silentbob comments on Understanding LLMs: Insights from Mechanistic Interpretability