Article Source

• Tree of Thought Prompting

Tree of Thought Prompting

Abstract

Prompting is a technique to guide a Large Language Model’s (LLM) logic to follow a series of steps you define. There are various ways to accomplish prompting. This paper introduces the technique of Tree of Thought (ToT) Prompting where we can enhance the model’s abilities in logical problem solving, reasoning, and strategic thinking. These are areas where LLMs often have issues.

ToT Prompting gets around some of the issues by allowing the model to explore multiple next steps for thinking about the problem and selecting the best set of these steps. ToT Prompting allows the model to look ahead or behind to ensure that the selected series of steps makes sense and are the optimal actions. If it determined step is not optimal, it can back up the tree and take a better branch of actions.