TY  - RPRT
TI  - Attention Degradation, Function Token Anchoring, and the Limits of Attention-Based Intervention in Large Language Models
AU  - Sagar Dangal
AU  - Manoj Shakya
PY  - 2026
UR  - https://arxiv.org/abs/2607.20524
ID  - 2607.20524
ER  -