Tree of Thoughts generalizes CoT by exploring multiple candidate reasoning paths instead of committing to a single chain.
Why It Matters for Agents
For open-ended tasks (planning, strategy selection, decompositions), an agent often benefits from evaluating alternatives before taking an irreversible action.
Key Ideas
- Treat a “thought” as a state.
- Expand multiple candidate next thoughts.
- Use a scoring function (often model-based) to prune and backtrack.
Further Reading
- Tree of Thoughts: Deliberate Problem Solving with Large Language Models --- Yao et al. (2023)