“Pedro Franceschi states that a major pitfall of LLMs is the user's inability to know how much training data the model has seen for a specific query, which affects the reliability of its output.”