Making a better Social Media
Introduction What is a ‘better’ social media? My last post on recreating old Facebook surprisingly made it to the front page of Hacker News, so it’s safe to assume there’s some interest, or nostalgia, for something in the social space that perhaps existed or c…
A Greyscale iPhone Setup that Works in Everyday Life ⋅ Prof. Dr.-Ing. Fabian Hemmert
I play on a private Aion server. For those unfamiliar, Aion is an MMO from 2009 that NCSoft has largely abandoned in the West, but a community of private servers keeps it alive. The server I play on uses a third-party anticheat called Active Anticheat, a comme…
One day I watched a video where David Pruess goes through an analysis of a chess position (David is an amazing chess teacher and commentator), and he goes through like a dozen moves before pointing out one random little detail that reverses the conclusion. Thi…
MTG Bench tests how well LLMs can play Magic.
How a New DSL Survives in the Era of LLMs
A few months ago I wrote an article in which I considered the idea of LLMs being the “new high level language”: something on a higher abstraction layer that allows to build software.
Cultural idiosyncrasies reveal what a company actually does, not just what it says. They can have an outsized impact on both public company image and internal perception and job satisfaction.
Have you bought a car lately? • Ninjas and Robots
Single people can be as productive as whole teams. People are freaking out. There’s so much angst about how their jobs are changing with AI. I get it. It does feel like the things you used to do aren’t as valuable anymore. There’s new things to... | Ninjas and…
Malware Analysis Report: deps Report date: 2026-06-11 VT Link Triage Link Note: The following report was very hastily written by Codex. (I have fact-checked it against the IDA decompilation though 🐉) Scope and Handling This report summarizes static rever…
Once I wrote users.select { |u| u.admin? }.map(&:email) and realized I’d written Lisp.
In this blog post, I will share the adventures I had creating my own LLM, from (almost) scratch, trained only on old texts.I made my own base-training and fine-tuning scripts, data processing pipelines and custom datasets.("almost from scratch" means I did use…
In this post, I will present an algorithm that was able to compute an optimal tokenizer in some settings. This result is cool because optimal tokenization is theoretically intractable, but seems to be solvable in practice. My fi…
Reading the tea leaves
There is a growing call for sustainable open source, non-profits, and adjacent organisations that appears to align with the very strong culture of radical anti-capitalist organising and contribution in either free or open source software projects1. There is al…
Code got cheap. Review didn’t. What queueing theory says about the AI-era backlog.
Neural Cellular Automata with WebGPU
A WebGPU demo of Growing Neural Cellular Automata. Each cell runs inference through a shared pretrained neural network entirely in compute shaders
6.0.0
Today, I’m proud to announce Homebrew 6.0.0. The most significant changes since 5.1.0 are a new tap trust security mechanism, the new faster, smaller, default internal Homebrew JSON API, sandboxing on Linux, better defaults informed by our user survey, many br…
A case against measuring AI-assisted engineering by code volume and other vanity metrics instead of outcomes.
AI agent runs amok in Fedora and elsewhere
Agentic AI systems can be used to do a variety of things autonomously on behalf of a human user [...]
With the release of Claude Fable, it is clear that Anthropic is progressing from poems to enterprise-scale narrative objects.