In the context of a Large Language Model (LLM), a hallucination is when the model generates information that is factually incorrect, fabricated, or not grounded in the input or training data—even though it may sound plausible.

Example:

Prompt: “In what year did Einstein discover oxygen?” Hallucinated Answer: “Albert Einstein discovered oxygen in 1920.”

Causes:

• The model tries to “fill in the blanks” based on patterns, not actual facts.
• Lack of access to real-time or verified knowledge.
• Ambiguous or incomplete prompts.

Types:

• Factual errors (wrong dates, names, etc.)
• Fabricated sources or citations
• Inaccurate code, logic, or math

Hallucinations are a key limitation in using LLMs for tasks requiring high accuracy or trust.