Hallucination
When an AI model generates plausible-sounding but factually incorrect or fabricated information. A major challenge in deploying AI for business.
In-Depth Explanation
Hallucination in AI refers to generated content that sounds confident and coherent but is factually wrong or entirely made up. Unlike human lies, hallucinations aren't intentional - they're an inherent behaviour of how language models generate text.
Why models hallucinate:
- They predict probable next tokens, not truth
- Training data may contain errors or contradictions
- They lack real-time knowledge or verification
- They try to be helpful even when uncertain
- Rare topics have less training signal
Common hallucination types:
- Factual errors: Wrong dates, statistics, or attributions
- False citations: Inventing papers, quotes, or sources
- Fabricated details: Making up product features or policies
- Confident uncertainty: Answering definitively when unsure
- Plausible fiction: Creating believable but false scenarios
Mitigation strategies:
- RAG to ground responses in verified sources
- Clear instructions to say "I don't know"
- Temperature reduction for factual tasks
- Citation requirements in prompts
- Human review for high-stakes content
Business Context
Hallucinations can damage customer trust and create legal liability. RAG and grounding techniques reduce hallucinations by 80-95%.
How Clever Ops Uses This
Hallucination mitigation is central to our AI implementations for Australian businesses. We use RAG, careful prompting, and verification systems to ensure AI outputs are trustworthy.
Example Use Case
"AI confidently citing a policy that doesn't exist or inventing product features that your business doesn't offer."