Adversarial In-Context Learning (adv-ICL) optimizes prompts for large language models through a two-player game, resulting in significant performance improvements across various tasks.
