Hallucination

Hallucination in AI refers to when a model generates information that sounds plausible but is factually incorrect, fabricated, or nonsensical. The term borrows from psychology but describes a distinct technical phenomenon.

Why Hallucinations Happen

• LLMs predict statistically likely next tokens, not verified facts
• Training data contains errors and inconsistencies
• Models lack access to real-time information
• Pressure to provide an answer even when uncertain

Examples

• Citing nonexistent academic papers with plausible-sounding titles
• Inventing quotes from real people
• Creating fictional case law or legal precedents
• Generating false historical events or dates

Mitigation Strategies

• Retrieval-augmented generation (RAG)
• Chain-of-thought prompting
• Uncertainty quantification
• Human verification workflows
• Training models to say “I don’t know”