arXiv cs.AI / cs.LG / cs.CL·10d agoRethinking Circuit Evaluation: Do Circuits Explain Model Errors?#mechanistic-interpretability#circuits#gpt-2AI research