From Latent Signals to Reflection Behavior: Tracing Meta-Cognitive Activation Trajectory in R1-Style LLMs

TL;DR

This paper investigates the internal layer-wise activation process of R1-style LLMs during self-reflection, revealing a structured progression from latent control to overt reflection, and demonstrates causal mechanisms through targeted interventions.

Contribution

It uncovers the layer-wise activation trajectory and causal chain underlying self-reflection in R1-style LLMs, providing insights into their meta-cognitive process.

Findings

Structured layer-wise activation progression identified

Causal chain linking prompt semantics to reflection tokens

Interventions demonstrate control over reflection behavior

Abstract

R1-style LLMs have attracted growing attention for their capacity for self-reflection, yet the internal mechanisms underlying such behavior remain unclear. To bridge this gap, we anchor on the onset of reflection behavior and trace its layer-wise activation trajectory. Using the logit lens to read out token-level semantics, we uncover a structured progression: (i) Latent-control layers, where an approximate linear direction encodes the semantics of thinking budget; (ii) Semantic-pivot layers, where discourse-level cues, including turning-point and summarization cues, surface and dominate the probability mass; and (iii) Behavior-overt layers, where the likelihood of reflection-behavior tokens begins to rise until they become highly likely to be sampled. Moreover, our targeted interventions uncover a causal chain across these stages: prompt-level semantics modulate the projection of activations along latent-control directions, thereby inducing competition between turning-point and summarization cues in semantic-pivot layers, which in turn regulates the sampling likelihood of reflection-behavior tokens in behavior-overt layers. Collectively, our findings suggest a human-like meta-cognitive process-progressing from latent monitoring, to discourse-level regulation, and to finally overt self-reflection. Our analysis code can be found at https://github.com/DYR1/S3-CoT.