I’ve created a new interactive resource: A Gentle Introduction to Neural Language Models, a web-based guide that I built vibe-coding with Claude, with the idea of having an interactive companion to use for teaching. The guide walks through the full arc of language modeling — from counting n-grams and the chain rule decomposition, through word embeddings and neural networks, all the way to the Transformer architecture, pre-training and fine-tuning, modern LLMs, and the interpretability of their internal representations. It’s structured as 11 chapters, each covering a key concept in the history and mechanics of how machines learn to process language.
In order to make it different from a standard set of lecture slides, I decided to pair every concept with an interactive widget. You can step through how the chain rule assigns probability to a sentence word by word, manipulate embedding vectors and watch analogy arithmetic unfold, or walk through the five steps of self-attention on a real example — inspecting the actual numbers at every stage. The idea is that these mechanisms become much more intuitive when you can poke at them directly, rather than just reading a formula on a slide.
The resource is freely accessible (link) and can be used as a standalone introduction or as a companion to a more traditional course.