Back to full roadmap
topicadvanced★ Pro
Tree-of-Thoughts (Planning)
Build a plan tree, try alternative branches at each node, follow the best.
3 hours1 resources1 prereqs
Planning variant of ToT: root = goal, each node = intermediate state, edge = action. BFS/DFS + LLM-as-evaluator scores each node, follow the most promising branch.
Expensive: 10 branches × 5 depth = 50 LLM calls. Worth it only for hard tasks (math, puzzle, complex code).
LATS (Language Agent Tree Search): hybrid of ToT + Reflexion + Monte Carlo Tree Search — agent learns from environment interaction.