Large language models (LLMs) like ChatGPT have shown impressive abilities, but can we trust them? A new research paper dives into a critical question: how can we tell if an LLM is truly confident in its answers? The study explores two ways LLMs express confidence – through the statistical probability of generated words (think of it like an internal confidence score) and through verbal statements like "I'm certain" or "I'm uncertain." Researchers found that an LLM's internal probability scores are often a better indicator of its true knowledge than its verbalized confidence. Interestingly, LLMs seem to be more honest about their uncertainty when tackling less common questions. This might be because they haven't encountered these questions as much in their training data and are therefore less likely to hallucinate or bluff. While LLMs have a general sense of what they know and don't know, the study revealed a disconnect between their internal confidence and their ability to express it accurately in words. This research is important because it highlights the challenge of interpreting LLM confidence and emphasizes the need for further research into making AI's internal reasoning processes more transparent. As LLMs become increasingly integrated into our lives, knowing when to trust them is paramount. Future research could explore ways to bridge the gap between an LLM's internal confidence and its verbal expression, potentially leading to more reliable and trustworthy AI systems.