DynaThink: Fast or Slow? A Dynamic Decision-Making Framework for Large Language Models

TL;DR

DynaThink is a dynamic framework enabling large language models to choose between fast and slow reasoning pathways, optimizing efficiency and accuracy across complex tasks by assessing confidence levels and task complexity.

Contribution

It introduces a novel decision-making framework that allows LLMs to autonomously select between fast and slow inference methods based on task complexity and confidence.

Findings

Outperforms baseline methods on five reasoning benchmarks.

Effectively balances inference speed and accuracy.

Adapts reasoning pathways dynamically based on task difficulty.

Abstract

Large language models (LLMs) have demonstrated emergent capabilities across diverse reasoning tasks via popular Chains-of-Thought (COT) prompting. However, such a simple and fast COT approach often encounters limitations in dealing with complicated problems, while a thorough method, which considers multiple reasoning pathways and verifies each step carefully, results in slower inference. This paper addresses the challenge of enabling LLMs to autonomously select between fast and slow inference methods, thereby optimizing both efficiency and effectiveness. We introduce a dynamic decision-making framework that categorizes tasks into two distinct pathways: 'Fast', designated for tasks where the LLM quickly identifies a high-confidence solution, and 'Slow', allocated for tasks that the LLM perceives as complex and for which it has low confidence in immediate solutions as well as requiring more reasoning paths to verify. Experiments on five popular reasoning benchmarks demonstrated the superiority of the DynaThink over baselines.