ModelRefs / Mechanistic Interpretability — AI Glossary
Mechanistic Interpretability — AI Glossary
A research program reverse-engineering the specific circuits and algorithms implemented by neural network weights.
Overview
Mechanistic interpretability (Elhage et al. 2021, Anthropic) identifies 'circuits'—small subgraphs of attention heads and MLP neurons implementing specific algorithms (indirect object identification, greater-than, key-value recall). Techniques: activation patching, logit lens, sparse autoencoders for feature decomposition. Goal: fully understand transformer computations.
Reference details
| Topic | safety |
|---|---|
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Mechanistic Interpretability — AI Glossary.
Frequently asked questions
What is Mechanistic Interpretability?
A research program reverse-engineering the specific circuits and algorithms implemented by neural network weights.
What concepts are related to Mechanistic Interpretability?
Closely related concepts include interpretability, alignment, xai.