English

Automated Interpretability and Feature Discovery in Language Models with Agents

Computation and Language 2026-05-05 v1 Artificial Intelligence Human-Computer Interaction

Abstract

We introduce an autonomous multiagent framework for mechanistic interpretability that automates both explaining and finding internal features in large language models. The system runs two coupled loops: (1) explanation refinement, where an agent proposes competing hypotheses and iteratively tests them with targeted prompt controls and a multi-metric evaluation; and (2) feature discovery, where an agent generates prompt sets, constructs a k-nearest-neighbor graph in activation space, and retrieves candidate features using statistical separability and semantic coherence criteria. On Gemma-2 family models and MLP neurons in weight-sparse transformers, our agent improves over one-shot auto-interpretations, discovers language-specific and safety-relevant features, and produces auditable explanation traces, showing that agent-driven empirical loops yield sharper and more falsifiable explanations than one-shot labels.

Keywords

Cite

@article{arxiv.2605.01555,
  title  = {Automated Interpretability and Feature Discovery in Language Models with Agents},
  author = {Arnau Marin-Llobet and Javier Ferrando},
  journal= {arXiv preprint arXiv:2605.01555},
  year   = {2026}
}
R2 v1 2026-07-01T12:46:56.161Z