Unveiling the Latent Directions of Reflection in Large Language Models

TL;DR

This paper explores the internal mechanisms of reflection in large language models by analyzing activation directions, demonstrating how reflective behavior can be systematically identified and controlled through activation steering techniques.

Contribution

It introduces a novel activation steering methodology to characterize and manipulate reflection levels in LLMs, advancing understanding of their internal reflective processes.

Findings

Reflection can be systematically induced or suppressed via activation interventions.

Steering vectors effectively differentiate reflection levels in model activations.

Suppressing reflection is easier than stimulating it in LLMs.

Abstract

Reflection, the ability of large language models (LLMs) to evaluate and revise their own reasoning, has been widely used to improve performance on complex reasoning tasks. Yet, most prior works emphasizes designing reflective prompting strategies or reinforcement learning objectives, leaving the inner mechanisms of reflection underexplored. In this paper, we investigate reflection through the lens of latent directions in model activations. We propose a methodology based on activation steering to characterize how instructions with different reflective intentions: no reflection, intrinsic reflection, and triggered reflection. By constructing steering vectors between these reflection levels, we demonstrate that (1) new reflection-inducing instructions can be systematically identified, (2) reflective behavior can be directly enhanced or suppressed through activation interventions, and (3) suppressing reflection is considerably easier than stimulating it. Experiments on GSM8k-adv and Cruxeval-o-adv with Qwen2.5-3B and Gemma3-4B-IT reveal clear stratification across reflection levels, and steering interventions confirm the controllability of reflection. Our findings highlight both opportunities (e.g., reflection-enhancing defenses) and risks (e.g., adversarial inhibition of reflection in jailbreak attacks). This work opens a path toward mechanistic understanding of reflective reasoning in LLMs.