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