Me, Myself, and $π$ : Evaluating and Explaining LLM Introspection

2026-03-17

AI Summary

Current evaluations struggle to distinguish genuine metacognitive capabilities of large language models (LLMs) from superficial behaviors stemming from reliance on world knowledge or textual self-simulation. This work formally defines LLM introspection as latent computational operations over the model’s own strategies and parameters, and introduces Introspect-Bench, a novel evaluation framework. Through causal mechanism analysis, attention diffusion modeling, and multidimensional benchmarking, the study demonstrates that state-of-the-art models—without explicit training—leverage attention diffusion to gain privileged access to their own internal states, significantly outperforming baselines in predicting their own outputs. These findings provide empirical validation of authentic introspective capacity in contemporary LLMs.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Natural Language Processing: (Large) Language ModelsComputer Vision: Large Vision Models

Application Category

User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactionsSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

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.

Problem

• Introspection
• Large Language Models
• Meta-cognition
• Evaluation
• Self-awareness

Innovation

• introspection
• large language models
• meta-cognition
• attention diffusion
• Introspect-Bench