LLM Hallucinations & Evaluation
AI · Advanced
Definition
Hallucinations are confident but false LLM outputs; evaluation measures accuracy, grounding and safety.
Reliable AI products must detect and reduce hallucinations.
Diagram
Flowchart
Example: Exact-match score
pred = ['Paris', 'Delhi', 'Rome']
gold = ['Paris', 'Mumbai', 'Rome']
print(round(sum(p == g for p, g in zip(pred, gold)) / 3, 3))Output: 0.667
Key points
- RAG and citations reduce hallucinations
- Lower temperature gives more consistent answers
Interview questions
Why do LLMs hallucinate?
They predict likely text, not verified facts.