Tree of Thoughts (ToT) is a 2023 paper by Princeton University and Google DeepMind titledTree of Thoughts: Deliberate Problem Solving with Large Language Models" was proposed in the "

This paper proposes a new language model reasoning framework, the Thought Tree (ToT), which generalizes the popular thought chain method to prompt the language model and enables the exploration of coherent text units (thoughts) as intermediate steps in problem solving. ToT allows the LM to make thoughtful decisions by considering multiple different reasoning paths and self-evaluation options to decide the next move, and looking forward or backtracking to make global choices when necessary. The experiments of the study showed that ToT significantly enhanced the problem-solving ability of the language model on three new tasks that require non-trivial planning or search: 24-point game, creative writing, and mini crossword puzzles. For example, in the "24-point game", GPT-4 with thought chain prompts only solved 4% tasks, while the method of this study achieved a success rate of 74%.