Andrej Karpathy's How I use LLMs, saved from YouTube.

The video sits at the top of the reader, and under it the transcript, in full: twenty four thousand words.

Scroll and the video comes with you.

It lifts out of the page and docks into a corner, snapping to the nearest one when you drag it.

The transcript is the talk written down, word for word.

If you want the short version, the page will make you a summary as well, and it sits in its own section.

In the rail, Video Strata: thirty-six moments, each carrying the stretch of time it covers.

The chips are the spoken vocabulary.

Intro, Story, Tip, Insight, Deep Dive.

Tokens and context window basics, a deep dive.

Start new chats by topic, a tip.

Final framework for evaluating LLM apps, an insight.

A talk is labelled the way a talk is built.

An article gets prose labels and a paper gets technical ones.

Each item keeps to one set, so the rail always means the same thing.

Click a section and both move together: the transcript scrolls to it, and the video seeks with it.

A talk you can read, and jump straight to the part you came for.