Transcript [EN]: Let's build GPT: from scratch, in code, spelled out.

The video is a step by step practical exploration of building a GPT style language model from scratch. It starts with the concept of language models and attention, then implements a tiny character-level transformer on Shakespeare data, and progressively scales to a full decoder-only transformer with self-attention, multi-head attention, feed-forward blocks, residual connections, layer normalization, dropout, and training loops. The presenter demonstrates training, generation, and evaluation, interleaving code explanations with architectural intuition. It culminates in a high-level discussion of how GPT-like models are pre-trained on large text corpora and later fine-tuned for alignment in applications like chat agents. The talk emphasizes practical coding, graphics of attention, and how the pieces fit into a GPT pipeline.