Profile & progress · Interpretable

Interpretabletransformers · LLMs · mech interp

Part 0 · Foundations

0.1Linear Algebra as Geometry 0.2Probability, Information & Optimization

Part 1 · Transformer Fundamentals

1.1Tokens & Embeddings 1.2Attention, Fully Understood 1.3The Full Block & the Residual Stream 1.4Training Dynamics & Scaling

Part 2 · How LLMs Are Actually Made

2.1Pretraining & Base Models 2.2Supervised Fine-Tuning & PEFT 2.3RLHF, Reward Models & DPO 2.4RLVR & Reasoning Models 2.5Inference, Performance & Reliability

Part 3 · Mechanistic Interpretability: The Core

3.1The Interp Mindset & Observational Tools 3.2A Mathematical Framework & Induction Heads 3.3Superposition: Toy Models 3.4Sparse Autoencoders & Dictionary Learning 3.5Causal Methods: Patching & Circuit Discovery

Part 4 · Frontier Interpretability

4.1Circuit Tracing & the Biology of LLMs 4.2Functional Emotions 4.3The Global Workspace & Introspection

Part 5 · Safety, Steering & Editing

5.1Steering Behavior & Making Models Enjoyable 5.2Editing Weights & Learning on the Fly 5.3The AI Safety Landscape

Capstone projects Profile & progress