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

2026/03/17 by Naphade, Atharv, Atharv Naphade, Samarth Bhargav +5 · 1 voice
Computer Science · #Cognition #Explainable Artificial Intelligence (XAI) #Frontier #Introspection #Malleability #Mechanism (biology) #Multimodal Machine Learning Applications #Suite #Taxonomy (biology) #Topic Modeling #cs.AI

paper · pdf · doi:10.48550/arxiv.2603.20276

openalex publication_date 2026/03/17 · arxiv published 2026/03/17 · arxiv updated 2026/03/17 · openalex created_date 2026/03/25 · openalex updated_date 2026/07/28

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.

Citations

• Emergent Introspective Awareness in Large Language Models
• RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?
• Because we have LLMs, we Can and Should Pursue Agentic Interpretability
• DiffusionAttacker: Diffusion-Driven Prompt Manipulation for LLM Jailbreak
• Looking Inward: Language Models Can Learn About Themselves by Introspection
• Fine-tuning Language Models for Factuality
• The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
• Large Language Models Cannot Self-Correct Reasoning Yet
• The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"
• The Internal State of an LLM Knows When It's Lying
• Self-Refine: Iterative Refinement with Self-Feedback
• Reflexion: Language Agents with Verbal Reinforcement Learning
• Discovering Latent Knowledge in Language Models Without Supervision
• Language Models (Mostly) Know What They Know
• Teaching Models to Express Their Uncertainty in Words
• Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift
• Theory of mind
• Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry.

Discussions

• frontier LLMs can predict their own behavior better than they can predict other models' behavior the mechanism is attention diffusion, not explicit training https://arxiv.org/abs/2603.20276 [bsky, 0 points, 0 comments]

Related

• Looking Inward: Language Models Can Learn About Themselves by Introspection
• Emergent Introspective Awareness in Large Language Models
• Introspection Fine-Tuning (IFT): Training Small LLMs to Introspect
• Does It Make Sense to Speak of Introspection in Large Language Models?
• Can LLMs Introspect? A Reality Check