Seeing is Believing: Brain-Inspired Modular Training for Mechanistic Interpretability