other8.0 / 10
Circuit Tracing: Revealing Computational Graphs in Language Models
Anthropic Interpretability Team (Ben Ainsworth, Amanda Askell, Tom Conerly, Nelson Elhage et al.)
Abstract
Methods companion to the 'Biology' paper. Introduces attribution graphs as a scalable technique for tracing computational flows through transformer circuits, enabling systematic reverse-engineering of model behavior on real tasks.
Research area
embedded agencymodel robustness
Published
2025
Source
other
Org
Anthropic
Sign in to read and join the discussion.