This research introduces a novel approach to generate synthetic medical records while preserving patient privacy by strategically prompting large language models.

• Enables hospitals to share sensitive data for ML training without privacy violations
• Eliminates need for costly model fine-tuning or specialized training
• Leverages prompt engineering rather than complex differential privacy mechanisms
• Delivers synthetic text that maintains utility for downstream medical applications

This breakthrough allows healthcare organizations to accelerate medical AI development while maintaining strict privacy compliance and reducing technical barriers to implementation.

Private Text Generation by Seeding Large Language Model Prompts