Learn LLMs: From Zero to Transformer
What this course covers
From Zero to Transformer is a free, self-paced course that explains how large language models work, starting from no prior machine learning knowledge and ending at the mathematics of the original Transformer paper. It runs about six hours across ten modules, each with interactive labs and a checkpoint quiz. No installation and no account are required.
Who it is for
The course is written for technology leaders, analysts, engineers, and policy staff who need an accurate mental model of large language models rather than a surface summary. The early modules assume no background. The later modules assume comfort with vectors and matrix multiplication, though every step is explained.
Course syllabus
- Orientation (Start here, 10 minutes). How the labs, callouts, and checkpoints work.
- The Big Idea: Prediction Machines (Beginner, 35 minutes). Next-token prediction, what makes a model large, the stochastic parrot argument, and why raw prediction alone is not enough.
- A Brief History and the Landscape (Beginner, 30 minutes). From ELIZA to Transformers, the main model families, and how AI, machine learning, and large language models relate.
- Words to Numbers: Tokens and Embeddings (Beginner, 40 minutes). Tokenization, embeddings, vector arithmetic, and similarity measured by the dot product.
- How LLMs Learn: Training, Fine-tuning and RLHF (Intermediate, 45 minutes). Pre-training, the scale of compute involved, fine-tuning a base model into an assistant, and reinforcement learning from human feedback.
- Inside the Transformer I: Architecture and Data Flow (Advanced, 45 minutes). What the letters in GPT stand for, and how data moves through the stack.
- Inside the Transformer II: Attention (Advanced, 50 minutes). Self-attention and cross-attention, and the masking that stops a model reading ahead.
- The Original Paper and the Math (Expert, 40 minutes). A guided reading of Attention Is All You Need with the equations worked through.
- Capabilities, Limits and the Frontier (Expert, 35 minutes). Context windows, mixture-of-experts, multimodality, and knowledge distillation.
- Capstone: Exam, Glossary and Resources (Finish, 20 minutes). A final exam, a glossary of terms, and where to go next.
For shorter explanations of individual ideas, see the twelve AI concepts every tech leader should know. For the applied side, see Learn Agentic AI and the AI project portfolio.