We study and build mechanistic interpretability: circuits, SAEs, patching, steering. Papers → replications → demos. Open to all backgrounds; bring curiosity.