Inference Trajectory Topology: A New Method for Uncertainty Quantification of Large Language Models Without Calibration

Introduction / Main Floor: Inference Trajectory Topology: A New Method for Uncertainty Quantification of Large Language Models Without Calibration

This article introduces an innovative method called "Inference Trajectory Topology", which achieves uncertainty quantification without additional calibration by analyzing the topological structure of the chain of thought during the reasoning process of large language models, providing new ideas for improving model reliability.