ML//tree of thought

Tree of thought is a prompting technique that generalizes chain of thought by exploring multiple reasoning paths with lookahead and backtracking, and it trades more compute for better accuracy on planning, search and multi-step reasoning. Instead of one linear chain, it generates a tree of possibilities and evaluates the branches before committing.

Tree of thought is a prompting technique that generalizes chain of thought by exploring multiple reasoning paths with lookahead and backtracking, and it trades more compute for better accuracy on planning, search and multi-step reasoning. Instead of one linear chain, it generates a tree of possibilities and evaluates the branches before committing.

As a prompting technique it shares its principle (branching exploration) with reasoning models, which implement it with RL training and dedicated compute: tree of thought is the idea at prompt level, and tree search in a reasoning model is its trained, RL-optimized form.