Global Divergence, Local Convergence: Representation Geometry in SSMs and Transformers EMNLP 2026 (Main) Amit Ben-Artzy, Roy Schwartz ArXiv Code SpeLLM: Character-Level Multi-Head Decoding Preprint, 2025 Amit Ben-Artzy, Roy Schwartz ArXiv Code Attend First: Consolidate Later: On the Importance of Attention in Different LLM Layers BlackBox NLP 2024 Amit Ben-Artzy, Roy Schwartz ArXiv Tweet Code