Extending the Context of Pretrained LLMs by Dropping Their Positional Embeddings Paper • 2512.12167 • Published Dec 13, 2025 • 4