Rethinking Circuit Evaluation: Do Circuits Explain Model Errors?
Mechanistic interpretability (MI) aims to explain a model's behaviour through analyzing its internal computations; circuit-based explanations aim to isolate these computations with compact subnetworks validated by ablating the rest of the model. We show that circuits validated this way may fail to recover the underlyin...
Li Zhang, Chu-Qin Geng, Mark Zhang et al.
· 0 citations