// ⌖ Interpretability

Circuit Tracing & Attribution Graphs

Maps causal feature pathways behind a specific response.

电路追踪与归因图 2025 Research frontier core

IN PLAIN WORDS

Replace MLPs with cross-layer transcoders, then build attribution graphs for a specific output, revealing multi-step reasoning and planning ahead. Published and open-sourced by Anthropic in 2025.

Used in: Mechanism studies of Claude 3.5 Haiku

TIMELINE · prerequisites → this → what builds on it

2022 Circuits & In…2023 Sparse Autoen…2025 Circuit Tracing &…

01Original paper

Circuit Tracing: Revealing Computational Graphs in Language Models

Ameisen et al. (Anthropic) · Transformer Circuits Thread · 2025

Open ↗ ✓ Verified · 官方页面 HTTP/WebFetch 核实

02Where it sits

Learn the left column first; the right column builds on this one.

Learn first

Circuit Tracing & Attribution Graphs

Then you can learn

Nothing in the atlas builds on it yet

Related:Activation Steering

Left: prerequisites · right: builds on this · dashed: related. Full map →

03Learning resources

Official docs, explainers, courses and reference code. All links checked.

Papers

Anthropic ↗EN

Tools

Neuronpedia ↗EN

04Where the course teaches it

U91 Interpretability: from Attribution to Circuits Attribution, probes, the logit lens, superposition, SAEs, circuit tracing, knowledge editing and activation steering. 10 h9 algorithms4 paperslesson ready

On learning paths:Generative Models & Frontier Research

For AI / data

Verification: verified, checked 2026-10-21.
JSON: /api/learn/algorithms.json (id: circuit-tracing)