AISuffer
llm

Hallucination

When an AI model generates confident-sounding but factually incorrect or fabricated information.

What Is AI Hallucination

A hallucination is when a model “invents” facts, quotes, references, or data that don’t exist, but presents them as real.

Examples

  • Fabricated scientific papers with real author names
  • Non-existent API endpoints
  • Fake legal precedents
  • Incorrect mathematical calculations

How to Reduce Hallucinations

  1. RAG — give the model access to real data
  2. Temperature 0 — reduce “creativity”
  3. Ask for sources — the model hallucinates less when asked to cite
  4. Verification — always verify critical information
  5. Narrow the domain — the narrower the topic, the fewer hallucinations

Related Terms