// the find
colah/colah.github.io
This is Christopher Olah's personal blog and presentation repo — the source for the influential neural network explainer posts (LSTMs, attention, feature visualization) that have been assigned reading in ML courses for years. It's not a library or tool; it's a static site with hand-crafted SVG diagrams and interactive D3 visualizations. Primarily useful as a reference or as a model for how to write clear technical content about ML.
The interactive D3 visualizations in posts like the neural network topology piece are genuinely well-done and explain things that static diagrams can't. The writing is unusually precise about intuitions — the LSTM post in particular is still the clearest walkthrough of gating mechanics that exists. The custom SVG diagrams are hand-tuned, not auto-generated, which shows. The DeepIntro slide decks bundle reveal.js locally, so they work offline without fetching CDN assets.
Last meaningful post was 2020, so anything touching transformers or modern architectures beyond attention basics is absent. The repo itself is a mess to navigate — reveal.js vendored in full twice across two slide directories, `.html~` backup files committed, `_DS_Store` files checked in. There's no build system for the blog posts themselves; they're raw HTML with inline styles, so contributing or forking is painful. The interactive visualizations use D3 v3, which is ancient and will break in subtle ways as browsers continue to tighten CSP rules.