Automated research systems increasingly run LLM agents over long horizons, but more inference does not by itself produce more progress: agents replay growing…
机构:Stanford
来源:arXiv 2610.07625 | AI4Papers 论文推荐平台