AlphaMaze: Enhancing Large Language Models' Spatial Intelligence via GRPO Paper โข 2502.14669 โข Published Feb 20 โข 14