Efficient Reasoning via Chain of Unconscious Thought

TL;DR

LRMs suffer token-inefficient reasoning due to explicit stepwise outputs. The authors propose CoUT, a training-free, unconscious thought–inspired paradigm that internalizes reasoning inside hidden layers via Reasoning Process Internalization (RPI) and applies Token-Efficient Strategies (TES) to curb outputs. They formalize the objective as minimize $\text{len}(\mathcal{R})$ while maximize $s$, where $s=\text{eval}(\hat{\mathcal{Y}}, \mathcal{Y})$. Across four math benchmarks and multiple LLMs, CoUT achieves substantial token reductions (e.g., up to ~47.6% vs CoT) with accuracy comparable to or slightly below CoT, demonstrating improved efficiency without sacrificing performance.

Abstract

Large Reasoning Models (LRMs) achieve promising performance but compromise token efficiency due to verbose reasoning processes. Unconscious Thought Theory (UTT) posits that complex problems can be solved more efficiently through internalized cognitive processes. Inspired by UTT, we propose a new reasoning paradigm, termed Chain of Unconscious Thought (CoUT), to improve the token efficiency of LRMs by guiding them to mimic human unconscious thought and internalize reasoning processes. Concretely, we first prompt the model to internalize the reasoning by thinking in the hidden layer. Then, we design a bag of token-efficient strategies to further help models reduce unnecessary tokens yet preserve the performance. Our work reveals that models may possess beneficial unconscious thought, enabling improved efficiency without sacrificing performance. Extensive experiments demonstrate the effectiveness of CoUT. Remarkably, it surpasses CoT by reducing token usage by 47.62% while maintaining comparable accuracy, as shown in Figure 1. The code of CoUT is available at this link: https://github.com/Rohan-GRH/CoUT

Figures (2)

• Figure 1: Average Performance and Tokens of CoUT and Baselines for 4 LRMs over 4 Benchmarks.
• Figure 2: Ablation Studies on Our Proposed CoUT. CoT denotes the Chain-of-Thought baseline. RPI denotes Reasoning Process Internalization. TES denotes Token-Efficient Strategies.