Reproduce “Uncertainty-Aware Budget Allocation for Efficient LLM Tree Search” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jianguo Huang, Zhiyi Lyu, Fuxiang Zhang, Yanchen Deng, Deheng Ye, Bo An

UAA adapts tree-search branching to estimated semantic diversity among sibling actions, while the existing value function ranks candidates. In a Qwen3-32B MATH-500 test with confidence-guided MCTS, accuracy rose from 88.00% to 89.00%, while average token use fell from 134.47k to 14.12k and wall-clock time from 3029.64 to 338.19 seconds. These empirical results do not guarantee accuracy is maintained in other settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.