Mechanistic Interpretability is Being Pursued for the Wrong Reasons — AI Alignment Forum