Large language models (LLMs) are impressive, but they can sometimes generate incorrect or nonsensical information, a problem known as 'hallucination.' Ensuring LLMs produce reliable, accurate content is a critical challenge, especially in fields like medicine where errors can have serious consequences. Researchers are exploring new ways to quantify and control these uncertainties, and a recent paper introduces a promising technique called 'Conformal Uncertainty' (ConU). ConU works by cleverly measuring the uncertainty within an LLM's output space, rather than relying on potentially misleading confidence scores. It leverages the idea that if an LLM generates many diverse outputs for the same question, it's likely less certain about the correct answer. ConU clusters similar answers together and calculates the uncertainty based on how diverse these clusters are. This 'uncertainty score' is then used to create a set of possible answers that are most likely to contain the correct one. This approach offers statistical guarantees about the accuracy of the generated answer sets, even without needing to understand the complex inner workings of the LLM. Tests show that ConU consistently outperforms other uncertainty methods and accurately predicts the correctness of LLM responses. More importantly, these sets of probable answers tend to be small, making it easier to identify the most likely correct answer. ConU also has the potential to improve the overall accuracy of LLMs by allowing them to abstain from answering questions with high uncertainty, or to present a range of possible responses when complete certainty isn't possible. While promising, ConU has limitations. One challenge is knowing if a correct answer was even sampled among the generated outputs, particularly in real-world applications. Future work will explore extending ConU to other NLG tasks like summarization and refining its application for non-standard situations. Overall, ConU represents a significant step toward making LLMs more trustworthy and reliable for critical applications.