Choose a learning path.
Pick a question you care about. We’ll guide you through the lessons, then an experiment. New here? Start with The big picture.
The big picture
Build a useful mental model without starting with the mathematics. Follow the path from text to a response, then learn where that picture stops.
Follow a token
Travel through the numerical pipeline, from text pieces to a probability distribution. Use the worked pipeline alongside the lessons.
How learning happens
Connect data, prediction error, gradients, and updates. Make one small weight learn before considering a large training run.
From model to assistant
Understand how demonstrations and preferences can shape a pretrained model, without treating one historical recipe as universal.
The math path
Work through vectors, dot products, normalized exponentials, and gradients. The labs keep the calculations small enough to inspect.
Context is not memory
Untangle learned parameters, the supplied prompt, temporary caches, and retrieved evidence. These are different mechanisms with different limits.
Beyond the model
Follow the boundary between a language model and the software around it: messages, retrieval, tool calls, and evaluation.
Limits & tradeoffs
Look beyond parameter counts and benchmark headlines. Examine compute, latency, compression, and the difference between generalization and recall.