This research introduces a novel approach to reduce hallucination risks when deploying large language models in sensitive healthcare settings, particularly women's health.

• Proposes semantic entropy as a reliable measure to identify potential AI hallucinations
• Demonstrates improved safety through specialized validation in obstetrics & gynecology contexts
• Achieves significant reduction in false or misleading medical information
• Establishes a framework for more trustworthy clinical decision support

Why it matters: In high-stakes medical domains, AI hallucinations can lead to harmful patient outcomes. This approach represents a critical step toward safer AI implementation in healthcare, where reliability and accuracy directly impact maternal and neonatal health.

Reducing Large Language Model Safety Risks in Women's Health using Semantic Entropy