Hallucination Detection.prompt

# Hallucination Detection

## Overview

Hallucination detection identifies when LLMs generate content that is factually incorrect, unsupported by evidence, or fabricated. Hallucinations are a critical safety concern for production LLM systems, especially in healthcare, legal, and financial applications where incorrect information has serious consequences. Detection methods include self-consistency checking, evidence verification, factual grounding, and confidence calibration.

## Key Concepts

- **Factual Hallucination**: Generating factually incorrect statements
- **Faithfulness Hallucination**: Generating content inconsistent with provided context
- **Self-consistency**: Checking if the model gives consistent answers across samples
- **Evidence Verification**: Checking claims against retrieved evidence
- **Citation Verification**: Ensuring cited sources actually exist and support claims
- **Confidence Calibration**: Aligning model confidence with actual accuracy

## Implementation Patterns

### Hallucination Detector