Deep Dive into LLMs like ChatGPT

The full version. Where the one-hour talk gives you the shape of a language model, this three-and-a-half-hour lecture opens the machine: how raw internet text becomes training data, what tokens are and why models see the world through them, what actually happens in pre-training, how a raw text-completer is turned into a helpful assistant, and why models hallucinate, struggle with arithmetic, and improve when allowed to "think" before answering. It ends with reinforcement learning and the reasoning models that followed.

Watch it after the intro talk, in sittings — it is a university course compressed into one video, built from the same understanding Karpathy brings to building these systems. The watching guide breaks it into segments with plain-language summaries, so you can treat it as a course syllabus rather than a marathon.