Large language models (LLMs) like ChatGPT are impressive, but they can sometimes confidently give wrong answers. This "hallucination" problem makes it hard to know when to trust them. New research tackles this issue by figuring out how uncertain an LLM is about its own answers. The research introduces a clever technique called "Directed Uncertainty Evaluation" or D-UE. Imagine asking an LLM a question and getting several different responses. Instead of just looking at how similar the answers are, D-UE analyzes the *logical relationships* between them. It checks if one answer implies another, creating a directed graph that captures the flow of reasoning. This approach is combined with a method to clarify vague or incomplete answers, making the uncertainty assessment even more accurate. The results are promising! D-UE significantly improves uncertainty estimates compared to existing methods. This is a big step towards making LLMs more trustworthy and reliable. It means future AI could be better at knowing – and telling us – when it’s not sure about something. This research opens up exciting possibilities for using LLMs in critical tasks, from healthcare to autonomous driving, where knowing the level of uncertainty is crucial. While more work is needed, this research provides a novel approach to quantifying uncertainty in LLMs, paving the way for a future where AI can better understand and communicate its limitations.