How AI models memorize data

LLMs can memorize training data, potentially revealing personal information. Differential privacy and data sanitization help prevent this leakage.

• LLMs can memorize and leak training data.
• Differential privacy reduces this risk.