RToT Prompt Enhancement: Unlocking the Key to New Potential of Fishery Large Models

With the development of artificial intelligence technology, Large Language Models (LLMs) have shown remarkable capabilities in solving complex reasoning tasks using methods like Chain-of-Thought (CoT) prompting. However, these methods often involve tedious reasoning processes that consume significant computational resources. In this paper, we introduce Reverse Tree-of-Thought (RToT) to improve the Fishery Large Model, a novel prompting strategy that reverses the traditional reasoning process. RToT starts from the desired outcome and works backward to identify the necessary steps and information required to reach that conclusion. This approach not only maintains the accuracy of reasoning but also significantly reduces the number of tokens used and the delay involved. Through extensive experiments on various reasoning benchmarks, we demonstrate that RToT outperforms standard CoT and other concise reasoning methods in terms of efficiency while preserving or even enhancing reasoning accuracy.