Language Models Can Control Their Own AttentionPaper Language Models Control Their Own Attention: Declarative AttentionTL;DR — Existing sparse attention methods still paid $O(N)$ per … September 7, 2026 14 min2609.02737v1