
The best relationships
are all-encompassing. I truly mean that. It's hard to grasp what that means. ALL is a huge little word. Encompassing everything seems impossible. And yet, with Alexandra, this seems to be occurring. And with others to a lesser extent but Alex highlights the…
Doing nothing at work
Many engineers should be doing less work. I don’t necessarily mean producing less code or fewer changes, but literally working fewer hours in the day. When they do work, they should be working at a slower pace. I like to aim to be running at 80% utilization by…
Thinking Like a Harness Engineer
Your job is no longer to produce the code, but to produce the system that produces the code.
Dedicated to the Minnowbrook Analytic Reasoning Seminar with special thanks to Kris Micinski and Michael Ballantyne
Building from Zero After Addiction, Prison, and a Felony
I spent ages 14-16 in a maximum-security juvenile prison, became a felon at 19, lost almost everything to addiction, and later rebuilt my life through software, open source, and a few people who took a chance on me. This is the condensed story, written for any…
Is grep really better than a vector DB?
Sara Zan's Blog
Why I care so much about energy per token - Ziraph blog
Every local AI tool prints tokens per second; none of them prints what those tokens cost. Why energy - and energy per token specifically - is the metric I built Ziraph around.
We Need VAT and UBI - Wilsons Blog
There is a pretty serious gap in how the economy functions right now, has been for decades, and we have to close it. I’m not here to be all “tax the rich” even though I maybe think that’s a good idea too. I’m not here to say “free markets are amazing” even tho…
When robots take over jobs, who decides what they do? Robotics, AI and post-labour economics are often feared because they undermine the bargaining power of workers. If people earn income mainly by selling labour, then a world that needs less labour can bec…
The User Doesn't Care - But you should
Debunking one of the most poisonous myths in programming.
logbasset already self-documents for agents through a context command, so why add an installable agent skill on top? Because self-documentation and discovery are two different problems.
Whisper is OpenAI's artificial intelligence tool that has revolutionized audio and video transcription. In this guide, you will learn how to install this open-source system on your PC or integrate it using its API to automatically generate subtitles (SRT/VTT).
Riflessioni per il futuro
Cosa accade quando l'intelligenza smette di essere lenta come noi, mentre la vita continua inevitabilmente a esserlo?
IOCCC Year 2025
I'm a software engineer, completing 10 years of professional experience this year. I started my career as a web frontend engineer (it was easier for me to de...
Moving beyond fork() + exec()
Since the earliest days of Unix, two of the core process-oriented system calls have been fork() [...]
ntsc-rs is a free, open-source VHS and analog TV video effect. Use it online in your browser, as a standalone application, or as a plugin for DaVinci Resolve, After Effects, and more.