Fine-tuning medical AI can improve diagnosis but also creates privacy risks

Qingyu Chen, PhD, and his squad acceptable retired to survey however artificial quality connection models are adapted for medicine and recovered that what these models memorize tin beryllium some utile and risky. A exemplary whitethorn clasp invaluable aesculapian knowledge, but successful a controlled survey utilizing existent infirmary records, the aforesaid fine-tuning—the added grooming that adapts a exemplary to a circumstantial task—that improved diagnostic show besides made it much apt to reproduce worldly it had seen during training, including delicate diligent information.