多轮对话代理只能“一刀切”蒸馏？这篇论文给出何时蒸馏、蒸馏什么的智能选择策略
arXiv:2605.19447v1 Announce Type: new Abstract: Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignme…
用非洲低资源语言玩多轮对话，成功绕过ChatGPT、Gemini等主流大模型的安全护栏，安全漏洞新发现。
arXiv:2605.18239v1 Announce Type: new Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails…
基于增量语义知识图谱的状态化多轮对话评估方法，提升对话系统评测的连贯性与深度。
arXiv:2605.16650v1 Announce Type: new Abstract: Evaluating multi-turn dialogue systems remains challenging because response quality depends not only o…
新方法通过多轮对话挖掘隐藏恶意意图，解决多智能体系统审核盲区。
arXiv:2605.12856v2 Announce Type: replace Abstract: The emergence of multi-agent systems introduces novel moderation challenges that extend beyond con…
多轮对话智能体易陷入重复模式？这篇论文提出缓解“对话惯性”的新方法。
arXiv:2602.03664v3 Announce Type: replace-cross Abstract: Large language models excel as few-shot learners when provided with appropriate demonstratio…