The syllabus · every curated resource, one sequenced path

Watch & read

Everything this primer recommends, gathered in one place and sequenced. Every link was verified live on 2026-08-07; every entry says what the resource teaches and why it earned its slot over the alternatives. Formats are tagged: video, interactive, written.

S.1If you only have three hours: the spine

Five picks that carry the whole arc. Watch/play them in this order, reading the matching chapters between sittings:

• Transformers, the tech behind LLMs · 3Blue1Brown · video, 27 min · after ch. 01. The full pipeline at a glance, with the embeddings-as-directions segment this primer's chapter 01 is built around.
• Attention in transformers, step-by-step · 3Blue1Brown · video, 26 min · after ch. 06. The finest visual explanation of attention in existence.
• Transformer Explainer · Polo Club, Georgia Tech · interactive, ~20 min of play · alongside ch. 03. A live GPT-2 in your browser: type, drag temperature, watch the distribution bend.
• Gradient descent, how neural networks learn · 3Blue1Brown · video, 21 min · after ch. 05. The canonical loss-landscape animation.
• LLM Visualization · Brendan Bycroft · interactive 3-D, ~30 min of exploration · after ch. 06. Fly through a working GPT; the architecture becomes a place you've been.

S.2Chapter 01 · Vectors and meaning

• The Illustrated Word2Vec · Jay Alammar · written. Embeddings from a "personality scores" analogy up through training; the canonical gentle on-ramp.
• Embedding Projector · TensorFlow team · interactive. Real embeddings in 3-D; click a word, see its cosine neighbors.
• Understanding the Dot Product · Kalid Azad · written, short. "Directional multiplication," with the solar-panel picture.
• Essence of Linear Algebra · 3Blue1Brown · video series, ~3 h total. The full rebuild; chapters 1, 2, and 9 (dot products) serve this chapter, chapters 3–4 serve the next.
• Embeddings: what they are and why they matter · Simon Willison · written. The engineer's tour: three lines of Python to search, RAG, and CLIP.
• Counterintuitive Properties of High Dimensional Space · Marc Khoury · written. Fifteen figures of 3-D intuition failing politely.
• How might LLMs store facts · 3Blue1Brown · video, 22 min. The superposition counting argument, animated (final third).
• Tiktokenizer · interactive. Two minutes that dissolve the "models read words" misconception.

S.3Chapter 02 · Matrices as transformations

• A Geometrical Understanding of Matrices · Gregory Gundersen · written. Columns-are-destinations, in essay form with figures.
• mm + Inside the Matrix · Basil Hosmer · interactive 3-D + written tour. Matmul as a navigable cube, with real GPT-2 attention-head presets.
• Immersive Linear Algebra · Ström, Åström & Akenine-Möller · interactive textbook. Every figure draggable; the linear-mappings chapter especially.
• Einsum is All you Need · Tim Rocktäschel · written. Index notation as a power tool, twelve worked examples ending at attention.
• Dive into Deep Learning · Zhang, Lipton, Li & Smola · free book + code. Chapter 2 pairs every operation here with live tensors.

S.4Chapter 03 · From scores to probabilities

• Transformer Explainer · Polo Club · interactive. The temperature/top-k/top-p sliders on a real model; this chapter made tactile.
• ArgMax and SoftMax · StatQuest · video, 14 min. The most concrete softmax arithmetic walk-through on video.
• Temperature, top-k, and top-p · Sebastian Raschka · written, short. The sampling dials with a worked five-token example.
• Seeing Theory · Brown University · interactive. The probability refresher, beautifully built.
• CS231n: Linear Classification · Stanford · course notes. Scores → softmax → loss, the classic bridge into chapter 04.

S.5Chapter 04 · Loss, surprise, and information

• The Key Equation Behind Probability · Artem Kirsanov · video, 26 min. Surprise → entropy → cross-entropy → KL, derived visually; ends exactly at the LLM loss.
• Visual Information Theory · Chris Olah · written. Entropy as code length; the definitive "what bits mean."
• Evaluation Metrics for Language Modeling · Chip Huyen · written. Cross-entropy, perplexity, and bits-per-character as models are actually scored.
• Entropy Clearly Explained · StatQuest · video, 17 min. The zero-prerequisite on-ramp.
• Perplexity, interactively · TensorTonic · interactive. Sharp-vs-flat sliders with live perplexity readouts.

