Tree of Thoughts (ToT)
Exploring Branches of Logic
For complex tasks that require strategic lookahead, traditional Chain-of-Thought falls short because it only pursues a single, linear path. Tree of Thoughts (ToT) solves this by maintaining a 'tree' of intermediate reasoning steps.
In a programmatic ToT framework, the language model generates multiple thoughts, self-evaluates them, and uses search algorithms (like Depth-First or Breadth-First Search) to explore promising paths and backtrack from dead ends.
However, you can simulate this powerful framework in a standard chat interface using 'Tree-of-Thought Prompting' (Hulbert, 2023).
By instructing the LLM to act as a 'Panel of Experts' that think one step at a time, share their thoughts, and drop out if they make a mistake, you force the model to self-evaluate and prune bad logical branches internally!
Reference Examples
Imagine three different experts are answering this question.
All experts will write down 1 step of their thinking, then share it with the group.
Then all experts will go on to the next step, etc.
If any expert realizes they're wrong at any point then they leave.