LLM Maths
Comics
Long-form deep-learning mathematics articles — comic-style.
AI slop. The dollar-bin stuff. Cheap thrills. Long-form articles to build mathematical intuition behind the WHYs of how the mathy ideas behind LLMs work.
Here’s where this is coming from. A while ago, I started working on a project for making LLM inference faster, so I realized I had to learn this stuff.
Learning HOW these LLMs things work was straightforward enough (thanks to Andrej Karpathy’s microGPT, for example). You take this vector and shove it there. One step after another, sure.
But building the intuition for WHY LLMs use these things is a totally different thing. It’s a bit less obvious that you can do that. It also takes more time. But ultimately it’s more rewarding and interesting.
So I started asking LLMs to explain me things and to teach me. We would do the bak and forths. Then I started saving the LLM responses. And now the best ones are up here on the internets.
Latest Dispatches
All issues →From Dead Insects to Deep Learning: How Logits Conquered Statistics and AI
A historical walk through the ideas connecting 1930s entomology to the output layer of your neural network — probits, logits, and the sigmoid function.
Word2Vec and The Geometry of Meaning
How Mikolov's 2013 breakthrough replaced brittle one-hot vectors with dense geometric embeddings — and why cosine similarity suddenly meant something.
Tech Stack & Build System
A detailed breakdown of how the almanac is built — Hugo, the custom theme, the two visual styles, pyplot execution, and GitHub Pages deployment.
Prompt: Word2Vec and the Geometry of Meaning
The prompt that generated Chapter 1 — embeddings, cosine similarity, and why dense vectors beat one-hot encoding.
Prompt: Logits and Probits — From Entomology to Deep Learning
The prompt that generated Chapter 2 — tracing logits from 1930s insecticide research to the neural network output layer.
Contributing Articles
How to write, format, and submit articles to the almanac — including the art of prompting an LLM to do the heavy lifting.