3 Weeks · 15 Days · Self-study

Mechanistic Interpretability Curriculum

The full study plan behind my Field Notes on AI Interpretability. Pick a day from the menu to jump to that session.

Overview

Week 1 · Foundations

Why It Matters, How Transformers Work, and the Superposition Problem

Week 2 · Modern Methods and Circuits

From Features to Mechanisms: Understanding How Models Think

Week 3 · Recent Updates, Advanced Topics, and Synthesis

Recent Updates, Safety Applications, Alternative Methods, Open Problems, and Final Synthesis

Reference

Legend

Critical reading
Open URL for visuals
Key concept
Hands-on exercise
Video resource
Reference resource

Mechanistic Interpretability: A 3-Week Self-Study Curriculum

Loading…

Loading curriculum…

Overview

Week 1 · Foundations

Why It Matters, How Transformers Work, and the Superposition Problem

Week 2 · Modern Methods and Circuits

From Features to Mechanisms: Understanding How Models Think

Week 3 · Recent Updates, Advanced Topics, and Synthesis

Recent Updates, Safety Applications, Alternative Methods, Open Problems, and Final Synthesis

Reference

Legend

Critical reading
Open URL for visuals
Key concept
Hands-on exercise
Video resource
Reference resource

Mechanistic Interpretability: A 3-Week Self-Study Curriculum

Goal: Develop strong conceptual and conversational fluency in mechanistic interpretability, with an emphasis on Anthropic's research agenda and the latest developments. Dive deep into the last several years of interpretability discoveries to build the knowledge required to meaningfully engage with discussions and prioritization of research.

Time Commitment: ~2-3 hours/day, 5 days/week

How to Use This Curriculum:

🔑 = Critical reading
📷 = Open the URL for diagrams/visualizations
💡 = Key concept to internalize
🖥️ = Hands-on exercise or interactive session
🎥 = Video resource
📚 = Reference resource

Loading…