LLM agents accumulate interaction histories that grow linearly with task length, causing quadratic inference cost scaling and performance degradation from at…
机构:Microsoft
来源:arXiv 2609.37590 | AI4Papers 论文推荐平台