AI Models Can Now Selectively ‘Forget’ Data After Training

Microsoft researchers use reinforcement and retraining to wipe a large language model’s memory of Harry Potter

ai models data harry potter language language model large language large language model memory microsoft reinforcement researchers training