Tree-of-thought prompting explores multiple solution paths by combining prompt chaining with self-evaluation. Each planning step produces alternatives that can be assessed before later steps continue. section 10.5 section 10.3.2
A tree-of-thought workflow typically:
- Generates several plans or answers at each level. section 10.3.2
- Evaluates each node before extending its path. section 10.3.2
- Short-circuits plans that fail evaluation. section 10.3.2
- Executes breadth-first in prompt flow; depth-first is described but not quickly implemented because of the DAG pattern. section 10.3.2
Time-travel planning problem
A prompt-flow run evaluates only selected branches. Several answer nodes remain empty after failed evaluation, while surviving branches produce differing totals, including 29 days and 9 days. section 10.3 section 10.3.2
Figure 10.10 Breadth-first vs. depth-first execution on a ToT pattern
Compare breadth-first and depth-first execution visually before examining the prompt-flow implementation. section 10.3.2
