On the Limits of Layer Pruning for Generative Reasoning in LLMs - Explained Simply | ArXiv Explained