返回全部动态
评估大语言模型有限理性策略深度的Level-k可区分机制
原标题：Level-k Distinguishable Mechanisms for Evaluating Bounded Rationality in LLMs
AI 摘要
本研究提出level-k可区分机制，用于评估大语言模型在有限理性环境中的策略推理深度。作者构建了满足level-k可区分性的新博弈结构，并测试了四个LLM在递归推理和对手行为归纳下的策略深度，发现模型在递归推理下表现准确，但在归纳推理下性能下降。研究还指出，错误源于推理深度步骤数错误而非最佳响应计算错误。
以上摘要由 AI 生成，可能存在误差。事实请以原文为准。
正文节选
Level-k Distinguishable Mechanisms for Evaluating Bounded Rationality in LLMs Abstract Strategic depth of reasoning is essential for human interaction of Large Language Models (LLMs) operating in boundedly rational environments. However, existing evaluations are primarily based on canonical games prevalent in pretraining corpora, making it difficult to disentangle true strategic reasoning from memorisation. To address this, we formalise a necessary level- distinguishability condition for strateg
发布时间：2026-08-24 12:00
抓取时间：2026-08-24 12:34
来源机构：arXiv