GUT: Quantifying and Optimizing the Reasoning Uncertainty of LLMs via Graph Complexity
The key idea of GUT is to characterize the potential branches of each reasoning chain with a directed acyclic graph, thereby ensuring that all potential branches are comprehensively covered within the graph space.
Shuang Liang, Xin-Miao Hu, Xiang-Jun Ou et al.
· 0 citations