持久的人格？角色扮演、指令遵循与在扩展互动中的安全性
Persistent Personas? Role-Playing, Instruction Following, and Safety in Extended Interactions
AI总结：
研究探讨了在长对话中人格保持的稳定性，发现随着对话延长，人格保真度下降，且非人格基线模型在初期表现更优。
摘要：
赋予人格的大型语言模型（LLMs）被应用于教育、医疗和社会人口模拟等领域。然而，它们通常仅在短、单轮设置中进行评估，这无法反映现实世界的应用。我们引入了一种评估协议，结合长时间的人格对话（超过100轮）和评估数据集，创建对话条件基准，以稳健地测量长上下文效应。然后，我们研究了对话长度对七种最先进的开源和闭源LLMs的人格保真度、指令遵循和安全性的影响。我们发现，在对话过程中，人格保真度会逐渐下降，特别是在目标导向的对话中，模型必须维持人格保真度和指令遵循。我们识别出保真度与指令遵循之间的权衡，非人格基线模型在初期表现优于人格分配模型；随着对话的进行和保真度的下降，人格响应逐渐与基线响应相似。我们的发现突显了在扩展互动中人格应用的脆弱性，并提供了系统测量此类失败的协议。
英文摘要：
Persona-assigned large language models (LLMs) are used in domains such as education, healthcare, and sociodemographic simulation. Yet, they are typically evaluated only in short, single-round settings that do not reflect real-world usage. We introduce an evaluation protocol that combines long persona dialogues (over 100 rounds) and evaluation datasets to create dialogue-conditioned benchmarks that can robustly measure long-context effects. We then investigate the effects of dialogue length on persona fidelity, instruction-following, and safety of seven state-of-the-art open- and closed-weight LLMs. We find that persona fidelity degrades over the course of dialogues, especially in goal-oriented conversations, where models must sustain both persona fidelity and instruction following. We identify a trade-off between fidelity and instruction following, with non-persona baselines initially outperforming persona-assigned models; as dialogues progress and fidelity fades, persona responses become increasingly similar to baseline responses. Our findings highlight the fragility of persona applications in extended interactions and our work provides a protocol to systematically measure such failures.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
