SINdex is a novel uncertainty-based framework that automatically detects hallucinations in Large Language Models without requiring external data.

• Leverages semantic clustering to identify inconsistencies in model outputs
• Works with standard LLMs without special modifications
• Provides a scalable solution for factual verification
• Improves security by detecting potentially harmful misinformation

This research addresses critical security concerns in AI deployment across domains like healthcare and financial services, where factual accuracy is essential for maintaining trust and preventing harmful outcomes.

SINdex: Semantic INconsistency Index for Hallucination Detection in LLMs