Topography / AI

54 lessons and 389 connections between them, based on the cross-references in each lesson. Lines show references. For reading order, seethe AI track page.

• What a Neural Network Is19 lessons link here
• How It Learns (loss, gradient descent, backprop)14 lessons link here
• Tokens & Embeddings19 lessons link here
• Attention & the Transformer19 lessons link here
• From Transformer to LLM13 lessons link here
• Inference (sampling, context, KV cache, hallucination)19 lessons link here
• Fine-tuning & Alignment16 lessons link here
• RAG (Retrieval-Augmented Generation)10 lessons link here
• Agents & Tool Use11 lessons link here
• Capstone: Anatomy of an LLM App6 lessons link here
• Inference at Scale (why an LLM is fast, and cheap)14 lessons link here
• Reasoning & Test-Time Compute (why "thinking longer" works)9 lessons link here
• Evaluating Models (how do you *know* it's good?)9 lessons link here
• Diffusion & Image Generation (sculpting from noise)3 lessons link here
• Multimodality (teaching an LLM to see)3 lessons link here
• Mixture of Experts (a big brain with a cheap forward pass)2 lessons link here
• Inference Engineering (serving an LLM is a systems problem)9 lessons link here
• The Two Clocks (TTFT, TPOT, and why goodput is the number)16 lessons link here
• Prefix Caching (the same start, computed once)14 lessons link here
• Chunked Prefill (one long prompt must not freeze the room)6 lessons link here
• Prefill / Decode Split (two pools, one KV in flight)4 lessons link here
• Constrained Decoding (the grammar holds the pen)9 lessons link here
• Routing and Cascades (not every request needs the big model)8 lessons link here
• What the Model Sees (one string, every turn, from the top)11 lessons link here
• Prompt Engineering (shaping the next-token distribution)7 lessons link here
• Context Engineering (what goes in the window, and who decides)10 lessons link here
• Structured Output and Tool Schemas (the contract between model and code)10 lessons link here
• The Loop (an agent is a while loop around one model call)10 lessons link here
• The Harness (everything around the loop that the model never touches)10 lessons link here
• Memory (four places a fact can live, and only one of them is the model)6 lessons link here
• Graphs and State Machines (when the shape of the task is known, draw it)7 lessons link here
• Multi-Agent (a second loop is a second window, not a second mind)5 lessons link here
• Evals for Apps (the test suite for a program that is a string)12 lessons link here
• Guardrails, Cost and Latency (the three budgets the harness enforces)7 lessons link here
• One Knob, One Slope3 lessons link here
• Vectors, Dot Products and Matrix Multiply7 lessons link here
• The Maths Floor, Part 2: Probability, Softmax and Temperature5 lessons link here
• Precision and Recall (where to set the spam filter's bar)2 lessons link here
• Why the Loss Won't Go Down6 lessons link here
• Convolutional Networks0 lessons link here
• RNNs and LSTMs0 lessons link here
• A LoRA Run0 lessons link here
• The GPU (thousands of cores, and the memory that feeds them)2 lessons link here
• Four Ways to Split a Model1 lesson links here
• Looking Inside the Model0 lessons link here
• Red-Teaming (breaking your own model before a customer does)1 lesson links here
• Autograd by Hand6 lessons link here
• The Bigram Model2 lessons link here
• The MLP Language Model3 lessons link here
• Attention by Hand2 lessons link here
• The GPT Block2 lessons link here
• Sampling and What Scale Buys0 lessons link here
• RAG Over Your Own Notes0 lessons link here
• The Returns Desk, For Real (a real model, a real harness, and a dress rehearsal that fails)0 lessons link here