Real Sparks of Artificial Intelligence and the Importance of Inner Interpretability