Textbook of AI

Abstract. Introduces Tree-of-Thoughts (ToT), a search-based prompting method that generalises chain-of-thought to a tree of partial solutions. At each node the LLM proposes several next reasoning steps, scores them with a self-evaluation prompt, and the search algorithm (BFS, DFS or beam) prunes and expands. ToT solves Game-of-24 puzzles, creative writing tasks and 5×5 crosswords substantially better than greedy CoT or self-consistency. The paper inspired the Monte-Carlo-Tree-Search-over-reasoning-steps line of work that culminated in the o1-style training pipeline.