AIExplainer

What is a tree-of-thought prompting?

A sophisticated prompting strategy that encourages an LLM to pursue and refine the most promising intermediate solutions and to abandon the rest.

A sophisticated prompting strategy that encourages an LLM to pursue and refine the most promising intermediate solutions and to abandon the rest. Tree-of-thought prompting uses an algorithm like the following: 1. Divide a complex problem into different branches (potential strategies), each comprised of multiple steps. 2. Prompt the LLM to work on each branch independently. 3. Ask the LLM to evaluate the quality of the solution to each branch after each step. 4. Continue refining the most promising branch(es); abandon the rest. 5. If a promising branch eventually fails, backtrack and try other promising steps.

Practitioners refer to tree-of-thought prompting when building, training, or evaluating machine learning systems. It appears in research papers, product documentation, and technical discussions about AI capabilities and limitations.