Stage 2 · How LLMs Work
Inside LLMs: Interpretability
What models represent internally and the tools researchers use to look inside.
3 lessons1 ready
- 01 What Do LLMs Learn Inside? Representations and Interpretability Features, superposition, circuits and world models — what we know and what we don't. Advanced 25 min ✓
- 02 Circuits and Sparse Autoencoders Induction heads, dictionary learning and attribution graphs in practice. Expert 25 min ✎ In progress
- 03 Probing, Editing and Steering Models Linear probes, causal interventions, knowledge editing and activation steering. Expert 20 min ✎ In progress
Matching units in the new course
Each unit has goals, algorithms, must-read papers, a project and a self-test.