LLM Mastery — 50 articles, first principles

Each article has a mental model, runnable code, a named failure mode, and a checkable exercise. Read in order. Type the code. Capstone (50) is how you know it stuck.

How to use this course

1. Go in order — later articles assume earlier scars.
2. Type code yourself. Reading is not training.
3. Keep a notes file of tensor shapes and failure modes you hit.
4. Do article 50. If you skip it, you only rented the knowledge.

Curriculum map

Phase	Articles	Focus
Foundations	01–10	Tokens, loss, bigrams, MLP mindset
Neural guts	11–20	Autograd, optimization, attention dawn
Transformers	21–35	GPT internals, systems, scaling
Post-training	36–50	SFT, LoRA, RLHF/DPO, RAG, agents, capstone

All articles

#	Article	Level
01	What is a Language Model, Really?	Beginner
02	Tokens Are Not Words	Beginner
03	Next-Token Prediction Is the Game	Beginner
04	Your First LM: Bigrams	Beginner
05	Loss: Cross-Entropy Without the Fear	Beginner
06	Embeddings: Meaning as Geometry	Beginner
07	Neural Nets for Language: The MLP	Beginner
08	The Training Loop, End to End	Beginner
09	Overfitting, Underfitting, and Data	Beginner
10	The MakeMore Mindset: Build Tiny, Understand Deeply	Beginner
11	Backpropagation as Local Blame	Intermediate
12	Micrograd Energy: Autograd From Scratch	Intermediate
13	Tensor Shapes: The Hidden Curriculum	Intermediate
14	Softmax and Temperature, Carefully	Intermediate
15	SGD, Adam, and Why Adam Won LLMs	Intermediate
16	Initialization Is Not Optional	Intermediate
17	Dropout, Weight Decay, and Noise as Teachers	Intermediate
18	RNNs: The Old Kings and Their Pain	Intermediate
19	Seq2Seq and the Dawn of Attention	Intermediate
20	Reading ‘Attention Is All You Need’ Like an Engineer	Intermediate
21	Self-Attention Mechanics	Intermediate
22	Multi-Head Attention	Intermediate
23	Positional Information: Absolute, Relative, RoPE	Intermediate
24	Residuals and LayerNorm: The Stabilizers	Intermediate
25	The Transformer MLP: Where Facts Often Live	Intermediate
26	GPT Architecture: Decoder-Only Transformers	Intermediate
27	Implement a Tiny GPT (Conceptual Walkthrough)	Advanced
28	KV Cache: Why Chat Is Fast After the First Token	Advanced
29	Batching, Throughput, and the Economics of Tokens	Advanced
30	Scaling Laws: The Bitter Lesson, Quantified	Advanced
31	Tokenization Deep Dive: BPE Under the Hood	Advanced
32	Pretraining Data: The Real Model	Advanced
33	Training Parallelism: DDP, FSDP, Pipeline	Advanced
34	Mixed Precision Training	Advanced
35	Evaluation: Beyond Vibes	Advanced
36	Supervised Fine-Tuning (SFT)	Advanced
37	LoRA and Parameter-Efficient Fine-Tuning	Advanced
38	Preferences: RLHF and DPO	Advanced
39	Prompting as Programming	Intermediate
40	RAG: Retrieval-Augmented Generation	Advanced
41	Agents and Tool Use	Advanced
42	Hallucinations: Why They Happen	Advanced
43	Interpretability: Looking Inside	Advanced
44	Quantization and Local Serving	Advanced
45	Multimodal LLMs: Vision Enters the Context	Advanced
46	Diffusion vs Autoregressive: Two Generative Religions	Advanced
47	Safety, Security, and Prompt Injection	Advanced
48	Build Eval-Driven: A Practical Workflow	Advanced
49	Research Taste: How to Read Papers	Master
50	Capstone: Train Your Own Tiny LLM	Master

Start

→ 01. What is a Language Model, Really?