S.6Chapter 05 · Gradients: how models learn

• Gradient descent (DL Ch. 2) and Backpropagation, intuitively (DL Ch. 3) · 3Blue1Brown · video, 21 + 13 min. The canonical animations of descent and blame-routing.
• CS231n: Backpropagation, Intuitions · Stanford · course notes. Circuits with local gradients; "add distributes, max routes, multiply swaps and scales."
• TensorFlow Playground · Smilkov & Carter · interactive. Train a tiny net live; feel the learning rate matter.
• Building micrograd · Andrej Karpathy · video, 2.5 h. A working autograd engine from an empty file; the most demystifying two hours available to an engineer.
• Calculus on Computational Graphs · Chris Olah · written. Why the backward direction is the cheap one.
• Yes you should understand backprop · Andrej Karpathy · written, 7 min. The production-scars argument for looking under the hood.
• Why Momentum Really Works · Gabriel Goh, Distill · interactive. Optimizer dynamics with sliders.

S.7Chapter 06 · The transformer block

• Attention, step-by-step · 3Blue1Brown · video, 26 min. The centerpiece recommendation of the whole syllabus.
• The Illustrated Transformer · Jay Alammar · written. The classic diagram walkthrough; the field's shared mental pictures.
• LLM Visualization · Brendan Bycroft · interactive 3-D. Every matrix of the block, drawn to scale, animated mid-inference.
• Rotary Embeddings: A Relative Revolution · EleutherAI · written. The definitive RoPE story, derivation and code.
• The math behind Attention · Luis Serrano · video, 36 min. The gentlest full attention derivation; ideal second angle.
• Let's build GPT · Andrej Karpathy · video, 2 h. The block in working PyTorch, √d experiment included.
• Spreadsheets Are All You Need · Ishan Anand · interactive. GPT-2 in Excel; audit every formula of chapter 06 cell by cell.
• An Interactive Guide to RoPE · Sunil Dhaka · interactive. Draggable rotations; the relative-distance property, felt.
• Rotary Positional Embeddings · Efficient NLP (Bai Li) · video, 11 min. The concise RoPE explainer if you'd rather watch than read.
• Patterns and Messages · Chris McCormick · written series, 2025. The residual-stream-as-bus reframing; the on-ramp to interpretability's vocabulary.
• The Annotated Transformer · Harvard NLP · written + code. The 2017 paper reproduced line by line in PyTorch.

S.8Chapter 07 · The moves papers make

• Singular Value Decomposition series · Steve Brunton · video series; first ~4 videos (~40 min) suffice. Rotation-stretch-rotation and optimal truncation, from the best SVD teacher on video.
• SVD Image Compression Demo · Tim Baumann · interactive. The rank slider that makes low-rank permanent.
• Parameter-Efficient Finetuning with LoRA · Sebastian Raschka · written. W + BA, the intrinsic-dimension bet, honest experiments.
• LoRA explained visually + from scratch · Umar Jamil · video, 27 min. Picture and PyTorch together.
• How DeepSeek Rewrote the Transformer · Welch Labs · video, 18 min. MHA → MQA → MLA: the low-rank move in a shipping frontier model.
• A Visual Guide to Quantization · Maarten Grootendorst · written, 60+ figures. Every number format, then GPTQ and GGUF.
• float.exposed · Bartosz Ciechanowski · interactive. Flip real fp16/bf16/fp32 bits.
• Notation · Goodfellow, Bengio & Courville · written, 4 pages. The field's de-facto symbol key; print it.

S.9The reference shelf and the graduation reads

• Mathematics for Machine Learning · Deisenroth, Faisal & Ong · free book (Cambridge UP). The rigorous backstop for every chapter here; use as a lookup, not a read-through.
• Dive into Deep Learning · Zhang et al. · free book. Math with runnable tensors, through attention and beyond.
• The Hundred-Page Language Models Book · Andriy Burkov · free book. The best single written companion covering this primer's full arc, foundations to LLMs.
• The Matrix Calculus You Need for Deep Learning · Parr & Howard · written. Jacobians and the vector chain rule, built from calc-1.
• A Visual Guide to Attention Variants in Modern LLMs · Sebastian Raschka, 2026 · written. What attention became: GQA, MLA, sliding windows, hybrids.
• A Primer on the Inner Workings of Transformer-based Language Models · Ferrando et al. · survey. The bridge into interpretability research.
• A Mathematical Framework for Transformer Circuits · Elhage et al., Anthropic · research. The deep end; read it last and recognize everything.

Contents · Glossary · All links verified 2026-08-07.