中文

Course orientation

The Road to the Next Token

Follow one real generation through tokens, probability, training, RNNs, attention, and Transformers.

8 parts

41 lessons; 29 published

Not slideware

Narrative, interaction, sources, and next questions each have a job

Mechanism to code

The path returns to real models, formulas, and source code

How you move through this course

The sidebar mirrors the complete P0–P7 tree exposed by the course manifest. The narrative, companion playground, dissection, and code workbench remain in the course's own runtime; the platform supplies discovery, return paths, and cross-course connections.

This is a highly interactive course that evolves and deploys independently. DeeptoAI uses a public course manifest and Platform Rail to recognize whether you are reading, dissecting, or inspecting code. Version one shares navigation context only; opening a page never masquerades as mastery.

The course-runtime contract

  • The course tree gives position; narrative supplies the teacher's explanation; interaction only tests causality that text cannot replace.
  • A lesson is not a card: it states a question, example, mechanism, boundary, exit task, and next question.
  • The platform standardizes URLs, return paths, indexing, state, and safety contracts—not whether a course uses MDX, split panes, canvases, or an independent app.
Full courseOpen in the original course runtimeTransformer dissectionOpen in the original course runtimeReal model codeOpen in the original course runtime
Enter through the first real questionC01 · 机器怎样读一句话,又怎样写出下一块?