Provoking Multi-modal Few-Shot LVLM via Exploration-Exploitation In-Context Learning | Read Paper on Bytez