๐—ง๐—ฟ๐—ฒ๐—ฒ ๐—ผ๐—ณ ๐—ง๐—ต๐—ผ๐˜‚๐—ด๐—ต๐˜๐˜€: ๐— ๐—ฎ๐—ธ๐—ฒ ๐—Ÿ๐—Ÿ๐— ๐˜€ ๐—˜๐˜…๐—ฝ๐—น๐—ผ๐—ฟ๐—ฒ

Chain-of-Thought makes a model think step by step. It follows one path. If the first step is wrong, the entire answer fails.

Tree of Thoughts fixes this. It turns reasoning into a search.

Instead of one line of reasoning, you treat each part of a solution as a node. You generate several next steps from that node.

Follow these steps:

The model evaluates itself. You do not need an external solver.

Repeating this cycle allows for real backtracking. If the best path hits a dead end, you move to the next best option. Chain-of-Thought cannot do this.

Tree of Thoughts works best for:

It requires more LLM calls than standard methods. Do not use it for simple questions. Use it when your problem needs a search process.

See how it works on the Game of 24 puzzle: https://dev48v.infy.uk/prompt/day6-tree-of-thoughts.html

Full post: https://dev.to/dev48v/tree-of-thoughts-how-to-make-an-llm-explore-instead-of-guess-3kn1

Optional learning community: https://t.me/GyaanSetuAi