LLMs are powerful but incur substantial inference cost, which can be exploited by malicious users to induce overly long outputs, increasing latency and opera…
机构：清华
来源：arXiv 2611.11261 | AI4Papers 论文推荐平台