Training language-model agents to play long-horizon strategy games is hard because deterministic feedback is sparse and often arrives only at the end of a ga…
机构:南加州大学
来源:arXiv 2610.41264 | AI4Papers 论文推荐平台