Scaling Latent Reasoning via Looped Language Models - Explained Simply | ArXiv Explained