Detecting the Disturbance: A Nuanced View of Introspective Abilities in LLMs

TL;DR

This paper investigates whether large language models can detect internal perturbations, revealing that they show partial, layer-dependent introspective abilities, especially in early layers, through systematic experiments and mechanistic explanations.

Contribution

It demonstrates that LLMs possess layer-dependent introspective capabilities, clarifies previous misconceptions, and provides mechanistic insights into how these abilities manifest.

Findings

Models localize injected signals with up to 88% accuracy.

Detection accuracy drops to chance in later layers.

Introspection is confined mainly to early layers.

Abstract

Can large language models introspect, that is, accurately detect perturbations to their own internal states? We systematically investigate this question using activation steering in Meta-Llama-3.1-8B-Instruct. First, we show that the binary detection paradigm used in prior work conflates introspection with a methodological artifact: apparent detection accuracy is entirely explained by global logit shifts that bias models toward affirmative responses regardless of question content. However, on tasks requiring differential sensitivity, we find robust evidence for partial introspection: models localize which of 10 sentences received an injection at up to 88\% accuracy (vs.\ 10\% chance) and discriminate relative injection strengths at 83\% accuracy (vs.\ 50\% chance). These capabilities are confined to early-layer injections and collapse to chance thereafter -- a pattern we explain mechanistically through attention-based signal routing and residual stream recovery dynamics. Our findings demonstrate that LLMs can compute meaningful functions over perturbations to their internal states, establishing introspection as a real but layer-dependent phenomenon that merits further investigation. Our code is open-sourced here: https://github.com/elyhahami18/llama-introspection-new