TY  - RPRT
TI  - The Policy Cliff: A Theoretical Analysis of Reward-Policy Maps in Large Language Models
AU  - Xingcheng Xu
PY  - 2025
UR  - https://arxiv.org/abs/2507.20150
ID  - 2507.20150
ER  -