LLMs Can Annotate Attribution Graphs

2026/07/28 by Ameen Patel, Max Zhang, Nathan Hu
Computer Science · #cs.LG

paper · pdf

Poster at ICML Mechanistic Interpretability Workshop 2026

arxiv created 2026/07/28 · arxiv updated 2026/08/05

Abstract

Circuit tracing is an exciting technique for revealing the internal computation of language models, but it requires a time-intensive manual step of grouping individual features or MLP neurons into supernodes. We present a simple pipeline for automating this step: directly presenting feature descriptions to a language model that groups them into supernodes. Using automated interpretability metrics, we confirm that supernodes generated by our pipeline are as interpretable as those generated by human annotators. On a two-hop Capitals task, our pipeline recovers a supernode corresponding to the intermediate hop in 97 of 100 prompts. Finally, we present a simple proof of concept using our pipeline for open-ended exploration, where we automatically annotate 1000 attribution graphs from Wikipedia prompt completions and then use an LLM judge to flag interesting graphs worth human review. We hope this work demonstrates that even simple automation can produce meaningful attribution graph annotations, motivating further work on automated circuit tracing.

Citations

• What's the plan? Metrics for implicit planning in LLMs and their application to rhyme generation and question answering
• Automated Attribution Graph Interpretation via Probe Prompting
• Detecting and Characterizing Planning in Language Models
• Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
• Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models