探究大模型自我反思能力，揭示其无需显式训练也能内省的机制。
Me, Myself, and $π$ : Evaluating and Explaining LLM Introspection
- 提出形式化框架，将内省定义为对模型策略与参数的隐式计算操作。
- 在自评估任务中，前沿模型能更准确预测自身行为表现。
- 发现注意力扩散是模型自发产生内省能力的关键机制，适合关注模型可解释性者。
人类智能的一个显著特征是内省——对自身认知过程进行评估和推理的能力。内省已成为大型语言模型（LLMs）中一个有前景但存在争议的能力。然而，现有评估方法往往难以区分真正的元认知与仅依赖通用世界知识或基于文本的自我模拟。本文提出一种原则性分类体系，将内省形式化为对模型策略和参数的潜在计算操作。为分离广义内省的组成部分，我们构建了Introspect-Bench，一个多层次评估套件，用于严格测试模型能力。结果表明，前沿模型对其自身策略具有特权访问权，在预测自身行为方面优于同类模型。此外，我们提供了因果性和机制性证据，解释了为何大模型能在无显式训练的情况下学习内省，以及注意力扩散如何促成内省机制的涌现。
原文摘要 · Abstract (English)
A hallmark of human intelligence is Introspection-the ability to assess and reason about one's own cognitive processes. Introspection has emerged as a promising but contested capability in large language models (LLMs). However, current evaluations often fail to distinguish genuine meta-cognition from the mere application of general world knowledge or text-based self-simulation. In this work, we propose a principled taxonomy that formalizes introspection as the latent computation of specific operators over a model's policy and parameters. To isolate the components of generalized introspection, we present Introspect-Bench, a multifaceted evaluation suite designed for rigorous capability testing. Our results show that frontier models exhibit privileged access to their own policies, outperforming peer models in predicting their own behavior. Furthermore, we provide causal, mechanistic evidence explaining both how LLMs learn to introspect without explicit training, and how the mechanism of introspection emerges via attention diffusion.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。