This research addresses the critical problem of hallucinations in large language models when applied to molecular analysis and drug design.

• Identifies the knowledge shortcut phenomenon as a key source of hallucinations in molecular comprehension
• Develops a benchmark specifically designed to detect and measure hallucinations in molecular applications
• Proposes detection and mitigation strategies to improve LLM reliability in scientific domains

Why it matters: Reducing hallucinations in LLMs is essential for safe and effective applications in drug discovery, medication analysis, and other medical applications where molecular accuracy is critical.

How to Detect and Defeat Molecular Mirage: A Metric-Driven Benchmark for Hallucination in LLM-based Molecular Comprehension