Andrej Karpathy reproduces the 124M-parameter GPT-2 from scratch — a four-hour, no-shortcuts training run that shows every choice that goes into a modern LLM.