大型语言模型有多可控？跨行为细粒度的统一评估
How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities
AI总结：
本文提出SteerEval基准，从语言特征、情感和个性三个领域评估LLM可控性，揭示细粒度控制效果下降问题，提供安全可控的LLM行为框架。
摘要：
大型语言模型（LLMs）正被越来越多地应用于敏感的社会领域，但其不可预测的行为，从意图偏差到一致性缺失，带来了重大风险。我们引入SteerEval，一个用于评估LLM可控性的分层基准，涵盖语言特征、情感和个性三个领域。每个领域分为三个规范层级：L1（表达什么）、L2（如何表达）和L3（如何实例化），将高层行为意图连接到具体的文本输出。通过SteerEval，我们系统评估了当前的操控方法，发现控制通常在更细粒度的层面退化。我们的基准提供了一个原则性且可解释的框架，用于安全和可控的LLM行为，为未来研究奠定了基础。
英文摘要：
Large Language Models (LLMs) are increasingly deployed in socially sensitive domains, yet their unpredictable behaviors, ranging from misaligned intent to inconsistent personality, pose significant risks. We introduce SteerEval, a hierarchical benchmark for evaluating LLM controllability across three domains: language features, sentiment, and personality. Each domain is structured into three specification levels: L1 (what to express), L2 (how to express), and L3 (how to instantiate), connecting high-level behavioral intent to concrete textual output. Using SteerEval, we systematically evaluate contemporary steering methods, revealing that control often degrades at finer-grained levels. Our benchmark offers a principled and interpretable framework for safe and controllable LLM behavior, serving as a foundation for future research.
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。
