This research introduces a novel privacy-preserving framework for personalizing large language models while maintaining user data security.

• Employs evolutionary model merging to tailor LLMs to individual preferences without direct access to sensitive user data
• Optimizes for both task-specific performance and privacy protection simultaneously
• Achieves superior personalization compared to standard fine-tuning approaches while minimizing privacy risks
• Demonstrates practical implementation with measurable privacy-utility tradeoffs

For businesses, this breakthrough enables AI customization without compromising user trust or data security compliance, paving the way for responsible personalized AI systems.

Personalized Language Models via Privacy-Preserving Evolutionary Model Merging