andrej karpathy

Karpathy shares his personal, practical workflow for using LLMs day-to-day — which models he uses for which tasks, how he prompts them, and the mental model he applies to get consistently better results from tools like ChatGPT and Claude.

Karpathy builds a GPT model from scratch in Python/PyTorch, live on camera — from the basic bigram model all the way to a working transformer with self-attention. The clearest hands-on explanation of how GPT architectures actually work.

A crisp 1-hour introduction to LLMs for a general technical audience. Karpathy covers what language models are, how they work at a high level, the emerging 'System 2' reasoning capabilities, and the security challenges they introduce — without requiring a machine learning background.

A 3.5-hour end-to-end walkthrough of how large language models like ChatGPT are actually built — from tokenization and pretraining through supervised fine-tuning and RLHF. Karpathy explains everything in plain engineer-to-engineer language with no hype.