[LG] Because we have LLMs, we Can and Should Pursue Agentic Interpretability
[Google DeepMind]
https://arxiv.org/abs/2506.12152