Reproduce “Tree of Uncertain Thoughts Reasoning for Large Language Models” 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
Shentong Mo, Miao Xin
The Tree of Uncertain Thoughts (TouT) framework enhances Large Language Model reasoning by quantifying and integrating local uncertainties in intermediate steps, extending the Tree of Thoughts approach. This method improves problem-solving accuracy, achieving a 78% success rate on the Game of 24 and 67% on Mini Crosswords, representing substantial relative improvements over previous methods.
Key insights
- Quantifying uncertainty in LLM intermediate responses, for instance via Monte Carlo Dropout, provides critical information about the reliability of each reasoning step.
- Integrating uncertainty estimates into global search algorithms allows for a risk-aware strategy, prioritizing reasoning paths that are both promising and reliable.
- Dynamically adjusting the LLM's temperature parameter during reasoning balances initial high-quality reasoning with later exploration of diverse solutions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.