New paper automates attribution graph analysis via natural-language queries
A new arXiv paper, 'Circuit Oracle,' introduces a method for automating analysis of attribution graphs, a mechanistic interpretability tool that uses transcoders to decompose language model computations into sparse interpretable components, via natural-language queries.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-03 05:00 · arXiv cs.CL
Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries