Media by module

One or two recordings per module, then the lab. Core items play on the module page. The full catalogue, with verification notes, is the media library.

Module 02: Math and classical ML as table stakes

• Watch: V01 Essence of linear algebra (playlist, 16 videos) (core, about 180 min).
• Watch: V03 But what is a neural network? (Deep learning chapter 1) (core, about 19 min).
• Watch: V04 Machine Learning Fundamentals: Bias and Variance (core, about 7 min).
• Watch: V05 Machine Learning Fundamentals: The Confusion Matrix (core, about 7 min).
• Watch: V06 ROC and AUC, Clearly Explained! (core, about 16 min).
• Watch: V07 The spelled-out intro to neural networks and backpropagation: building micrograd (core, about 146 min).
• Watch: V02 Essence of calculus (playlist, 12 videos) (optional, about 180 min).

Module 03: Tokens and the next-token objective

• Watch: V08 The spelled-out intro to language modeling: building makemore (core, about 118 min).
• Watch: V09 Building makemore Part 2: MLP (core, about 76 min).
• Watch: V11 Let's build the GPT Tokenizer (core, about 134 min).
• Watch: V12 CS336 Spring 2025, Lecture 1: Overview and Tokenization (core, about 79 min).
• Watch: V13 [1hr Talk] Intro to Large Language Models (core, about 60 min).

Module 04: Attention and the Transformer

• Watch: V15 Transformers, the tech behind LLMs (Deep Learning chapter 5) (core, about 27 min).
• Watch: V16 Attention in transformers, step-by-step (Deep Learning chapter 6) (core, about 26 min).
• Watch: V17 Let's build GPT: from scratch, in code, spelled out (core, about 116 min).
• Watch: V47 Building LLMs from the Ground Up: A 3-hour Coding Workshop (recommended, about 165 min).
• Watch: V53 How might LLMs store facts (Deep Learning chapter 7) (recommended, about 23 min).
• Watch: V10 Building makemore Part 3: Activations and Gradients, BatchNorm (optional, about 116 min).

Module 05: Training, scaling laws and inference

• Watch: V18 Let's reproduce GPT-2 (124M) (core, about 241 min).
• Watch: V19 CS336 Spring 2025, Lecture 9: Scaling laws 1 (core, about 65 min).
• Watch: V20 CS336 Spring 2025, Lecture 10: Inference (core, about 83 min).
• Watch: V41 CS336 Spring 2025, Lecture 5: GPUs (core, about 74 min).
• Watch: V46 Build A Reasoning Model From Scratch 2: Loading a Base Model, Text Generation, and KV Caching (recommended, about 97 min).

Module 06: Post-training: SFT, preference and RL

• Watch: V14 Deep Dive into LLMs like ChatGPT (core, about 211 min).
• Watch: V21 CS336 Spring 2025, Lecture 15: Alignment, SFT and RLHF (core, about 75 min).
• Watch: V22 CS336 Spring 2025, Lecture 16: Alignment, RL 1 (RLVR) (core, about 81 min).
• Watch: V23 Build A Reasoning Model From Scratch 6: Implementing GRPO for RLVR (core, about 87 min).
• Watch: V52 A Taxonomy for Next-gen Reasoning (Nathan Lambert) (recommended, about 19 min).

Module 07: Retrieval done properly

• Watch: V24 The unreasonable effectiveness of BM25 for agentic search (core, about 18 min).
• Watch: V25 How to look at your data (Jeff Huber and Jason Liu) (core, about 19 min).

Module 08: Evals as a discipline

• Watch: V06 ROC and AUC, Clearly Explained! (core, about 16 min).
• Watch: V25 How to look at your data (Jeff Huber and Jason Liu) (core, about 19 min).
• Watch: V26 Why AI evals are the hottest new skill for product builders (core, about 107 min).
• Watch: V27 LLM Evals: Common Mistakes (core, about 28 min).
• Watch: V49 CS336 Spring 2025, Lecture 12: Evaluation (recommended, about 81 min).
• Watch: V28 Evals in Action: From Frontier Research to Production Applications (optional, about 22 min).

Module 09: Agents and context engineering

• Watch: V29 How We Build Effective Agents (Barry Zhang, Anthropic) (core, about 15 min).
• Watch: V48 Building more effective AI agents (Erik Schluntz with Alex Albert) (recommended, about 19 min).
• Watch: V30 Prompting 101 (optional, about 25 min).

Module 10: Cost, latency and security

• Watch: V31 Prompt Injection, explained (core, about 12 min).
• Watch: V32 Mastering LLM Inference Optimization: From Theory to Cost-Effective Deployment (Mark Moyou) (optional, about 34 min).

Module 11: Fine-tuning for behaviour

• Watch: V33 Build an LLM from Scratch 7: Instruction Finetuning (core, about 106 min).
• Watch: V53 How might LLMs store facts (Deep Learning chapter 7) (recommended, about 23 min).

Module 12: The internal move and public work

• Watch: V34 From Software Engineer to AI Engineer, with Janvi Kalra (core, about 70 min).

Module 13: Interviews the 2026 way

• Watch: V34 From Software Engineer to AI Engineer, with Janvi Kalra (core, about 70 min).

Module 14: Track A, applied AI engineer

• Watch: V35 AI Engineering 101 with Chip Huyen (core, about 83 min).

Module 15: Track B, inference and GPU engineering

• Watch: V37 Lecture 1: How to profile CUDA kernels in PyTorch (core, about 56 min).
• Watch: V39 Lecture 3: Getting Started With CUDA for Python Programmers (core, about 78 min).
• Watch: V40 Lecture 4: Compute and Memory Basics (core, about 57 min).
• Watch: V41 CS336 Spring 2025, Lecture 5: GPUs (core, about 74 min).
• Watch: V42 Fast LLM Serving with vLLM and PagedAttention (core, about 32 min).
• Watch: V43 Lecture 22: Hacker's Guide to Speculative Decoding in vLLM (core, about 69 min).
• Watch: V50 CS336 Spring 2025, Lecture 6: Kernels, Triton (recommended, about 80 min).
• Watch: V32 Mastering LLM Inference Optimization: From Theory to Cost-Effective Deployment (Mark Moyou) (optional, about 34 min).
• Watch: V38 Lecture 2: Chapters 1 to 3 of PMPP (optional, about 52 min).
• Watch: V51 GPU MODE Lecture 35: SGLang (optional, about 45 min).

Module 16: Track C, research engineering

• Watch: V23 Build A Reasoning Model From Scratch 6: Implementing GRPO for RLVR (core, about 87 min).
• Watch: V44 Neel Nanda: Mechanistic Interpretability, A Whirlwind Tour (core, about 22 min).

Module 17: Expert markers and staying current

• Watch: V45 Andrej Karpathy: Software Is Changing (Again) (core, about 40 min).
• Watch: V54 2025 in LLMs so far, illustrated by Pelicans on Bicycles (Simon Willison) (recommended, about 18 min).
• Watch: V36 Keynote: Build and keep your context window (PyData Amsterdam 2023) (optional, about 38 min).