Reasoning Strategies in Large Language Models: Can They Follow, Prefer, and Optimize?
This work addresses the limitation of LLMs favoring a single reasoning strategy, which is an incremental step for improving their effectiveness in diverse reasoning tasks.
The study investigated whether prompting can control reasoning strategies in large language models (LLMs) for logical problem-solving, finding that no single strategy consistently improved accuracy, but performance could be enhanced through adaptive strategy selection.
Human reasoning involves different strategies, each suited to specific problems. Prior work shows that large language model (LLMs) tend to favor a single reasoning strategy, potentially limiting their effectiveness in diverse reasoning challenges. In this work, we investigate whether prompting can control LLMs reasoning strategies and assess its impact on logical problem-solving. While our experiments show that no single strategy consistently improves accuracy, performance could be enhanced if models could adaptively choose the optimal strategy. We propose methods to guide LLMs in strategy selection, highlighting new ways to refine their reasoning abilities.