Tree of Thoughts
An LLM reasoning framework that explores multiple reasoning branches and evaluates which paths are most promising.
Overview
Tree of Thoughts (ToT) extends chain-of-thought reasoning by allowing LLMs to explore multiple reasoning paths in a tree structure, evaluate intermediate states, and backtrack from unpromising branches. This mimics human deliberative thinking where we consider alternatives and choose the best approach.
Key Details
The framework uses the LLM both as a generator (proposing next reasoning steps) and an evaluator (assessing how promising each path is). Search strategies include breadth-first search and depth-first search through the reasoning tree. ToT significantly improves performance on tasks requiring exploration and planning, such as creative writing, mathematical problem-solving, and puzzle-solving, at the cost of more LLM calls per problem.
Related Concepts
chain of thought reasoning • monte carlo tree search • reasoning