● Status · updated 2026-10-02
A work in progress
rumblr is being built in the open. Every lesson already reads end to end; the illustrated ones add a hand-built figure for each chapter. This page counts them from the site's own files, so it is never out of date.
17 of 54 lessons illustrated · 17 fact-checked against primary sources
What remains
- Illustrate the next lessonsHardware, structured output, long context and efficient architectures, loss functions, and on through the course in reading order.
- Sign-in and members-only lessonsSome lessons will be free and the rest for members. Everything is open while the course is being built.
- More illustrated papersOne landmark paper is explained so far, with 136 more on the map.
- Easier diagrams on a phoneWide left-to-right diagrams are hard to read on a small screen.
Every lesson
Illustrated means a hand-built, interactive figure for every chapter. Text and diagrams means the full lesson, built from the AI Primer, with its own diagrams but without those figures yet.
- 00Math notation, from zeroIllustratedFact-checked
- 01The big pictureIllustratedFact-checked
- 02Neural networksIllustratedFact-checked
- 03OptimizersIllustratedFact-checked
- 04Training deep networksIllustratedFact-checked
- 05AttentionIllustratedFact-checked
- 06Positional informationIllustratedFact-checked
- 07The transformerIllustratedFact-checked
- 08TokenizationIllustratedFact-checked
- 09Training stagesIllustratedFact-checked
- 10Pretraining at scaleIllustratedFact-checked
- 11Fine-tuning in practiceIllustratedFact-checked
- 12Reinforcement learningIllustratedFact-checked
- 13Reasoning modelsIllustratedFact-checked
- 14Alignment and safetyIllustratedFact-checked
- 15The hardware underneathText and diagramsNot yet fact-checked
- 16InferenceIllustratedFact-checked
- 17Structured outputText and diagramsNot yet fact-checked
- 18Long context and efficient architecturesText and diagramsNot yet fact-checked
- 19Loss functionsText and diagramsNot yet fact-checked
- 20MetricsText and diagramsNot yet fact-checked
- 21Reading benchmarksText and diagramsNot yet fact-checked
- 22Overfitting and regularizationText and diagramsNot yet fact-checked
- 23Trees and boostingText and diagramsNot yet fact-checked
- 24CNNs and RNNsText and diagramsNot yet fact-checked
- 25Looking inside the modelText and diagramsNot yet fact-checked
- 26Word embeddingsText and diagramsNot yet fact-checked
- 27SimilarityText and diagramsNot yet fact-checked
- 28Training embedding modelsText and diagramsNot yet fact-checked
- 29Dimensions and compressionText and diagramsNot yet fact-checked
- 30Vector indexesText and diagramsNot yet fact-checked
- 31RetrievalText and diagramsNot yet fact-checked
- 32Clustering and matchingText and diagramsNot yet fact-checked
- 33Embeddings in productionText and diagramsNot yet fact-checked
- 34Autoencoders and VAEsText and diagramsNot yet fact-checked
- 35GANsText and diagramsNot yet fact-checked
- 36Diffusion and flow matchingText and diagramsNot yet fact-checked
- 37Multimodal modelsText and diagramsNot yet fact-checked
- 38Talking to a modelText and diagramsNot yet fact-checked
- 39OrchestrationText and diagramsNot yet fact-checked
- 40The agent loopText and diagramsNot yet fact-checked
- 41ToolsText and diagramsNot yet fact-checked
- 42Coding and computer-use agentsText and diagramsNot yet fact-checked
- 43Model Context ProtocolText and diagramsNot yet fact-checked
- 44Retrieval-augmented generationIllustratedFact-checked
- 45Context engineeringText and diagramsNot yet fact-checked
- 46MemoryText and diagramsNot yet fact-checked
- 47PlanningText and diagramsNot yet fact-checked
- 48EvaluationText and diagramsNot yet fact-checked
- 49GuardrailsText and diagramsNot yet fact-checked
- 50Cost and latencyText and diagramsNot yet fact-checked
- 51ObservabilityText and diagramsNot yet fact-checked
- 52Safe deploymentText and diagramsNot yet fact-checked
- 53Why the hard ones failText and diagramsNot yet fact-checked
Built from the open-source AI Primer at c8d5c21.