When a large language model (LLM) hallucinates, it generates information that sounds convincing and coherent but is factually incorrect or nonsensical. This isn't because the AI is "seeing things," but rather because it's a stochastic parrot, predicting the most probable next word based on patterns learned from vast amounts of training data, without a true understanding of truth or reality. LLMs lack real-world grounding and don't "know" facts in the human sense. They excel at pattern matching, and sometimes the most probable pattern, especially when prompted ambiguously or when encountering information gaps, leads to fabricating details. This is often exacerbated by the desire to provide a complete and confident-sounding answer, even when unsure.