Aware First, Think Less: Dynamic Boundary Self-Awareness Drives Extreme Reasoning Efficiency in Large Language Models

2025-08-15

AI Summary

To address excessive computational overhead and low efficiency in complex reasoning tasks caused by lengthy chain-of-thought (CoT) generation in large language models (LLMs), this paper proposes DR.SAF—a Dynamic Reasoning Self-Aware Framework. DR.SAF abandons conventional paradigms relying on handcrafted difficulty priors and instead introduces three core mechanisms: boundary self-aware alignment, reinforcement learning–driven adaptive reward management, and inference path constraints—enabling LLMs to dynamically perceive and regulate their own reasoning depth. Experiments demonstrate a 49.27% reduction in total response tokens, a 6.59× improvement in token efficiency, a 5× decrease in training time, and over 16% accuracy gain under extreme training conditions. The key contribution is the first realization of endogenous, adaptive perception of reasoning difficulty by LLMs, coupled with controllable optimization of reasoning boundaries.

Technology Category

• Natural Language Processing: Safety and Robustness
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Recent advancements in large language models (LLMs) have greatly improved their capabilities on complex reasoning tasks through Long Chain-of-Thought (CoT). However, this approach often results in substantial redundancy, impairing computational efficiency and causing significant delays in real-time applications. To improve the efficiency, current methods often rely on human-defined difficulty priors, which do not align with the LLM's self-awared difficulty, leading to inefficiencies. In this paper, we introduce the Dynamic Reasoning-Boundary Self-Awareness Framework (DR. SAF), which enables models to dynamically assess and adjust their reasoning depth in response to problem complexity. DR. SAF integrates three key components: Boundary Self-Awareness Alignment, Adaptive Reward Management, and a Boundary Preservation Mechanism. These components allow models to optimize their reasoning processes, balancing efficiency and accuracy without compromising performance. Our experimental results demonstrate that DR. SAF achieves a 49.27% reduction in total response tokens with minimal loss in accuracy. The framework also delivers a 6.59x gain in token efficiency and a 5x reduction in training time, making it well-suited to resource-limited settings. During extreme training, DR. SAF can even surpass traditional instruction-based models in token efficiency with more than 16% accuracy improvement.

Problem

• Reducing redundancy in LLM reasoning processes
• Aligning model self-awareness with difficulty assessment
• Dynamically adjusting reasoning depth for efficiency

Innovation

• Dynamic reasoning depth adjustment for efficiency
• Self-awareness alignment with adaptive reward management
• Boundary preservation mechanism balancing accuracy efficiency