Tree of Thought
Letting a model branch into several lines of reasoning and explore them like a decision tree instead of one straight path.
Think of it like
Like planning a chess move by imagining several candidate moves and their follow-ups, not just the first one you see.
Example
For a logic puzzle, the model proposes a few possible first steps, evaluates each, and pursues the most promising branch.
How it actually works
Where plain chain-of-thought commits to a single reasoning path, tree of thought generates multiple candidate steps, scores them, and can backtrack from dead ends. It trades a lot more compute for better results on problems that need search or planning. It’s more a prompting-and-orchestration strategy than a change to the model’s weights.
For product teams
A way to squeeze harder problems out of a model, at a steep compute multiplier.
For engineers
A search procedure over branching reasoning steps with evaluation and backtracking.
Related
- Chain of thought — The single-path version it extends.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome