Tree Search
A search method that constructs reasoning paths through branching and backtracking, such as Tree of Thoughts (ToT), in which decisions to expand or backtrack are made dynamically based on intermediate node evaluations.
Timeline (last 90 days)
The study compared three methods—independent sampling, a nominally configured tree search, and multi-operator search—on two model families while controlling the inference budget.
The so-called tree search configuration in the study actually implements non-adaptive full-workflow sampling, and does not implement adaptive branching or pruning.
Independent sampling with extended prompts reproduces the same candidate discovery results as the nominal tree search without building a tree.
In the discovery stage, the gains of the nominal tree search may come from prompt design rather than the search algorithm itself.
A popular class of autonomous AI agents abstracts model building as a code search problem and solves it via tree search, where each node is a candidate program and the tree grows by generating improved child programs from parent nodes.
All Facts (5)
The study compared three methods—independent sampling, a nominally configured tree search, and multi-operator search—on two model families while controlling the inference budget.
50%UnverifiedThe so-called tree search configuration in the study actually implements non-adaptive full-workflow sampling, and does not implement adaptive branching or pruning.
50%UnverifiedIndependent sampling with extended prompts reproduces the same candidate discovery results as the nominal tree search without building a tree.
50%UnverifiedIn the discovery stage, the gains of the nominal tree search may come from prompt design rather than the search algorithm itself.
50%UnverifiedA popular class of autonomous AI agents abstracts model building as a code search problem and solves it via tree search, where each node is a candidate program and the tree grows by generating improved child programs from parent nodes.
50%