'Fundamental' vs 'applied' mechanistic interpretability research — AI Alignment Forum