Hallucinations happen because language models generate statistically likely text, not verified facts. They're more likely when the model lacks relevant training data or when the question is outside its knowledge cutoff. Grounding responses with RAG, asking models to cite sources, and running evals against known-correct answers all help reduce hallucination rates.
ELI5
Think of Hallucination as the brain part that guesses or decides. When an AI model confidently states something that is factually incorrect.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.